Food accessibility, affordability, cooking skills and ... · United Kingdom ..... 41 Overall quality and extent of the evidence for socioeconomic differences in-store ... purchase

Elisabeth A. H. Winkler, BHlthSc(Nutr & Diet)(Hons)

Institute of Health and Biomedical Innovation School of Public Health

Queensland University of Technology

A thesis by publication submitted for the degree of Doctor of Philosophy, 2008

Food accessibility, affordability, cooking skills

and socioeconomic differences in fruit and vegetable

purchasing in Brisbane, Australia

II

Keywords

Socioeconomic inequalities

Socioecological model

Health risk behaviours

Diet

Fruits

Vegetables

Food accessibility

Food affordability

Cooking skills

Associations

III

Abstract

Across Australia and other developed nations, morbidity and mortality follows a

socioeconomic gradient whereby the lowest socioeconomic groups experience the

poorest health. The dietary practices of low socioeconomic groups, which are

comparatively less consistent with dietary recommendations, have been thought to

contribute to the excess morbidity and mortality observed among low socioeconomic

groups, although this phenomenon is not well understood. Using a socioecological

framework, this thesis examines whether the local food retail environment and

confidence to cook contribute to socioeconomic differences in fruit and vegetable

purchasing. To achieve this, four quantitative analyses of data from two main sources

were conducted. The food retail environment was examined via secondary analysis

of the Brisbane Food Study (BFS) and confidence to cook was examined in a cross-

sectional study designed and carried out by the author.

The first three manuscripts were based on findings from the BFS. Briefly, the BFS

was a multilevel cross-sectional study, designed to examine determinants of

inequalities, that was conducted in Brisbane in the year 2000. A stratified random

sample was taken of 50 small areas (census collection districts, CCDs) and 1003

residents who usually shopped for their households were interviewed face-to-face

using a schedule that included a measure of fruit and vegetable purchasing and three

socioeconomic markers: education, occupation and gross household income. The

purchasing measure was based on how often (never, rarely, sometimes nearly always

or always) participants bought common fruits and vegetables for their households in

fresh or frozen form, when in season. Food shops within a 2.5 km radius of the

CCDs in which survey respondents lived were identified and audited to determine

their location, type, their opening hours, and their price and availability of a list of

food items.

The first publication demonstrated there was minimal to no difference in the

availability of supermarkets, greengrocers and convenience stores between areas that

were most and least disadvantaged, in terms of the number of shops, distance to the

nearest shop, or opening hours. Similarly, the second publication showed the most

IV

disadvantaged and least disadvantaged areas had no large or significant difference in

the price and availability of fruits and vegetables within supermarkets, greengrocers

and convenience stores, but small differences were consistently apparent, such that

on average, low socioeconomic areas had lower prices but also lesser availability

than more advantaged areas. The third submitted manuscript presents results of

multilevel logistic regression analyses of the BFS data. While there were some

associations between environmental characteristics and fruit and vegetable

purchasing, environmental characteristics did not mediate socioeconomic differences

in purchasing the fruit and vegetable items since there was no substantial

socioeconomic patterning of the price or availability of fruits and vegetables.

The fourth submitted manuscript was based on the cross-sectional study of cooking

skills. A stratified random sample of six CCDs in Brisbane was taken and 990

household members ‘mostly responsible’ for preparing food were invited to

participate. A final response rate of 43% was achieved. Data were collected via a

self-completed questionnaire, which covered household demographics, vegetable

purchasing (using the same measure employed in the BFS for continuity), confidence

to prepare these same vegetables, and confidence to cook vegetables using ten

cooking techniques. Respondents were asked to indicate how confident they felt

(ranging from not at all- to very- confident) to prepare each vegetable, and to use

each technique. This fourth study found respondents with low education and low

household income had significantly lower confidence to cook than their higher

socioeconomic counterparts, and lower confidence to cook was in turn associated

with less household vegetable purchasing.

Collectively, the four manuscripts comprising this thesis provide an understanding of

the contribution of food accessibility, affordability and cooking skills to

socioeconomic differences in fruit and vegetable purchasing, within a

socioecological framework. The evidence provided by this thesis is consistent with a

contributory role of confidence to cook in socioeconomic differences in fruit and

vegetable purchasing, but is not definitive. Additional research is necessary before

promoting cooking skills to improve population nutrition or reduce nutritional

inequalities. An area potentially useful to examine would be how cooking skills

integrate with psychosocial correlates of food and nutrition, and socioeconomic

V

position. For example, whether improvement of cooking skills can generate interest

and knowledge, and improve dietary behaviours, and whether a lack of interest in

food and nutrition contributes to a lack of both fruit and vegetable consumption and

cooking skills. This thesis has demonstrated that an inequitably distributed food

retail environment probably does not contribute to socioeconomic variation in fruit

and vegetable purchasing, at least in contemporary Brisbane, Australia. Findings are

unlikely to apply to other time periods, rural and regional settings, and perhaps other

Australian cities as residential and retail development, and the supply and pricing of

produce vary substantially across these dimensions. Overall, the main implication for

public health is that interventions targeting the food supply in terms of ensuring

greater provision of shops, or altering the available food and prices in shops may not

necessarily carry a great benefit, at least in major cities similar to Brisbane. Future

studies of equitable food access may need to look beyond mapping the distribution of

shops and prices, perhaps to more personal and subjective facets of accessibility and

affordability that incorporate individuals’ perceptions and ability to access and pay

for foods.

VI

Table of contents Abstract……………………………………………………………………………...III List of figures……………...………………………………………………………...XI List of figures……………...………………………………………………………XIII Chapter 1: Introduction ................................................................................................ 1

1.1. Background - Socioeconomic position, diet and health ............................... 1 1.2. Description of the research problem ............................................................ 3

1.2.1. Theoretical framework ......................................................................... 3 1.2.2. Framing and operationalising the research problem ............................ 7

1.3. Objectives and aims of the study .................................................................. 7 1.4. Account of research progress linking the research papers ........................... 9

Chapter 2: Literature Review ..................................................................................... 12 2.1. Overview .................................................................................................... 12 2.2. Background ................................................................................................ 12

2.2.1. Fruits and vegetables and health ........................................................ 15 2.2.2. Fruits and vegetables and socioeconomic position ............................ 17 2.2.3. Fruits and vegetables, other dietary markers and socioeconomic context 19 2.2.4. Factors that may contribute to socioeconomic differences in diet ..... 20

2.3. Environmental factors and socioeconomic position ................................... 23 2.3.1. Access to shops .................................................................................. 30

United States .................................................................................................. 30 Canada ............................................................................................................ 33 United Kingdom ............................................................................................. 33 Australia ......................................................................................................... 35 Alternate measures of access .......................................................................... 35 Extent and quality of the evidence of a socioeconomic gap in shop accessibility .................................................................................................... 36

2.3.2. In- store availability of fruits and vegetables ..................................... 40 United States .................................................................................................. 40 United Kingdom ............................................................................................. 41 Overall quality and extent of the evidence for socioeconomic differences in-store availability of fruits and vegetables ....................................................... 41

2.3.3. In-store prices fruits and vegetables ................................................... 43 Case studies .................................................................................................... 43 Ecological, cross-sectional studies ................................................................. 43 Cross-sectional studies of subjectively-measured affordability ..................... 44 Extent and quality of the evidence connecting prices and socioeconomic position ........................................................................................................... 45

2.4. Environmental features and dietary behaviours ......................................... 49 2.4.1. Retail infrastructure ............................................................................ 53

Presence of local shops .................................................................................. 53 Distance to shops ............................................................................................ 54 Shop patronised .............................................................................................. 56 Car ownership ................................................................................................ 57

VII

Extent and quality of evidence connecting shop access to fruit and vegetable intake .............................................................................................................. 57

2.4.2. In- store availability ........................................................................... 60 Diet and objective measures of food availability ........................................... 60 Subjective measures of availability and fruit and vegetable intake ............... 61 Extent of the evidence connecting in-store availability to fruit and vegetable consumption ................................................................................................... 62

2.4.3. In-store prices and affordability ......................................................... 64 Price elasticities: evidence from economics .................................................. 65 Evidence from public health .......................................................................... 67 Overall extent and quality of the evidence connecting price and fruit and vegetable intakes ............................................................................................ 71

2.4.4. Summary: accessibility, affordability and socioeconomic differences in fruit and vegetable intake ............................................................................... 73

2.5. Cooking skills and fruit and vegetable consumption across socioeconomic groups 76

2.5.1. Cooking skills, nutrition and health ................................................... 77 2.5.2. Cooking skills and socioeconomic position ....................................... 81 2.5.3. Summary: cooking skills, socioeconomic position and fruit and vegetable intake .................................................................................................. 82

Chapter 3: Overview of research hypotheses and methods ....................................... 84 3.1. Research questions ..................................................................................... 84 3.2. Concepts and measurement issues ............................................................. 85

3.2.1. Socioeconomic position ..................................................................... 85 3.2.2. Accessibility and affordability ........................................................... 86 3.2.3. Cooking skills .................................................................................... 87 3.2.4. Fruit and vegetable purchasing .......................................................... 89

3.3. Hypotheses ................................................................................................. 92 3.4. Overview of research methods ................................................................... 93

3.4.1. Brisbane Food Study .......................................................................... 93 Brisbane Food Study measures ...................................................................... 95 Methodological critique ................................................................................. 97

3.4.2. Cooking Skills Study ......................................................................... 98 Sampling ........................................................................................................ 99 Data collection ............................................................................................. 105 3.4.2.1. Instrument development ........................................................... 106 Methodological critique ............................................................................... 113

3.4.3. Integration of the studies .................................................................. 114 4. Chapter 4: Does living in a disadvantaged area mean fewer opportunities to purchase fresh fruit and vegetables in the area? Findings from the Brisbane Food Study ........................................................................................................................ 115

4.1. Abstract .................................................................................................... 116 4.2. Introduction .............................................................................................. 116 4.3. Methods .................................................................................................... 121

4.3.1. Setting .............................................................................................. 121 4.3.2. Sampling of areas ............................................................................. 122

Areas of varying socioeconomic disadvantage ............................................ 122 Shopping catchments ................................................................................... 122

4.3.3. Data collection ................................................................................. 124 Shop data ...................................................................................................... 124

VIII

4.3.4. Measures ........................................................................................... 126 4.3.4.1. Independent variables ............................................................... 126 Dependent variables ..................................................................................... 126

4.3.5. Analysis ............................................................................................ 127 Number of shops .......................................................................................... 127 Distance to the nearest shop ......................................................................... 128 Opening hours .............................................................................................. 128

4.4. Results ...................................................................................................... 129 4.4.1. Number of shops .............................................................................. 129 4.4.2. Opening hours .................................................................................. 131 4.4.3. Distance to nearest shop ................................................................... 132

4.5. Discussion ................................................................................................ 134 4.5.1. Number of shops .............................................................................. 134 4.5.2. Opening hours .................................................................................. 135 4.5.3. Distance to nearest shop ................................................................... 136 4.5.4. Overseas contrasts – methodological explanations .......................... 137 4.5.5. Overseas contrasts – social/economic explanations ......................... 138

4.6. Acknowledgements .................................................................................. 139 Chapter 5: Does living in a disadvantaged area entail limited opportunities to purchase fresh fruit and vegetables in terms of price, availability, and variety? Findings from the Brisbane Food Study .................................................................. 140


5.3.1. Geographical coverage and sampling of areas ................................. 143 5.3.2. Shopping catchments and data collection ........................................ 144 5.3.3. Measurement .................................................................................... 145 5.3.4. Analysis ............................................................................................ 146

5.4. Results ...................................................................................................... 147 5.4.1. Availability ....................................................................................... 147 5.4.2. Variety .............................................................................................. 147 5.4.3. Price .................................................................................................. 150

5.5. Discussion ................................................................................................ 150 5.6. Acknowledgements .................................................................................. 153

Chapter 6: Local food retailing and the purchase of fruits and vegetables by socioeconomic groups .............................................................................................. 154

6.1. Summary .................................................................................................. 155 6.2. Introduction .............................................................................................. 156 6.3. Material and methods ............................................................................... 157

6.3.1. Sampling ........................................................................................... 157 6.3.2. Data collection .................................................................................. 158 6.3.3. Outcome variables ............................................................................ 159 6.3.4. Independent variables ....................................................................... 159

Household-level variables ............................................................................ 159 Area-level variables ...................................................................................... 160

6.3.5. Analysis ............................................................................................ 160 6.4. Results ...................................................................................................... 161 6.5. Discussion ................................................................................................ 168

Chapter 7: Confidence to cook vegetables and the buying habits of Australian households ................................................................................................................ 172

IX


7.3.1. Study sample and data collection ..................................................... 175 7.3.2. Measures .......................................................................................... 176

7.3.2.1. Socio-demographic characteristics........................................... 176 7.3.2.2. Outcome variables .................................................................... 177

7.3.3. Statistical methods ........................................................................... 178 7.4. Results ...................................................................................................... 178

7.4.1. Sample characteristics ...................................................................... 179 7.4.2. Confidence to cook among socio demographic subgroups .............. 182 7.4.3. Confidence to cook and household buying habits............................ 186 7.4.4. Household buying habits among sociodemographic subgroups ...... 186

7.5. Discussion ................................................................................................ 189 Chapter 8: Discussion .............................................................................................. 193

8.1. Overall findings ........................................................................................ 193 8.2. Findings in context with other research ................................................... 194

8.2.1. Socioeconomic differences in shop access, in-store availability and price 194 8.2.2. Relationships between shop access, in-store availability and price and fruit and vegetable purchasing, intake, or other dietary outcomes .................. 197 8.2.3. Confidence to cook .......................................................................... 200

8.3. Strengths and limitations .......................................................................... 202 8.3.1. Secondary analysis of Brisbane Food Study .................................... 202

Design and analysis ...................................................................................... 202 Measures ...................................................................................................... 205 Generalisabilty ............................................................................................. 207

8.3.2. Cooking Skills .................................................................................. 209 Study design ................................................................................................. 209 Measures ...................................................................................................... 210

8.4. Nutritional implications of findings ......................................................... 211 8.5. Future directions for health promotion and research ............................... 213 8.6. Concluding remarks ................................................................................. 217

I. Appendix I: Details of major studies included in the literature review ............... 2 II. Appendix II: Relevant materials used in the BFS data collection ...................... xl

II.1. Household survey – relevant sections ........................................................ xli II.2. Audit tools and instructions ....................................................................... liv

III. Appendix III: Relevant materials used in the cooking skills survey .............. lxi III.1. Ethical clearance .................................................................................... lxi III.2. Cooking Skills Study questionnaire ..................................................... lxiv III.3. Invitation letter to potential participants ............................................... lxx III.4. Invitation letter for repeatability sub-study .......................................... lxxi III.5. Thank-you cards .................................................................................. lxxii

IV. Appendix IV: Questionnaire development, validity and reliability ........... lxxiii IV.1. Methods .............................................................................................. lxxiii

IV.1.1. Face validity ................................................................................... lxxiii IV.1.1.1. Content validity ...................................................................... lxxiii IV.1.1.2. Repeatability: test-retest ......................................................... lxxiii IV.1.1.3. Internal consistency ................................................................ lxxiv

IV.2. Results ................................................................................................. lxxv

X

IV.2.1. Face validity .................................................................................... lxxv IV.2.2. Content validity ............................................................................. lxxvii IV.2.3. Internal consistency ........................................................................ lxxxi IV.2.4. Test-retest reliability ...................................................................... lxxxi

IV.2.4.1. Representativeness and Missing Data .................................... lxxxi IV.2.4.2. Categorical data .................................................................... lxxxiii IV.2.4.3. Continuous data .................................................................... lxxxiii

V. Appendix V: Other relevant data ................................................................. lxxxix V.1. Maps of Brisbane and other capital cities ............................................ lxxxix V.2. Spatial patterning of socioeconomic disadvantage (IRSD) across census collection districts in the Statistical Subdivisions of Australian Capital Cities in 2000 …………………………………………………………………………xcii V.3. Evidence of change in food prices from the Consumer Price Index (CPI) cv

VI. Appendix VI: Tests for spatial autocorrelation ............................................. cvi VII. Appendix VII: Checking assumptions of a priori sample size calculations. cix VIII. References ...................................................................................................... cx

XI

List of tables Chapter 2: Literature Review

Table 2.1: Associations between fruits and vegetables and low

socioeconomic position (SEP)……………………………………….............. 18

Table 2.2: Findings of studies assessing the relationship between

socioeconomic position and features of the food retail environment…........... 23

Table 2.3: Findings of studies assessing the relationship between features of

the food retail environment and dietary measures............................................ 51

Chapter 3: Overview of Research Hypotheses and Methods

Table 3.1: Characteristics of eligible and excluded CCDs from 2001

Census………………………………………………………………………... 104

Table 3.2 Response rates by area and housing type………………………..... 105

Table 3.3: Test-retest reliability of survey items pertaining to confidence… 111

Table 3.4: Test-retest reliability of survey items pertaining to purchasing

and demographics…………………………………………………………… 112

Table 3.5: Test-retest reliability, Intra-class correlation coefficients……….. 113

Chapter 4: Does living in a disadvantaged area mean fewer

opportunities to purchase fresh fruit and vegetables in the area?

Findings from the Brisbane Food Study

Table 4.1: Categories and definition of food outlets………………………... 125

Table 4.2: Selected Characteristics of Disadvantaged, Medium and

Advantaged areas…………………………………………………………….. 130

Table 4.3: Crude and adjusted relative rates of shops by socioeconomic

disadvantage…………………………………………………………………. 130

Table 4.4: Daily opening hours of greengrocers, convenience stores and

supermarkets…………………………………………………………………. 131

Table 4.5: Odds ratios for shops opening ‘short’ hours for areas of varying

socioeconomic disadvantage………………………………………………… 132

Chapter 5: Does living in a disadvantaged area entail limited

opportunities to purchase fresh fruit and vegetables in terms of price,

availability, and variety? Findings from the Brisbane Food Study

Table 5.1: Hypothetical baskets of widely consumed fruits and vegetables… 146

XII

Table 5.2: Availability, variety and price of fruits and vegetables in

catchments surrounding areas varying in socioeconomic characteristics…… 148

Chapter 6: Local food retailing and the purchase of fruits and

vegetables by socioeconomic groups

Table 6.1: Not purchasing common fruits and vegetables by household

income among participants of the Brisbane Food Study 2000………………. 162

Table 6.2: Odds of not buying fruits and vegetables by socioeconomic

groups in participants of Brisbane Food Study 2000: influence of local

socioeconomic and built environmental characteristics……………………... 164

Table 6.3: Odds of not buying fruits and vegetables in participants of

Brisbane Food Study 2000: effect modification of local environmental

characteristics by household income………………………………………… 167

Chapter 7: Confidence to cook vegetables and the buying habits of

Australian households

Table 7.1: Sample characteristics and descriptive results…………….……... 180

Table 7.2: Bivariate predictors of Respondents’ Confidence to Prepare

Vegetables, Confidence to use a variety of cooking techniques and

Vegetable Purchasing………………………………………………………... 181

Table 7.3: Multivariable adjusted sociodemographic differences in the odds

of having very low confidence to prepare vegetables and to use a variety of

cooking techniques…………………………………………………………... 183

Table 7.4: Age differences in confidence to cook among men and women

who usually cook for their household, adjusted for other demographic

characteristics …………………………………………...………………… 185

Table 7.5: Socio-demographic characteristics and confidence to cook as

predictors of household vegetable purchasing among adults who prepare

food for their households …………………….……………………………. 187

XIII

List of figures Chapter 1: Introduction

Figure 1.1: Conceptual diagram – socio-ecological model of socioeconomic

differences in diet and health………………………………………………… 6

Figure 1.2: Diagrammatic overview of the thesis research………………….. 10

Chapter 2: Literature Review

Figure 2.1: Conceptual diagram of local accessibility and affordability and

cooking skills as mediators of socioeconomic differences in fruit and

vegetable purchasing, based on socio-ecological framework……………….. 75

Chapter 3: Overview of Research Methods and Hypotheses

Figure 3.1: Conceptual model: Fruit and vegetable purchasing and total

household fruit and vegetable intake………………………………………… 91

Figure 3.2: Sample Size Requirement……………………………………….. 100

Figure 3.3: Costing for cooking skills study………………………………… 101

Figure 3.4 Sampling Process………………………………………………… 103

Chapter 4: Does living in a disadvantaged area mean fewer

opportunities to purchase fresh fruit and vegetables in the area?

Findings from the Brisbane Food Study…………………………………..

Figure 4.1: a) (left) Sampled Census Collection Districts (CDs) across the

Brisbane Statistical Subdivision; b) (right) One sampled CD with its

surrounding shopping catchment and local shops …………………............... 123

Figure 4.2: Distance from disadvantaged, medium socioeconomic, and

advantaged census collectors’ districts (CDs) to the nearest shop (km)a

a)(top) supermarkets; b) (middle) greengrocers; c) (bottom) convenience

stores…………………………………………………………………………. 133

XIV

Statement of original authorship

XV

Acknowledgements

The author would like to acknowledge the support of QUT and the Australian

government in provision of the necessary resources to complete this thesis including

an Australian Postgraduate Award scholarship. I would like to gratefully

acknowledge the invaluable support and guidance of the late Dr Carla Patterson

(Associate Professor of Public Health). I also wish to thank my principal supervisor,

Associate Professor Gavin Turrell for his contribution and patient willingness to read

countless drafts of manuscripts and chapters. I would also like to thank my associate

supervisor, Professor Michael Dunne, for his insight and continual reassurance and

good humour. I would like to extend an enormous thank-you to Professor Beth

Newman for assisting to supervise this thesis near its final stages, at late notice, and

for her near magical ability to double the readability of a draft with a few deft strokes

of a pen. I would like to express my gratitude to Dr Diana Battistutta for her

statistical support and answers to my scores of statistical queries.

I would also like to thank my colleagues for their assistance with the practicalities of

conducting research and general camaraderie: in particular Dr Wenbiao Hu, Julie-

Anne Carroll, Trish Gould, Tracey Dutton, Glenn Draper, Clare MacAuley, Loretta

McKinnon, Katrina Giskes and Lee-Ann Wilson. I would also like to thank the often

forgotten administration staff who have helped along the way, especially Carmel

Stafford, Mary-Anne Watts, Genny Carter and Meika Yaxley. On a more personal

note, I would like to thank my family and friends for their encouragement, emotional

support and prayers.

Finally, to my partners in rhyme:

To keep writer's block abated A limerick board was created With prose and with rhyme To waste up our time All nonsense was duly debated!

1

Chapter 1: Introduction

1.1. Background - Socioeconomic position, diet and health

There is a growing body of evidence linking diet to health outcomes. While much of

the focus in the developing world is on prevention of associated problems of ‘under-

nutrition’, in Australia the major nutritional focus is prevention of diseases and risks

related to ‘over-nutrition’. Coronary heart disease, diabetes and cancer are the more

prevalent, high-cost, diet-related illnesses in Australia (3). Insufficient dietary intake

of fruits and vegetables is arguably the single biggest independent dietary contributor

to ill health in Australia (4). The most recent national dietary survey indicates nine

in ten Australians do not consume sufficient fruits and vegetables (5). In recent years,

fruit and vegetable intake has shown no major improvement or deterioration in

Australia (3, 6) the United States (7) or the United Kingdom (8) and intakes remain

well below recommended levels. More generally, unfavourable secular trends in diet

and lifestyle have paralleled increases in overweight, obesity and associated

metabolic syndromes (9, 10). Fruits and vegetables are thought to reduce the risk of

heart disease, some cancers, and diabetes (11, 12). The mechanisms by which they

may promote health include their comparatively low content of substances

considered detrimental to health (sodium, saturated fats, energy) and high content of

substances considered beneficial (vitamins, anti-oxidants, phytochemicals and fibre)

(11, 12).

Morbidity and mortality follows a socioeconomic gradient in many nations including

the United States (13-16), the United Kingdom (17), and Australia (4, 18, 19). In

these countries and elsewhere, the lowest socioeconomic groups experience the

poorest health (20). Important diet-related diseases and disease risks, such as

cardiovascular disease (19, 21), stroke (22-27), some cancers (19, 21), and

overweight/ obesity (21, 28-30) all display this same socioeconomic patterning. On

the whole, Australia is experiencing increases in diet-related diseases (3) and

increasing socioeconomic disparities in health and diet-related health outcomes (3,

2

18). Concern regarding rises in chronic disease and socioeconomic inequalities is an

international phenomenon.

Several pathways through which socioeconomic inequalities in health are generated

and maintained have been put forward. Explanations for inequalities in health were

initially categorised as being artefactual, or relating to material deprivation, access to

health services, working conditions and environmental exposures, cultural factors

(including health behaviours such as diet), education, or alienation and normlessness

(31). These pathways are considered complementary rather than competing (15, 16,

32). Researchers have speculated that diet contributes to some of the socioeconomic

differences in mortality and morbidity (20, 21) because low socioeconomic groups

are the most likely to report diets that are consistent with higher risk of developing

many diseases that also most commonly affect low socioeconomic groups. Lower

fruit and vegetable intake among low socioeconomic groups has been documented in

numerous studies (33) and evidence suggests these dietary differences have persisted

over several decades (34).

Despite the relatively consistent description of the existence of socioeconomic

disparities in diet and diet-related outcomes, the comparative lack of population-

based studies into what underlies these disparities provides a poor evidentiary basis

on which to develop effective interventions that reduce inequalities in diet-related

disease outcomes. A conceptual model depicting likely mediators of socioeconomic

differences in diet-related behaviours (such as fruit and vegetable purchasing),

subsequent dietary intake and health outcomes is presented in Figure 1.1. As

outlined in more detail later (Chapter 2), there are several potential contributors to

socioeconomic differences in diet, and perhaps fruit and vegetable intake

specifically, including nutritional knowledge, attitudes and beliefs, food preferences,

food accessibility and affordability and cooking skills. Mostly, mediation of

socioeconomic differences in fruit and vegetable intake (or diet more generally) has

not been examined directly, however these factors have been associated with both

socioeconomic position (SEP) and fruit and vegetable intake (or other dietary

measures) and therefore could partially mediate socioeconomic differences in fruit

and vegetable intake.

3

1.2. Description of the research problem

This thesis examined the potential contributory roles of the food retail environment

and cooking skills to socioeconomic differences in the purchasing of fruits and

vegetables. Accessibility, affordability and cooking skills form part of nutrition

policy (35) but are under-researched, particularly in the Australian context.

Importantly, these potential mediators of socioeconomic differences in diet-related

behaviours may be modifiable through partnerships with the food retail and

education sectors. Fruit and vegetable purchasing was examined rather than dietary

intake of fruits and vegetables as accessibility, affordability and cooking skills are

likely to operate by influencing the behaviours that culminate in overall dietary

intake (such as purchasing, preparation and eating) (Figure 1.1). Purchasing was

examined specifically as this behaviour occurs within the food retail environment,

often for the purposes of later preparation and consumption, and therefore has

relevance to the possible operation of accessibility, affordability and cooking skills.

1.2.1. Theoretical framework The social-ecological perspective has been chosen as a theoretical framework for this

research because its characteristics enable a comprehensive and realistic view of both

dietary behaviour and socioeconomic inequalities. Giles-Corti and Donovan argue

that a social-ecological perspective is consistent with a focus on the broader

determinants of health behaviour (36). This perspective assumes that multiple facets

of the physical and social environment, as well as personal attributes (including

behaviour patterns), influence health and health behaviours through a dynamic

interplay. Rather than assuming exclusively that the environment determines

behaviour or vice versa, the social-ecological perspective considers that behaviour

and environments are related ‘cycles of mutual influence’. Further, this perspective

does not assume that the environment has the same impact on all individuals (37).

Accessibility, affordability and cooking skills can be examined within this

framework as it is suited to examining both individual and environmental

determinants of behaviour. The assumptions of a social ecological perspective lend

themselves well to studying social determinants of health behaviours, such as healthy

4

eating practices, where diet can be examined as a health-promoting behaviour that is

influenced by interdependent personal and environmental factors.

There are numerous advantages to applying the social-ecological perspective to the

research problem. It integrates multiple perspectives and theories (Sorensen et al.,

1998) and incorporates multiple levels of analysis and diverse methodologies

(Stokols, 1992). Thus, evidence which has arisen from a broad range of disciplinary

areas, having been driven from different perspectives and theories, can be

incorporated into a social-ecological perspective. For example, determinants of

dietary behaviour described in literature that has been informed by the other

psychosocial models can be considered as individual factors in a social-ecological

framework. Also, a socio-ecological framework allows people to be studied as

individuals, small groups or populations. This provides methodological advantages

in examining relationships between environmental and individual attributes and

socioeconomic variation in behaviour.

The use of a social ecological framework may also increase the relevance of findings

to health promotion efforts. There has been a call for redirection in health promotion

on the basis of ecological models of research and community intervention, as

research and practice have tended to focus on psychological factors, with less input

from sociology, anthropology, economics and political science (Stokols, 1992).

Often behavioural risk factors have been treated as individual attributes, ignoring the

surrounding environment (Von Korff et al., 1992). Some authors believe the

"individualisation of risk" has perpetuated the idea that risk is individually rather

than socially determined, and "behaviours" are dissociated from social context and

regarded as matters of free individual choice (Diez-Roux, 1998). Attention is turning

to the connections between individuals and their contextual settings (Duncan et al.,

1998) (Diez-Roux, 1998). Development of integrative models that address the joint

influence of personal and environmental factors in health promotion has been

highlighted as a challenge for public health research, including the ways in which

environmental factors contribute to the development, modification and maintenance

of health-promoting behaviour (Stokols, 1992). There is also increasing agreement

that health interventions are most effective when change occurs at many levels

(Grzywacz and Fuqua, 2000) and there is some evidence that targeting interventions

5

at multiple levels of the Social Ecological model may improve intervention

effectiveness (Sorensen et al., 1998).

Models other than the socio-ecological framework have been applied to

understanding dietary behaviours, however none offers a better alternative for this

research problem. For example, the Health Belief Model (38, 39), the theory of

planned behaviour (40, 41), market segmentation (42) and the transtheoretical model

(43) have all been applied to investigate or promote consumption of fruit and

vegetables. These models are largely psychosocial, focusing on individual attributes

and cognitive factors, such as specific knowledge, attitudes and beliefs. These more

individually-focused models have at best predicted only one-third of dietary

behaviour (44), probably due to missing important dietary determinants in addition to

the inherent difficulty in comprehensively measuring psychosocial factors.

Compared with the socio-ecological framework, these models are too individually-

focused to adequately examine both individual and environmental factors. Another

reason these models are ill-equipped to apply to the research problem is that while

they have had some success, albeit limited, in understanding dietary behaviours,

these models are not widely used in the sociological literature in examining

socioeconomic position, perhaps as they often treat socioeconomic position as a

direct, external influence on behaviour. This theoretical incompatibility makes them

poor choices to explain the relationship between diet and socioeconomic position

compared with a social-ecological perspective, which is suitable for focussing on

broader determinants of health behaviour (Giles-Corti and Donovan, 2002).

6

Figure 1.1: Conceptual diagram – socio-ecological model of socioeconomic differences in diet and health

7

1.2.2. Framing and operationalising the research problem

The potential contribution of food accessibility and affordability to socioeconomic

differences in fruit and vegetable purchasing was framed in terms of whether fruits

and vegetables are comparatively more or less accessible or affordable to people of

low socioeconomic position in the local environment, and whether the accessibility

and affordability of foods in the local environment has the same relationship with

behaviour within different socioeconomic groups. A socioeconomically patterned

food supply or a lesser ability to access the food supply among low socioeconomic

groups could contribute to socioeconomic differences in dietary behaviours.

Accessibility was conceptualised as the availability of supermarkets, greengrocers

and convenience stores in proximity to the neighbourhood (including the number of

shops, their distance from people, and opening hours) as well as the availability of

fruit and vegetable items within local shops. The affordability of fruits and

vegetables was examined in terms of the price of fruits and vegetables in local shops.

Fruits and vegetables were examined separately in this thesis as accessibility,

affordability and cooking skills would not be expected to influence their purchase to

an equal degree. Accessibility and affordability were examined for their possible

contribution to socioeconomic differences in both fruit and vegetable purchasing,

whereas cooking skills were examined only in connection with vegetable purchasing,

as fruits are often consumed with little preparation. Although it was not feasible to

assess cooking skills directly (45) a proxy measure is a person’s self-reported

confidence to cook (46). Confidence to cook captures people’s subjective

perceptions of their own skills, and focuses on how easy or achievable people find

the activity of cooking vegetables rather emphasising the quality of results of

cooking endeavours.

1.3. Objectives and aims of the study

This thesis aims to examine whether cooking skills and the accessibility and

affordability of fruits and vegetables in the local environment mediate

socioeconomic differences in fruit and vegetable purchasing among Brisbane

8

residents. Statistical models explore the impact of adjustment for the potential

mediators as well as the presence of effect modification, which may suggest they

operate differently in different socioeconomic strata. Consistent with the underlying

social-ecological model, such effect modification is examined as the environment

does not necessarily impact all individuals equally, and a greater influence on low-

socioeconomic groups might reasonably be posited to contribute to inequalities.

Mediation would suggest that accessibility, affordability and cooking skills might act

as causal intermediates of the relationship between socioeconomic position and fruit

and vegetable purchasing. The presence of effect modification would indicate there

may be merit in targeting any accessibility, affordability or cooking skills related

interventions to those groups for whom it would have the greatest impact on

behaviour. Due to the cross sectional data used, direct evidence of causal

associations are lacking, but patterns of results may yield useful insight into potential

mechanisms.

The specific aims were to:

1. Describe the relationship between accessibility and affordability and:

a. neighbourhood socioeconomic disadvantage

b. fruit and vegetable purchasing

2. Determine the whether accessibility and affordability:

a. mediate the relationship between SEP and fruit and vegetable

purchasing

b. modify the relationship between SEP and fruit and vegetable

purchasing

3. Describe the relationship between cooking skills and:

a. SEP

b. vegetable purchasing

4. Determine whether cooking skills mediate the relationship between SEP and

vegetable purchasing

9

1.4. Account of research progress linking the research papers

To meet the research aims, four quantitative analyses were conducted, based on two

data sources (Figure 1.2). A secondary data analysis of the multilevel Brisbane Food

Study was undertaken to meet aims one and two. The secondary analysis constitutes

original work, as substantial amounts of data preparation, all analytical planning,

statistical analyses and interpretation of results were conducted by the candidate. A

cross-sectional study was designed and implemented to meet aims three and four. A

a conceptual model of the role of accessibility, affordability and cooking skills in

socioeconomic differences in fruit and vegetable purchasing was developed from a

detailed literature review (Chapter 2). Following the review, the research hypotheses

are identified, and an overview of the methods employed is provided, including

details regarding specific aspects of the research not included in subsequent

manuscripts (Chapter 3). Findings from the studies are presented and discussed

within four manuscripts (Chapters 4-7).

The first two manuscripts examine the relationships between accessibility and

affordability and socioeconomic position, using data from the Brisbane Food Study.

The first manuscript was published in Health & Place and examines whether lower

and higher socioeconomic areas have equal access to shops in terms of numbers of

shops, distance to the nearest shop and opening hours(47). The second manuscript,

also published in Health & Place, examines whether lower and higher

socioeconomic areas have similar prices and in-store availability of fruits and

vegetables in local shops (48). The third manuscript, in preparation for submission

to the International Journal of Epidemiology, examines whether local accessibility

and affordability mediate socioeconomic differences in fruit and vegetable

purchasing. This manuscript examines whether shop access and the price and

availability of fruits and vegetables relate to residents’ fruit and vegetable purchasing

overall, whether they relate similarly to food purchasing within socioeconomic

groups and whether they statistically mediate the relationship between

socioeconomic position and food purchasing.

10

Figure 1.2: Diagrammatic overview of the thesis research

11

The fourth manuscript, submitted to the Journal of the American Dietetic

Association, examines whether the level of confidence to prepare vegetables is

similar among low and high socioeconomic groups, relates to household vegetable

purchasing, and mediates socioeconomic differences in vegetable purchasing.

A general discussion of the overall research findings can be found in Chapter 8,

including strengths and limitations, the public health implications of the findings, and

recommendations for future research. Collectively, this research contributes to an

understanding of why low socioeconomic groups tend to purchase and consume

comparatively fewer fruits and vegetables on a regular basis compared with their

higher socioeconomic counterparts.

12

Chapter 2: Literature Review

2.1. Overview

This chapter consists of two sections. The first section describes the process of the

review and outlines background material relevant to the relationship between

socioeconomic position and fruit and vegetable intake (or relevant associated

behaviours), most of which was collected and examined to inform topic selection.

The second section is the formal literature review, which seeks to examine the state

of the current evidence connecting the food retail environment and cooking skills,

two theoretical enablers of a healthy diet, with the dietary patterns of socioeconomic

groups. The state of the current evidence is examined in terms of quantity, quality

and consistency, and the need for and direction of future research is identified. From

the literature, a conceptual model was developed, depicting how accessibility,

affordability, cooking skills may mediate socioeconomic differences in fruit and

vegetable purchasing, and subsequent dietary intake. This review is limited to

studies conducted in developed Western nations (Australia, New Zealand, the United

States, Canada, the United Kingdom, and Europe). The review emphasises studies of

fruits and vegetables, but also includes studies that have used other dietary outcomes,

particularly where the methodology is of interest or where the literature is sparse.

2.2. Background

Multiple databases were searched (Science Direct, Informit Online, Info Track,

Proquest and Swets Wise) and manual searches of ‘Nutrition Abstracts’ were also

conducted. An initial search for all literature pertaining to explanations for

socioeconomic differences in diet was performed. Initially, in 2003, searches were

conducted for all scholarly literature that contained any of the following terms for

socioeconomic position (SES, socioeconomic, income, education, occupation, or

poverty) and any of the following terms that might indicate a focus on fruit or

vegetables specifically or diet more generally (diet, dietary, food, nutrition, fruit(s),

or vegetable(s)). Irrelevant studies were discarded, such as those focusing on the

13

education process rather than education as a socioeconomic marker, or in which

socioeconomic position had been mentioned only as confounder of some other

dietary relationship. Most of the relevant literature documented the relationship

between diet and socioeconomic position or pertained to the role of diet in explaining

socioeconomic disparities in health. A small amount of literature was found that

sought to explain socioeconomic differences in diet.

Later, as the research problem was defined and accessibility, affordability and

cooking skills were chosen as specific foci, more specific searches were conducted

for additional relevant materials on these issues. Searches were conducted for terms

relating to access (shop(s), retail, distance, access, accessibility, environment,

supermarket(s), availability), affordability (price(s), affordability, cost(s)), cooking

skills (cook, cooking, food preparation, culinary) in intersection with terms for either

socioeconomic position or diet used previously. Given the focus on fruit and

vegetable intake, studies of shops where fruits and vegetables are not typically sold

for at-home consumption, such as fast-food outlets, were deemed to be outside the

scope of this review. Relevant material from the initial search is summarised in an

indicative fashion in the background section that follows, while the literature

pertaining to the research problem is critiqued more comprehensively in the

subsequent literature review. The topic-specific searches were periodically updated

during the thesis to capture recent additions to the literature for inclusion in the

manuscripts. Where possible, the more recent studies were integrated into the

literature review, but unfortunately could not inform the formation of hypotheses and

study design. While this review attempts to be comprehensive, some literature may

have been overlooked if it used synonyms for the key terms other than those utilised

in the search, if the journal was not indexed within the university library’s databases

accessible by the search engines employed, or was published after the manuscripts

dealing with that topic were produced. Due to the timing of the thesis and variation

in how early pre-publication versions are made online means that literature from

2007 onwards may be incomplete.

Understanding social inequalities in health and promoting the consumption of fruits

and vegetables are significant policy and public health issues. The National Health

Priorities Initiative (49) describes where health efforts ought to focus in Australia,

14

and the major goals of this initiative are reducing both the burden of ill-health in the

population and inequalities in health between sub-groups of the population, such as

low socioeconomic groups. Consequently, public health nutrition efforts relating to

the priority areas of cardiovascular disease, cancer, and diabetes need to address

overall population health and health inequalities to be compatible with the

overarching health priorities framework.

One major nutrition effort in Australia is the continuing development and

dissemination of solid, evidence-based nutrition advice to the population. The

Australian Dietary Guidelines (50) recommend that in order to avert diet-related

diseases, adults should “enjoy a wide variety of nutritious foods” and “eat plenty of

breads and cereals (preferably wholegrain), vegetables (including legumes) and

fruits.” The Australian Guide to Healthy Eating informs adult consumers that they

should consume at least 5 serves of vegetables and at least 2 serves of fruit each day

(where each serve is approximately half a cup of cooked vegetables, 1 cup of raw

vegetables, or a medium-sized piece of fruit). Similar recommendations are made

for adolescents and older Australians. Official dietary recommendations are for a

high intake of a wide variety of fruits and vegetables on a regular basis (51), as

variety, quantity, nutritional quality and frequency of consumption of fruits and

vegetables are important in their disease prevention mechanisms. The 1995 National

Nutrition Survey indicates that Australians consume on average only 3.5 serves of

vegetables and 1.5 serves of fruit daily (52), leaving a substantial proportion of the

population consuming fewer than the recommended quantities. Inadequate

consumption of vegetables and fruits is estimated to account for 2.7% of the total

burden of disease in Australia (52).

Consequently, the National Nutrition Strategy, Eat Well Australia, emphasises the

importance of increasing the consumption of a wide range of fruits and vegetables in

the Australian population, and particularly in vulnerable groups (including people

who are socioeconomically disadvantaged) (35), as does the state nutrition strategy,

Eat Well Queensland (53). A national action plan has been formulated which aims to

increase the vegetable consumption of the population by at least one serve per day

15

within five years (54). The need to create supportive environments is clearly

outlined in these nutrition policy documents, and has been a focus of public health

since the Ottawa Charter for Health Promotion (55). This is in line with the

increased attention towards structural and environmental influences on health

behaviours within the “new public health” paradigm (56). Eat Well Australia and

Eat Well Queensland do not solely focus on individuals eating well, but also focus on

the food supply as an important aspect of ensuring good nutrition. The first of five

major objectives listed in the national action plan is to “increase and sustain access to

high quality, safe, affordable vegetables and fruit” (54).

Effective implementation of these policies requires supportive research. Eat Well

Australia (35) and Eat Well Queensland (53) both stress the need to base policy and

action on good information, in order to understand better the social, physical and

economic barriers to eating more vegetables and fruit. In recent years, major

breakthroughs have been made in monitoring the food supply in much of Australia.

Initially monitoring the price and availability of foods occurred in isolated regions

through efforts like the Kimberley Market Basket Surveys (57). Now much broader

sections of the food supply (the Northern Territory, Western Australia, South

Australia, and Queensland) are monitored through the Healthy Food Access Basket

Survey(58, 59), and recently efforts have been made to harmonise these surveys

across states to provide better national monitoring of the food supply (60). However,

this monitoring ideally needs to coexist with additional research that seeks to

uncover the way the food supply affects nutrition in the population and sub-

populations. Understanding why social groups differ in fruit and vegetable

consumption, focusing both on people and the environment, should contribute to the

evidence base for development of nutrition interventions with a focus on social

inequalities in health.

2.2.1. Fruits and vegetables and health

Fruits and vegetables contain nutrients, fibre, and phytochemicals, which are either

shown or speculated to be beneficial for health (61, 62). Consumption of fruits and

16

vegetables have been associated with reduced risk of stroke (63-65), heart disease

(66-69) (65) and some cancers (70-73), including mouth, oesophageal, lung,

stomach, colorectal, laryngeal, pancreatic, breast and bladder cancers (74). Higher

intakes of fruits, vegetables and both combined have been associated with reduced

risk of most of these diseases, although some associations are more convincing for

fruit intake (e.g. lung cancer) or vegetable intake (e.g. colorectal cancer) (74).

Consumption of cruciferous vegetables in particular has been associated with

reduced risk of lung, colorectal, breast, and prostate cancer (75). Causality is not

definitively established, and a number of biases could contribute to the observed

epidemiological associations (76, 77). However, a number of biologically plausible

mechanisms have been identified by which fruit and vegetable intake may prevent

some cancers and coronary heart disease (77, 78).

Reduced risk of stroke and coronary heart disease may operate through the effect of

fruits and vegetables on reducing LDL cholesterol (79), oxidative modification of

LDL (79, 80), platelet activation and aggregation (81), blood pressure (82, 83) and

homocysteine (84). Vitamins C and E, folic acid, coenzyme Q-10 (85, 86), fibre,

magnesium, potassium (87) appear to contribute to the protective effect of fruits and

vegetables against coronary heart disease and stroke. The relative lack of fat

(especially saturated fat), cholesterol and low energy-density of fruits and vegetables

also may be important.

Protection against cancer may come from a number of substances contained in fruits

and vegetables that are currently in clinical trials for cancer prevention, including

folate, vitamin E, vitamin C, calcium, selenium, lycopene, genstein, panthocyanidins

and resveratrol (78). Other constituents of fruits and vegetables may also contribute

to cancer prevention, including derivatives of chlorophyll (88) and derivatives of

glucosinolates (that occur in cruciferous vegetables) (75). The antioxidant

hypothesis is a popular explanation for the chemo-preventive action of fruits and

vegetables (whereby antioxidant substances are purported to reduce DNA damage

from reactive oxygen species and therefore prevent cancer initiation) (77, 78).

Constituents of fruits and vegetables may also reduce cancer initiation through the

detoxification of xenobiotic substances (75, 78, 88) and carcinogens (78). Fruit and

vegetable consumption may reduce cancer promotion and progression by altering

17

oestrogen metabolism (75), increasing the death of mutated cells (apoptosis) (75, 88),

activating tumor suppressor genes, improving immune response and reducing the

growth of blood vessels around tumours (angiogenesis) (78).

2.2.2. Fruits and vegetables and socioeconomic position

Studies from Australia, the United States and the United Kingdom and Europe show

that measures of fruit and vegetable intake or purchasing tend to be less consistent

with dietary recommendations among people who are less educated (33, 89-99), have

less skilled or manual jobs (89, 100-102) (33, 94, 97-99, 103) or have lower incomes

(89-91, 93-95, 99, 104, 105) compared with those with more education, more highly

skilled jobs and higher incomes. Details of these studies are presented in Table 2.1.

A systematic review estimated the magnitude of socioeconomic differences in fruit

and vegetable intake for men and women in the top compared with bottom quintiles

of education and occupation (33). Vegetable intake is higher among more highly

educated men (17.0 g, 95% CI: 8.6 to 25.5g) and women (17.1 g, 95% CI: 9.5 to

24.8g) and for men and women with more highly skilled occupations (20.1g (95%

CI: 9.6 to 30.5g) and (9.6g (95% CI: 1.2 to 18), respectively) compared with their

lower socioeconomic counterparts. Fruit intake is also higher among more highly

educated men (24.3g (95% CI: 14 to 34.7)) and women (33.6g (95% CI: 22.5-44.8g)

and for men and women with more skilled occupations (16.6g (-8.3 to 41.5g) and

11.4g (95% CI: 6.1 to 16.6g), respectively) (33). The net health effect of these

socioeconomic differences in fruit and vegetable intake has not been quantified (to

the knowledge of the author) however, socioeconomic differences in fruit and

vegetable intake have been described as a potential contributor to inequalities in

health outcomes (20, 21). Other dietary differences between socioeconomic groups

have also been observed (e.g. (106)), and these differences have been noted to also

generally show that the diets of low-socioeconomic groups are least consistent with

dietary recommendations (107).

18

Table 2.1: Associations between fruits and vegetables and low SEP Study Country Income Education Occupation Other SEP (89) AUS ↓ F ↓ V (variety

purchased) ↓ V ↓ F (purchase) ↓ F ↓ V

(purchase) -

(100) AUS - - Traditional V (+); Ethnic V (-)

-

(105) AUS ↓ F ↓ V; ↓ vit C, A, folate rich V; = vit C rich V (women); = folate rich V (men)

- - -

(90) AUS - ↓ F juice; ↓tomatoes; ↑potatoes; (=other)

- -

(101) AUS - - ↓ F ↓ V (women); = F = V (men)

-

(102) UK - - ↓ F & V ↓ F & V (98) UK - ↓ V ↓ F ↓ V ↓ F - (103) UK - - ↓ F & V - (108) UK - - ↓ F ↓ V (109) Sweden - - - = F; ↓F juice; ↓

V (97) - ↓ V; = F; = potatoes

(↑ for low/middle occupation men)

↓ Ves; = F; = potatoes

-

(33) b Europe - ↓ F ; ↓ V ↓ F ↓ V - (99) Norway ↓ F & V (quant) ↓ F & V ↓ F & V - (110) Ned - - - ↓ F; = V ; ↑

potatoes (91) US (1965) = F; =V

(1994-1996) Only high income increased F & V intake over time

(1965) ↓ F; ↓ V (1994-1996) only high income↑ F & V over time

- -

(104) US ↑ % improved F; ↑ % improved V ****

= - -

(93) a US = Fuit; = V ↓ Fuit ; ↓ V - (111) US - - - ↓ F ;↓ V area

SEP (112) US ↓ F; ↓ V / 1000

kJ (n.s) (92) US - ↓ F; ↓ V - - (94) US ↓ F & V ↓ F & V ↓ F & V - (113) US - - - (1965) ↓F; ↓ V;

(1965 - 89/91) ↑ F & V over time for high & middle SEP

(96) US - = F ; = spinach & kale; =potatoes; ↓ carrots& broccoli

- -

(95) US ↓ F ↓ V ↓ F ↓ V - - Legend: ↓ or ↑ = significant decrease or increase with lower SEP; +/ - = positive or negative association with low SEP; = no association; a non-probability sample; b meta-analysis; F= fruit; V= vegetables; n.s= p>0.05; ‘=’= no association

19

2.2.3. Fruits and vegetables, other dietary markers and socioeconomic context

Living in a low socioeconomic context has been associated with lower fruit and

vegetable intake (108) and having a less healthy diet according to other dietary

measures (111, 114, 115), independently of personal socioeconomic position. Not all

studies have replicated these findings (116). The absence of significant area

socioeconomic effects in the Brisbane-based study (116) compared with other studies

could relate to the methodological discrepancies between the studies but might also

indicate a true difference between Brisbane city and other locations. The Brisbane-

based study used census collection districts, not equivalent to boundaries used

elsewhere, and area effects may depend on the spatial unit employed (56). Some

positive findings could have been influenced by the purposive sampling of areas

(115, 117). Alternatively, the Brisbane-based study may have been too small to

detect the small area effects noted in other studies, having a smaller sample size

(n=50 areas, n=1003 individuals) compared with a UK based study (n=52 areas,

n=3039 individuals) (114) and the ARIC study in the United States (n=13095

individuals in an unreported number of census blocks across n=4 counties) (111).

The conclusions to be drawn from these studies are that area socioeconomic

characteristics might influence selected aspects of residents’ diets, either

independently or through other associated factors, such as the local shopping

infrastructure or social capital.

These studies have attempted to separate the effect of socioeconomic context from

that of personal socioeconomic position by using ‘contextual-effects’ models or

‘multi-level’ models to partition the variance in dietary measures that occurs across

areas from the variance that occurs between individuals. The argument is made that

relationships between area-level socioeconomic position and diet after statistically

adjusting for individual-level socioeconomic characteristics indicate area effects.

From one perspective, these studies may overstate the contribution of areas, as

effects attributed to areas could be partially or wholly an artefact of unmeasured

characteristics, measurement error or improper partitioning of individual and area-

level variance (118). The latter explanation for findings is unlikely as the

partitioning of variance is more problematic in contextual-effects than multi-level

20

models (118), and two (111, 114) of the three studies (111, 114, 116) that used multi-

level techniques found area socioeconomic characteristics were associated with some

dietary measures. From another perspective, this approach may understate the

contribution of context, firstly as the approach gives theoretical priority to the

individual-level characteristics, and secondly as socioeconomic context may be more

imprecisely measured than individual-level socioeconomic position. The assessment

of socioeconomic position of individuals and households is historically more

developed and simpler than measuring socioeconomic context (119), and the

geographical aspect of exposure may have been inadequately captured by these

studies. Boyle and Willms (120) argue it is necessary to define areas to maximise

between-area differences in exposure (such as socioeconomic disadvantage) or

response (such as diet), although the appropriate definition of area suitable for study

depends on the issue being studied, and there is no agreement about how best to

define a geographical area in terms of socioeconomic position (121). These studies

have generally used administrative boundaries as proxies for neighbourhoods (census

tracts, postcodes, regions), which MacIntyre et al. (56) point out may be poor proxies

for neighbourhood.

2.2.4. Factors that may contribute to socioeconomic differences in diet

From the literature, a number of factors were associated both with diet and with

socioeconomic position, and could possibly contribute to dietary differences across

socioeconomic groups. These are presented in the conceptual model of mediators of

dietary and diet-related health inequalities alluded to in the previous chapter (Figure

1.1). There are too many potential contributory factors to examine in any depth, so a

brief overview is provided here. Being the focus of this thesis, accessibility,

affordability and cooking skills are examined comprehensively further on.

The potential exists that household composition and dynamics may vary across

socioeconomic groups, which could contribute to dietary differences among

socioeconomic groups. The acquisition of foods has been linked to the number of

people who live in a household (122-124), and their characteristics (122, 125-129).

21

These characteristics include the age, gender, and occupation of the main food

procurer, the absence or presence of a partner and his/her occupation, and the

presence of children and their number and ages. It is usually women who report

bearing the major responsibility for food purchasing and cooking in the United

Kingdom (130), the United States (131), and contemporary Australia (116). The

preferences of the person who purchases food and of other household members form

part of the food choice process (132). People sometimes mention other family

members as 'barriers' to making healthy choices in studies of general populations

(133) as well as in low socioeconomic groups (134, 135). Family members can also

have a positive role on diet, for example in terms of household food rules that shape

children’s diets, and these household rules have been shown to vary according to

household socioeconomic position (136).

Food preference also has been shown to relate strongly to food choice (137-140).

Differences in food preferences among socioeconomic groups have been proposed as

a mechanism to explain the dietary differences observed between socioeconomic

groups (141, 142). Food preferences are related to exposure (143) which could

provide a cycle whereby low income groups remain unexposed to healthy foods and

therefore do not select them.

Knowledge of dietary recommendations, the relationship between diet and health,

attitudes and beliefs towards the importance of nutrition, and specific nutrition

advice have long been studied in relation to dietary behaviours and dietary intake

(92, 133, 144-149). A relationship exists in which cognitive factors partly relate to

or predict dietary behaviours or intake. Studies have shown a relationship between

socioeconomic position and cognitive factors (150, 151) and demonstrated that

differences in the measured cognitive factors partially mediate socioeconomic

differences in diet.

The skills to purchase and prepare foods may contribute to dietary differences

between socioeconomic groups, although these topics have been seldom studied.

Multifaceted interventions that include provision of food skills (usually covering

both food purchasing and preparation) have been used to improve the dietary

behaviours of low-income groups, for example in Australia (152, 153) and the

22

United States (154, 155). The evaluations of these programs generally point to some

degree of success, however the direct attribution of success to cooking or budgeting

skills is not possible due to the multifaceted approach of these interventions. It is

doubtful that a lack of budgeting skills explains socioeconomic differences in fruit

and vegetable consumption, as studies of expenditure indicate that low-income

families purchase food more efficiently than higher income families, at least from a

monetary perspective (156, 157).

Evidence suggests travel-related resources and practices vary between

socioeconomic groups, and therefore could possibly contribute to socioeconomic

differences in dietary behaviours. Lack of resources (for example, car ownership)

impairs access to food (158) and the way in which people travel is considered a facet

of food access (159). Mode of travel to shops varies across socioeconomic groups,

as low-income shoppers are more likely to lack private transport and rely on taxis

(which are expensive), walking (which imposes logistic constraints on shopping) or

public transport (which can be problematic in both these manners) (160, 161).

Census data for 1996 (162) indicate that despite high overall car ownership rates

(87%), 41% of households in which the weekly income was less than $159 reported

that they had no motor vehicle compared with only 1% of households in which the

weekly income was more than $1,500.

A number of features have been associated with both diet and socioeconomic

position, including household composition and dynamics, food preferences,

budgeting skills, nutritional knowledge and other cognitive factors (Figure 1.1). This

thesis, and the literature that follows, focuses specifically on accessibility,

affordability and cooking skills as potential contributors to the differences between

socioeconomic groups in their consumption of fruits and vegetables. Accessibility

and affordability may be a particularly relevant focus of study as attention is

increasingly focusing on the role of the environment in shaping health and health-

related behaviours (including diet) and as a contributor to socioeconomic inequalities

(163). Cooking skills are also an important focus of study in view of contemporary

trends towards convenience foods, which have prompted some researchers to

examine the possible role of cooking skills in enabling healthy dietary choices, for

low-socioeconomic groups in particular (164, 165).

23

2.3. Environmental factors and socioeconomic position

Accessibility and affordability depend on the external environment and individual

characteristics which facilitate or hinder people in procuring food from the

environment. The accessibility of foods relates to the provision of local shops (eg.

their abundance, opening hours, proximity to people and public transport services,),

the types of foods available within shops, and also relates to individual factors (such

as mobility constraints and resources for private or public transportation) (159). The

cost of food is a key component of affordability, as is the purchaser’s ability to meet

food costs (which depends on income and other budgetary costs). Among the studies

included in this review, the accessibility of fruits and vegetables and other foods

necessary for a healthy diet has mostly been studied objectively on the basis of

access to shops, the availability of food items within shops and food prices, which

have been defined and measured in varying ways. Subjective measures that focus on

how available or affordable people perceive foods to be have also been used by some

researchers.

The studies included in this review have mostly examined whether environmental

measures of accessibility and affordability vary according to area-level

socioeconomic characteristics (such as median income, poverty rates or indexes that

focus on multiple aspects of socioeconomic disadvantage). Some studies have

examined whether individual measures of accessibility and affordability are

associated with to individual-level socioeconomic position. Details of the studies

examining the relationship between socioeconomic position and access to shops, in-

store availability and price are provided in Table A1.1 of Appendix 1. Findings for

each measure of shop access, availability and price are presented in the sections that

follow, and are summarised in Table 2.2.

24

Table 2.2: Findings of studies assessing the relationship between socioeconomic position and features of the food retail environment Study Details Access to shops In-store

availability Price Quality

Location Spatial, SEP Type of shop used

Shop access Distance Other access

MacDonald & Nelson 1991

multi-city US

zip codes, poverty rates

central city (store size) (chain status)

non-central (store size) (chain status)

non-central (Food basket)

central city x (Food basket)

Kaufmann (1999)

Lower Missisipi Delta (US)

rural high poverty counties, household income

study counties vs average for Arkansas, Louisiana, Mississipi

nst (s'mkt/ sq km)

nst (accessibilty ratio based on food stamp use)

Alwitt & Donley (1997)

Chicago, US

zip-codes, multiple indicator

zip-codes R ns (all shops) R (small groc)

(large groc) (s'mkt)

within 2 miles & within 3 miles

ns (all) x (small groc)

(lge groc) (s'mkt)

Per $mil purchasing power R (all) R (sml groc) x (lge groc) x (s'mkt)

Finke et al 1997

multi-city US

Households, household income

Price paid (expenditure) ns (overall)

(urban) x (suburban)

(Black) x (White)

25

Study Details Access to shops In-store availability

Price Quality Location Spatial, SEP Type of

shop used Shop access Distance Other access

(Urban Black) ns (Urban White)

Chung & Myers Jr 1999

Hennepin & Ramsey counties, US

zip code, poverty rates

(% chain stores) x (grocery & produce items)

ns (Food Basket)

Fisher & Strogatz 1999

New York, US

zip-codes, median income

low fat milk (% shelf

space)

Hayes, 2000

multi-city US

median household income assigned to store

R nst (stores) R nst (per capita)

nst (store size)

R (Food Basket) R ns (oranges)

(lettuce)

Frankel & Gould 2001

multi-city US

cities, poverty rates, median income & change over time

Basket of grocery items, price

nst (poverty) nst (income)

Basket of grocery items, price change over time

nst (income) nst (poverty)

Morland et al., 2002b

multi-city US

census tracts, wealth

(s'mkt) R (groc)

x (conv)

Topolski Boyd-Bowman, & Ferguson 2003

mid-sized southern city' US

6 stores in three strata of zip code median household income

Fruit

(appear-ance) (taste)

Both overall & by chain

Horowitzet al., 2004

Harlem New York, US

Upper East Side vs East Harlem

R nst (all shops per capita)

(shop size) 'desirable' stores

bread milk green V fruit

26




R (lacking stores) x diet soda Gallagher (2005)

Chicago, US

community areas, annual per capita income

"major player" groc stores per capita

nst (all) nst (Jewel) nst (Dominick)

R nst (Aldi)

Zenk et al., 2005

Detroit, US Census blocks, poverty rate

(sup)

Baker et al., 2006

St Louis, US

zip-codes, median household income

nst (78 F & V items)

Jetter & Cassady, 2006

Sacramento & Los Angeles, US

zip code & within 5 km (median household income)

s'mkts & independent groc R ns (Regular Basket) R (Healthy Basket)

19 groc items nst

Moore & Diez - Roux (2006)

multi-city US

census tracts, median houshold income

per capita R (groc)

(sup) R (conv) x (F&V)

Powell et al., (2007)

multi-city US

zip-codes, median household income

overall (chain s'mkt)

R (non-chain s'mkt) R (groc)

(conv) urban subsample

(chain s'mkt) R (non-chain s'mkt) R (groc) R (conv

Travers et al., 1997

Nova Scotia (Canada)

counties, income x (thrifty food basket) x (regular basket) x (healthy substitute basket)

27




x (other healthy basket)

Smoyer-Tomic et al., 2006

Edmonton, Canada

Postal areas (Y/N meeting low-income cut-off)

R (s'mkt) R (s'mkt)

Latham & Moffatt 2007

Ontario, Canada

"Uptown" vs "Downtown"

nst (% s'mkt) per capita R nst (variety) R nst (groc)

nst (s'mkt) nst (specialty)

per sq km R nst (variety) R nst (groc) R nst (s'mkt) . R (specialty)

fresh F & V x nst (variety) x nst (groc) x nst (s'mkt)

Food Basket x nst (variety) x nst (groc) R nst (s'mkt)

Sooman et al., 1993

Glasgow, UK

2 regions (Low vs high SEP)

nst (most items)

nst (healthy basket) x (less healthy basket) x (F & V basket)

F & V nst

(quality rating)

MacIntyre & Ellaway 1998

Glasgow, UK

4 wards, varying disadvantage & occupation of household head

Low SES less likely to shop in own ward

(except lowest SEP ward)

ns (by occupation)

Cummins & MacIntyre 1999

Glasgow, UK

Postcode, composite deprivation index

nst (F&V) nst (independent

groc) nst (freezer) nst (discount s'mkt) nst (multiples s’mkt)

Cummins &

Glasgow, UK

postcode sectors, composite

all stores x 46 /57 items

x (fruit) x (V) by item

28




MacIntyre, 2002

deprivation index 10/57 items R 1/57 items (coca cola)

x (fish & meat) & mixed results item by item Branded method Multiples x 47 / 57 items 7/57 items

R 3/57 items Independent x 45/57 items 5/57 items R 7/57 items Cheapest price Multiples 52 /57 items

1/57 items R 4/57 items Independent x 50/57 items

2/57 items R 5/57 items

Furey et al., 2002

Ireland 4 wards, 2 high, 2 low SEP (multiple indicators)

all nst R nst (food basket) x nst (fresh green) x nst (fresh other V) x nst (frozen V) . x nst (processed V) x nst (fresh fruit) x nst (26 items)

Dibsdall et al., 2003

East Anglia, UK (pubic housing )

personal SEP (occupation)

(perceived transport difficulties)

nst (own car)

(perceived affordabiltiy)

Guy et al., 2004

Cardiff, UK

electoral divisions; composite deprivation index

nst (closures) nst (openings) nst (number shops)

Composite access (shop attractiveness &

29




distance) nst (baseline) nst (end) nst (change)

Dibsdall et al., 2003

East Anglia, UK (pubic housing residents)

personal SEP (occupation)

(perceived transport difficulties)

nst (car ownership)

(perceived affordabiltiy)

White et al., 2004

Newcastle, UK

enumeration district, compsite deprivation index (of residence or store)

ns (shop type) distance to shop selling…

R (10 F&V) R (10 F&V hi

quality) R (14 F&V) R (21 healthy) R (10

unhealthy)

(reported difficulty shopping) x (opening hours)

In shops mostly used by respondents x (10 F&V) x (14 F&V) x (21 healthy items) x (10 unhealthy items)

x (price 33 items) R (price F&V) expenditure on food

(% income) R (absolute)

x (F&V )

household, composite household SEP index

R (shop used) distance to shop selling... x (10 F&V)

R (10 F&V hi quality) x (14 F&V) x (21 items healthy)

R (10 items unhealthy)

(reported difficulty shopping)

x (10 F&V) x (14 F&V) x (21 healthy items) x (10 unhealthy items)

expenditure on food (% income)

R (absolute)

Burns & Inglis 2007

Melbourne, AUS

census collection districts, SEIFA

(comparative distance s'mkt vs takeaway)

F = fruit; V= vegetable; s’mkt = supermarket; groc= grocery store; conv= convenience store = association in the expected direction: more shop access, better availability and lower price is associated with more healthy diet; R = association in the opposite to expected direction x = no association ns = association is present qualitatively but not statistically significant at p<0.05; nst = association is not statistically tested

30

Reporting has not been consistent across studies. Some studies have reported only p-

values while others have reported associations without performing statistical tests.

Where possible, measures of spread (such as standard errors and confidence

intervals) are presented, however their reporting in the literature was limited. In this

review, findings are referred to as being statistically significant if p<0.05, non-

significant if p>0.05. Unfortunately, some subjectivity is inherent in the review as

there is insufficient literature to develop useful a priori definitions of meaningful

effect sizes. Accordingly, wherever available, this review presents the magnitude of

associations that did not reach statistical significance, to enable the reader to develop

their own judgement of the study findings. Details of the measurements and results

are available in Appendix 1, Tables A1.1-A1.4.

2.3.1. Access to shops

United States

All included studies from the United States (166-174) and a government report (175)

have found differences in provision of food shops according to area socioeconomic

characteristics. In general, these studies have shown a pattern of fewer supermarkets,

chain or large stores, and more small or independent stores in low socioeconomic

areas.

MacDonald and Nelson (166) defined poor zip-codes were those with >10% poverty

rates, and found a higher proportion of grocery stores were independent (rather

chain-operated) in poor (36%) compared with all other zip-codes (12.5%). They also

found the stores in poor zip-codes were smaller in terms of floor space (11600 vs

19500 square feet), and found a significant, positive correlation between median

neighbourhood income and store size (B=0.680, p<0.01 for the correlation between

median income store size (square feet) on a log scale). Similarly, Chung and Meyers

(168) found 40% of all non-chain grocery stores, but only 11% of all chain stores,

were located in the poor zip-codes (>20% poverty) in Hennepin and Ramsey

counties, Minnesota.

31

Alwitt and Donley (167) found that on average, the most disadvantaged zip-codes in

Chicago (based on poverty, education and employment) contained over two times

fewer supermarkets and 55% fewer large grocery stores, but 55% more small grocery

stores compared with non-poor zip-codes, irrespective of zipcode size and population

density (p<0.05 for all comparisons). The Atherosclerosis Risk in Communities

(ARIC) study conducted in Mississippi, North Carolina, Maryland and Minnesota,

found wealthy, relative to less wealthy census tracts, contained significantly more

supermarkets (Prevalence ratio (PR): 3.3, 95% CI: 1.4, 7.9) and fewer grocery stores

(PR: 0.6, 95% CI: 0.3, 0.9), but contained similar numbers of convenience stores

(PR: 1.0, 95% CI 0.6, 1.8), adjusted for population density and ethnicity (170).

(Prevalence ratios reported here are for the top versus bottom quintiles of

neighbourhood wealth, measured by median housing prices.) According to a report

by Gallagher (175), there were more major chain grocery stores shops per 100, 000

residents in higher compared with lower income community areas of Chicago. Areas

with lowest compared with highest quartile of income contained 2.2 times fewer

grocery stores overall, but 83% more discount grocers (Aldi).

As part of the Multiethnic Study of Atherosclerosis in Maryland, North Carolina and

New York (specifically in northern Manhattan and the Bronx), Moore and Diez-

Roux (172) found low-income census tracts contained similar numbers of fruit and

vegetable markets (PR: 0.9, 95% CI: 0.6, 1.4) , but significantly more grocery (PR:

4.3, 95% CI: 3.6, 5.2) and convenience stores (PR: 2.4, 95% CI: 1.8, 3.2) and fewer

supermarkets per 100,000 population (PR: 0.5, 95%CI: 0.3, 0.8), compared with

high-income census tracts, having accounted for tract size and population size. A

study of metropolitan Detroit (176) found greater accessibility of supermarkets in

high-poverty compared with low-poverty census blocks using three separate

definitions of poverty: Manhattan Block distance to the nearest supermarket, sum of

the distances to all Detroit supermarkets, and the number of supermarkets within a

three-mile radius. Results were presented only for distance to the nearest

supermarket and reflected an excess distance to supermarkets of 0.7 miles in high-

poverty areas compared with low-poverty areas, having adjusted for population

density, ethnicity and spatial autocorrelation.

32

Powell et al (173) examined 28,050 zip-codes across the United States which had

available grocery store data from private business listings (Market Place) and data

from the 2000 US Census. Unlike other studies, they used middle income zip-codes

(defined as the middle three quintiles of median household incomes) as the basis for

comparison and found low income zip-codes contained fewer chain supermarkets

(RR=0.75) and slightly more non-chain supermarkets (RR=1.10) and grocery stores

(RR=1.18) and a similar number of convenience stores (RR=0.96). All differences

were statistically significant, perhaps in view of the size of the study. High-income

zip-codes contained significantly fewer shops of all types (RRs from 0.62 to 0.84)

compared with middle-income zip-codes. Most of their results held true for a sub-

sample of 4404 urban zip-codes, and were independent of ethnicity, population and

region (South, West, Midwest and North East).

Two studies that used a different approach to measuring shop accessibility also noted

socioeconomic patterning. Kaufman (169) examined 36 low-income counties in the

lower Mississippi Delta and found a greater proportion of low-income households

were located in zip-codes classed as having low accessibility compared with what

would be expected from the distribution of accessibility by zip-code within these

counties (using accessible food stamp redemptions as a proportion of food stamp

sales as a measure of shop accessibility). Accessibility ratios > 1, which indicated

poor accessibility, were found for 30.8% of the study sample compared with 22.5%

of all zip-codes in the Lower Delta core counties overall. Their sample of low-

income counties had fewer supermarkets per square mile than the average for rural

counties in Arkansas, Louisiana and Mississippi (1 per 190.5 vs 1 per 153.5 square

miles). Horowitz et al (171) found significantly different access to different types of

shops in census blocks within two contrasting areas. Compared with residents of the

lower socioeconomic, predominantly non-white, area of New York (East Harlem),

residents of the higher socioeconomic, predominantly white, area (the Upper East

Side) had more stores classed as ‘desirable’ (stocking at least one of the items the

authors described as potentially useful to the diet of a person with diabetes mellitus,

including fresh fruit and fresh green vegetables) (RR: 3.2, 95% CI: 2.2, 4.6). In part,

these results reflected the greater proportions of all stores that were medium (RR:

3.0, 95% CI: 1.5, 6.1) and large (95% CI: 2.8, 1.4, 5.8) and lesser proportion of small

stores (RR: 0.7, 95% CI: 0.7, 0.9) in the upper socioeconomic area. However,

33

accounting for shop-type by stratification, small stores were more likely to be classed

as desirable when located in the upper socioeconomic area (5.3, 95% CI: 3.1, 9.1)

which may be important as small stores were the least likely to be classed as

desirable overall.

Canada

Few Canadian studies were located (177). As part of a mixed methods study,

Latham and Moffat (177) compared shop access in one high and one low

socioeconomic area (based on income, education, unemployment, lone-parent

families and ethnicity), which they dubbed “Uptown” and “Downtown” Hamilton.

Compared with Uptown, Downtown had 43% fewer supermarkets but had 40% more

specialty stores, 60% more grocery stores, and 2.8 times the number of convenience

stores on a per capita basis. However, Downtown was an inner-city area with higher

population density, and Downtown actually had approximately 6 times the number of

all types of shops per square kilometre than Uptown. These findings were supported

by a stronger study conducted in Edmonton (178) that found a weak-to-moderate,

positive correlation (Spearman’s R=0.35, p<0.001), between the number of

supermarkets within a one-kilometre radius of neighbourhoods, defined by postal

areas, and the percentage of low-income households. Similarly, the percentage of

low-income households was correlated with less distance to the nearest supermarket

(Spearman’s R= -0.387, p<0.001). Interestingly, the low-income areas in this study

tended to cluster towards the population-dense inner-city, a finding which may be

unique to the study setting, but which, unlike the Hamilton study, cannot be

explained by the purposive sampling of neighbourhoods. This may have contributed

to the results obtained, as inner-city neighbourhoods in this study had more

supermarkets and lesser minimum distances to supermarkets than the study area as a

whole.

United Kingdom

In contrast to the U.S. studies, research from the United Kingdom has been more

heterogeneous, in terms of method, measures and findings, with one study finding

34

less access to shops in low socioeconomic areas (179), one finding greater access to

shops in low socioeconomic areas (180), and one finding both greater and less shop

access to low income households, depending on the access measure employed (181).

A study in Glasgow (180) most resembled the U.S. studies in methodology but least

resembled the U.S studies in results. This study reported that the most disadvantaged

post-code districts (according to the Carstairs-Morris Deprivation Category) had

more stores of all types than the more advantaged post-code districts. (Numeric

descriptions and statistical tests not reported).

A preliminary study identifying food deserts in Cardiff pointed towards a

socioeconomic patterning in shop accessibility, since four out of five of the “food

deserts” identified had lower proportions of high socioeconomic households (social

classes A and B) compared with Cardiff as a whole (182). This socioeconomic

pattering was explored in subsequent research that found comparatively less

accessibility to shops (based on residents’ expenditure, shop attractiveness to

consumers, and the distance between shops and residents) in more deprived electoral

districts (based on the Welsh Index of Multiple Deprivation) (179). Overall,

accessibility scores increased over time, however accessibility scores were lower for

the 50 most deprived compared with the 50 least deprived areas, by 33% in the 1990s

and by 38% in 2001. Changes in accessibility corresponded with the

disproportionate store closures occurring in the most disadvantaged areas.

A comprehensive study of food access was conducted in Newcastle upon Tyne in the

United Kingdom, which used multiple measures of access to shops and included a

specific focus on fruits and vegetables (181). The study included 5044 individuals

from 3153 households, and all 560 food retail outlets in the study area that agreed to

participate (85% participation). Two socioeconomic measures were employed,

Townsend Deprivation Scores (TDS) at the enumeration district level (which was

applied to the location of households and shops), and a composite household measure

that incorporates income, standard of living, housing quality and tenure, and welfare

receipt. Mostly, the associations with socioeconomic position were noted for the

TDS measure, although some also were noted for the household socioeconomic

measure. Low socioeconomic position was associated with shopping at discount

rather than multiple supermarkets, and with reporting experiencing difficulty in

35

carrying shopping home, but not with mode of travel to stores or shop opening hours.

Areas in which convenience stores were located had higher average disadvantage

(TDS=10.55) compared with the locations of discount supermarkets (TDS=7.85) and

freezer centres (TDS=7.27), however the differences were not statistically

significant. On average, low socioeconomic households lived closer to various shops

than high socioeconomic households, including the shop mainly utilised by

respondents (difference in median distance: 1094 m), shops that sell a wide range of

fruit and vegetables (294 m), and shops that sell high-quality fruit and vegetables

(353 m).

Australia

Only one Australian study on this topic was located. Burns and Inglis (183) used

council data on shop locations, road networks, slopes and boundaries and GIS

software to map travel distances to supermarkets and major fast-food chain stores in

Casey, located in South-East Melbourne, Australia. They classed census collection

districts as being relatively closer (by road travel time) to supermarkets, fast-food

shops or equally close to both and found a significant linear trend between increasing

disadvantage, as measured by SEIFA scores, and less relative closeness to

supermarkets. Areas that were closer to takeaway shops than supermarkets were the

most disadvantaged (mean (SE), 957.9 (75.9)), areas equidistant to both were less

disadvantaged (988.0 (54.2)) and areas that were closer to supermarkets than

takeaway shops were the least disadvantaged (1016.2 (81.6)). Based on the measure

used, it is uncertain whether the results are reflecting a socioeconomic difference in

the location of supermarkets, takeaway stores, or both.

Alternate measures of access

Other studies have taken a more subjective approach to examining access. An early

case study in Glasgow had found residents in poorer areas were less likely to shop in

their own locality compared with residents in a higher socioeconomic area (West

End) (Garscaddan OR: 0.50, p<0.05, Mosspark OR: 0.39, p<0.05, Greater Pollock

OR:0.69, p>0.05) (115). In light of later work showing greater shop access in low

36

socioeconomic areas of Glasgow (180), these findings may not relate to access

issues. Another study noted occupational differences in difficulties with transport for

shopping, even among an exclusively low-income population (184). Lower attitude

ratings, indicating more perceived difficulty with transport for shopping, were found

among jobseekers and retirees (Mean (SE): 4.9 (1.5) and 5.3 (1.4), respectively)

compared with other occupational groups (i.e. 5.7 (1.2) employed full-time, 5.6 (1.2)

employed part-time, 5.7 (1.3) on sick leave 5.5 (1.2) family care). This might

represent an age effect rather than an effect of occupation as a socioeconomic

marker.

Extent and quality of the evidence of a socioeconomic gap in shop accessibility

Overall, less access to shops in low-income areas has been found consistently in

studies conducted in the United States, but inconsistently elsewhere, and too few

studies have been conducted to draw conclusions within an Australian context. It is

uncertain whether the varied findings reflect true contextual differences or are a by-

product of the varied array of methods and measures employed. For example, the

differences between the U.K. and the U.S. findings may owe to the tendency of U.K

studies to classify area socioeconomic characteristics based on indices of

socioeconomic disadvantage, or deprivation, and a tendency for the U.S. research to

focus on income-, wealth-, or poverty-based measures. (This is not necessarily a

technical issue so much as a reflection of the different experience of disadvantage

across contexts). It is possible that findings of each study may not be generalisable

beyond their locations, as the study settings were not randomly chosen to reflect

entire states, countries, rural or urban settings in general, and the patterning of

socioeconomic characteristics and retail development may vary across locations.

Also, findings may not be generalised across time periods, as suggested by the

changes over time in Cardiff (179). Whether socioeconomic differences in shop

provision would be expected to occur in all contexts depends on the rationale for

expecting relationships between socioeconomic characteristics and the availability of

shops.

37

Literature from economics shows that residents’ income forms part of market

potential, which is used to decide where to place food shops (185). A process of

‘redlining’, whereby investors avoid locating in low-income areas (186), and the

lower purchasing power of people with low incomes (187) could directly produce a

relative lack of large shops in low-income areas. The study designs have not been

able to rule out reverse causal relationships. For example, the availability of nearby

facilities such as shops might be a comparatively more or less influential

consideration for low socioeconomic groups in deciding where to live. Any of these

processes could produce socioeconomic differences in shop access that would be

expected to be fairly consistent across different contexts. However, the comparative

lack of shop access in poor areas could also operate indirectly through the spatial

distribution of socioeconomic characteristics across cities, in which case

socioeconomic differences in shop access would be expected in some contexts, but

not others. In many U.S. cities, affluent families have migrated from the inner city to

the suburbs (186), and studies from the U.S. have also noted that the low income

areas have tended to cluster towards the population-dense, inner-city area (187, 188).

The tendency for low socioeconomic areas to cluster towards the inner city may have

contributed to the results, as only one study (166) accounted for the inner-city

location of areas (by stratification), and non-chain (189), independent and smaller

stores (188) have been disproportionately located in inner-city areas. Socioeconomic

differences in shop location may partly, but not entirely, depend on population

density, as studies have still had positive findings despite adjustment for population

density (172, 173, 187). Future studies need better description of context, and

consideration of the processes by which socioeconomic differences in shop access

may occur is necessary to move this research area forward.

Apart from contextual differences, the methodological approaches used in these

studies may have contributed to their findings. The studies in this review have tested

the relationship between access to shops and socioeconomic characteristics using

areas, shops, and individuals as units of analysis. Each comparison offers a different

perspective on access and may have been affected by different biases. Studies that

have used shops as a unit of analysis (166, 181, 188) provide information about the

socioeconomic characteristics of areas in which different types of shops are located,

but cannot fully quantify the level of access to shops in different socioeconomic

38

areas as they do not examine the socioeconomic characteristics of areas that lack

shops entirely. However, findings from these studies have been consistent with the

majority of studies that have compared access across areas (169-173, 175, 179, 180,

187). Using individuals as a unit of analysis (115, 181, 184) is the most problematic

approach, due to the difficulties in obtaining an unbiased sample representative of

residents in the area studied. The different units of analysis could contribute to the

discrepancy between the different shop access measures in the Newcastle study

(181), as the distance findings may have been affected by the low survey response

(18% for households, and 83% for individuals within participating households), if

participating households were not geographically representative.

The majority of research, and the most direct evidence, comes from studies that have

compared shop access across areas, however the way in which they have measured

shop access may have affected results. Measurement of access to shops has occurred

on a per capita, per square kilometre, or per neighbourhood basis. There is no

consensus as to which way shop provision is best measured, however the approaches

are not equivalent and do not always yield the same results (177). The per capita

approach is problematic, as areas with high population density may show low shop

provision per capita (but still have similar shop numbers proximal to residents as

other areas with less population density). In contrast, areas of low population density

may appear well served by shops per capita (although they may contain only one

shop servicing a comparatively large geographic area). In the studies reviewed, the

variation in population size generally exceeds the variation in shop numbers per

administrative boundary, and therefore per capita measures of shop provision may be

overly sensitive to population density. Measuring access per square kilometre is

likely to best represent access in terms of how far or long people need travel to

access shops; however, providing equal access to a larger population might require

more shops per square kilometre. Measuring access per neighbourhood using

administrative boundaries, with or without standardizing per capita, was the

approach most commonly taken (170, 172, 173, 175, 180, 187) and probably

introduced bias into the studies, as outlined below.

Administrative boundaries (such as zip-codes or postcodes, census tracts, electoral

divisions or wards) are not uniform in land area, and greater land areas may contain

39

more shops. In the ARIC study, the least wealthy census tracts were much smaller

on average than the most wealthy census tracts (8 km2 vs 20km2) (170). The size

differences could occur possibly due to the clustering of low-income areas around

the population-dense inner city where boundaries may be smaller since they are

designed to contain similarly sized populations rather than similarly sized

geographical areas (190). Shop access was still associated with socioeconomic

position in two studies that did adjust for size (172, 187), however confounding by

boundary size may have overestimated the lack of large shops in low-income areas

for many studies (170, 173, 175, 180).

A further limitation of the studies is that observations have been treated as

independent, while clustering may have been present, as areas and shops that are near

to each other may have been more similar than shops or areas more distant. The

likely effect of such clustering is to underestimate standard errors and make

confidence intervals artificially small. For the studies that did find significant

differences between low and high socioeconomic areas, there is a greater possibility

that the findings arose by chance than p-values would indicate. Studies that

conducted analysis as per a simple random sample, having clustered data that had

been pooled from multiple sites (170, 172, 173, 188) may have incurred additional

clustering and further sources of bias. Since site-specific findings were not presented

and the distribution of low socioeconomic areas across the multiple sites was not

held constant, the socioeconomic differences in shop access may have been

reflecting differences across the study sites. For example, in the Multiethnic Study

of Atherosclerosis (172), the low income tracts may have had fewer shops because

they were most often located in Manhattan and the Bronx, New York and least often

in Forsyth County, North Carolina. The differential shop access, attributed to SEP,

may reflect other differences between Manhattan and the Bronx and Forsyth County

relevant to shop access, such as degree of urbanization.

In summary, within some contexts, but not others, low socioeconomic areas tend to

have fewer large and chain-operated stores, and more small independent stores

compared with higher socioeconomic areas. There is a clear need for more studies to

be conducted, particularly outside of the United States. Biases present in the studies

available to date indicate future studies need to employ improved methods that either

40

avoid defining neighbourhoods by administrative areas, or compensate for the biases

these boundaries can introduce. The method of assessing shop access has generally

been driven by convenience rather than theory, and different approaches have

yielded different results, so future studies should include a variety of access

measures, both subjective and objective. Furthermore, future studies need greater

consideration of context and the mechanisms by which socioeconomic differences in

shop access might be present or absent from different contexts, particularly with

regard to the spatial patterning of socioeconomic characteristics.

2.3.2. In- store availability of fruits and vegetables

In addition to examining whether low socioeconomic areas lack access to shops,

studies have also examined whether the stores located in low socioeconomic areas

stock fewer items, compared with stores in higher socioeconomic areas, using a

variety of approaches. This review focuses mostly on the studies of fruits and

vegetables, but alludes to studies of other food items where necessary, for example,

to illustrate a methodological point.

United States

Studies from the United States have reported socioeconomic differences in the

availability of fruits and vegetables (171, 191), but not within random or

representative samples of areas. In two contrasting areas of New York, shops in a

higher socioeconomic, predominantly white area (the Upper East Side) were

significantly more likely to sell fresh fruit (OR:1.2 , 95% CI: 1.1, 1.4) and green

vegetables (OR: 1.3 , 95% CI: 1.1, 1.5), compared with stores in a lower

socioeconomic, non-white area (East Harlem) (171). Similarly, across the eight “Hi-

Five, low Fat” intervention sites in St Louis, US, lower levels of area-level income

were reportedly associated with less selection of fruits and vegetables (figures not

presented) (191).

41

United Kingdom

In the United Kingdom, an early case study found lesser availability of fruits and

vegetables in low- compared with high- socioeconomic areas (192), however

subsequent studies with stronger study designs did not (181, 193). In a case-study in

Glasgow, shops in the low-socioeconomic area had lesser mean availability of 10

fruit and vegetable items (7.4) compared with shops in a high-socioeconomic area

(8.5) (192). A later study that used a probabilistic sample of areas in Glasgow, (193)

reported finding reported no significant differences in the availability of 13 out of 15

fruit and vegetable items assessed in the study between disadvantaged compared with

advantaged postcode sectors. Fewer shops in the disadvantaged compared with

affluent areas stocked new potatoes (43%vs 62%) and tinned tomatoes (56% vs

74%), and the size of any availability gap overall is uncertain as figures were

presented only for significant findings. In the Newcastle study (181), there was no

difference in the median availability of 14 fruits and vegetables for higher and lower

socioeconomic respondents within the shops they reported using for food shopping

(14 vs 14). For more than 90% of respondents, this was a multiple1 or discount

supermarket.

Overall quality and extent of the evidence for socioeconomic differences in-store availability of fruits and vegetables

One rationale for expecting lesser in-store availability in low socioeconomic areas is

that the lesser demand for fruits and vegetables (as evidenced by their comparatively

low consumption) could affect the degree to which they are stocked in local shops,

particularly for perishable items (195). The evidence from the United States tends to

show less fruit and vegetable availability in low- compared with high socioeconomic

areas (171, 191) while the UK research has produced more mixed findings (181, 192,

193). All the studies with positive findings in both the U.S and UK have been case 1 Mulitples are large, chain-operated supermarkets. The UK Fair Trading Act defines ‘multiple’ supermarkets as “supermarkets with 600 sq metres or more of grocery sales area, where the space devoted to the retail sale of food and non-alcoholic drinks exceeds 300 sq metres and which are controlled by a person who controls ten or more such stores.” 194. U.K. Competition Commission. Supermarkets: a report on the supply of groceries from multiple stores in the United Kingdom: Department of Trade; 2000.

42

studies (171, 191, 192), so a fair comparison across contexts is difficult. However, a

probabalistic U.S study did find less availability of a mixed array of grocery and

fresh produce items in low socioeconomic areas (189), perhaps indicating there is a

true difference between the U.K and U.S in the socioeconomic patterning of fruit and

vegetable availability within shops. As with studies of shop access, this may, or may

not, owe to differences between the U.S and U.K in the nature and experience of

disadvantage, and the associated emphasis on economic or combined social and

economic disadvantage in the U.S and U.K research, respectively.

Socioeconomic differences in average in-store availability may arise from the

predominance of smaller shops in lower socioeconomic areas within some contexts,

as outlined in the previous section. Smaller stores, independent and inner-city stores

have less availability of fresh fruits and vegetables compared with larger, chain-

operated stores and stores in the inner city (181, 189). A similar problem could have

arisen in the Newcastle study (181), which examined the shop mostly patronised by

respondents, as low socioeconomic households were comparatively more likely to

shop at discount supermarkets, which had lower availability of food items than

multiple stores. However, a socioeconomic patterning of availability may exist apart

from shop type, as a study that did control for the chain-ownership and inner-city

location of shops found the availability of grocery and fresh produce items was lower

by 21% in poor areas (189).

In-store availability has also been measured more subjectively. One study examined

respondents’ perceptions of the availability of fruits and vegetables in stores (with a

measure that also included a small focus on the ease of accessing a supermarket and

the range of shops available locally) (184). The authors reported there were no

differences in the perceived availability of fruits and vegetables among occupational

groups within a low-income population (figures not presented). It would be useful

for future studies to determine whether socioeconomic differences in perceptions of

availability exist more widely.

43

2.3.3. In-store prices fruits and vegetables This review emphasises results specific to fruits and vegetables, but also considers

the broader food price literature. The literature from the U.K and U.S has been

pooled, as only one U.S study has reported fruit- and vegetable-specific results. In

view of the in-store availability results, and the paucity of U.S studies reporting fruit-

and vegetable-specific results, studies have been reported together based on their

method, rather than country.

Case studies

Case studies comparing the price of foods across different socioeconomic areas have

had mixed findings. A basket of nine fruit and vegetable items was similarly priced

in low and high low and high socioeconomic areas of Glasgow (£3.54 vs £3.59), but

were of a lower average quality on a 1-5 scale (2.6 vs 3.3) (192). A study in Ireland

found small variations in the prices of fresh fruit, frozen vegetables, processed

vegetables, fresh green and other fresh vegetables across two low and two high

socioeconomic areas. Most items were more expensive than average in one of the

high socioeconomic areas (z-scores 0.20 to 0.72) and most were cheaper than

average in the other three areas (z scores from -0.01 to -0.25) (196). Since these

abovementioned studies did not use random samples of areas, it is not certain

whether the price differences that were present or absent resulted from selection bias.

Ecological, cross-sectional studies

Other studies that have used probabilistic samples of areas, or shops, have also had

mixed findings. Two of the studies examining cost differences in fruit and

vegetables found no difference across socioeconomic areas. In Glasgow, average

prices were similar for the most deprived areas (i.e. DEPCAT 7) compared with

other areas (i.e. DEPCAT 1-6) for fruit (3% cheaper) and vegetables (5% cheaper)

(p>0.05 for all comparisons) (193). The Newcastle study found greater material

44

deprivation (measured by Townsend Deprivation scores) was associated with lower-

priced fruits and vegetables within enumeration districts (r=-0.42, p=0.002) (181). A

study using data from the U.S. Bureau of Labor Statistics found only small

differences between poor zip-codes (defined as having >20% poverty rates)

compared with other zip-codes in the mean price of lettuce (81c vs 76 c per pound

p<0.05) and oranges (68c vs 83c per pound, p>0.05). The results for these items

were atypical of other items in the study, as poor zip-codes had significantly lower

food prices overall (by approximately 6%) (197).

Cross-sectional studies of subjectively-measured affordability

In addition to the studies that have examined price differentials across socioeconomic

areas, researchers have also examined the relationship between socioeconomic

position and affordability in a broader sense. The concept of affordability includes

both price and the ability of the purchaser to pay for foods (159). For low-income

groups, the same prices may not be affordable, due to their lower incomes (156).

Low socioeconomic households have often reported spending less on food in

absolute terms, but more as a proportion of their income, compared with higher

socioeconomic households in the Newcastle study (181), and in expenditure studies

in Australia (156) and the United States (122). Socioeconomic differences in

affordability may extend beyond income. Among low-income public housing

residents in the U.K (184), there were significant differences in perceived

affordability of fruits and vegetables among occupational groups. In descending

order, fruits and vegetables were perceived as most available by full-time employees

(mean (SE): 3.6 (1.6)), retirees (3.4 (1.7)), part-time employees (3.3 (1.7), people

looking after families (3.1 (1.5)), people on sick leave (3.0 (1.5)) and jobseekers.

Low socioeconomic groups sometimes report placing a greater emphasis on price in

purchasing foods compared with higher socioeconomic groups (198, 199). Based on

data from the Diet and Health Knowledge Survey (1994-1996) and Continuing

Survey of Food Intake by Individuals routinely conducted by the U.S. Department of

Agriculture (USDA) (198) believing price to be very important in food purchases

45

was associated with socioeconomic position. More women with a high school

education or less reported believing price to be ‘very important’ (56.6%, 95%CI:

52.6, 60.6) compared with women with some college (39.2%, 95% CI: 34.8, 43.6) or

5 years or more years of college education (26.4, 95% CI: 20.4, 32.4). Similarly,

women from low-income households more commonly reported believing price to be

very important (69.0%, 95 CI: 64.0-74.0) than those from middle-income (50.9%,

95% CI: 46.4, 55.4) or high-income households (31.7%, 95% CI: 27.5, 35.9). Trends

were less consistent by employment, however women who were not employed

(51.5%, 95 CI: 48.9, 54.1) were also more likely to report food price to be ‘very

important’ compared with women who were employed full-time (31.3%, 95% CI:

27.7, 34.9). Food price was more often reported as ‘not very important’ by women

in professional /technical jobs (40.8%, 95% CI: 36.2, 45.2) than by service workers

(13.0%, 95% CI: 9.6, 16.4). Another study of American men and women (199)

reported finding no association between the importance people attached to price in

purchasing food and their income level. The extent to which their findings are

genuinely null is questionable as figures were not reported, other than p=0.051).

Extent and quality of the evidence connecting prices and socioeconomic position

As with the other literature reviewed, findings have been mixed. The price of fruits

and vegetables have been similar for low- and high socioeconomic areas (192, 193,

196), lower in low socioeconomic areas (181) and higher in low-income areas (197).

Mixed results were also found for studies that have not reported fruit and vegetable

specific results, but have often included fruits and vegetables. Based on baskets

designed to meet dietary needs, constructed from population consumption patterns,

overall dietary costs have been similar for low- and high- socioeconomic areas (177,

188, 200, 201), more expensive for low socioeconomic areas (166) or cheaper in

low-income areas (201) (196). Three studies, all from the United States, studied

price variation in small numbers of common food items that vary minimally in

quality and all found small, significant differences in food price for low- and higher

socioeconomic consumers (197, 202, 203). Based on commercial and government

price data, lower income was associated with significantly lower prices in one study

46

(at the zip-code level) (197), while low-income, and reductions in income over time

were associated with significantly higher prices in another study (at the city-level).

Based on prices paid for food (as measured by expenditure), low-income consumers

paid more than their higher income counterparts for the same items (202). Overall,

the variation in findings could have occurred from differences in the items included

in studies, differences in the types of shops in which prices were measured and other

methodological limitations of the studies, or may reflect true differences across

contexts or time-periods.

There may have been a tendency to misclassify prices by measuring price in audits,

usually at one time point only (181, 189, 192, 196, 204), since prices fluctuate,

especially for fresh produce (205, 206). As there is no immediate reason to expect

any misclassification would be differential by socioeconomic disadvantage, results

may have been biased towards the null. Interestingly, studies that used averages

from routinely collected price data (166, 197, 203) did find significant associations

between food price and socioeconomic disadvantage. Perhaps this owes to

comparatively better classification of prices, or alternatively to biases in routinely

collected data, which are less under the control of the researchers.

A number of other methodological issues could have contributed to the mixed

findings in these price studies. As with the studies of shop location, the clustered

nature of the data and the pooling of multiple sites could have affected results, and

the inner-city location and population density may have been important factors that

were largely overlooked. Most studies did not examine other factors that might

contribute to price differentials observed, either as additional causes, or confounders.

Hayes et al. (197) found a number of factors that influenced the associations between

socioeconomic characteristics and food prices, including neighbourhood

characteristics (population density and the presence of unoccupied dwellings), the

market structure (nearby shops per square kilometre and per capita, store size, store

sales and market share), discretionary costs (use of coupons, demonstrations,

circulars and frequent shopper programs), “search” factors (rates of car ownership,

shops outlet density, education) and shop quality (services offered, e.g., deli).

47

Average prices within areas are affected by proportions of different shops from

which price measures are taken (207). Most studies comparing fruit and vegetable

prices across areas have not accounted for shop type or location relative to the inner

city, which may have contributed to the presence and absence of price differences

that were observed. In the Irish case-study, the absence of a consistent

socioeconomic difference in fruit and vegetable prices may not hold true for all types

of shops, as overall food prices were cheaper in the low- compared with high

socioeconomic areas within symbol stores, but not within multiples (196). Possibly,

the greater shop access in low compared with high socioeconomic areas of Glasgow

(180) could contribute to the lack of large, significant price differences there (192,

193). In the Newcastle study, which found socioeconomic disadvantage was

associated with significantly cheaper fruit and vegetable prices, shop-type was

partially standardized, as prices were only compared for shops selling all items

(mostly supermarkets) (181). Food prices are higher in smaller, independent and

inner-city stores compared with larger, chain stores and stores in more suburban

areas (167, 189, 203). The tendencies within the U.S. discussed earlier for low-

income areas to be located in the inner city, and have a disproportion of independent,

small stores therefore may have contributed to the higher food prices in low-

compared with high socioeconomic areas that have been observed in U.S. studies

exclusively. Adjusting, or stratifying, for location and shop-type has attenuated

socioeconomic differences in food prices dramatically, often to the point of no

difference, where this has been performed (166, 189, 201). For example, crude price

differences between poor and non-poor zip-codes were estimated as US$5.15

(approximately 5%), compared with only 1.5% after adjusting for the store’s type

and location (189). Future studies need to present stratified results or otherwise

account for shop type and location to understand socioeconomic differences in price

when they are observed and to avoid missing socioeconomic price differentials that

may occur in selected types of shops.

Another key issue with price studies is quality. Quality is one driver of price among

many (208), and lower prices may not be much of a behavioural incentive if they are

coupled with lower quality. An early case study (192) found fruits and vegetables in

Glasgow were of a lower average quality in low socioeconomic areas (but were

equally priced), while the Newcastle Study (181) found no relationship between fruit

48

and vegetable quality and socioeconomic position (despite finding price differences).

Thus, in low-socioeconomic areas, fruits and vegetables appeared better value for

money in the Newcastle study (181) and worse value for money in the Glasgow

study (192). Too few studies have been conducted to know whether quality varies

across socioeconomic areas. Quality assessment of fruits and vegetables may have

been biased in these studies as raters were not blind to stores’ locations (or

socioeconomic characteristics) when assessing quality. A well-controlled study

found the fruits sourced from shops in the low-income areas were of significantly

lower quality compared with the fruit sourced at the same time from higher income

areas in a mid-sized metropolitan city in the United States (209). The study was

unbiased in its estimation of quality, in using a blinded, standardized procedure for

rating fruit, and a random procedure for sampling fruit within stores. However, the

authors do not outline their method of sampling areas, nor stores, leaving the results

potentially unrepresentative and potentially biased if the reputation for quality of

stores or areas formed any part of the selection process. Studies using both random

sampling of areas and blinded quality assessment would be needed to examine

whether produce is of a lower quality in low socioeconomic areas, or whether the

value of produce (price at a consistent quality) varies according to socioeconomic

position.

Overall, the body of evidence suggests that prices are often similar in areas

irrespective of their socioeconomic characteristics, but in some contexts may be

comparatively cheaper, or more expensive, in low socioeconomic areas, and there are

no studies from which to draw conclusions within the Australian context. Apart

from the methodological limitations of the studies, the variation in findings could

reflect contextual differences, especially as studies have not used samples

representative of urban settings, rural settings, states or countries, with one exception

(197). Relationships between price and socioeconomic characteristics could vary

across contexts due to the multiple processes that could lead to higher or lower prices

occurring in low socioeconomic areas. Price discrimination could lead to higher

prices in poor areas, as could greater operating costs (for example, for security and

insurance related to higher crime rates) (197). Frankel and Gould (203) found the

higher prices in low-income areas in their study were associated with the reduced

presence of middle-income households, which they argued occurs because marketers

49

compete mostly for the patronage of middle-income consumers, who can most afford

both the time and money to search around for the cheapest store from which to

purchase foods (203). Poor areas could experience lower prices if marketers in these

areas are more likely to reduce their prices to increase sales among their low-income

market, based on their comparatively greater price sensitivity (210, 211). The studies

were observational and usually cross-sectional, so it is possible that the causal

direction is the reverse: people living on low incomes might choose to live in areas

where the cost of living is low, which might partially include food prices. These

various processes could occur more in some settings than others, contributing to the

varied findings.

2.4. Environmental features and dietary behaviours

The implications for socioeconomic inequalities of the findings from previous

studies regarding socioeconomic differences in retail provision, prices and in-store

food availability depend on whether or not these factors influence dietary behaviours.

There is a clear assumption that accessibility and affordability of foods act as

important determinants of their purchase in the rationales for these studies and in

food and nutrition policy (53, 158). The food system is considered a key component

of food security through diverse factors including the location of food retail outlets,

availability in stores, price of food, quality of food and food promotion (212).

However the evidentiary basis is not strong for the assumption that these factors

influence dietary behaviours, within contemporary urban settings in developed

nations. People include issues of accessibility and affordability among the reasons

for their dietary choices, including fruit and vegetable choices e.g. (135, 198, 213).

However, little empirical research examines to what extent accessibility and

affordability contribute to dietary behaviours and outcomes. In particular, few

studies have examined the issues in such a way as to be able to predict the impact on

dietary or health inequalities of the variation in shop access, in-store availably and

prices that have at times been observed. The possibility exists that shoppers devise

strategies to cope with limited availability in such a way that accessibility and

affordability as measured in previous studies may have minimal or no effect on

50

dietary practices. Therefore, the potential usefulness of improving the accessibility

and affordability of fruits and vegetables as a strategy to improve population

nutrition (or reduce inequalities) is unknown, even if accessibility and affordability

vary along socioeconomic lines, as has been found in some of the research reviewed

in the previous section.

A tabular summary of the findings of studies examining the relationship between

accessibility and affordability and dietary behaviours is provided in Table 2.2, and

relevant details about the studies are provided in Appendix 1, Tables A2.1-A3.1.

Here, fruit and vegetable-specific results are emphasised, however other dietary

studies have been considered, particularly where the literature is sparse. Contextual

differences are considered, but studies are grouped by their measurement of access

rather than by country, unlike previous sections. The presence or absence of

socioeconomic differences in shop access, price and availability appeared to be

driven by the spatial patterning of socioeconomic disadvantage, which varied across

contexts. However, context may be less important than the way in which access the

relationship between accessibility and affordability has been quantified in

understanding whether access to shops, in-store availability and prices are associated

with dietary behaviours.

51

Table 2.3: Findings of studies assessing the relationship between features of the food retail environment and dietary measures Study Details Access to shops

In-store availability Price Population Outcome

Type of shop used

Shop density/ presence Distance Travel time

Car use / ownership Other access

Fisher & Strogatz, 1999 adults (US)

% households with low-fat milk

(% shelf space)

Shankar & Klassen, 2001

African American (urban US)

Fast-food consumption

x (s'mkt, self-report)

x (own)

Edmonds et al., 2001

African American 11-14 y.o boy scouts, (US)

F intake census tract, grocery stores

(shelf space & Y/N)

V intake x F juice intake x

Morland et al., 2002

white adults (US) F & V intake census tract x (s'mkt) x (groc)

black adults (US) F & V intake (s'mkt) x (groc)

Laraira et al., 2004

Lower-middle income pregant women (US)

Diet Quality Index for pregancy

s'mkt, groc & conv x (census tract) x (0.5 km buffer)

(s'mkt) (conv)

x (groc)

Rose & Richards., 2004

Food Stamp Participants (US)

F household use per adult equivalent

(shop used)

ns (shop used)

ns (own) (use supermarket & travel time)

V household use per adult equivalent

ns (shop used)

ns (shop used)

ns (own) ns (use supermarket & travel time)

Zenk et al., 2005

African American (urban US)

F & V intake Shop used (perceived)

x (perceived affordability)

Jetter & Cassady, 2006

Adults Sacramento & Los Angeles, (US)

Healthy Eating Index

x (travel 10 min) R (travel & shopping time)

x (use) R (perceived) ns (perceived)

52

Bodor et al., 2007

adult food shoppers (US)

F intake x (s'mkt 1km)

ns (small, 100m)

within 100m, small stores a ns (shelf-space)

V intake x (s'mkt 1km) (small, 100m)

Wrigley et al., 2002

residents of 1 low SES area (UK)

F & V intake nst (pre vs post store construction)

F & juice intake

nst

Dibsdall et al., 2003 low income (UK) F & V intake (perceived) (perceived) White et al., 2004

adults (UK) F & V intake ('shop') x (use)

Pearson et al., 2005

adult food shoppers (UK)

F intake x (shop used) x (report difficulties)

V intake x (shop used) x Ball et al., 2006

adult women (AUS)

F intake per captia, suburbs x (s'mkt) x (F&V)

V intake x (s'mkt) (F&V) (crude

only)

Giskes et al., 2007

adult food shoppers (AUS)

healthier vs regular grocery items

shop used x (actual)

(perceived)

shop used x (actual $ difference)

(perceived $ difference)

F = fruit; V= vegetable; s’mkt = supermarket; groc= grocery store; conv= convenience store = association in the expected direction: more shop access, better availability and lower price is associated with more healthy diet;

R = association in the opposite to expected direction x = no association ns = association is present qualitatively but not statistically significant at p<0.05; nst = association is not statistically tested

53

2.4.1. Retail infrastructure

Presence of local shops Two studies examined the relationship between the availability of local shops and

fruit and vegetable intake (214, 215). In the ARIC study (214), living in a census

tract that contained a supermarket was associated with meeting the dietary guideline

for fruit and vegetable intake among African Americans (RR: 1.54, 95% CI: 1.11,

2.12), independently of the presence of other types of food stores and food service

places, income, education and age. For white Americans, this relationship was less

substantial and non-significant (RR: 1.08, 95% CI: 0.89, 1.30) (214). Small, non-

significant relationships between the presence of a grocery store and fruit and

vegetable intake were observed for both African- (RR: 1.07, 95% CI: 0.83, 1.38) and

White Americans (0.93, 95% CI: 0.78, 1.10). A study in Melbourne, Australia (215),

found the number of fruit and vegetable shops available per capita (at the suburb

level) significantly correlated with women’s intake of vegetables (Pearson’s R=0.06

p<0.05) but not fruits (R=0.04, p>0.05). The availability of supermarkets showed

similar, but smaller relationships with vegetable (R=0.04, p>0.05) and fruit intake

(R=-0.03, p>0.05). After adjustment for age, marital status and education, the small

correlation between greengrocer availability and vegetable intake was attenuated by

approximately 50%, and to non-significance.

A quasi-experimental study (216) found intake of fruits and vegetables increased

after the construction of a large superstore in a low socioeconomic area that

previously had little food retail provision. The mean intakes increased slightly

overall, from 2.88 to 2.93 portions per day. Among participants with low baseline

intakes, the increases were much more pronounced, possibly indicating a ceiling

effect. For example, among participants who consumed less than one portion of

fruits and vegetables daily before the new store’s development (ie < 7 portions per

week), the mean intake of fruit and vegetables increased from just over 4 to just

under 10 portions per week. The results tend to support a relationship between food

store accessibility and fruit and vegetable intake, however this study provides only

low-quality evidence as there was no comparison group, so factors other than the

54

store development could have contributed to the dietary changes. Also, there is no

quantification of the probability the changes occurred by chance.

Further results from this study added some strength to the former findings by

examining change separately for respondents who did and did not switch to the

newly provided store to somewhat control for non-intervention-related change (217).

For low and intermediate consumers of fruit, dietary change was similar regardless of

whether the new store was used for shopping or not. However, among respondents

who consumed three or more portions of fruits and vegetables per day, there was a

significant decline in fruit and vegetable consumption on the order of half a serve per

day among those who did not switch to the new store while no such decline occurred

among those who switched to the new store.

A subsequent “natural experiment” in Glasgow (218) extended on the approach used

in Seacroft by incorporating a nearby control site to account for concurrent dietary

change. As in Seacroft, positive dietary change followed the construction of the

superstore, however after accounting for confounding differences between the study

sites and baseline consumption levels, there was no additional improvement in fruit,

vegetable or combined fruit and vegetable intake in the intervention area compared

with the control group. However, this study provides only limited evidence against a

positive impact of the superstore in view of the very low response rate (15%), high

attrition rate (32%), and the differential attrition between intervention (29%) and

control groups (35%).

Distance to shops

Distance to the nearest shop has been associated with fruit and vegetable intakes, but

not for all types of shop (219, 220). In the Seacroft intervention, living within 500

metres of the newly opened store was positively associated with increases in fruit and

vegetable intake, independent of baseline consumption and changes in the type of

store utilised for shopping (217). Among adult residents (n=426) of four wards in

South Yorkshire (219), greater distance to the nearest supermarket was associated

with small, non-significant increases in fruit intake (Beta coefficient (β)=0.05, 95%

55

CI: -0.02, 0.12) and vegetable intake (β =0.01 -0.05, 0.07), adjusted for age, gender

and area SES. Bodor et al. (220) studied the fruit and vegetable intakes of 102 adult

residents of four contiguous census tracts in New Orleans selected for their high

racial and socioeconomic variability. Participants who lived close (<100m),

compared with further (>100m) from the nearest grocery store had significantly

higher intakes of vegetables (mean (SD): 3.3 (2.3) vs 2.4 (1.6)) and also had slightly

higher intakes of fruit (2.4 (1.8) vs 1.8 (1.4), p=0.08). However, fruit and vegetable

intakes were similar among participants living less than one kilometre from a

supermarket (2.0 (1.4) and 2.5 (1.5), respectively) or more than one kilometre from a

supermarket (mean (SD): 2.1 (1.9) and 2.9 (2.1), respectively). The difference in

findings could indicate the greater relevance of shops within a short distance for

dietary behaviours, or could indicate a greater relevance of grocery stores compared

with supermarkets.

Associations between fruit and vegetable intake and distance to shops have

sometimes been observed in studies that have not examined distances to a specific

type of shop (181, 221). In the Newcastle study (181), distance to the nearest shop

(of any type) was reported as not being significantly associated with participants’

fruit and vegetable intakes (figures not reported), adjusted for age, gender, physical

activity, ethnicity, weekly food expenditure, being a ‘safe’ alcohol drinker and BMI.

Among Food Stamp participants across the US (221), living further from the store

mostly used for shopping was associated with lower fruit and vegetable usage2,

independently of urbanisation, household income, size, race/ ethnicity, schooling,

single parent status, and employment. Compared with participants living less than

one mile from shops, those living more than five miles from shops had significantly

lower fruit usage (mean difference= -62, 95% CI: -117, -7) and lower vegetable

usage (-36, 95% CI: -108, 35) and participants living 1-5 miles from shops also had

lower usage of fruits (-15, 95% CI:-64, 34) and vegetables (-20, 95%CI:-101, 61),

although to a lesser extent. Respondents who reported travelling less than thirty

minutes to and from shops had higher usage of fruits (23 (-41, 88)) and vegetables

(30 (-22, 81) compared with those who travelled 30 minutes or more, although the

differences were not statistically significant.

2 Food usage was calculated as foods bought or grown for the supply of foods to be prepared and/or consumed at home in grams per adult male equivalent per day.

56

Shop patronised

Some studies have reported associations between the type of stores used for food

shopping and fruit and vegetable intake. In the study of Food Stamp recipients

(221), participants who shopped at supermarkets had more daily usage of fruits (70 g

per adult equivalent, p>0.05), and vegetables (35g per adult equivalent, p>0.05) than

participants who shopped elsewhere. The differences were small, but could be

important in view of the overall low fruit and vegetable intakes in this population

group. Greater fruit and vegetable intake was associated with shopping at a

supermarket rather than smaller independent stores types among African-American

women living in Detroit (174). This study used a structural equation modelling

approach, and thus could assess the inter-relationships between some of the

accessibility measures. The models indicated that higher income was indirectly

associated with fruit and vegetable intake (via shopping at supermarkets), which

suggests that accessibility might play some role in socioeconomic differences in fruit

and vegetable intakes. White et al. (181) also found men and women living in

Newcastle upon Tyne who shopped at multiple rather than discount supermarkets or

department stores consumed more fruits and vegetables. However, this study found

no relationship between the type of store used and fruit and vegetable intake after

adjustment for an array of demographic, socioeconomic, health-related and other

environmental factors, including measures of socioeconomic position. Given the

findings from the Detroit study (174), the results are possibly ‘over-adjusted’ if the

differences in fruit and vegetable intake of socioeconomic and other demographic

groups operate through the choice of shop utilised.

These studies might be capturing an effect of greater in-store availability or lower

prices on fruit and vegetable intake, as supermarkets are often shown to have lower

prices and greater availability of food items than other shop types in the price studies

previously examined. Alternatively, findings may also be reflecting a process in

which people select where to shop on the basis of its provision of their food choices.

57

Car ownership

Owning a car, or using a car to shop, mostly has not been associated with fruit and

vegetable intakes. In the study of Food Stamp recipients, participants who reported

not owning a car consumed similar amounts of fruits (mean difference=-13, 95% CI:

-63, 38) and vegetables compared with car owners (-30, 95% CI: -78, 19) (221).

Similarly, residents in South Yorkshire who reported potential difficulties shopping,

based on both car ownership and mobility, had similar intakes of fruits (mean

difference=0.02, 95% CI: -0.36, 0.39) and vegetables (0.01, 95% CI: -0.05, 0.07)

compared with those who reported no difficulties (219). In the New Orleans study

(220), respondents from households that owned a car consumed similar amounts of

fruits compared with respondents from car-less households (mean (SD): 2.1 (2.2) vs

2.0 (2.2), p>0.1), however they consumed slightly more vegetables (mean (SD):

3.0(2.1) vs 2.3(1.5), p=0.09). Similarly, in the Newcastle study, there were no

differences in fruit and vegetable intake associated with whether participants

shopped for food by bike or on foot, by taxi, by public transport, or by car (181).

Extent and quality of evidence connecting shop access to fruit and vegetable intake

Studies from the US and UK have reported associations between shop access and

fruit and vegetable intake. Associations observed have generally been in the

direction of access to shops being positively associated with fruit and vegetable

intake (181, 214, 216, 220, 221). Associations have mostly been small, and not all

have been statistically significant. This may or may not have a follow-on effect to

health, as access to shops has sometimes been associated with diet-related outcomes,

such as body mass index (BMI) (172). Access to more shops, living closer to shops,

and the type of shop patronised have all been associated with higher intake of fruits,

vegetables, or both. Within studies, the relationship between shop access and fruit

and vegetable intake differs depending on the type of shop examined. Unfortunately,

the overall evidence is not consistent with respect to whether which particular type of

58

shop, if any, is more strongly associated with fruit and vegetable intake.

Associations between fruit and vegetable intake and shop access were stronger for

supermarkets than other types of stores in the ARIC study (214), and shopping at a

supermarket, rather than other types of stores was often associated with higher fruit

and vegetable intake (174), however relationships between distance and fruit and

vegetable intake were found for other types of shops, but not supermarkets. The

difference in supermarket-related findings for distance compared with other access

measures might be due to the limited range of distances to supermarkets under

observation, as the studies had very little geographic coverage (n=4 wards , n=4

census tracts) (219, 220).

The relationship between shop access and fruit and vegetable intake appears to vary

among population sub-groups. Studies with positive findings have researched

vulnerable populations: African Americans but not White Americans (214), residents

in a low socioeconomic area (216), Food Stamp recipients (221) and African

American women (174). Studies with null findings have tended to be among more

socioeconomically varied populations, such as all women (215) or all adults (181,

219, 220). Authors have argued that socioeconomic disadvantage may constrain

people to their local food environments, as low-income families are most likely to

lack access to private transport (160) and are most likely to use alternate transport

methods to shop for groceries (161). Lower rates of car-ownership among African

Americans might explain the positive findings for African Americans, but not white

Americans in the ARIC study (214). However, differences between population

groups in the relationship between shop access and fruit and vegetable intake appear

more complex, since using a car to shop or owning a car mostly has not been

associated with the intakes of fruits and vegetables (181, 220, 221) or other dietary

outcomes (135, 181). Future studies that do not focus on population sub-groups may

need to examine interactions, or stratify results, to avoid missing relationship

between access and diet among vulnerable population groups.

A key issue in the studies connecting objective measures of accessibility to dietary

behaviours is the scale at which access has been measured. Studies have used a

variety of units (census tracts, zip-codes, post-codes) and distance cut-offs. The

scale on which access to the food environment might affect behaviour is not known

59

and the studies in this review have varied along too many dimensions to discern

whether any spatial unit tends to show more association than others. Beyond the size

at which effects may operate, arbitrary boundaries most likely do not delineate

neighbourhoods as perceived by residents, and do not consider natural boundaries

and obstacles, such as rivers, which may affect where residents shop and which area

might reasonably act as their neighbourhood (56, 163). Administrative boundaries

may be particularly poor at capturing the actual area where people shop for those

who live near area boundaries. Zip-codes were a common proxy for neighbourhood,

and Fisher and Strogatz reported that only half of their respondents actually shopped

for milk within their zip-code of residence (222). The non-uniformity in size of

administrative boundaries is unlikely to be differential by dietary patterns. The

studies that have examined retailing within a particular distance from residents’

homes manage to achieve uniformity but still lack meaningful neighbourhood

definition, and therefore may still be subject to non-differential misclassification

bias. Access to shops according to their presence or absence in census tracts, or zip-

codes, or within particular distances, might have effectively been misclassified,

which could have contributed to some of the null findings.

Confounding is less likely to be problematic in these studies connecting the shop

access to dietary behaviours, compared with the studies investigating shop access

across areas of varying socioeconomic characteristics. Results have usually been

adjusted for a limited range of demographic factors that influence fruit and vegetable

intake (gender, age, and socioeconomic position being the most common).

Unmeasured demographics are not likely to have strongly affected findings, as these

are not known to be associated with shop access. The area-level features associated

with shop access that were problematic in their co-occurrence with area-level

socioeconomic characteristics (such as central city location and population density)

have less obvious association with diet than with area-level socioeconomic

disadvantage. However, confounding could still have arisen from unmeasured area-

characteristics or individual demographics, especially since the literature in this area

is relatively new.

Overall the findings of these studies are suggestive of a relationship between the

local food environment and the dietary behaviours of residents, which might operate

60

more strongly among some population sub-groups than others. Taken together with

the findings of studies examining the relationship between shop accessibility and

socioeconomic position, an unequally distributed food supply might contribute to the

differences in fruit and vegetable intake among socioeconomic groups in the United

States. Like the previous studies indicating less shop access in socioeconomic areas,

most of the studies finding relationships between access to shops and diet have been

conducted in the United States, and these may, or may not, generalise to the

Australian experience. It is possible that differences in the spatial patterning of

shops, population density and associated travel patterns, and differences in retailing

practices may affect the type of ‘exposures’ to shop access experienced in different

contexts, and the way in which people respond to access to shops. The only

Australian study to date (215) has focused on women.

2.4.2. In- store availability

Diet and objective measures of food availability

The relative availability of fruits and vegetables in grocery stores was associated

with the intakes of fruits and vegetables in one study of adults (220), but not in a

study of children (223). Some additional support, albeit limited, that in-store

availability may be associated with fruit and vegetable intake comes from studies

that have found associations between the in-store availability of ‘healthier choice’

items, such as low-fat milk or non-white bread, and their purchase or intake. Within

zip-codes or larger community areas, the relative in-store availability of ‘healthier

choice’ items, such as low-fat milk has shown correlations with the average intake or

proportion of residents using these items (222, 224, 225).

In the New Orleans study (220), the availability of fruits and vegetables in nearby

grocery stores was measured as the total amount of shelf space dedicated to fresh,

frozen, canned, bottled and dried fruits and vegetables in all grocery stores within

100 metres of respondents’ homes: 0 m , >0 to <3m, or >3m. Adults who had

greatest in-store fruit and vegetable availability had significantly higher intakes of

vegetables (mean (SD): 4.5 (2.4)) compared with those with less in-store availability

61

(3.3 (2.4)) or no in-store availability within 100m of their homes (2.4 (1.6)), adjusted

for age, gender, income (poverty income ratio which is specific to household size),

and race/ ethnicity. Smaller, non-significant, differences in fruit intake

(approximately half a serve per day) were also observed.

Edmonds (223) studied the intake of fruits, vegetables and juices as measured by 24-

hour recall of boys aged 11-14 years recruited from 12 predominantly African-

American (>80%) boy scout troops in Texas, USA. Boys’ fruit, vegetable and juice

intakes were not associated with the relative availability of vegetables, fruits and fruit

juice in grocery stores in the census tracts where they lived. Availability was based

on both shelf space and whether or not items were available for a checklist of fresh,

frozen, bottled, canned and dried vegetable, fruit and juice items. Despite its

comprehensiveness, the study’s measure may have poorly classified neighbourhood

availability in grocery stores, as only one, randomly selected, grocery store was used

to represent each census tract, and the treatment of tracts that lacked grocery stores is

unclear. However, since this type of misclassification is likely to be non-differential

with respect to boys’ dietary intakes, bias towards the null could have contributed to

the lack of association between availability and intake in this study.

Subjective measures of availability and fruit and vegetable intake

Food availability has been examined on the basis of respondents’ perceptions as well

as the objective measures previously outlined. Perceptions of food availability have

been associated with fruit and vegetable intake in two out of three studies on this

topic (174, 184). Associations between the subjective measures of the availability of

other food items and their purchase have also been reported (226).

Welfare recipients living in public housing in East Anglia in the U.K. (184) who

consumed five or more portions of fruits and vegetables daily perceived greater

availability of fruits and vegetables compared with those who consumed two or

fewer portions of fruits and vegetables per day. A population-based telephone

survey of a representative sample of 592 low- to middle-income black, Hispanic, and

62

white (non-Hispanic) American adults aged 18-60 years (227) examined the

relationship between respondents “having difficulty finding fruits and vegetables of

choice” and their fruit and vegetable consumption according to a six-item measure

adapted from the Behavioural Risk Factor Surveillance Survey. The authors did not

present the actual figures but reported finding no significant associations between

respondents’ perceptions of availability and fruit or vegetable consumption for any

ethnic group, having adjusted for demographic characteristics, social roles, food

skills, dietary change for health and food upbringing. Among African-American

women living in East Detroit (174), fruit and vegetable intakes were positively

associated with the degree of selection of quality fresh produce in the stores women

mostly used for shopping (which women nominated as being somewhere on four-

point scale from poor to excellent). Furthermore, perceptions of produce availability

mediated the relationship between fruit and vegetable intake and shopping in

different types of stores, or shopping at suburban compared with city stores.

Extent of the evidence connecting in-store availability to fruit and vegetable consumption

There have been too few studies to determine whether the availability of fruits and

vegetables in local shops influence residents’ purchasing and consumption of fruits

and vegetables. Only one study examined in-store availability and fruit and

vegetable purchase in relation to the dietary behaviour of adults (220). This study

found a relationship, however its findings may not be widely generalisable, being

based on a small, non-randomly sampled setting. A study of young African-

American boys (223) did not support these findings, although differences in both

method and the population group could have contributed to the discrepancy.

Ecological studies have given additional evidence that availability in local shops is

sometimes associated with residents’ dietary practices (222, 224, 225) however these

studies are of a low evidentiary quality and have not focused on fruits and vegetables

specifically. Findings of the studies focusing on perceptions of availability (174,

184, 227) have generally been supportive of a relationship between availability and

fruit and vegetable intake (174, 184), at least among vulnerable sub-populations.

63

From the literature, it appears possible that both access to shops, and the availability

of fruits and vegetables within shops are associated with fruit and vegetable intake.

Unfortunately, most of the current evidence does not distinguish whether the in-store

availability of fruits and vegetables is important above and beyond the availability of

shops, as the measures have not always separated the availability of items from the

availability of shops themselves. In the New Orleans study (220) availability was

based on total shelf space in all grocery stores within 100 metres. Therefore, the

findings could reflect associations between fruit and vegetable intake and having

more shops nearby, or shops nearby that sell a greater range of fruits and vegetables.

In the study of welfare recipients (184), perceptions of fruit and vegetable

availability were examined in a way that emphasised item availability within shops

but also extended to the availability of food shops from which to purchase fruits and

vegetables. The Detroit study does provide some evidence specific to in-store

availability, at least for African American women (174).

The studies were all observational, and all but one ecological study (not focused on

fruits and vegetables) was cross-sectional. The relationships that were observed

might reflect an effect of the local food environment on food choice. Alternatively,

store owners may stock their shelves in response to consumer demand, as argued by

Cheadle et al (224). The latter scenario is quite plausible, as stocking practices are

carefully managed to maximise profits and minimise wastage. Therefore, the relative

availability (e.g. shelf space) of highly perishable items, such as fruits and

vegetables, may well be reduced in areas where residents consume and purchase few

fruits and vegetables. Reverse causality cannot explain the findings of intervention

studies, however interventions that have increased the availability of fruits and

vegetables or other ‘healthy choice’ food items e.g (228, 229) have been conducted

in confined, immediate consumption contexts, such as school or workplace

cafeterias, and have limited applicability to grocery stores or supermarkets and

overall dietary intake or the at-home consumption setting. A supermarket-based

experiment (230) found that doubling display space for fruit and vegetable items led

to a 44% increase in the sales of those items. The findings do not necessarily reflect

dietary change among individuals, as the increased overall sales could have occurred

at the expense of other fruit and vegetable items, or from the venue attracting new

customers. Existing studies cannot adequately demonstrate nor disprove whether

64

community-level availability influences residents’ dietary behaviours, or mediates

socioeconomic inequalities in dietary behaviours, but are sufficient to justify further

study.

2.4.3. In-store prices and affordability

Some researchers dismiss outright the notion of a role of food prices in improving

population nutrition, at least in the United States, on the grounds that foods are cheap

relative to other countries (231). However, this assertion is somewhat simplistic, and

overlooks the possible impact for vulnerable groups, including low-income

households. Even within countries where food is comparatively cheap, qualitative

work has shown prices form part of the food choice process in complex ways (232,

233).

Furst (232) interviewed a diverse sample of shoppers and found their purchasing

decisions were based on a process of value negotiations between a number of factors

they believed to be important (including monetary concerns, quality, convenience,

sensory perceptions, relationships, and health and nutrition). Their monetary

considerations extended beyond the ‘cheapest’ to including concepts such as ‘value

for money’ and ‘deals’. Owen et al (233) examined the role of price in the food

choice processes employed by 24 consumers shopping for fruits and vegetables by

examining statements they made into a dictaphone as they verbalised their decision-

making process. Their results indicated price formed part of the purchasing

decisions for some, but not all, consumers as participants displayed either “price –

indifferent”, “value-conscious” or “price-based” shopping approaches. Participants

monitored prices, compared products based on price and reacted to unexpectedly

high and low prices. Encountering high prices, participants sometimes elected not to

purchase certain items at that place and time, and they weighed other factors against

prices: need for a product, varietal preferences, the overall quantity needed, and the

pleasure derived from the product, being less price sensitive for ‘basic’, ‘needed’

items and those purchased in small quantities. When encountering low prices,

participants sometimes ‘stocked up’.

65

Price elasticities: evidence from economics Most of the quantitative evidence linking food prices with food purchase comes from

economics and marketing literature, which quantifies these relationships in terms of

“price-elasticities” which represent the change in average “consumption” (generally

measured as sales or expenditure) that coincides with change in average food prices

(211, 234-237). These studies have shown sales of fruits and vegetables decreases

by a small amount when prices increase. These correlations between prices and

purchasing are typically ecological and may not capture the true relationship between

prices and individual behaviours nor represent the effect of prices on people’s overall

diet. For example, stores may note an inverse correlation between their prices and

sales of fruit items, however people’s overall purchase of fruits and vegetables may

not alter in response to price, if they alter where and when they shop in response to

price. This store-switching phenomenon is consistent with the observation that there

was less price sensitivity in isolated stores compared with stores nearer to

competitors (Hoch et al., 1995).

Price elasiticities highlight some important considerations for examining the possible

effect of local food prices on socioeconomic differences in fruit and vegetable

purchasing, in that they show the relationship between food prices and purchase

varies along several dimensions. Firstly, the food item itself appears important.

Purchase of staple foods, including many common fruits and vegetables, does not

vary in response to price changes as much as non-staple foods (such as chocolate)

(238), and purchase of some fruit and vegetable items vary more in response to price

than others (230, 234). Changes in prices of one food can affect purchasing of other

foods (as measured by cross-price elasticities). For example, fruit and vegetable

purchases are influenced by prices of meats and cereals (238, 239). When food prices

increase, purchasing of substitute foods increases and purchasing of complementary

foods decreases (234). The income of the purchaser appears to modify the

relationship between price and purchase, as food prices are more elastic in low-

income countries and for low-income households compared with their higher income

counterparts (210, 211, 236-238). The context of the purchase appears important, as

both price elasticities and the difference in price elasticities between high- and low-

66

income consumers vary depending on whether foods are being purchased for a social

(rather than private) context or for luxury (rather than necessity) (211). Finally, price

may impact purchase in several ways. Looking at consumer responses to price in

more detail, Richards (235) found increases in the price of apples could prompt

several responses: purchasing other fruits, purchasing another variety of apples, or

purchasing a smaller quantity of the same variety of apple. Importantly, the tendency

for each response varied for different consumer groups, further supporting the notion

that the price – purchase relationship varies according to personal characteristics.

Using own-price and cross-price elasticities, Kinsey and Bowland (239) attempted to

estimate the dietary impact in the population of changes in prices of meats, dairy

foods, grains, fruits, vegetables, fats and sweets. They modelled the effect of

changes in price of each food group to changes in the purchase of Food Pyramid

Guide Groupings, having used food composition and apparent consumption data to

translate data about individual items to food groups. They estimated reductions of

10% in the prices of fruits could result in 6% increase in fruit intake, and 10%

reductions in vegetable prices would increase vegetable intake by 2% at a population

level, assuming the relationships they modelled were causal and valid. Notably, 10%

reductions in meat prices were associated with increased consumption of meat,

vegetables, fruits, grains and dairy foods. They further quantified the overall dietary

impact of food price changes by examining predicted effects on overall Healthy

Eating Index Scores (HEI), which are a measure of dietary quality based on the

degree to which diets consumed match nutritional recommendations for dietary

variety and consumption of Food Guide Pyramid groups, total fat, saturated fat,

cholesterol, and sodium. Reduction in fruit prices of 10% was not associated with

improvement in overall HEI scores, as increases in fruit consumption were balanced

by concomitant increases in sodium intake. Reductions of meat prices resulted in

lower HEI scores (by increasing consumption of total fat, saturated fat and

cholesterol). Their stratified analyses showed the lack of overall dietary

improvement with reductions in fruit prices and meat prices held true across all

educational groups. Unfortunately, analyses stratified by income, or area, were not

presented. However, the predictions made by such research are only as robust as the

assumptions on which they are based and cannot determine the effect of prices on

individuals or low socioeconomic groups.

67

Evidence from public health Public health research that empirically links price and purchasing comes mostly in

the form of interventions that aim to increase the purchase of ‘healthy foods’ through

reductions in price in an absolute sense or relative to less healthy options. Results

from a randomised controlled trial that examines the effect of price discounts in

supermarkets trial will become available in the near future (240). Existing

interventions in grocery stores and supermarkets have been education-based (241),

with the exception of two studies that examined the effect of price coupons (242) and

price reductions (230) on purchase of fresh produce sales. Only Curhan’s (230)

study included a separate price component that could be evaluated, and found that

price reductions (of 10%) led to no changes in the sales of hard fruit, cooking

vegetables or salad vegetables, and led to a small but significant increase in the sales

of soft fruit, particularly soft fruits usually sold in high volumes and at high prices.

Most interventions have occurred in immediate consumption contexts (such as

cafeterias in workplaces or schools) e.g. (228, 241, 243-248). While these

interventions generally support a relationship between price and purchase, their

findings may not apply to grocery retail settings, and are of low-quality overall since

they have generally tracked venue-level sales rather than individuals and it is not

possible to determine which elements (price, purchasing, or promotion) were

important when they have been simultaneously altered. With intervention studies, it

is possible the reduction in prices may be important in attracting notice, rather than

prices per se, especially since there has been limited long-term follow up, and where

follow-up has occurred, intervention effects did not appear sustained.

Many approaches to framing the relationship between price and purchase have been

employed, including examination of whether consuming a diet high in fruits and

vegetables is comparatively expensive (and therefore cheaper diets are adopted).

This is not a straightforward hypothesis to test, though substantial support for the

idea can be found. Greater weekly expenditure on food was associated with

consuming more fruits and vegetables per week in the Newcastle study (181) though

this does not mean consuming fruits and vegetables is necessarily more expensive.

68

Energy-dense diets (which tend to be low in fruits and vegetables) tend to be

comparatively cheaper (249). Fruits and vegetables are cheaper per kilogram than

many other foods, and are a comparatively cheap source of nutrients, but are more

expensive per kilojoule of energy, meaning that it is more expensive to satiate hunger

with fruits and vegetables than many other foods (250, 251). Researchers have

hypothesised that this may encourage people with low incomes to purchase more

energy-dense foods and fewer fruits and vegetables (249-253). An approach called

“linear programming” showed the introduction of a cost constraint altered dietary

patterns to include more energy and fewer nutrients in a manner consistent with this

hypothesis (249, 254).

Prices in shops and diet

Only recently has there been an emergence of studies that operationalise the

relationship between price and diet, or diet-related health outcomes, to the level of

neighbourhoods and residents, which is the scale on which much of the research and

discussion about social determinants of nutrition and inequalities occurs. This is

important, as researchers have attempted to draw public health and policy

implications for population health and nutrition and socioeconomic inequalities from

the spatial patterning of retail outlets and retail prices, and their implications could be

flawed by a reliance on assumptions about the nature of the relationship between

food prices and food purchasing and consumption. Studies have examined the

geographic dispersion of prices in regard to diet-related outcomes, but not fruit and

vegetable intake or other measures of diet. Geographic variation in the price of fruit

and vegetables (at the zip-code level) has been associated with gain in body mass

index over a three-year period among kindergarten-aged children in the U.S. (255)

and state-level variation in average grocery store food prices has been associated

with rates of obesity in the United States adult population (256). The relationships in

these studies may operate through diet, however body weight is influenced by many

other factors and more direct studies are needed.

One study, conducted in the UK, was located that examined whether adults’ intakes

of fruits and vegetables (219) were associated with price of fruits and vegetables

69

(carrots, onions, cauliflower, and potatoes, lettuce, tomatoes, apples, bananas,

oranges) in the stores where they usually shopped. Higher prices (each additional £ /

pound) had no significant or substantial association with the number of portions of

fruits (Price (£) β= -0.01 (-0.52, 0.50)) or vegetables (β=-0.26 (-0.69, 0.16))

consumed daily, adjusted for sex, age, area SES, distance, difficulties shopping and

ward of residence. The absence of an association between price and purchase of

fruits may have arisen from a mismatch between the fruits consumed and the limited

range of fruit items for which prices were measured. This mismatch may also have

affected the estimated price – vegetable purchase relationship, though to a lesser

extent due to the greater inclusion of vegetable items. Some non-differential

misclassification of dietary intake could also have been present as fruit and vegetable

intakes were calculated by a single item about consumption in the previous 24 hours.

Using discrete choice analysis, a study of households with school-aged children

(n=1355) in Birmingham, Alabama (257), showed that fruit, vegetable and juice

items that were more expensive per serving (based on expenditure data) were

significantly less likely to be have been present in respondents’ homes in the

fortnight prior to the survey. Categorisation of prices showed a threshold effect,

whereby prices of 30c per serve or more were associated with lower odds of fruits

and vegetables being purchased relative to items costing less than 20c per serve

(OR=0.67, 95% CI: 0.63, 0.71), while lesser price differences were not (ie. for 20-

29c per serve OR=0.99, 95% CI: 0.94, 1.03). The relationship between price and

household vegetable availability was stronger for white than African-American

respondents but existed for both groups (OR: 0.72, 95% CI: 0.70, 0.75 and 0.89, 95%

CI: 0.82, 0.96, respectively). Rather than reflecting prices in any particular shop, or

any particular area, prices were measured as the national average prices paid for

foods according to expenditure data. Prices of each item were compared with the

prices of other fruit and vegetable items. Accordingly, this study demonstrates that

cheaper fruits and vegetables are more likely to be part of regular household

consumption, but does not examine whether there is any relationship between

neighbourhood variation in prices and the purchasing or consumption of residents, or

whether price is associated with lower dietary intake, or dietary variety of fruits and

vegetables overall.

70

Affordability and perceptions of price

Studies of fruits and vegetables have included either objective, or subjective

measures of price, but not both. For studies of other foods, different findings have

been obtained for objective and subjective measures of price (226, 258). Perceptions

of the affordability of fruits and vegetables have been associated with intake of fruits

and vegetables within a low-income population in the UK (184) but not African-

American women in Detroit (174) or a population-based study of African-American

men and women in North Carolina (259). In the UK study of public housing

residents (184), respondents who consumed five or more portions of fruits and

vegetables daily perceived fruits and vegetables as more affordable compared with

respondents who consumed fewer than two portions of fruits and vegetables per day

(mean attitude rating (SD): 3.8 (1.8) vs 3.0 (0.6), p<0.05). In the Detroit study of

African-American women (174), there was no substantial or significant relationship

between women’s fruit and vegetable intake and how affordable women described

fresh produce to be (on a four-point scale ranging from “very-” to “not at all-”

affordable), either directly or indirectly (via the type of shop patronised, or suburban

vs city location). Watters et al (259) found no significant or substantial differences

in fruit, vegetable, or combined fruit and vegetable intakes between African

Americans according to whether they reported feeling they “can afford to purchase

healthy foods, such as fruits and vegetables”, after accounting for age, gender,

educational level, BMI and other psychosocial constructs. (The largest difference in

mean intake was 0.2 serves per day). The discrepancy between the studies could

have occurred through differences between the study populations in income, race, or

location, however a greater relevance of affordability in purchasing fruit and

vegetables among low-income groups is consistent with the economic literature

previously discussed.

Food assistance in the United States in the form of food stamps may make foods

more affordable by effectively lowering the price consumers pay for food out of their

general budget. There is no prescription that food stamps must be spent on fruit and

vegetables specifically, however in the study of food access in New Orleans (220),

71

respondents who received food assistance consumed an extra serve of fruit per day

on average (mean (SD): 2.8 (2.1)) compared with respondents who did not receive

assistance (1.9 (1.7), p<0.05), after adjustment for income differences. A smaller,

non-significant tendency was also evident for greater vegetable consumption in this

study (3.2 (2.1) vs 2.8 (2.0)).

Attaching a greater importance to price in choosing foods has been associated with

lesser intake of fruit and vegetables (198), but not consistently (42). A population-

based study of American women (198) showed considering price to be “very

important” in purchasing food (as opposed to somewhat important, not too important

or not at all important) was associated with lower daily intakes of fruits and

vegetables, without adjustment for demographic characteristics. A population-based

study of adults in the United States (42) reported finding no significant relationship

between fruit and vegetable consumption and the importance that people attached to

cost in purchasing food, after adjusting for a variety of sociodemographic

characteristics and lifestyle factors. The authors did not present the size or direction

of any relationship, only the estimate of statistical significance. The apparent

discrepancy between the two studies could have arisen from the focus on women

versus all adults, but more likely resulted from the fact that the study reporting null

associations adjusted for income and other demographic characteristics (42) while

the study with positive findings did not (198). Since this review is considering cost

concerns as a possible causal intermediate of the relationship between income and

fruit and vegetable intake, the unadjusted findings (198) might be of greater

relevance.

Overall extent and quality of the evidence connecting price and fruit and vegetable intakes

Overall, the evidence shows that prices are associated with the purchasing or

consumption of fruits and vegetables. Associations between objectively measured

prices and purchase or other dietary measures have been found, although not at a

neighbourhood level. Price elasticities indicate that overall purchase of food items,

including fruits and vegetables, increases when prices of these items decrease (230)

72

(234, 238, 239). Relationships between price and dietary measures may operate at a

threshold (257) and are likely to be stronger among low-income groups (210, 211,

236-238). Cheaper fruit and vegetable items are more likely to be part of

households’ consumption than more expensive fruit and vegetable items (257).

Subjective measures of affordability (174, 184), and the importance of price in

buying food (198, 199) have been associated with fruit and vegetable intake (184,

198), although not in all studies (174, 199). Price perceptions have also been

associated with other dietary outcomes, (226).

The way in which relationships between price and purchase are conceptualised and

operationalised appears important in understanding the ways in which socioeconomic

differences in diet may be affected by price, or in which price changes might affect

diet. Yet studies have not examined the relationship between variation in

neighbourhood prices and fruit and vegetable consumption. The available evidence

is therefore difficult to extrapolate to predicting the dietary effects of local food

prices, and to determining whether the higher prices, or lower prices, of foods in low

socioeconomic areas discussed previously are likely to contribute to the different

dietary practices of socioeconomic groups. Furthermore, if the associations between

price and purchase are the result of reverse causality, regardless of the ways in which

the relationships between price and dietary measures have been examined, then price

differences across socioeconomic areas may not necessarily contribute to

socioeconomic differences in fruit and vegetable purchasing. In connecting price

and purchase at the neighbourhood level, store owners may adjust their pricing

policies according to demand provided by residents (203). Relationships between

people’s diets and prices in the stores in which people shop could be the result of

people choosing to shop in stores where the items they consume are offered at low

prices. In connecting average prices to the types of vegetables people are most likely

to consume (257), the high demand for these items could create economies of scale

and drive down prices. The positive findings of intervention studies suggest price

may, in fact, influence purchase or consumption, however this evidence has been

limited both in quantity and quality.

There are further general limitations to the evidence, which may or may not have

affected the findings. Studies have generally not considered food quality in

73

examining the relationship between prices and markers of fruit and vegetable

consumption. Studies have generally standardized packaged grocery items for size

and branding, however significant quality variation could still exist across fruit and

vegetable items, which at best have been standardized for their variety. Better

quality might increase the likelihood of purchase, and may be associated with higher

prices, thereby negating some of the possible relationships observed between prices

and purchase of fruits and vegetables. Also, prices have been considered in isolation

from other facets of accessibility and affordability, which are likely to be inter-

related, and could have acted as confounders. Competition forms part of food

retailers’ marketing strategies (260) and therefore the presence of other shops may

affect prices. For example, Walden (261) showed a relationship between food prices

and the presence of other shops, as did Thomadsen (262) in examining fast-food

chain stores. Ideally, future studies need to examine multiple facets of accessibility,

both in order to be comprehensive, and to neatly separate each effect.

2.4.4. Summary: accessibility, affordability and socioeconomic differences in fruit and vegetable intake

Overall, the evidence has shown that low-socioeconomic areas have less access to

shops in some contexts, but not others. A number of processes appear to contribute to

the presence or absence of socioeconomic differences in shop location, including the

spatial patterning of socioeconomic disadvantage across cities (gentrification and

degree of spatial polarisation). There also appear to be socioeconomic differences in

the availability of fruits and vegetables within shops, which may or may not be a

function of the types of shops contained in more and less disadvantaged areas.

Greater shop access has in turn been associated with small increases in fruit and

vegetable intakes, at least among low socioeconomic groups. It appears likely that

any effect of the availability of shops and fruits and vegetables on fruit and vegetable

purchasing might be modified by income.

Similarly, studies have at times noted small differences in price between lower and

higher socioeconomic areas, but not consistently. While there is a lot of general

evidence showing associations between price and purchase or intake of fruits and

74

vegetables, the evidence to date is too limited to discern whether these small

geographic variations in price are associated with residents’ purchasing or intakes of

fruits and vegetables. Again, the existing evidence is suggestive that any effect of

price on purchasing or intake is modified by income. However, effect modification

has been seldom studied outside of the economics literature.

Based on the existing evidence, a detailed conceptual model of the way in which

socioeconomic differences in fruit and vegetable purchasing might be mediated by

accessibility and affordability was developed (Figure 2.1). The evidence to date has

mostly examined relationships between accessibility and affordability and

socioeconomic position or fruit and vegetable intake, but has seldom examined direct

mediation of socioeconomic differences in fruit and vegetable purchasing or intake

by accessibility and affordability. Very few Australian studies have been conducted,

and appear to be necessary in view of the notable differences in findings between the

United States and the United Kingdom. Shop access, fruit and vegetable price and

availability have generally shown small or moderate relationships with

socioeconomic disadvantage and fruit and vegetable intake. Accordingly, it is

unlikely that these factors entirely mediate socioeconomic differences in fruit and

vegetable purchasing. Other individual-level factors are likely to be important.

75

Figure 2.1 : Conceptual diagram of local accessibility and affordability and cooking skills as mediators of socioeconomic differences in fruit and vegetable purchasing, based on socio-ecological framework

NB: Arrows do not denote a unique causal direction.

76

2.5. Cooking skills and fruit and vegetable consumption across socioeconomic groups

Nutrition policies in Australia and elsewhere (53, 54, 158) include a focus on

cooking skills as a means of improving population nutrition generally (158) and fruit

and vegetable intakes specifically (54). The National Nutrition Strategy aims to

“educate and skill our population to be able to choose a healthy diet” (158) p. 14, and

the National Action Plan to increase the consumption of fruits and vegetables

includes an aim to “increase the proportion of the population with knowledge, skills

and confidence to select and prepare convenient, low cost, tasty vegetable and fruit

dishes” (54) p.2. The rationale for improving population cooking skills is a belief

that a decline in cooking skills is one reason for the inadequate consumption of fruits

and vegetables in Australia (158) p. 56. However, the evidentiary basis is very

limited on which to support the idea that cooking skills contribute to the population’s

inadequate fruit and vegetable intake or that promoting cooking skills will improve

population nutrition.

The dearth of empirical literature on cooking skills has been noted by key researchers

in the area. In 2001, Lang (46) claimed there was no literature available on the

influence of cooking skills upon choice of healthy foods and cooking methods and

described the literature available at the time as largely conceptual and qualitative.

At a similar time, Caraher argued that the absence of empirical data hinders the

development of a coherent theory of the role of cooking and its relationship to health

(165). To a large extent these statements are still reflective of the current literature,

although some new research has emerged since this time. Very little of the literature

about cooking skills has examined their association with dietary behaviours or with

socioeconomic position. Mostly the cooking skills literature has focused on either

gender issues e.g. (232, 263) or promoting independence via cooking skills in

rehabilitative, geriatric and mental health contexts e.g.(264, 265).

77

2.5.1. Cooking skills, nutrition and health

Most of the nutrition and health-related cooking skills literature has been based on

concerns about a decline in cooking skills in contemporary western populations that

might exist based on trends in convenience food consumption e.g. (266, 267).

However, there are no long-term population studies documenting this purported

decline, which instead could relate to false beliefs about previous educational

practices and skill levels (46, 268). Regardless of whether the trends towards

convenience foods may signal a loss of cooking skills, concern has arisen that

concurrent trends of rising convenience foods and de-skilling may put control of

nutrition and health out of consumers’ hands and into those of food manufacturers

(266). In the context of the removal of cooking skills from compulsory education in

the United Kingdom, Stitt (164) describes the co-occurrence of a decline in cooking

skills and a rise in convenience foods as a concern for the population,

“Deskilling and ‘McDonaldization’ are parallel developments, both

interacting and feeding off each other. One could not flower (or deflower as

the case might be) without the other. We are invited to ponder on the

possibility that the deskilling experience in food education might generate

another type of product – the ‘McChild’.”

Cooking skills and the confidence to use them are argued to be important for health

in a variety of ways. For example, foods prepared at home can be more nutritious

and cheaper than foods purchased pre-prepared (267). Cooking skills may form a

part of general health outlook and behaviour (165), empower people to prepare

nutritious foods, aid people in making sound purchasing decisions through providing

an understanding of the preparation of ready-prepared foods (165), and to enable

people to participate in the food culture, thereby reducing a sense of food

insufficiency (46). Low-income families have been described as particularly at risk

from a lack of food skills, as they also lack the financial resources to compensate for

their lack of food skills in paying extra money for healthy convenience foods (165,

269).

Despite the impassioned rhetoric warning of the dire consequences of a loss of

cooking skills, few studies have actually quantified the nutritional or health impact of

78

cooking skills. People have reported that a lack of cooking skills influences their

food choice in qualitative studies of low-income African-American women (135),

New York residents (270) and older English men (271). A population based study in

the UK in the early 1990s found 10% of the general population indicated not

knowing how to cook was one of the factors that limited their choice of foods (165).

Some evidence that cooking skills relate to nutritional outcomes can be gleaned from

interventions that have aimed to improve nutrition through using cooking skills,

often among low-income groups in Australia e.g. (153, 272) the United States e.g.

(154, 155, 273, 274) and the UK (275). In absence of long-term follow-up it is not

known whether any dietary benefit is sustained and it is not certain whether the

cooking skills per se, the educational process, the other educational components of

these interventions (including general nutrition education and budgeting) have

produced any dietary improvement observed.

A variety of measures of cooking skills have been associated with dietary outcomes

(196, 271, 276-278), although some studies have reported finding no relationship

(279, 280). Mostly, the direction of results has indicated that cooking skills are

associated with more healthy dietary practices (196, 271, 276, 278) but not always

(277). Differences between the measures used and the populations examined could

explain the discrepancy between findings.

Food preparation practices (used as a measure of cooking skills) have been

associated with dietary measures (276, 278), including fruit and vegetable intake

(276). A recent study of adolescents in the United States (276) found helping

prepare food at home more often was associated with higher fruit intake among boys

and with higher fruit and vegetable intake among girls. Comparing intakes of those

who prepared dinner most often with those who never helped prepare dinner, average

intakes of fruit were higher by approximately 0.7 serves per day while average

intakes of vegetables were higher by approximately 0.5 serves per day for both boys

and girls. A study of young adults showed their food preparation behaviours

(making salad, preparing dinner with chicken or fish and vegetables, and preparing a

dinner for two or more people) were associated with significantly higher intakes of

fruits and vegetables, among other indicators of a healthy diet. People with the

highest food preparation scores more often met the recommended five serves of

79

fruits and vegetables daily (31%) compared with those with the lowest preparation

scores (3%), and were also more likely to meet the recommendations for deep-yellow

or green vegetables (18% vs 2%) (279). However, food preparation practices may

not reflect cooking skills as they are likely influenced by other factors, such as

resources, opportunities and attitudes.

Attitudes and beliefs towards cooking and cooking skills have been associated with

fruit and vegetable intake (277), though not always positively (277) and not among

all population sub-groups (227). In an Australian study (277), attitudes towards

cooking were associated with vegetable intake, but not always in the expected

direction. Total vegetable intake was positively associated with cooking for

convenience (B (SE): 3.73 (0.99)), and was inversely associated with looking for

new ways to cook and seeing cooking as a woman’s task (-3.03 (1.06) and -2.94

(0.95), respectively). The relationship between attitudes to cooking and intake varied

for different types of vegetables. Convenience cooking and planning ahead for meals

had small, inverse associations with intake of green or boiled vegetables (-0.06 (0.02)

and -0.04 (0.02), respectively). Intake of salad vegetables had small associations

with convenience cooking and looking for new ways to cook (0.02 (0.0) and -0.02

(0.00)). Devine et al (227) studied representative samples of low- to middle-income

Hispanic, Black, and White Americans from a north-eastern U.S. city and reported

that food preparation attitudes were independently associated with higher intakes of

both fruits and vegetables, but among Hispanic women only. Figures were not

presented, so the magnitude of the relationships could not be ascertained. In this

study, food preparation attitudes included having enough time to prepare vegetables,

and beliefs about fruit and vegetable spoilage, in addition to thinking fruits and

vegetables are easy to prepare and considering oneself a good cook.

Studies of self-assessed cooking skills have reported associations with dietary

outcomes indicative of a healthy diet (266, 271, 281), but not unanimously (279,

280). Better self-rated cooking skills were associated with higher vegetable intake,

and combined fruit and vegetable intake among older men living alone in the United

Kingdom. Unfortunately, the magnitude of these associations could not be assessed

based on the figures reported (271). A study of university students in the United

Kingdom (280) reported finding no association between cooking skills and fruit and

80

vegetable intake. As these findings were reported only in abstract form, it is

difficult to assess their reliability. In the study of young adults in the U.S (279) the

degree to which respondents believed their cooking skills and resources were

adequate was reportedly not associated with fruit and vegetable intake. Again,

figures were not presented. Both studies with null findings were conducted among

young adults, so it is possible that the population groups account for the differences

in findings for these studies. Differences between the measures of cooking skills

could also have contributed. The evidence regarding fruits and vegetables is limited,

however, self-rated cooking skills have been associated with other dietary outcomes.

Among adult ‘household heads’ in Northern Ireland, self-assessed cooking skills

were associated with lower take-away consumption. (266). A Canadian study

found having lower self-rated cooking skills was associated with a greater chance of

household food insecurity, after accounting for income, among households with pre-

school aged children in inner-city Vancouver (281).

The extent to which western populations, including Australians, possess sufficient

skills to enable a healthy diet is not known, and many of the studies previously

described did not systematically measure the prevalence of cooking skills. Where

figures have been presented it has usually been among population sub-groups. A

population-based study of U.S. adolescents reported that 23% of males and 18% of

females rated their skills and resources for cooking as very inadequate or inadequate

(279). A study of U.S college athletes, presumably a health-conscious population,

reported only 61% were confident in their ability to cook food (282). A survey of

young people in the United Kingdom conducted by the National Food Alliance in the

in the early 1990s showed 50% of young people reported knowing how to boil food,

59% reported knowing how to grill and 75% knew how to microwave foods. Even

basic skills relating to preparing fruits and vegetables were not universal, as only

79% reported knowing how chop or slice them and only 85% reported knowing how

to peel them (164).

81

2.5.2. Cooking skills and socioeconomic position

There is some evidence that low socioeconomic groups lack cooking skills relative to

higher socioeconomic groups, although none from the Australian population. In a

population-based study in Ireland, Furey et al. (266) reported a significant

association between occupation and self-reported cooking skill level, although they

did not elaborate on the size or direction of this relationship. In published conference

preceedings, Strugnell, Furey and Farley (283) reported “low income family units …

exhibited lower levels of cooking skills” in a study of principal food shoppers in

Northern Ireland, although very little detail was provided. One population-based

study of cooking skills (284) formed the basis for much of the literature surrounding

cooking skills, health and socioeconomic position (46, 165, 267, 285). The National

Health and Lifestyles Study in the United Kingdom (284) in the early 1990s assessed

self-reported confidence to use different cooking techniques and to cook particular

foods and found most measures of confidence to cook were higher among

respondents of higher socioeconomic position (as measured by occupational status,

education and income), women and older respondents. Confidence to prepare both

fresh green vegetables and root vegetables was significantly higher among

respondents with more education and higher income, with income showing the

stronger relationship. Higher socioeconomic groups were more likely to report being

confident to use all the cooking techniques assessed (steaming, shallow frying,

grilling, poaching, microwaving, stir-frying, oven baking or roasting, and stewing,

braising or casseroling) except for deep frying, although each marker of

socioeconomic position was not associated with every cooking technique.

Outside of the United Kingdom, studies have focused on adolescents (276) and

young adults (279) and have found socioeconomic differences in food preparation

practices. Adolescents of low socioeconomic position (as defined by highest

parental education) helped to prepare food for dinner more often per week (mean

(SE): 2.4 (0.08)) than adolescents of higher socioeconomic position (1.7 (0.08))

(276). This finding does not appear to be related to working hours, as helping to

prepare dinner was not associated with maternal employment. In the U.S study of

young adults (279), there was a small, non-significant (p=0.06) tendency towards

lower food preparation scores associated with lower socioeconomic position (as

82

defined by highest parental education). High food preparation scores were found

more commonly among high socioeconomic respondents (25.1%) than low

socioeconomic respondents (15.0%), while moderate, low and very-low food

preparation scores were more commonly found among low, compared with high

socioeconomic respondents (39.8, 39.4 and 5.5% vs 33.6, 36.3 and 5.0%

respectively). In this same study, there were no reported socioeconomic differences

in perceived adequacy of cooking skills and resources. However, the size and

direction of the relationship were not reported despite the result approaching

statistical significance (p=0.065).

2.5.3. Summary: cooking skills, socioeconomic position and fruit and vegetable intake

In summary, very few population studies have examined the relationship between

cooking skills and dietary measures. Mostly, studies have supported a relationship

between various cooking-related measures and diet-related outcomes. Among the

limited literature, it appears that cooking-related practices, attitudes, or skills are

linked with healthier dietary behaviours or dietary intakes, including fruit and

vegetable intake. The current evidence is insufficient to determine if any association

between cooking skills and fruit and vegetable intake is causal. Alternate

explanations include the possibility that dietary choices influence the degree to which

cooking skills and attitudes are developed, and cooking-practices are performed. It is

possible that the high-level cooking skills are associated with generally high self-

efficacy and confidence. The Australian study used a structural equation modelling

approach, and was thus able to determine that the cooking related attitudes and

practices were inter-related with a number of other psychosocial constructs and had

an independent relationship with fruit and vegetable intake (277).

The broader population-based studies have tended to support a socioeconomic

patterning of self-assessed cooking skills or confidence (266, 284), however

population-based studies on this topic are lacking outside of the United Kingdom. If

the relationship between cooking skills and diet is causal, then the comparative lack

of cooking skills among low socioeconomic groups could contribute to

83

socioeconomic differences in fruit and vegetable intake. In view of the way cooking

skills are acquired, socioeconomic differences in cooking skills and nutritional

behaviours could be transmitted across generations, as the U.K. Health and Lifestyles

Study (284) showed that most people of all social classes learned to cook primarily

from their mother, and correlations exists between individuals’ personal

socioeconomic position and that of their parents and children (286).

To date, the inter-relationships between socioeconomic position, cooking skills and

dietary behaviours have been discussed, but have not been empirically examined.

Existing studies tentatively point to the idea that dietary differences among

socioeconomic groups might be mediated by socioeconomic differences in cooking

skills. Based on the existing literature, a conceptual model of how socioeconomic

differences in fruit and vegetable purchasing might be partially mediated by cooking

skills was developed (Figure 2.1).

84

Chapter 3: Overview of research hypotheses and methods

3.1. Research questions

Based on the knowledge gaps identified in the literature review and the conceptual

model that was developed (Figure 2.1), this thesis aims to examine whether

accessibility, affordability or cooking skills mediate socioeconomic differences in

fruit and vegetable purchasing among Brisbane residents.

In determining whether socioeconomic differences in fruit and vegetable purchasing

are mediated by accessibility and affordability this thesis examines -

• the relationship between neighbourhood socioeconomic characteristics and

accessibility and affordability of fruits and vegetables;

• the relationship between accessibility and affordability of fruits and vegetables

and fruit and vegetable purchasing;

• whether the relationship between accessibility and affordability and fruit and

vegetable purchasing varies according to household income; and

• the effect of adjusting for any differences in accessibility and affordability on

the relationship that exists between socioeconomic position and fruit and

vegetable purchasing.

To determine whether socioeconomic differences in vegetable purchasing are

mediated by cooking skills, this thesis examines:

• whether socioeconomic position is related to cooking skills;

• whether cooking skills are related to vegetable purchase;

• the effect of adjusting for any differences in cooking skills on the relationship

that exists between socioeconomic position and vegetable purchase

85

3.2. Concepts and measurement issues

A brief précis of the key concepts and measurement issues encountered in the

literature is discussed below, as a prelude to how the general research questions have

been translated into testable hypotheses. Details regarding aspects of measurement

are discussed within the methods.

3.2.1. Socioeconomic position

Socioeconomic position has been described as “the social and economic factors that

influence what position(s) individuals and groups hold within the structure of

society” (287) p.14. Researchers have not consistently defined and measured

socioeconomic position (288). In contrast with this more general term, the terms

“social class” and “socioeconomic status” have arisen from sociological theories of

Marx and Weber (respectively) and should be used in conjunction with these theories

for conceptually sound research (119), however the literature often uses the terms

interchangeably. The research in this thesis has focused on socioeconomic position,

based on the conceptual model of how accessibility, affordability and cooking skills

might contribute to socioeconomic differences in fruit and vegetable purchasing,

which involve social as well as economic mechanisms (Figure 2.1).

No one measurement captures all aspects of socioeconomic position. Income,

education, and occupation are common markers that describe aspects of individuals’

or families’ positions in the social hierarchy. Household income is believed to

represent resources available to acquire goods and services conducive to health

(119), while education is believed to reflect an ability to assimilate information and

access services (289) and to accumulate wealth over the lifecourse (288).

Occupation-based measures are believed to represent social standing and may also

involve occupational cultures (290). Area-level measures generally come in three

types (119) (p40): as aggregate measures of the socioeconomic characteristics of

residents (such as median income), including composite aggregated measures (like

the Socio-Economic Indexes For Areas (SEIFA) constructed by the Australian

Bureau of Statistics (190)); as environmental measures (which are measured in the

86

environment, but have an analogue at the individual level, such as disadvantage); and

as global measures (which have no individual analogue) such as level of social

disorganisation (119).

Each measure has distinct methodological and conceptual issues associated with their

use, and currently, researchers recommend measuring multiple aspects of SEP and

not treating them as interchangeable for purposes of explaining socioeconomic

inequalities (291). For example, area-based measures are appropriate for their

intended use, but are less suitable as proxies for individual SEP (292). The

controversial theoretical basis of some markers of women’s SEP, particularly

occupation-based measures, underscores methodological debate as to the ideal

manner in which to operationalise SEP measures for women, based on their own and

their partner’s SEP (290). In this thesis, the socioeconomic measures have been

chosen on a theoretical basis, within the constraints of using secondary data. Area-

level socioeconomic position has been utilised when examining accessibility and

affordability across socioeconomic contexts. Where mechanisms are expected to

operate through material resources, income has been used as a socioeconomic marker

and where mechanisms expected to operate through acquisition of information,

education has been used (Figure 2.1).

3.2.2. Accessibility and affordability

Accessibility and affordability are multifaceted, complex, and inconsistently

measured. Access encompasses both environmental and individual facets. The way

in which people travel (293), their available resources (car ownership or money for

transport), frailness and chronic illness are all considered within the broad banner of

access to food (158). By a typical dictionary definition, a food is accessible if it is

“capable of being used or seen: available”, and affordable if the cost of it can be

borne “without serious detriment” (294). Accessibility and affordability therefore

depend both on the external environment and individual characteristics that facilitate

or hinder people in procuring food from the environment. The accessibility of foods

relates to the provision of local shops (eg. their abundance, opening hours, proximity

to people and public transport services), the types of foods available within shops,

87

and individual factors (such as mobility constraints and resources for private or

public transportation) (293). The cost of food is a key component of affordability,

as is the purchaser’s ability to meet food costs (which depends on income and other

budgetary costs). Studies using objective measures of accessibility and affordability

have tended to focus on environmental characteristics, while other studies have used

subjective measures that capture individual and environmental characteristics to

some degree in assessing how available or affordable people perceive foods to be

(184).

In this thesis, accessibility and affordability have been examined on the basis of the

food retail environment, using multiple indicators to capture diverse aspects, and

allow for the possibility that some aspects are comparatively more influential to

dietary behaviours than others, in view of the heterogeneous results within the

reviewed literature. Accessibility was conceptualised as consisting of the provision

of different types of shops that usually sell fruits and vegetables (supermarkets,

greengrocers and convenience stores) in neighbourhoods and the provision of fruits

and vegetables within those shops. (This thesis did not examine the individual-level

facets of accessibility and affordability, but did examine modification of the effects

of environmental characteristics by SEP to allow for differences between

socioeconomic groups in their ability to access the same food environment, as

explained further on). Access to shops was measured on the basis of the number of

shops in neighbourhoods, the distance residents live to the nearest shop, and shop

opening hours. The availability of fruits and vegetables within shops was also

examined, on the basis of their absolute availability (yes/no) and relative availability

(number of varieties available). Affordability was examined on the basis of average

prices of fruits and vegetables within neighbourhoods, allowing for modification of

the effect of price by SEP, whereby price may have a differential impact on

behaviour among low-income groups.

3.2.3. Cooking skills

Domestic cooking skills are complex, poorly defined and people’s perceptions of

what constitutes cooking, and skilled cooking, vary from person to person and do not

88

necessarily adhere to definitions of researchers or culinary professionals (45).

People make clear distinctions between concepts like ‘proper cooking’, ‘cooking

from scratch’ and cooking with convenience foods in their discussions of cooking

skills (269). To the knowledge of the author, there was no gold standard, or even

validated published tool, available in the academic literature for measuring adults’

cooking skills for research purposes at the initiation of this thesis. Anderson (295)

developed a perceived ‘confidence in cooking skills’ measure for children, which is

not applicable to adults, as this tool rated children’s ability to prepare foods by

themselves, with a little help, with a lot of help, or not at all. A tool for adults was

developed, but measures cooking attitudes rather than skills (277).

Observation is the most obvious way to objectively and accurately measure skills,

however, it fails to capture the important tacit skills used in cooking practise, such as

planning (45). Studies have used self-reported knowledge, self-rated skills and self-

reported confidence when discussing ‘cooking skills’. Knowledge may be somewhat

disconnected from cooking practise. For example, in one study chefs scored poorly

on a basic food science quiz (296), and in another study, dieticians scored poorly in

knowledge of food preparation facts(297). In a marketing survey, the majority of

respondents described themselves as good or excellent cooks, yet the majority failed

a basic food knowledge test (298). Knowledge is unlikely to represent skill, and self-

assessed skills are unlikely to match objective measures if people do not have the

knowledge to accurately self-assess their skill level.

Confidence to prepare food may have an advantage over actual skill levels as the

subjectivity is integral to the concept rather than a source of error in measurement,

and arguably, actual skill level might be expected to influence how well a task is

performed rather than whether or not someone is going to perform it at all. Stead et

al (269) conducted focus groups among residents of low-income areas in Scotland,

asking about their enthusiasm for cooking, confidence in cooking and claimed ability

and developed a typology of three cooking approaches in which confidence played

an important role: respondents were confident, basic but fearful, or “useless” and

“hopeless”. Confidence to cook was utilised as a cooking skills measure in the UK

Health and Lifestyles Survey (284). Self-reported ‘confidence to cook’ closely

resembles the concept of self-efficacy, a powerful predictor of dietary behaviours in

89

many psychosocial models (43, 44, 299). According to these behavioural theories,

the confidence a person has that they can successfully perform a specific action

increases the likelihood of performing that action (300). Self-efficacy needs to be

measured according to the dimensions of magnitude (performing a number of tasks,

ordered according to their level of difficulty), strength (level of confidence in

performing each specific task) and generality (number of domains of functioning in

which people judge themselves as efficacious) (300).

Accordingly, confidence to cook was chosen as a measure of cooking skills, and the

principles of measuring self-efficacy were applied in constructing the measure.

Confidence was considered as being allowed to vary by degrees, was measured

across two domains (repertoire of vegetables and repertoire of cooking techniques),

and a number of vegetable items and techniques are included which respondents are

likely to find vary in difficulty.

3.2.4. Fruit and vegetable purchasing

There is sound conceptual basis for using fruit and vegetable purchasing as a dietary

focus, even though the choice of this outcome was based on its availability within

secondary data (described further below). Studies mostly focus on dietary intake of

food and nutrients, however a holistic view of diet considers also the behaviours

(purchasing, preparing and consuming) that result in dietary intake. Most existing

studies have quantified dietary intake (either by food frequency questionnaire or

dietary recall), and most have not reported the validity or reliability of the measures

they employ. Generally, consecutive, weighed records are considered the gold

standard for measuring dietary intakes, followed by dietary recall or dietary history

using trained personnel and visual aids, with food frequency questionnaires being of

the lowest quality with respect to precision and accuracy e.g. (301-304).

As indicated in the overarching model of determinants of socioeconomic differences

in diet and subsequent health inequalities (Figure 1.1), mediators of socioeconomic

differences in diet are likely to operate by influencing the behaviours that culminate

in dietary intake. A behavioural, food-based (not nutrient-based) approach to diet

90

provides a simpler conceptual basis on which to examine why people buy and

consume what they do, as opposed to a dietary intake approach, which would require

asking why intake of particular foods or nutrients varies across socioeconomic

groups. A diagrammatic representation of the contribution of the fruits and

vegetables purchased as part of grocery food shopping to fruit and vegetable intake is

presented in Figure 3.1. Though seldom measured, purchasing for at-home

consumption is a behaviour of particular interest. The act of purchasing requires an

intersection between people and their local physical environment which is useful in

examining the effect of accessibility and affordability on dietary behaviour, and the

‘food-at-home’ context of purchasing has relevance for examining cooking skills.

Foods purchased for at-home consumption have nutritional relevance as they

contribute strongly to total consumption by the Australian population (305) and

grocery purchasing (as quantified by receipts) can discriminate between lean and

overweight households (306).

91

Figure 3.1: Conceptual model: Fruit and vegetable purchasing and total household fruit and vegetable intake

92

3.3. Hypotheses

The research questions were translated into eight hypotheses, here expressed in the

null by statistical convention. Results of the tests of these hypotheses are

communicated in the four manuscripts that form Chapters 4-7.

1. The number of local shops, their opening hours, and the distance to the

nearest shop are similar across low-, middle- and high- socioeconomic areas

of Brisbane. (Chapter 4)

2. The availability and prices and of fruits and vegetables are similar in low-,

middle- and high- socioeconomic areas of Brisbane. (Chapter 5)

3. Prices and availability of fruits and vegetables, and the availability of local

greengrocers and supermarkets (in terms of abundance and proximity):

a. do not relate to the purchase of fruits and vegetables of Brisbane

residents;

b. have the same relationship with purchase of fruits and vegetables

among low-, middle- and high- income households in Brisbane; and

c. do not mediate the difference between low- and high- income

households in purchasing fruits and vegetables. (Chapter 6)

4. Among Brisbane residents who usually prepare food for their households,

confidence to prepare vegetables overall and using a variety of cooking

techniques:

a. are equal across socioeconomic groups (as measured by education and

income);

b. are not associated with the variety of vegetables purchased regularly

by households; and hence,

c. do not mediate socioeconomic differences in buying vegetables.

(Chapter 7)

93

3.4. Overview of research methods

Most of the relevant details of the research methods are presented in the individual

manuscripts and extra details that are not reported in manuscripts due to space

constraints are presented here. This thesis employs a quantitative framework to

answer the research questions, and draws on data from two studies (see Figure 1.2).

The first study is a secondary analysis of the Brisbane Food Study (BFS) (307),

which was used to test the first three sets of hypotheses. Unlike many secondary

analyses, there was no mismatch between the research objectives and the original

design. The BFS was specifically designed to examine individual and environmental

determinants of inequalities, and the gain in using this study was that it offered a

larger, better-equipped study than could have been reasonably designed and

implemented within the scope of a PhD program. The second study is a cross-

sectional study that was conducted specifically for this thesis, and is used to examine

whether cooking skills mediate socioeconomic differences in vegetable purchasing

(i.e. hypothesis 4).

3.4.1. Brisbane Food Study

The BFS is a multilevel, cross-sectional study conducted in Brisbane City Statistical

Subdivision (SSD) during the year 2000, in which environmental data were collected

by identifying, classifying and auditing all shops within catchments surrounding 50

sampled census collection districts (CCDs), and household data were collected via a

face-to-face survey from sampled participants who lived in the study CCDs.

Participants were randomly selected using proportional-to-size sampling from 50

small areas (CCDs) that were themselves sampled randomly from within deciles of

socioeconomic disadvantage, as measured by the Australian Bureau of Statistics

(ABS) Index of Relative Socioeconomic Disadvantage (IRSD) (190). This sampling

method was employed to facilitate achieving a probabilistic sample of households

and areas with sufficient representation across the spectrum of socioeconomic

disadvantage. The probabilistic sampling of areas across Brisbane in this study

94

represents a methodological advantage over previous studies examined in the

literature review that have compared purposively sampled areas (307). The overall

response rate to the household survey was 66.4%. Reasons for non-participation

were not collected, however some participants filled in a non-response card that

collected data on age, gender, education and their usual type of bread purchase.

Detailed analysis by the study team of response rates across socioeconomic strata

and the data from the non-response cards indicated some possible socioeconomic and

dietary differences between respondents and non-respondents that could indicate

selection bias, as non-respondents were less educated and less likely to consume

wholemeal or multigrain than white bread. The biases were determined to

underestimate socioeconomic differences in dietary outcomes to a small degree

(308). All food shops within 2.5 kilometres of the study CCDs were sampled, and

selection bias is unlikely to affect the price and availability components of the

survey, as data were obtained for 94% of the supermarkets, grocery and convenience

stores eligible for inclusion in the study. (Data were unavailable for a small number

of shops where managers had refused entry to data collectors.)

In the BFS, all shops in the study catchments were identified through a multi-step

process using council listings, observation, and data from survey respondents cross-

checked with telephone directory listings. The BFS conducted an audit of the

shopping catchments during July through October and collected data on shops’

locations, their type, and their opening hours. Shops were classified based on their

primary activity, floor size and number of checkouts using a tool developed

specifically for the BFS, which had good inter-rater and test-retest reliability (Kappa

> 0.8 for both). For this thesis, only supermarkets, greengrocers and convenience

stores are considered, as these are likely to be the chief sources of fruit and vegetable

purchase for at-home consumption (rather than meat and fish shops, baker and cake

shops, takeaway shops, specialty food shops or other food shops). In November

through December, the shops identified in the audit were surveyed for their price and

availability of a list of food items, including 10 fruits and 10 vegetables. Copies of

the audit tool and the price and availability survey appear in Appendix 2.

95

Brisbane Food Study measures

From the environmental data collected in the BFS, several measures of accessibility

and affordability were available. Shop accessibility measures included the number

of shops, the distance to the nearest shop and opening hours of shops. Several in-

store availability and price measures were also available for the 10 fruit and

vegetable items: their availability (yes/ no); their number of available varieties; and

their price (per unit or per kilogram). During the BFS pilot, four independent data

collectors examined the same convenience store and showed good agreement in their

observations of availability, variety and price (Kappa > 0.8 for all three measures).

Quality could not be examined in this thesis, as a measure of fruit and vegetable

quality showed poor inter-rater reliability during the pilot (Kappa (SE): 0.07 (0.26))

and was subsequently dropped from the BFS (307). The BFS did not establish the

test-retest reliability of the price and availability measures, although it is likely that

price, availability and variety measures fluctuate over time, while shop locations and

their opening hours are more stable. Fluctuation in price and availability is likely to

be random, but might be greater in the CCDs containing fewer observations, i.e. for

CCDs with fewer shops.

From the household survey, a marker of fruit and vegetable purchase was available.

Two questions asking participants how often (never, rarely, sometimes, nearly

always, or always) they bought 19 common fresh fruits (when in season) and 21

common vegetables, either in fresh or frozen form, for the household. The fruit and

vegetable items were widely available according to the 1996-1997 Apparent

Consumption of Foodstuffs (309) and widely consumed according to the National

Nutrition Survey (310). All fruit and vegetable items examined in the audit were

included in the survey, along with additional items that had been deemed too costly

to examine within the environmental data collection. Previous papers from the BFS

(89, 116, 311) have reported on indices formed from the sum of responses to these

items, which are semi-quantitative, and provide an indication of how regular and

varied respondents’ household fruit and vegetable purchasing patterns are. The total

quantity purchased by a household is shared among individuals, however the types of

vegetables purchased are not shared in the same manner. Therefore, in absence of a

fully quantitative measure, fruit and vegetable purchasing were not standardised to

96

household composition, for example in terms of adult equivalents. The internal

consistency and test-retest reliability of these measures were unknown and were

therefore tested as part of the cooking skills study, described in the next section.

Several socioeconomic measures were available within the BFS. The socioeconomic

characteristics of the census collection districts were known from the Socioeconomic

Indexes for Areas produced by the ABS, based on census data. The census

immediately prior to the BFS occurred in 1996, so the 1996 IRSD which had been

used in sampling was retained for analyses, rather than using figures from the 2001

census. The IRSD is a composite index constructed by the ABS, who used a

principal components analysis of variables that reflect the social and economic

disadvantage of an area’s residents, such as proportions of low-income families, one-

parent families with dependents, and proportions of people who are Indigenous,

unemployed, lack fluency in English or have relatively unskilled occupations, to

create the composite measure (190). The household survey additionally collected

several personal socioeconomic markers. This thesis uses those socioeconomic

measures that showed an association with dietary outcomes in the BFS (311):

education (the highest level of schooling or post-school training achieved by the

primary food shopper), and gross household income (income including pensions,

allowances and investments received by all household members). Household

socioeconomic position was examined on the basis of income within this thesis.

Based on the literature review, the mechanisms that would be expected to explain

any relationships between socioeconomic position, the food retail environment and

food purchasing are largely economic and income is the most economically-focused

of the socioeconomic measures.

The conceptual model (Figure 2.1) lists likely confounding factors that were

identified from the literature review. Other co-variates that were confounders of the

relationship between SEP and fruit and vegetable purchasing in previous BFS papers

were also used in this thesis, specifically the gender and age of the respondent (in

completed years). Unlike previous BFS papers, which have treated age

continuously, categories were created, as preliminary analyses showed a more

complex than linear relationship between age and fruit and vegetable purchasing.

The BFS collected data on the other household members: their presence, ages,

97

gender, and relationship to the respondent. From these data, the number of adults

and the number of minors present in household were calculated and examined as

further potential confounders of the relationship between household income and fruit

and vegetable purchasing. In using gross household income (which is influenced by

the number of wage earners) and household fruit and vegetable purchasing (which

might be more varied on a regular basis in households containing more people), there

was a clear need to adjust for the composition of the households. An alternative to

the approach would be to standardize income on a per person, or per adult equivalent

basis. However, to do this properly would have required a continuous measure of

household income and was unnecessary in view of the non-standardized household-

level outcome measure.

As alluded to in the literature review, there are a number of household and individual

factors associated with diet, such as the food preferences of household members and

nutritional knowledge. Many of these were available for use from the BFS, however

these were not considered here, as they deserve more detailed examination as causal

intermediates of socioeconomic differences in purchasing by future studies. Insofar

as these factors are unlikely to be associated with accessibility and affordability, their

omission is unlikely to affect the main research hypotheses of this thesis. For

example, the number of supermarkets in an area is unlikely to be causally associated

with respondents’ nutritional knowledge.

Methodological critique

This study was larger, and better-resourced than could have otherwise been

developed within a PhD program, however there were several limitations to using

this study for the purposes of this thesis. These are listed briefly here, but are

discussed in more depth in the manuscripts and discussion chapters that follow, and

have been considered in interpreting and reporting results. The measures available in

the study did not allow for examination of issues pertaining to dietary intake,

quantities purchased, the dietary habits of individuals within households and fruit

and vegetable quality. The sample size of the BFS was not designed to test the

hypotheses of this thesis, but rather was based on “a range of considerations,

98

including costs and operational constraints, the aims of the study, the level of

disaggregation and the accuracy of the survey estimates… the ‘pioneering nature’ of

the study and its emphasis on description and explanation rather than hypothesis

testing”(307). The sample is small relative to many previous similar studies

examined in the literature review (n=1000 individuals within n=50 areas) and, as will

be covered in the discussion chapter, was underpowered to test the hypotheses.

Accordingly, interpretation is based on effect sizes and confidence intervals, with

effects similar in magnitude to those of critiqued studies being treated as noteworthy.

The approach to examining socioeconomic position was selected to be the most

compatible with the theoretical framework within the confines of using secondary

data. While income is the most appropriate choice of marker for this topic, it does not

represent the full experience of position in both the social and economic hierarchy,

and is particularly limited as a measure for women (119). Given the focus of the

research at the household level, this should not be overly problematic. The process

of mutually adjusting socioeconomic markers can impact the results that are obtained

and their interpretation. In this thesis, estimates of income differences in fruit and

vegetable purchasing were not adjusted for education and occupation, as these are

assumed to operate along the same causal pathway rather than acting as confounding

factors. The one exception is that mutual adjustment of education and income was

utilised as a means of comparing and assessing the independence of education-based

and income-based socioeconomic differences in cooking skills. Area socioeconomic

disadvantage has been examined in this study without adjustment for a full array of

individual and household socioeconomic measures. This differs from usual practice,

because unlike most studies, the purpose of its usage was not to examine contextual

effects above and beyond compositional effects.

3.4.2. Cooking Skills Study The cross-sectional survey of cooking skills and buying practices was designed

specifically for this thesis. Ethical approval for the study was obtained from the

QUT Ethics Committee.

99

Sampling A priori calculations indicated 592 persons were necessary to achieve 80% power to

detect a minimum difference of 6.3 units in vegetable purchasing among any two

income groups at a two-tailed p<0.05 (Figure 3.2) (1). Data from the BFS (89),

which used the same income and purchasing variables, were used to derive the

estimated population standard error (15.2), and the minimum difference of interest

(6.3 units) was set at approximately half the difference found between top and

bottom income groups in vegetable purchasing. There was insufficient data to

conduct a priori calculations based on cooking skills, as the cooking skills measures

had to be constructed specifically for this thesis. Based on an anticipated response

rate of 60%, it was estimated that 990 persons would need to be sampled to recruit

the requisite number of respondents. Initial cost calculations (Figure 3.3) were

performed to determine whether a larger sample could be taken to allow for the fact

that the number of participants needed to test the hypotheses was unknown and the

possibility that the response rate might not reach the target (60%). The available

funding was insufficient to recruit additional participants or to employ time-saving

measures that would have made recruiting a larger sample more feasible. The final

response rate achieved was only 43%, hence the study may have been underpowered

to detect some effects of interest at the p<0.05 level of significance.

100

Figure 3.2: Sample Size Requirement

Sample size estimation based guidelines from Kirkwood and Sterne (1) and

including modifications for cluster sampling based on Carlin et al (2).

Where ‘I’ is 2.8 for two-tailed α =0.05 and one-tailed β=0.2, ‘s’ is the population

standard error, and ‘diff’ is the minimum difference of interest. Based on the BFS,

s=15.2, diff=6.3.

Number per group = (2 * I2 *s2)/ diff2

= (2 * 2.82 * 15.22)/6.32

= 91.275

+ statistical adjustments (15%)

& measurement error (20%) = 123.2

x 1.6 for cluster sampling(Design Effect) = 197.15

Overall number (for 3 groups) = 591.45

Number needed to sample

(with 60% response rate) ≈ 990

Design Effect (DEFF) = 1 + (ICC* (k-1)) = 1 + 0.0094*(61.6-1) ≈1.6

where k is the average number of individuals in each cluster

k=(123.2 per SEP group x 3 SEP groups / 6 areas) = 61.6,

and ICC is intraclass correlation (0.0094 from BFS).

101

Figure 3.3: Costing for cooking skills study

The cooking skills study was costed firstly to examine whether a larger sample

was feasible in the event that the response rate did not meet the targeted 60%,

and secondly whether a mailing method would be affordable (to reduce time

costs). However, even without mailing out replacements exceeded the student

budgetary allowance ($1000), even at a sample size of n=990.

Anticipated costs for cooking skills study by delivery and mailing method

No. of Units

Units Unit cost ($)

Total cost ($)

Envelopes 1400 each 0.03 36.96Cover letter 1400 each 0.08 109.34Consent information 1400 each 0.08 109.34Questionnaire 1400 each 0.24 336.00Reply paid envelopes 1400 each 0.14 196.00Thank-you/ reminder cards 990 each 0.08 74.29Fuel 736 / km 0.07 53.53Postage – returned questionnaires 600 each 0.50 300.00Total costs (delivery method) 1215.46+ 20% sample (n=1188) 1458.55+ 30% sample (n=1287) 1580.10+ 50% sample (n=1485) 1823.19Postage – follow-up surveys 400 each 1.00 400.00Postage – repeatability 85 each 85.00Postage – thank-you cards 990 each 0.50 495.00Total costs (mailing method) 2195.46

NB - Stationery costs are based on quotes from QUT Printing Services. Postage

is based on Australia Post fees. Fuel costs were estimated using 80c/litre with

11km/litre fuel efficiency (the candidates’ car) over an expected distance of

736km (which reflected three return trips from Kelvin Grove to a random

sample of six areas, plus driving within the study areas).

102

A diagram of the sampling process is provided in Figure 3.4. Households were

sampled from within a random sample of CCDs in the Brisbane SSD, stratified by

tertiles of Index of Relative Socioeconomic Advantage/ Disadvantage to ensure

sufficient participation across the socioeconomic spectrum while retaining

probabilistic sampling methods. For feasibility reasons, the sample included only six

CCDs, with 165 households being sampled per CCD to achieve a total sample of 990

households. Accordingly, only CCDs containing at least 180 occupied private

dwellings (according to the 2001 Census) were eligible for inclusion in the study, to

enable 165 households to be sampled with a margin to allow for changes since the

2001 census. Table 3.1 indicates there were no substantial differences between the

residents in the eligible and ineligible CCDs, beyond what could be explained by

greater regression towards the mean in the CCDs with larger populations.

Households were randomly sampled by dwellings to avoid the use of the electoral

roll or telephone directories with known socioeconomic biases (312). Eligibility of

dwellings was determined through observation by the candidate during data

collection. Dwellings were considered to be ineligible if they did not appear to be

private residential dwellings (for example, by signage that indicated the property was

a child-care facility or school) or if the dwelling appeared unoccupied (for example,

by being under construction or visibly vacant without furniture). A number of

apartments/units were sampled that may have been vacant, as this was more difficult

to determine for apartments (n=303) than for houses (n=687), which is reflected in

the lower response rates from apartments (36.7%) compared with houses (52.9%)

(Table 3.2). Sampling of vacant apartments could explain the lower response rate in

one of the middle socioeconomic areas which contained a substantial proportion of

this housing type (44.2%).

103

104

Table 3.1: Characteristics of eligible and excluded CCDs from 2001 Census Total Pool a

(n=1667) Included (n=1007)

Excluded b (n=660)

% Indigenous Australians 0.96 (0, 14.29) 0.97 (0, 11.52) 0.95 (0, 14.29) % Australian Born 71.93 (10.12) 72.08 (9.62) 71.71 (10.85) % Overseas Born 21.89 (8.33) 22.38 (8.36) 21.13 (8.23) % In private dwellings 96.61 (11.47) 98.06 (6.73) 94.40 (15.98) % Enumerated elsewhere 0 (0, 99.57) 0 (0, 67.16) 0 (0, 99.56) % Dwellings privately occupied 93.47 (0, 100) 93.87 (42.79, 100) 92.64 (0, 100) % no schooling 0 (0, 19.49) 0.48 (0, 7.66) 0 ( 0-19.49) Males Unemployment rate 7.24 (0, 41.79) 7.21 (0, 27.85) 7.29 (0, 41.79) Females Unemployment rate 5.77 (0, 45.45) 5.81 (0, 27.38) 5.71 (0, 45.45) Persons Unemployment rate 6.55 (0, 42.09) 6.54 (0.90, 27.61) 5.57 (0, 42.09) Mean Household size 2.47 (0.47) 2.50 (0.46) 2.42 (0.47) Occupied Private Dwellings 199 (0, 576) 238 (180, 576) 139 (0, 179) Total Persons 532.41 (232.97) 652.71 (191.62) 348.86 (159.06) Median Age 34.88 (5.50) 34.79 (5.28) 35.02 (5.82) Median house loan repayment

$200-599/ month 1.98 % 1.59 % 2.58 % $600-799/ month 12.12 % 12.21 % 11.97 % $800-999/ month 36.77 % 43.1 % 27.12 %

$1000-1199/ month 27.5 % 27.41 % 27.73 % $1200-1399/ month 13.44 % 11.52 % 16.36 %

$1400 or more/ month 5.76 % 3.68 % 8.94 % n.a 1.62 % 0.1 % 3.94 %

Median weekly rent $149 or less 19.7 % 16.5 % 24.7 %

$150-199 52.4 % 55.1 % 48.2 % $200-299 24.9 % 26.5 % 22.7 % $300-399 2 % 1.5 % 2.6 %

$400 or more 0.5 % 0.4 % 0.75 % Na 0.4 % 0 % 1.1 %

Median individual income/week $80-299 11.3 % 8.9 % 14.58 %

$200-399 29.4 % 30.1 % 28.5 % $400-599 54.2 % 57.5 % 49.1 %

$600 or more 5.2 % 0.9 % 2.6 % na 0.1 % 0 % 0.2 %

Median family income/week $499 or less 1.2 % 0.3 % 2.5 %

$500-999 40.8 % 43.5 % 36.7 % $1000-1499 45.3 % 46 % 44.2 %

$1500 or more 12.6 % 10.3 % 15.9 % na 0.3 % 0 % 0.8 %

Median household income/week $399 or less 1.5 % 0.7 % 2.6 %

$400-699 23.1 % 22.6 % 24.1 % $700-999 42.8 % 44.1 % 40.9 %

$1000-1199 19 % 19.8 % 17.7 % $1200 or more 13.4 % 12.9 % 14.1 %

n.a. 0.2 % 0 % 0.6 % Table presents Median (minimum to maximum), mean (standard deviation) or percentages. a i.e all CCDs of n=1680 in the Brisbane SSD with a population greater than zero b Exclusion criteria: fewer than 180 occupied private dwellings

105

Table 3.2: Response rates by area and housing type All Unit / Apartment Houses n % response n % response n % response Low SES area 1 165 39.4 % 0 - 165 39.4 % Low SES area 2 165 44.5 % 0 - 165 44.5 % Middle SES area 1 165 36.8 % 73 26.0 % 92 45.7 % Middle SES area 2 165 50.3 % 0 - 165 50.3 % High SES area 1 165 54.8 % 91 41.8 % 75 70.7 % High SES area 2 165 42.4 % 139 42.5 % 26 40.3% Average a 44.7% 36.7% 52.9% a Average response rates for the three areas that contained units/apartments only Within households, the person who was mostly responsible for preparing food was

invited to participate and a cross-check item was included in the survey to ensure the

appropriate person had completed the survey. The household chef was the intended

participant as their confidence to cook is arguably the most relevant to the household

dietary practices. Where other participants volunteered information (but the

household chef declined), the surveys (n=20) were discarded. Examination of the

differences between household chefs and ineligible respondents was not performed

due to the small number of surveys and the existence of alternate explanations for

any differences detected (i.e. volunteer bias and true differences between people who

do and do not cook).

Data collection

In each CCD, a household was selected at random from council BIMAPs3. In a

clockwise pattern spiralling from the randomly sampled starting address (block-by-

block), the next 164 spatially sequential eligible dwellings within the study CCDs

were invited to participate. The survey package containing a cover letter, consent

information, a self-administered questionnaire, and a reply-paid self-addressed

envelope was distributed to the sampled dwellings (Appendix 3). Addresses for the

sampled households were collected and matched to survey identification numbers to

enable follow-up. The data collection instrument and associated materials are

provided as Appendix 3.

3 BIMAP = Brisbane’s Integrated Map of Assets and Property, electronically available listings of properties in Brisbane

106

Based on recommendations by Dillman’s Tailored Design method (313) (p 151), a

thank-you card (Appendix 3) was left at these same premises one week after the

initial questionnaire delivery to thank participants, and to subtly prompt non-

responders to participate. Two weeks after the initial delivery, replacement surveys

were distributed to addresses from which a completed survey or an indication of

refusal had not been obtained. Due to the overall low response, an extra step to boost

response rates, using either a gratuity ($1 scratch-it) enclosed in the survey package,

or face-to-face initial contact with respondents, was trialled for all non-respondents

in the two areas with the lowest response rates. Both methods yielded a small

number of additional responses. Since these would have been insufficient to reach

the estimated necessary sample size and would have added significantly to either the

time or budgetary costs, neither was extended to the whole study sample.

3.4.2.1. Instrument development

The survey instrument to assess cooking skills was a four-page self-completed

questionnaire that was developed specifically for this research. The survey was kept

brief to minimise respondent burden and maximise response rates. It covered

household demographics, vegetable purchasing (using the same measure employed

in the BFS for continuity), confidence to prepare the same 21 vegetables, and

confidence to cook vegetables using 10 cooking techniques. Techniques chosen

were those used in the UK National Health and Lifestyles Survey (boiling, steaming,

stir-frying, microwaving, stewing, poaching, shallow-frying, deep-frying, roasting,

and grilling (i.e broiling)). Respondents were asked to indicate how confident they

felt (not at all confident/ not confident/ not quite confident/ a little confident/

confident/ quite confident/ very confident) to prepare each vegetable, and to use each

technique. Responses were summed to form two indices. Fruits are usually

consumed with little preparation, hence this study focused only on confidence to

prepare vegetables and vegetable consumption.

A detailed description of both the methods and results of pre-testing the instrument is

provided in Appendix 4. Briefly, the face validity of the measure was assessed by

asking a convenience sample of adults of varied ages, educational backgrounds and

107

gender to complete the questionnaire and provide comment. Content validity was

examined by passing the questionnaire to an expert panel comprised of a researcher

whose work has included a focus on cooking skills and a dietician and by presenting

the questionnaire for comment to a group of research students and research officers

at a measurements forum held within the QUT School of Public Health. The

repeatability and internal consistency of the instrument were tested among a subset

of early survey respondents (n=57, 67% response rate), chosen at random, who

received an additional re-test survey two weeks after the initial survey. An overview

of the main results is provided here, and further detail is provided in Appendix 4.

Face and content validity There was a degree of overlap between issues raised by participants and by experts.

No concern was expressed by experts over whether the questionnaire was adequately

measuring the desired constructs, but a level of caution was expressed as to the

extent to which the constructs measured in the questionnaire may or may not relate to

other commonly used constructs in this field of research.

Regarding the confidence measures, the expert researcher offered the opinion that the

questionnaire looked adequate but expressed concern that these psychometric

measures may not correlate with actual skill level or behaviours. This is a major

issue, particularly if this questionnaire is to be used for purposes other than to

determine whether confidence (not actual skill) relates to purchasing behaviours.

From the measurements forum, it was suggested that the term “food preparation”

may carry different meanings for participants, which may undermine the validity of

their responses. Respondents preferred a Likert-Style format with verbally described

categories to a 0-100% format for measuring confidence. Participants’ own words

generally matched the intended meaning of the questions, and they clearly

communicated that ‘confidence’ related to whether they thought they could do

whatever was being asked about, and the scaling referred to how certain or uncertain

they were that they could do it.

108

There are some validity issues with the purchasing measure. The expert dietician’s

comments focused mostly on the non-equivalence of purchasing and dietary intake.

Respondents raised similar concerns in pointing out that purchasing does not equate

to consumption and purchasing in bulk means a low frequency of purchase but not

necessarily a low frequency of consumption. An additional concern worthy of

consideration was the lack of a definable time period on the purchasing measures.

Dietary patterns are not fixed, and although questions focused on respondents’ usual

purchase, dietary responses are often biased toward current behaviour. Data on the

quantities purchased were not collected. Furthermore, the categories were described

qualitatively (e.g. “sometimes”, “nearly always”) as opposed to quantitatively (e.g

number of times per month standardized for how often food shopping is performed).

Thus the measure should probably be best interpreted as an ordinal measure of

respondents’ perceptions of the current typical range of fruits and vegetables

purchased for their households.

Overall, from the face and content validity assessments, there were no obvious

concerns with how the questionnaire measures self-reported purchasing and

confidence levels of participants. However there were numerous concerns with the

validity of trying to equate these measures to dietary intake or skill levels. For

consistency with the BFS, the purchasing measure was retained despite concerns

about its validity. Based on the pre-testing, the Likert-style measurement approach

was utilised, and a definition of food preparation was added:

“For this survey, ‘preparing’ food means anything you might do to make the

food suitable to eat (for example, make a salad from it).”

Internal consistency

The purchasing measure was treated as an index rather than a scale as it was not

expected to be internally consistent. It would not be reasonable to assume that the

frequent purchase of one vegetable item makes the frequent purchase of the other

items any more likely, especially as the literature shows a clustered nature to food

consumption (314). The properties of this measure were consistent with

expectations. The index was not entirely uni-dimensional, as the average inter-item

correlation was low 0.23 (95% CI: -0.10 to 0.51) and some items showed poor

109

correlation (less than 0.4) (315) with the sum of the other items (peas, Chinese

cabbage and eggplant).

The confidence measures were treated as scales as it was expected that for any item,

a low-level of confidence would indicate a greater likelihood of also having less

confidence for the other items. The 21-item confidence to cook vegetables scale

showed good internal consistency (Cronbach’s alpha=0.94) according to the rating

system described by George and Mallery (316), however this could partly relate to

the number of items in the scale. On average, items were weakly to moderately

inter-correlated 0.47 (95% CI: 0.19 to 0.89) and were correlated moderately with the

sum of the other items (range: 0.54 to 0.73). The 10-item cooking technique

confidence scale showed good internal consistency (Cronbach’s alpha=0.83).

Average inter-item correlation was weak to moderate 0.35 (95% CI: 0.15 to 0.59)

and most items correlated well with the sum of the other items (range: 0.45 to 0.69),

except for the confidence to microwave vegetables (0.29). The 10-item scale was

retained as removal of the microwave item resulted in only slightly improved internal

consistency (Cronbach’s alpha=0.84) and inter-item correlation 0.38 (95% CI: 0.16

to 0.59). The small degree of inconsistency present in these scales may stem from

respondents having or lacking confidence depending on their familiarity with

different vegetables and techniques. This slight inconsistency may not be

problematic in aiming to associate confidence with the variety of vegetables

purchased as respondents who are most confident to cook with the greatest variety of

vegetables, using the widest variety of techniques would score most highly on the

confidence scales.

Test-Retest reliability

Test re-test reliability was firstly examined for individual items (to examine whether

any individual items may be problematic) and later for the overall scales and index,

to provide easily reportable summary statistics for publication. Agreement between

test and retest values of individual items were examined using kappa statistics,

weighted linearly for the amount of disagreement between categories using Cichetti-

Allison based weights (i.e. wij= 1 – (|Ci – Cj| / CC-C1) where Ci is the score for

110

column and C is the number of categories or columns) . The test-retest reliability of

the confidence scales and purchasing index was examined using Bland-Altman plots

(see Appendix 4) and intra-class correlations.

The individual items comprising the confidence indices mostly showed ‘moderate’

agreement or better, however lettuce and capsicum showed only ‘fair’ agreement

(0.36 and 0.28, respectively) (Table 3.3) (317). For all items, discrepancies did not

favour the test or re-test period beyond what could have occurred by chance

(according to Bowker’s test of symmetry) (318). The items comprising the

purchasing index mostly showed substantial agreement, with weighted kappas

ranging from 0.53 to 1.00 (Table 3.4). There was no significant evidence that

discrepancies in purchasing items systematically favoured the test or re-test period.

The overall indices showed some differences between the test and re-test periods that

appeared non-systematic and minimal relative to overall variation. Any systematic

bias in the Purchasing, Vegetable Confidence and Technique Confidence Indices was

small, as mean differences were small (2.31, 1.75 and 0.08, respectively) and were

well within the limits of agreement. Limits of agreement indicated substantial non-

systematic measurement error in the Purchasing Index (-12.70 to 17.31), the

Vegetable Confidence Index (-8.94 to 12.43) and the Technique Confidence Index (-

6.63 to 6.79), however the intra-class correlations indicated the imprecision in the

indices was small compared to the general variation in these measures, ranging from

0.85 to 0.88 (Table 3.5).

111

Table 3.3: Test-retest reliability of survey items pertaining to confidence

Kappa (95% CI)a

Symmetry b

Confidence to prepare: Carrots 0.78 (0.55, 1.00) *** S=1.00, df=3, p=0.80 Mushrooms 0.78 (0.64, 0.91) *** S=1.20, df=10, p>0.99 Chinese Cabbage 0.71 (0.56, 0.86) *** S=10.80, df=15, p=0.77 Squash 0.71 (0.53, 0.88) *** S=4.33, df=15, p>0.99 Eggplant 0.69 (0.56, 0.82) *** S=13.33, df=15, p=0.58 Silver beet 0.68 (0.50, 0.85) *** S=8.00, df=15, p=0.92 Brussels Spouts 0.67 (0.50, 0.84) *** S=4.14, df=15, p>0.99 Pumpkin 0.65 (0.38, 0.91) *** S=3.57, df=3, p=0.32 Green Beans 0.64 (n/c) n/t Potatoes 0.64 (0.37, 0.90) *** S=0.33, df=3, p=0.95 Tomatoes 0.62 (0.38, 0.86) *** S=4.33, df=6, p=0.63 Cabbage 0.58 (0.10, 0.76) *** S=3.53, df=10, p=0.97 Cauliflower 0.57 (0.37, 0.78) *** S=3.14, df=6, p=0.79 Peas 0.57 (n/c) n/c Cucumber 0.56 (n/c) n/c Celery 0.55 (n/c) n/c Broccoli 0.48 (0.16, 0.80) *** S=1.11 df=3, p=0.77 Zucchini 0.47 (n/c) Onion 0.46 (0.14, 0.79) *** S=2.00, df=3, p=0.57 Lettuce 0.36 (n/c) n/c Capsicum 0.28 (n/c) n/c

Confidence to prepare vegetables by:

Deep frying 0.73 (0.61, 0.84) S=5.73, df=15, p=0.98 Steaming 0.68 (n/c) n/t Stir-frying 0.65 (0.52, 0.79) S=7.00, df=15, p=0.96 Micro-waving 0.62 (0.44, 0.80) S=8.57, df=15, p=0.90 Stewing 0.61 (0.45, 0.76) S=4.00, df=15, p>0.99 Poaching 0.59 (0.46, 0.72) S=7.50, df=15, p=0.94 Shallow frying 0.54 (n/c) n/c Roasting 0.52 (n/c) n/c Grilling 0.41 (n/c) n/c

a Kappa, with Cichetti-Allison based weights for level of disagreement(319) b McNemar-Bowker test of symmetry: that probabilities in the square table are

symmetrical (null-hypothesis). (I.e. discrepancy from perfect agreement does not systematically favour the tested or re-tested measure)

* p<0.05 ** p<0.01 *** p<0.001 n/a not assessable – no discordant pairs n/c not calculated: Armitage and Berry (320) (p.446) approach to calculating

weighted kappas for non-square tables adopted

112

Table 3.4: Test-retest reliability of survey items pertaining to purchasing and

demographics

Kappa (95% CI)a

Symmetry b

How often do you buy? Tomatoes 1.00 (1.00, 1.00) *** n/a Brussels Spouts 0.84 (0.51, 0.77) *** S=11.83, df=10, p=0.30 Potatoes 0.83 (0.70, 0.97) *** S=3.80, df=10, p=0.96 Cabbage 0.72 (0.60, 0.84) *** S=4.67, df=10, p=0.91 Capsicum 0.72 (0.58, 0.86) *** S=7.20, df=10, p=0.71 Green Beans 0.72 (0.59, 0.84) *** S=4.29, df=10, p=0.93 Silver beet 0.71 (0.59, 0.84) *** S=7.57, df=10, p=0.67 Zucchini 0.71 (0.58, 0.84) *** S=3.00, df=10, p=0.98 Chinese Cabbage 0.70 (0.56, 0.83) *** S=9.20, df=10, p=0.51 Squash 0.69 (0.54, 0.83) *** S=10.67, df=10, p=0.38 Celery 0.66 (0.53, 0.79) *** S=5.47, df=10, p=0.86 Broccoli 0.66 (0.50, 0.82)*** S=3.48, df=10, p=0.97 Cucumber 0.65 (0.52, 0.78) *** S=8.13, df=10, p=0.62 Peas 0.64 (0.47, 0.81) *** 13.00, df=10, p=0.22 Cauliflower 0.63 (0.48, 0.78) *** S=8.40, df=10, p=0.59 Lettuce 0.63 (0.47, 0.78) *** S=2.14, df=10, p>0.99 Pumpkin 0.59 (0.41, 0.78) *** S=2.73, df=10, p=0.99 Onion 0.58 (0.37, 0.79) *** S=5.82, df=10, p=0.83 Carrots 0.57 (0.31, 0.84) *** S=1.82, df=6, p=0.84 Eggplant 0.53 (0.33, 0.73) *** S=4.11, df=10, p=0.94

Demographics Gender 1.000 n/a Disability 1.000 n/a Australian, born 0.96 (0.88, 1.00) S=1.00, df=1, p=0.32 Household Income 0.99 (0.98, >0.99) S=11.00, df=78, p=1.00 Education 0.88 (0.78, 0.98) S=5.00, df=15, p=0.99 Non, English Speaking Background (NESB) 0.65 (0.19, 1.00) S=0.00, df=1, p=1.00 Aboriginal or Torres Strait Islander (ATSI)

n/a n/a

Aboriginal, Torres Strait or South Sea Islander (ATSthSI1)

n/a n/a

a Weighted kappa for ordinal variables, unweighted kappa for nominal variables (Gender, Australian- born, ATSI, ATSthSI1, Disability, NESB, Who cooks?)

b Bowker’s test of symmetry: that probabilities in the square table are symmetrical (null-hypothesis). (I.e. discrepancy from perfect agreement does not systematically favour the tested or re-tested measure)

* p<0.05 ** p<0.01 *** p<0.001 n/a not assessable, eg. no discordant pairs or only one response category

113

Table 3.5: Test-retest reliability, Intra-class correlation coefficients

Intra-class Correlation (95% CI)a

(Individual level) Vegetable Purchasing Index 0.85 (0.729, 0.919)*** Confidence to cook vegetables scale 0.87 (0.761, 0.929)*** Confidence to use cooking techniques scale 0.88 (0.79, 0.94)*** Age (completed years) 1.00 (1.00, 1.00)*** Number of persons in the household Adults (>18 years) 0.97 (0.94, 0.98) *** Teenagers (13-17 years) 0.95 (0.92, 0.97) *** Children (6-12 years) 0.93 (0.87, 0.96) *** Young children (2-5 years) 0.65 (0.46, 0.78) *** Infants (<2 years) 0.79 (0.66, 0.88) *** People (total) 0.96 (0.93, 0.98) *** * p<0.05 ** p<0.01 *** p<0.001 a ICC (2, 1) - Single measures ICC for absolute agreement between test and retest measure using two-way mixed effects model: people (random), items (fixed) (321).

Methodological critique

This study is one of the few population-based studies of cooking skills within

Australia. Its strengths include the use of confidence-based measures with

demonstrated test-retest reliability and internal consistency, and the use of sampling

and measures that are all consistent with a focus on behaviours occurring in the

household context. Its main limitations are a low response rate (43%) with some

associated non-response bias, a limited outcome measure and a lack of validation

against behaviours other than self-reported purchasing. These issues are considered

in interpreting and reporting findings within the manuscript and are outlined in more

detail in the discussion (Chapter 8).

114

3.4.3. Integration of the studies

In summary, analyses of the secondary data and the original study were conducted to

answer the research hypotheses in a manner consistent with the theoretical

framework developed from the literature review (Figure 2.1). In keeping with the

conceptual framework (Figure 2.1), the socioeconomic and purchasing measures

used in the BFS were retained in the cooking skills study. However, the nature of the

expected relationship between the mediators (shop access, availability, price and

cooking skills) and purchasing necessitated some analytical differences in how the

purchasing outcome was treated. The literature review indicated that the price-

purchase relationship would be likely to vary across items. To accommodate this,

fruits and vegetables were examined item by item. However, the literature did not

provide the same basis by which to expect heterogeneity in the confidence-purchase

relationship, so there was no need to examine items separately and the confidence

scales and purchasing index could be utilised in a manner consistent with their

design.

Further details, particularly those pertaining to technical aspects of the study

measures and statistical analyses can be found in the methods sections of the

manuscripts that follow. Chapter 4 examines the hypotheses pertaining to shop

access across socioeconomic contexts, while Chapter 5 examines the hypotheses

pertaining to in-store prices and availability across socioeconomic contexts. Chapter

6 examines the hypotheses pertaining to the relationships between shop access,

availability and price and fruit and vegetable purchasing among socioeconomic

groups. Chapter 7 examines all hypotheses pertaining to cooking skills.

115

4. Chapter 4: Does living in a disadvantaged area mean fewer opportunities to purchase fresh fruit and vegetables in the area? Findings from the Brisbane Food Study4

4 This manuscript has been published in Health and Place journal 2006, 12 (3):306-319

116

4.1. Abstract

Understanding the particularly low intake of fruits and vegetables among

socioeconomically disadvantaged groups is an important issue for public health.

This study investigated whether access to retail outlets is similar across areas of

varying socioeconomic disadvantage in an Australian urban setting, in terms of

distance, the numbers of local shops, and their opening hours. This ecological cross-

sectional study used 50 randomly sampled census collection districts and their nearby

shopping environment (i.e. within 2.5km), and generally found minimal or no

socioeconomic differences in shopping infrastructure. Important methodological and

social/economic issues may explain this contrast with overseas findings.

4.2. Introduction

Low intake of fruits and vegetables is associated with heart disease (322, 323), stroke

(324), and some cancers and many of these diseases and disease risks follow a

socioeconomic gradient (19, 21, 28, 29, 325). ‘Diet’, shows graded socioeconomic

differences, such that people or families who are less affluent, less educated, or

employed in less prestigious jobs have diets which are least concordant with official

recommendations, both in general, and specifically in relation to fruits and

vegetables in Australia (89, 90, 100, 105, 326-328), the United States (92-94, 112),

the United Kingdom (102, 103) and Europe (99, 109, 329). Evidence suggests that

diet, particularly lower intake of fruits and vegetables, partially explains the higher

rates of cardiovascular disease in low socioeconomic groups (20, 21). Understanding

reasons for the difference between socioeconomic groups in fruit and vegetable

intake is therefore an important public health issue.

Explanations for why socioeconomic dietary differences exist have generally

focussed on the individual level, and occasionally on the environmental level.

However, a need has been underscored for research to examine both environmental

characteristics and individual factors in order to understand health-related behaviour

117

(330) and socioeconomic inequalities (331). To explain socioeconomic differences

in diet, it is necessary to consider the way the environment impacts on the diets of

people who live in disadvantaged areas as well as on disadvantaged individuals.

The environment may be important in determining socioeconomic dietary

differences, since studies from the United States and United Kingdom have found

that living in a socioeconomically disadvantaged area is an independent risk factor

for poor diet, or low fruit and vegetable intake. In the United States, a recent multi-

level study (111) demonstrated that people living in the least wealthy areas had the

lowest fruit and vegetable intakes, irrespective of their race, personal income or

education. Similarly, in the United Kingdom a recent multi-level study (114) found

residents of deprived areas to have significantly worse diets irrespective of their

personal socioeconomic position, according to a ‘bad diet’ score, but not according

to a ‘good diet’ score. Earlier precursors to these studies in the UK had not measured

area socioeconomic status, but did find that many nutrient intakes related to where

people lived, independently of personal socioeconomic position (98, 332). Other

early work from the UK did not use multi-level analyses and may not have

appropriately partitioned variance between individual and environmental factors.

These studies found that living in a low socioeconomic area relates to many dietary

outcomes including fruit and vegetable intake (108), and that neighbourhood income

relates independently to lower ‘healthy’ and higher ‘unhealthy’ dietary scores (115).

The effects of area socioeconomic disadvantage on diet are unlikely to be entirely

direct, and could be mediated by differences in the food supply between areas which

vary in socioeconomic characteristics (108).

Issues pertaining to the food supply, such as the accessibility of shops which sell fruit

and vegetables at affordable prices, are important environmental determinants of

food purchasing and consumption, and have the potential to explain some of the

observed socioeconomic differences in diet. Accessibility is a complex concept that

is difficult to quantify, relating to the numbers/choice of local shops present, their

opening hours, their distance from shoppers, and other factors that influence how

people arrive there such as parking and public transport availability (207).

Furthermore, the same shopping environment may not be equally accessible to all

individuals, as factors such as motor vehicle ownership, and disposable income can

118

make the environment more (or less) accessible (158 p. 48). A recent multi-level

study (214) showed that the presence of a supermarket in a neighbourhood

significantly increased the chances that African American residents met the dietary

guidelines for fruits and vegetables, but this association was not significant for white

Americans. Interestingly, in this study, private vehicle ownership was much lower

for African Americans than white Americans. It is therefore plausible that the local

food environment has the greatest effect on those most constrained by lack of

transportation, resources, or physical mobility impairment. Socioeconomic

disadvantage may constrain people to their local food environments, as low income

families are most likely to lack access to private transport (160), and are most likely

to use alternate transport methods to shop for groceries (161). A recent case study in

a socioeconomically disadvantaged area of the UK found that the provision of a large

food retailer in an area that previously had few food retail outlets was associated with

some improvement in the consumption of fruits and vegetables, particularly for those

who lived close to the new store, who began to shop at the new store regularly and

those who had the lowest fruit and vegetable consumption prior to the intervention

(216).

Some overseas evidence shows that the food environment in socioeconomically

disadvantaged areas is less conducive to healthy purchasing and consumption than in

socioeconomically advantaged areas. Studies from the United States and the United

Kingdom have assessed whether provision of shops to people who live in deprived

urban areas is systematically worse than provision in wealthier areas. These studies

have generally focussed on the numbers or presence/absence of shops, the foods

available or the prices of foods. No identified studies have addressed other aspects

of accessibility such as opening hours, or distances to the nearest food shop in

relation to socioeconomic disadvantage.

In the United States, a recent paper (170) found three-fold higher prevalence of

supermarkets in the wealthiest compared to the poorest neighbourhoods. An earlier

study (188) found chain stores were more likely to locate outside the inner city areas,

and more likely to locate in more affluent neighbourhoods. A study in Chicago (187)

found significantly fewer large grocery stores and supermarkets and more small

grocery stores in poor compared with other neighbourhoods. The difference in the

119

number of stores was found to relate to the lower purchasing power of poor areas,

but the difference in the size of stores, particularly small grocery stores could not be

explained entirely through the reduced purchasing power in poor neighbourhoods.

Interestingly, a study in New Jersey (333) found that fast food chains were

disproportionately located more often in low income areas, and in areas with a

greater proportion of African Americans.

A central focus of the social exclusion and health inequalities debate in the United

Kingdom was the supposed existence of ‘food deserts’ (areas where foods integral to

a healthy diet were inaccessible to low-income households in poor neighbourhoods

in British cities) (216). However, there has been insufficient systematic research to

support the existence of ‘food deserts’, (334) or the claim that poor areas of cities in

the UK have relatively fewer food retail outlets than more advantaged areas. A study

by Cummins and MacIntyre (180) actually found relatively more large supermarkets

in deprived than affluent areas of London. However, a later study in Cardiff (179),

which utilized a different methodology, showed a substantial gap in accessibility to

shops between deprived and affluent areas which is likely to widen over time as

accessibility is improving overall, but more so for affluent than deprived areas.

Differences between findings from the UK and US, as well as differences between

countries in social and economic factors, illustrate the difficulty of translating

findings across countries, and perhaps cities, and time points.

Australian studies into the food supply have generally focussed on rural/urban

differences in price and availability, as foods are often more difficult to procure and

more expensive in rural areas (335). No identified Australian studies have addressed

systematic differences in supermarket location within urban areas which is

concerning as the majority of the Australian population live in urbanised areas,

particularly the capital cities (336). However, a recent study (337) found a relatively

larger prevalence of takeaway shops in socioeconomically disadvantaged areas of

Melbourne city.

In Australia, a recent multi-level study (116) found no significant differences in fruit

and vegetable purchasing between households in socioeconomically disadvantaged

and advantaged areas in Brisbane city other than differences explained by the income

120

of these households. It would be anticipated that in urban Australia, or at least

Brisbane, either there are no large differences in environmental factors likely to

influence fruit and vegetable purchasing for residents in disadvantaged areas, or

alternatively that individual factors, such as universal car ownership, reduce the

impact of the environment on dietary behaviours. Accordingly, this paper aims to

determine whether there are systematic differences in shopping infrastructure which

are likely to influence the fruit and vegetable purchasing patterns of socioeconomic

groups in an Australian, urban setting, and contrast findings with those from

overseas.

This study addresses whether area socioeconomic disadvantage relates to the

following three aspects of shop availability:

1) The number of nearby fruit and vegetable retail outlets

2) The opening hours of nearby fruit and vegetable retail outlets

3) The distance to the nearest fruit and vegetable retail outlet

The studies of shop location and socioeconomic position (170, 180, 187, 188, 333,

337) have all considered the number of shops located within administrative

boundaries (such as census tracts, postcodes, or census collection districts) which is

problematic if the aim is to compare residents’ access to retail outlets. Although

these studies have generally considered population size or density, finding more or

fewer shops within administrative boundaries still does not neatly translate to

residents within these administrative boundaries having comparatively better or

worse access to shops. Firstly, using the same spatial scale to measure

socioeconomic characteristics and the local food environment means that tradeoffs

must be made. The smaller the administrative boundary used, the more

homogeneous the area in terms of socioeconomic characteristics (120), thereby best

classifying the area as disadvantaged or advantaged. However, small spatial scales

may not meaningfully measure the local food environment, as residents are likely to

shop for food outside these boundaries (338), especially residents living near the

borders of boundaries (202). Larger administrative boundaries capture a more

realistic food environment for residents towards the centre of the boundary, but not

for residents at the boundary edges. Secondly, since administrative boundaries are

not uniformly sized, a different chance of finding shops within a particular

121

administrative boundary may exist as in relation to its size as well as its

socioeconomic characteristics.

This study aims to overcome these inherent limitations by using a different

methodology in which the local food environment is measured on a larger spatial

scale than the disadvantaged and advantaged areas of interest, looking at nearby

shops rather than just shops falling within administrative boundaries. While

analytically challenging, this approach more realistically represents the area in which

residents are likely to shop and therefore also more realistically captures whether

living in a socioeconomically disadvantaged area means fewer opportunities to

purchase fruits and vegetables.

It is important to distinguish between shop types when considering the number of

shops in areas, as the number and types of shops available may have a carry-over

effect to the price and availability of foods available to residents. Foods are

generally less expensive in large supermarkets than in smaller grocery stores (187).

In studies of food price and availability, the type of shop surveyed strongly relates to

the prices and availability of foods within shops (193) and the composition of local

shops, specifically having only one small food outlet, strongly relates to food being

relatively expensive in some areas (207). Consequently the classification of shops

needs to relate meaningfully to the types of foods likely to be sold, and to some

extent shop size, which might influence product ranges and prices. Accordingly, to

use meaningful shop categories, and minimise misclassification, this study classifies

shops by direct observation with a validated tool, using categories specifically

designed to reflect shops’ main products and activities and their size.

4.3. Methods

4.3.1. Setting

The study was conducted in 2000 in the Brisbane City Statistical Subdivision (SSD)

which covered approximately 1 200 km2 and had a population of approximately 806

800 at the most recent census prior to the study (190). When the study was

122

conducted, two-thirds of the Australian population lived in the state capital cities

(339), and Brisbane was the third most populous of these (336).

4.3.2. Sampling of areas

Areas of varying socioeconomic disadvantage

Census collection districts (CDs) are the smallest administrative unit of data

collection used by the Australian Bureau of Statistics, similar to a census tract in the

US or an enumeration district in the UK. They contain an average of 200

households, are socioeconomically homogeneous and cover varying spatial areas

(190). A stratified random sample of 50 CDs was selected from the 1517 CDs in the

Brisbane SSD which had pertinent census data available. These CDs are

representative of residential Brisbane, as the CDs without relevant census data were

areas with few people enumerated in private dwellings, being either industrial areas

with few residents, or areas with relatively large numbers of people enumerated in

non-private dwellings (often hotels, hospitals, army barracks and student

accommodation). CDs were stratified into deciles of socioeconomic disadvantage,

and within each decile, 5 CDs were sampled randomly without replacement, using

proportional to size sampling.

Shopping catchments

In order to overcome some limitations of only addressing shops within administrative

boundaries (such as areas having unequal sizes and little relevance to residents’

shopping habits), this study instead addressed shops ‘nearby’ to administrative

boundaries. ‘Shopping catchments’ were created as the area within a 2.5 km radius

of the centroid of the sampled CDs to represent the area where residents of sampled

CDs were likely to shop. When this paper refers to local shops, or nearby shops, it

refers to shops falling within the surrounding shopping catchments. Figure 4.1

shows the 50 sampled CDs across Brisbane, and one of these CDs with its

surrounding shopping catchment and local shops.

123

Figure 4.1: a) (left) Sampled Census Collection Districts (CDs) across the Brisbane Statistical Subdivision; b) (right) One sampled CD with its

surrounding shopping catchment and local shops

124

4.3.3. Data collection

Population and socioeconomic data were obtained using data from the 1996

Australian Census, from Socioeconomic Indexes for Areas 1996 (190).

Shop data

As part of an earlier pilot study (307), an eight category shop classification system

was developed based on government standards and shop classifications, broadly

relevant literature, and an expert panel including representatives in housing and urban

research, environmental health and local council licensing. The shop classifications

and their definitions are based on size, primary activity and merchandise, and are

presented in Table 4.1. During the pilot study, three data collectors used verbal and

written instructions to classify shops by these definitions on two occasions ten weeks

apart. This method of classifying shops showed a high inter-rater reliability

(kappa=0.81, se=0.04) and test-retest reliability (kappa=0.73, se=0.11), and was

retained for the main study.

Data pertaining to shops were chiefly obtained through an audit of the shopping

catchments which was conducted between July and October 2000. The audit process

involved initially identifying shops whose addresses placed them inside, or close to

catchment boundaries, visiting these shops and completing an observation worksheet.

Shops were identified and included in the study one of three ways. Shops were

initially identified through Brisbane City Council licensing lists and maps, which

proved to be incomplete. Unlisted shops that were included in this study were either

identified by data collectors during visits to listed shops, or through a later phase of

the Brisbane Food Study (89) in which respondents to a household survey indicated

where they usually shopped. The twenty-five shops which were identified by

respondents to the Brisbane Food Study Survey did not follow the complete audit

process, and lacked all data collected in the audit. These shops were verified to exist

through telephone directory listings, and included in the study with the listed address

and classification given by respondents. The shops that were on the initial listings or

noticed by data collectors were audited, and data were collected on the shop’s type,

125

exact address, and opening hours. The opening hours were either observed from

signage, or obtained from report by shop employees.

Table 4.1: Categories and definition of food outlets

Categories Operational Definitions Sub-categories and examples

Conventional Food Shop Major (CFS Major)

Mainly engaged in the sale of groceries (fresh foods, canned and packaged foods, dry goods) of non-specialised (conventional) food lines. May contain a butcher or baker. Usually have 5 or more checkouts and a floor area over 1000 square metres.

Woolworths Coles Bi-Lo Franklins (No Frills)

Conventional Food Shop Minor (CFS Minor)

Mainly engaged in the sale of groceries (fresh foods, canned and packaged foods, dry goods) of non-specialised (conventional) food lines. Usually have 4 or fewer checkouts and a floor area under 1000 square metres.

Seven Eleven, 727, Food Store, 4 Square, Night Owl, Petrol station that has a food shop or convenience store component. Independent corner shop, grocer or independent convenience store.

Meat and Fish Shop

Mainly engaged in the sale of fresh meat, fresh poultry, fresh fish, seafood and processed meat.

Conventional butchers, shops that exclusively stock fresh poultry, and fresh seafood shops.

Fruiterer and Greengrocer

Mainly engaged in the sale of fresh fruit and vegetables.

No sub-categories.

Baker and Cake Shop

Mainly engaged in the sale of bread, biscuits, cakes, pastries or other flour products with or without packaging.

Bakeries (eg Brumby’s, Bakers’ Delight) and shops that are mostly oriented towards the sale of cakes and pastries.

Takeaway Mainly engaged in the preparation and sale of meals or light refreshments that are ready for immediate consumption. Table service is not provided and the meal can be eaten on site or taken away.

Conventional takeaway fast food retailing (including McDonald’s, Hungry Jacks, KFC, Red Rooster, Fish & Chips, Pizza) and takeaway retailing in cut lunches, ice cream, milk or soft drinks. Takeaway located in a food court (sharing the same dining area).

Specialty Food Shop

Mainly engaged in the sale of groceries (fresh foods, canned and packaged foods, dry goods) of specialised/ethnic food lines, or mixed specialised lines.

Oriented towards the sale of ethnic food (eg. Asian, Vietnamese, Greek, Italian) Oriented towards the sale of health food (eg. health food stores, natural food stores, pharmacies that have a health food store component) Delicatessens and fine food stores (delicatessens, fine & imported food store) Food Halls (mixed specialties, if have counters that are dependent in a department store, shopping centre or mall).

Other Food Shop

Mainly engaged in the sale of food not elsewhere described.

Candy, nut and confectionary shops, tea and coffee shops, spice and herb shops.

126

4.3.4. Measures

4.3.4.1. Independent variables

Socioeconomic disadvantage Area level socioeconomic disadvantage was measured by Index of Relative

Socioeconomic Disadvantage (IRSD) scores of the sampled CDs. The IRSD is a

composite index that reflects the level of socioeconomic disadvantage in an area,

similar to the Townsend Index of Deprivation (340) or the Carstairs index (341). In

contrast to these other composite indices, the IRSD does not focus exclusively on

material deprivation. The ABS used principal components analysis to construct the

index scores from variables which relate to social and economic disadvantage of an

area’s residents, such as proportions of low income families, one parent families with

dependents, and proportions of people who are Indigenous, unemployed, lack

fluency in English or have relatively unskilled occupations (190). For analysis, the

scores were used to categorise the CDs into approximate tertiles which are referred to

as socioeconomically disadvantaged (n=17), medium (n=16) and advantaged (n=17)

areas.

Catchment population (density) Previous studies indicate that population density influences shop location, and may

relate to area socioeconomic characteristics (170, 188). Catchment population was

measured in lieu of catchment population density since all catchments had the same

spatial area (19.6km2). The population of a catchment was estimated by the sum of

the populations of the CDs enclosed within the catchment. Populations were

weighted by the fraction of the CD area falling inside the catchment area, such that

half of the population of a CD would be included if half of the CD lay within the

catchment.

Dependent variables

Number of shops To ensure an accurate count of shops within a 2.5km shopping catchment, shops

were geo-coded to their street address using MapInfo Professional 6.5 (342), and

those falling outside shopping catchment boundaries were excluded from the study.

Only those types of shops that were likely to sell fresh fruits and vegetables by virtue

of their definition were included. These were shops that fit the classification of

127

fruiterers/greengrocers, major conventional food stores, or minor conventional food

stores, which hereafter are referred to as greengrocers, supermarkets, and

convenience stores respectively.

Distance to the nearest shop Distances to the nearest supermarket, greengrocer, and convenience store were

measured as the straight line distance between the centroid of the CD and the

centroid of the nearest shop address, using MapInfo Professional 6.5.

Shop opening hours Shop opening hours were measured categorically as ‘closed’, or open ‘half a day (or

less)’, ‘all day’ (i.e. more than half a day but closing at 6pm or earlier), ‘all day and

closing after 6pm (but before 9pm)’, or ‘all day and closing after 9pm’ for each day

of the week.

4.3.5. Analysis

All statistical analyses were performed using SAS version 8 (343). Shops that were

located in more than one shopping catchment were included in analyses each time

they appeared5. Averages and spread are reported as medians and ranges in lieu of

means and standard deviations since shop counts, distances, and catchment

populations were not normally distributed.

Number of shops

Poisson regression was used to determine whether disadvantaged, medium and

advantaged areas had different numbers of nearby shops. To meet analytic

assumptions, supermarkets and greengrocers were pooled, while convenience stores

were counted separately. Since supermarkets and greengrocers are likely to be

similar in terms of the fruits and vegetables they sell, the information lost by pooling

is minimal. Where data were overdispersed relative to the Poisson distribution, a 5 (1) Since sampled CDs could be less than 2.5 km apart, sometimes shops were located in shopping catchments of more than one CD. If each shop were counted once, there would be 64 supermarkets, 81 greengrocers and 349 convenience stores in the study. However, the sum total of the number of shops in all catchments is 164 supermarkets, 196 greengrocers and 816 convenience stores.

128

scale factor was included to appropriately increase the standard error (estimated by √

Pearson’s chi-square/df). To be conservative, a scale factor was not included where

data were less dispersed than the Poisson distribution. Earlier bi-variate analyses

(not reported) had revealed population density to be highest around medium

socioeconomic areas, and to relate to the number of shops in catchments.

Consequently, medium socioeconomic areas could potentially have more nearby

shops as a function of population density rather than their socioeconomic

characteristics, so a second model was tested which included catchment population

as a covariate, to assess the mutually unconfounded effects of socioeconomic

disadvantage and population density, without modelling rates per se (such as by

including population as an offset term).

Distance to the nearest shop

Distances to the nearest shop were not normally distributed, and their distributions

were only comparable across disadvantaged, medium and advantaged areas for

supermarkets. The Kruskal-Wallis test was used to determine whether distance to

the nearest supermarket was different between the tertiles of socioeconomic

disadvantage. In absence of valid statistical tests, boxplots were used to compare

other distances ‘qualitatively’. CDs which did not have a particular shop in their

shopping catchments were excluded from analysis.

Opening hours

The association between socioeconomic disadvantage and the opening hours of

nearby shops was examined using logistic regression, with advantaged areas as the

reference group. Where more than two response categories existed, ordinal logistic

regression was used. Analyses were directioned to compare the odds of opening

short versus variable numbers of longer hours. (Score tests indicated no significant

disproportion of odds which would invalidate such comparisons.) To obtain stable

estimates, some categories were collapsed for analysis when response numbers were

small, and one analysis was reversed. (Because most supermarkets closed on

Sundays, the odds of supermarkets opening at all on Sundays were analysed.) Since

129

patterns were different for each shop type, supermarkets, greengrocers, and

convenience stores were analysed separately. Since results for Monday, Tuesday,

Wednesday and Friday were almost identical, logistic regression estimates are only

presented for Wednesday and Thursday, Saturday and Sunday. Nearly all

supermarkets opened the same hours through the week, so logistic regression was

restricted to Saturdays and Sundays.

4.4. Results

4.4.1. Number of shops

While all areas studied had at least one shop nearby, one disadvantaged and one

advantaged area had neither a supermarket, nor a greengrocer nearby, and another

disadvantaged area lacked a nearby supermarket. Table 4.2 presents the average

numbers of each shop type in socioeconomically disadvantaged, medium and

advantaged areas. Medium socioeconomic areas tended to have more shops than the

other areas. Table 4.3 presents the crude and adjusted relative rates of nearby

convenience stores, and greengrocers and supermarkets combined, for advantaged

and medium relative to disadvantaged areas. Relative rates indicate that numbers of

nearby shops for advantaged and disadvantaged areas were similar both before and

after the effect of population density was considered. The larger relative rates of

supermarkets and greengrocers, and convenience stores near medium socioeconomic

areas were at the border of statistical significance but related largely to the higher

population density in these areas, as these relative rates of shops became smaller

after adjusting for population density. Confidence intervals indicate that true

differences between the areas in either direction are likely to be quite small.

130

Table 4.2: Selected Characteristics of Disadvantaged, Medium and Advantaged areas

Disadvantaged (n=17) Median (min, max)

Medium (n=16) Median (min, max)

Advantaged (n=17) Median (min, max)

Catchment population 30237

(8122 , 46659) 35638

(22204, 45622) 32217

(14293, 46718)

Distance to CBD (km) 9.82

(4.05, 17.21) 9.59

(1.35, 15.91) 8.79

(2.89, 14.67) Number of supermarkets 3 (0, 7) 4 (2, 7) 3 (0, 5) Number of greengrocers 4 (0, 6) 4 (2, 9) 4 (0, 7) Number of convenience stores 12 (3, 33) 18.5 (7, 34) 14 (3, 34)

Nearest supermarket (km)a 0.48

(0.17, 1.72) 0.56

(0.65, 2.40) 0.55

(0.42, 1.96)

Nearest greengrocer (km)a 0.57

(0.17, 2.13) 0.47

(0.51, 2.07) 0.55

(0.51, 2.00) Nearest convenience store (km)a

0.42 (0.16, 1.59)

0.19 (0.16, 0.89)

0.41 (0.42, 1.76)

a Excluding areas where the nearest shop fell outside the 2.5km study area. Excluded areas: 2 disadvantaged, 1 advantaged (supermarkets); 1 disadvantaged, 1 advantaged (greengrocers); 0 (convenience stores)

Table 4.3: Crude and adjusted relative rates of shops by socioeconomic disadvantage

Supermarkets and Greengrocers

Convenience Stores

Crude RR (95% CI)

Adjusteda RR (95% CI)

Crude RR (95%

CI)

Adjusteda RR (95%

CI)

Disadvantaged 1.13

(0.83, 1.54) 1.13

(0.87, 1.46) 1.18

(0.80, 1.75) 0.97

(0.79, 1.20)

Medium 1.35

(1.00 , 1.82) 0.99

(0.76, 1.30) 1.45

(0.99, 2.12) 1.14

(0.93, 1.40)

Advantaged 1.00 (referent) 1.00 (referent) 1.00

(referent) 1.00

(referent) Population (‘000s) -

1.03 (1.02 to 1.04) -

1.05 (1.04 to 1.06)

a Adjusted for the effect of population density

131

4.4.2. Opening hours

Table 4.4 presents daily shop opening hours for convenience stores, greengrocers and

supermarkets. Shops were more likely to close on Sunday than Saturday, and usually

opened longer hours through the week than on the weekend. Convenience stores

tended to open longest. During the week and on Saturdays, supermarkets tended to

open longer than greengrocers, while on Sundays the opposite was true 6.

Table 4.4: Dailya opening hours of greengrocers, convenience stores and

supermarkets

Wednesday Thursday Saturday Sunday Greengrocers n=81

Closed / open ½ day 0 0 12 34 Open all day 41 27 40 18 Open all day and after 6pm/ 9pm 26 41 16 16

Convenience Stores n=349 Closed / open ½ day 0 0 11 22 Open all day 22 20 29 22 Open all day and after 6pm 167 169 149 148 Open all day and after 9pm 144 144 144 142

Supermarkets n=64 Closed 0 0 0 55 Open ½ day 0 0 0 1 Open all day 1 0 55 1 Open all day and after 6pm 58 59 4 2 Open all day and after 9pm 0 0 0 0 a Opening hours for Monday, Tuesday and Friday are not shown as these did not vary substantially to opening hours presented for Wednesday.

Table 4.5 presents odds ratios and 95% confidence intervals for the odds of local

shops opening shorter hours according to area level socioeconomic disadvantage.

Generally speaking, odds ratios indicated that the chance of local shops being

relatively unavailable to the public was similar for disadvantaged, medium

socioeconomic and advantaged areas. Medium socioeconomic areas showed some

tendency to have reduced opening hours, however confidence intervals were usually

wide, and none of the relationships were statistically significant.

6 Throughout Queensland at the time of the study, trading on Sundays was restricted for large but not smaller or independent food stores: only supermarkets in designated tourist zones were permitted to trade. As of August 2001, unrestricted Sunday trading was introduced.

132

Table 4.5: Odds ratios for shops opening ‘short’ hours for areas of varying

socioeconomic disadvantage

Disadvantaged Areasa Medium Socioeconomic Areasa OR 95% CI OR 95% CI

Supermarkets (n=153) Saturday 0.80 0.13 to 5.03 2.11 0.52 to 8.64 Sundayb 0.81 0.16 to 4.25 0.41 0.10 to 1.62

Greengrocers (n=171) Wednesday 1.05 0.49 to 2.27 1.01 0.49 to 2.11 Thursday 1.02 0.47 to 2.21 0.69 0.32 to 1.47 Saturday 1.00 0.54 to 1.88 0.99 0.36 to 2.55 Sunday 0.82 0.44 to 1.51 0.55 0.21 to 1.45

Convenience Stores (n=768) Wednesday 0.97 0.68 to 1.37 1.23 0.92 to 1.83 Saturday 0.97 0.69 to 1.37 1.24 0.89 to 1.73 Sunday 0.91 0.65 to 1.28 1.17 0.83 to 1.63 a Compared with advantaged areas b Compared the odds of being Supermarkets being open at all versus closed due to the low numbers of open supermarkets. These odds ratios are in opposite direction than for other comparisons.

4.4.3. Distance to nearest shop

The nearest supermarket tended to be farther from CDs than the nearest greengrocer,

and both of these types of shops tended to be farther from CDs than the nearest

convenience store. In all areas, a convenience store could be found within 2.5km.

Figure 4.2 presents boxplots of distances from disadvantaged, medium

socioeconomic and advantaged areas to the nearest supermarket, greengrocer and

convenience store. Average, minimum and maximum distances to shops have been

reported in Table 4.2. Different types of shops showed different patterns with

socioeconomic disadvantage. Distances to the nearest supermarket were

significantly different (χ2=6.98 df=2 p=0.03) across areas, on average being closest

to disadvantaged areas and farthest on average from medium socioeconomic areas.

However, all medium socioeconomic areas had a supermarket within 2.5km unlike

disadvantaged and advantaged areas. On average, the nearest greengrocer was also

closest to disadvantaged areas. The nearest convenience store was, on average,

approximately half a kilometre from areas of all levels of disadvantage, however

the spread of distances was much less for medium compared with disadvantaged and

advantaged areas (interquartile ranges 0.21, 0.39, 0.62 km respectively).

133

Figure 4.2: Distance from disadvantaged, medium socioeconomic, and advantaged

census collectors’ districts (CDs) to the nearest shop (km) a)(top) supermarkets; b)

(middle) greengrocers; c) (bottom) convenience stores

(CDs without a particular shop within the 2.5 km radius studied were excluded from

these figures)

134

4.5. Discussion

This study finds that while provision of fruit and vegetable retail outlets is not always

ideal in terms of large numbers, long opening hours or close proximity, these factors

are generally not systematically related to area socioeconomic characteristics. In

Brisbane (and perhaps other Australian cities), it is unlikely that living in a

socioeconomically disadvantaged urban area means having fewer opportunities to

purchase fruits and vegetables, however this does not necessarily mean that

individual socioeconomic differences in diet are not influenced by environmental

characteristics. In fact, the overseas studies provide some support for to the idea that

individual characteristics influence the degree to which the environment impacts on

behaviour, as the presence of a supermarket had the most noticeable positive effect

on the diets of African Americans, who were the most likely not to own a private

motor vehicle (170). Also, this study looked at retail outlets themselves but can

make no conclusions regarding the price, availability or quality of fruits and

vegetables within shops in areas of varying socioeconomic disadvantage.

4.5.1. Number of shops

The marginally higher number of shops surrounding medium socioeconomic areas

related strongly to these areas having higher populations for shops to service.

Whether the observed association is spurious, or reflects a general association

between population density and socioeconomic characteristics, is unknown. If the

relationship is simply an unintended by-product of sampling, then it can be inferred

that residents of medium socioeconomic areas do not have more shops in their local

areas. However, if the relationship reflects a more general pattern, then it could be

inferred that residents in medium socioeconomic areas may have slightly more

access to shops than residents of other areas since their areas are more densely

populated. The sample’s randomness supports the latter interpretation while its small

135

size suggests the former. In either case, differences were small, and did not reflect a

graded association between deprivation and availability of local shops.

These null findings may be genuine, or may relate to some important methodological

matters that must be considered. While the sampled CDs were independent

observational units, their midpoints were not always 5 km apart, and consequently

their shopping catchments sometimes overlapped. Areas with overlapping shopping

catchments are more likely to have similar numbers of local shops than are non-

overlapping areas, because a fraction of the shops in the catchments might be

common. Ignoring this is likely to result in confidence intervals that are narrower

than they ought to be, which has two important consequences: statistically significant

differences may still have resulted from sampling error (type I error); and in this case

of null findings, true differences may be more extreme than confidence intervals

indicate. Since only fifty areas were studied, the possibility of small differences

cannot be ruled out, as the study’s power would be insufficient to detect small

differences. Other studies have used much larger samples (170, 180, 187, 188, 333,

337). Based on current knowledge, it is uncertain whether small environmental

differences could produce large differences in the way residents purchase or consume

foods.

4.5.2. Opening hours

The overlapping shopping catchments has the same impact on interpreting the null

findings with opening hours as with shop numbers. Additionally, it must be

considered that the broad categorical measurement prohibited picking up differences

which could be meaningful to the way residents shop. A shop that closed at 8:45 pm

would be given the same classification as one that closed at 6:15 pm, yet the ability

of an individual to shop in each, say, after work, would be very different.

Additionally, the removal of many restrictions on retail trading hours that occurred in

Queensland in August 2001 may have substantially altered trading hours and any

socioeconomic patterning in them.

136

4.5.3. Distance to nearest shop

There are important concerns with treating distances in an absolute sense that are less

important when comparing distances between areas in a relative sense. Firstly, the

straight line distance from the CD centroid to its nearest shop does not exactly

represent the average distance from all CD residents to their nearest shops. When

multiple shops lie in or near CDs, different shops will be nearer to residents in

different parts of the CD. Allowing only one ‘nearest shop’ for the CD therefore

slightly exaggerates the average distance to the nearest shop. Secondly, straight line

distances are considerably shorter than road distances or travel times, which would

more accurately reflect the proximity of local shops to residents. Overall, the

distances measured in this study likely underestimate the true distances that residents

would need to travel to reach shops, but probably accurately reflect the relative

differences in distances to shops across areas varying in socioeconomic

characteristics.

Donkin et al (159) described 2km as a reasonable walking distance in relation to food

access, however no official definition exists. Considering the return trip, loaded with

enough groceries and domestic items to supply a family for a week or fortnight,

distances to the nearest greengrocer and supermarket found in this study probably

necessitate the use of a private motor vehicle, taxi or nearby public transport in many

areas of Brisbane to shop for food. If multiple round trips to shops were required,

this would impose time constraints. Literature has shown that socioeconomically

disadvantaged individuals are more likely than the socioeconomically advantaged to

lack a private motor vehicle or to have fewer vehicles in their household (160, 161),

and compounding this problem they also have the least disposable income with

which to pay for pubic transport, a taxi fare, or home delivery (158). Shops being of

similar proximity or closer to disadvantaged areas therefore does not equate to

socioeconomically disadvantaged individuals having equal or better access to fruit

and vegetable outlets.

Juxtaposing the results between the number of shops in catchments, and the distance

between CDs and their nearest shop highlights an interesting issue. If shops were

137

scattered evenly around the shopping catchments, then it would be expected the

closest shop would tend to be nearer when there are more local shops than when

there few local shops. However, the opposite tended to be the case. On average,

medium socioeconomic areas had the most local supermarkets and greengrocers, and

yet were farthest from supermarkets, and were farther from greengrocers than

disadvantaged CDs. Since shops were not evenly distributed within shopping

catchments, findings are likely to depend on the size of the area chosen to represent

‘nearby’ shops.

4.5.4. Overseas contrasts – methodological explanations

The findings of this study differ from those of overseas research, perhaps as a result

of true differences, or differences in methodology. Our findings cannot be compared

with those of Guy et al. (179), because their study amalgamates issues of population,

proximity, and shop attractiveness into the one measure of accessibility. Although

most studies (170, 180, 187, 193, 333, 337) controlled for population size or density,

none of the cited studies controlled for the unequal sizes of administrative

boundaries. For example one study from the U.S (170) found fewer shops in

socioeconomically disadvantaged neighbourhoods (which averaged eight square

kilometres in size) compared with advantaged neighbourhoods (which averaged 20

square kilometres in size). A study in Melbourne, Australia (337) found fewer

takeaway shops in high income administrative areas, which census data indicates

tend to be much smaller areas located in the densely populated areas around the inner

city (344). Other studies did not report size, however, U.S authors have noted a

tendency of some cities where poor neighbourhoods cluster around the inner city and

affluent neighbourhoods migrate towards the outer suburbs (186). Thus the use of

administrative areas as unequally sized neighbourhoods may lead to finding more or

fewer shops in poor neighbourhoods because their neighbourhood sizes do not allow

for fair comparison, as well as for socioeconomic reasons. This study used a

standardized area in which to count shops. Unlike overseas research, our study also

used different spatial scales to measure socioeconomic characteristics and

surrounding shops to account for the possibility that residents of particular CDs

138

shopped outside their CDs. What occurs on one spatial scale may not be directly

comparable to what happens on a different scale.

4.5.5. Overseas contrasts – social/economic explanations

Like the findings from the UK (180), this study did not support the claim that

socioeconomically disadvantaged areas present fewer opportunities to purchase fresh

fruits and vegetables, in direct contrast with the findings from the US (170, 187,

333). Non-methodological explanations can be made for differences between the

Australian and US findings regarding socioeconomic differences in shop provision,

and in diet. Firstly, studies have been confined to a small number of cities, and

findings may not be generalisable to all cities in a particular country, however it is

likely that findings are more similar within countries than between them.

In much of the United States, there has been a historical exodus of the upper

socioeconomic segment of society from inner urban areas into outer suburban areas,

occurring with the initial development of supermarkets in higher socioeconomic

areas. A process dubbed ‘supermarket redlining’ has been argued to explain why

poor urban areas lack large supermarkets, in which developers of supermarkets

choose to avoid poor urban areas based on a belief that locating there will not be an

economically viable investment (186). Conversely, in Australia, recent years have

seen an opposite process of ‘gentrification’ of the inner city areas, with an increase of

socioeconomically advantaged individuals moving to inner city areas. A move to

capture this section of the market by developing supermarkets in the inner city,

specifically catering to the perceived needs of this market has followed (345). As a

result spatial patterning of shops and socioeconomic characteristics are possibly

different in some areas of Australia and the United States. Notably, Guy et al. (179)

did not observe evidence of ‘supermarket redlining’ in their study in the UK in the

opening of new stores, although they observed that store closure occurred more

commonly in deprived areas. It has been argued that perhaps in Australia, unlike the

US and the UK, the nature and extent of the spatial segregation along social and

economic lines is not large enough to be detectable in people’s dietary behaviours

(116). Area level associations between socioeconomic characteristics and ‘diet’ are

139

observed where area level associations with provision of food retail outlets, or food

price and availability are observed. In Australia, it is possible that access to a

relatively equal shopping infrastructure assists in minimising socioeconomic

inequalities in diet.

4.6. Acknowledgements

We gratefully acknowledge the work of the Brisbane Food Study Project Team, and

Dr Diana Battistutta for feedback, support and statistical advice. Dr Turrell is

supported by a National Health and Medical Research Council/National Heart

Foundation Career Development Award (CR 013 0502).

140

Chapter 5: Does living in a disadvantaged area entail limited opportunities to purchase fresh fruit and vegetables in terms of price, availability, and variety? Findings from the Brisbane Food Study7

7 This manuscript has been published in Health and Place journal 2006; 12(4):741-748

141

5.1. Abstract

Understanding the role environmental factors may play in the dietary behaviours

of socioeconomic groups is relevant for efforts to reduce health inequalities. In

contrast with international research, earlier findings from the Brisbane Food Study

(BFS), Australia, found no relationship between area socioeconomic

characteristics and dietary behaviours or location of food shops. This paper

examines whether the price and availability of fruits and vegetables are

socioeconomically patterned using data from the BFS. Fifty census collection

districts were randomly sampled and all local (i.e. within 2.5km) supermarkets,

greengrocers and convenience stores were observed. Little or no differences in

price and availability were found on the basis of area socioeconomic

characteristics.

5.2. Introduction

Socioeconomic groups differ in their purchasing behaviours and consumption

patterns for fruits and vegetables, with disadvantaged groups being least likely to

have intakes that are consistent with healthy eating messages (89, 100, 105).

These dietary differences for fruits and vegetables are believed to contribute to

socioeconomic differences in mortality and morbidity for chronic disease (20, 21),

yet surprisingly, we know very little about why the dietary differences exist.

Recently, researchers in the US (111), UK (114) and elsewhere have focused on

the characteristics of the neighbourhood environment as a determinant of

socioeconomic differences in diet. This work shows that residents of

socioeconomically disadvantaged areas have poorer diets after adjustment for

individual-level socioeconomic position, suggesting that aspects of disadvantaged

neighbourhoods may act to hinder the procurement of healthy food. By contrast, a

multilevel study in Brisbane City (Australia) found a strong association between

household income and the purchase of fruits and vegetables but little evidence that

the purchase of these foods was influenced by neighbourhood-level factors,

142

implying that urban areas in Brisbane are not highly differentiated on the basis of

the availability and price of fruits and vegetables (116). This possibility was

supported by later research from the Brisbane Food Study (BFS) which showed

that socioeconomic areas varied little in terms of the number and type of shops

that sold fruits and vegetables (47). Brisbane, therefore, seemingly represents an

interesting counterpoint to that found in urban areas of other western countries:

strong associations between individual-level socioeconomic factors and the

purchase of fruits and vegetables exist even in the context of a food shopping

infrastructure that is not spatially divided along social and economic lines.

We would not necessarily expect any socioeconomic patterning of the price and

availability of fruits and vegetables in Australian cities to conform to patterns in

the U.S or U.K.. Although underlying issues of “supply and demand” are

common, important aspects of the spatial and socioeconomic patterning of food

retailing are country-specific. Across many U.S cities, an historical out-migration

of higher socioeconomic residential groups from the inner cities to the suburbs has

occurred (186), whereas in Australia, the reverse effect has been observed, with

wealthier and more educated social groups becoming increasingly attracted to

inner city living (345). Correspondingly, many cities in the US have experienced

a ‘redlining’ process whereby developers avoid placing supermarkets in low-

income inner city areas (345), while Australia has seen the development of

supermarkets in the inner city as part of a move to capture the upper

socioeconomic market (345). In the U.K., Guy et al. (179)observed store closures

occur more commonly in deprived areas, but did not observe evidence of

‘supermarket redlining’.

In this study we extend earlier work from the BFS by comparing

socioeconomically distinct urban areas in terms of whether fruits and vegetables

in the shops in these areas differ in availability and price. The overseas evidence

on these issues is both sparse and inconsistent. A case-comparison study in

Glasgow found that of two areas compared, food was less readily available in the

more deprived area (346). This same study also reported that a basket of ‘healthy

foods’ was more expensive in the deprived area, however, baskets of fruits and

vegetables were equally priced between the areas, although of lower quality in the

143

more disadvantaged area. Cummins and Macintyre (193) reported that some foods

were more expensive and others cheaper in disadvantaged areas of Glasgow,

while fruits and vegetables were similarly priced overall. Foods tended to be

equally available or less available in disadvantaged areas, although differences in

availability tended to exist for meat rather than fruits and vegetables or other

items. White et al. (181) reported fruits and vegetables to be significantly more

expensive in affluent areas of Newcastle. US research has shown that food prices

are lower in supermarkets (which are more often located in socioeconomically

advantaged areas) than in small stores (187, 188). A study of US consumers (202)

showed among urban residents, those on low incomes paid higher prices for food

than their higher income counterparts, but not among suburban residents.

No known Australian study to date has compared socioeconomically different

urban areas in terms of the availability and price of fruits and vegetables: the

major focus has been rural-urban comparisons (58, 335). However, given that

previous studies from the BFS have failed to find area-level differences in the

purchasing of these foods and the number and type of local food shops, we would

expect that socioeconomic areas in Brisbane are not greatly differentiated in terms

of the availability, variety, and price of fruits and vegetables.

5.3. Methods

5.3.1. Geographical coverage and sampling of areas

The BFS was conducted between June and December 2000 in the Brisbane City

Statistical Sub-Division (details published elsewhere) (47, 89, 116). The primary

area sampling unit was the Census Collection District (CCD), the smallest spatial

unit in the Australian Standard Geographical Classification which are relatively

socioeconomically homogeneous areas containing an average of 220 households

(190). To obtain a sample representing the whole spectrum of socioeconomic

characteristics, CCDs were selected using a stratified cluster design which

consisted of grouping the CCD into deciles based on each area’s Index of Relative

Socioeconomic Disadvantage (IRSD) score. The IRSD is constructed by the

144

Australian Bureau of Statistics and is derived from attributes such as low income,

low educational attainment, high levels of public sector housing, and high

unemployment (190), with higher values indicating lower levels of disadvantage.

Five CCDs were randomly selected from each decile using proportional to size

sampling (347) to yield a total sample of n=50 areas according to sample size

requirements for the multilevel components of the BFS (348).

5.3.2. Shopping catchments and data collection In order to examine the environmental characteristics of areas in terms of how

they constrained or facilitated the procurement of fruits and vegetables, we

devised a spatial region known as a ‘shopping catchment’ which comprised a

circular area of 2.5km radius (~19.6km2) that emanated from the geographic

centroid of each CCD. The exact spatial level at which effects on food purchasing

operate is not known, however the arbitrarily chosen 2.5km distance is similar to

the distances people typically travelled to stores by car, public transport, or taxi in

the Newcastle study (181), and the distance Donkin et al. (207) describe as a

reasonable walking distance. Using a combination of small CCDs and larger

shopping catchments has advantages over using administrative boundaries only,

as catchments are evenly sized and approximately symmetric around all residents,

not just residents living near the centre of an administrative area. From within

each shopping catchment, trained data collectors recorded detailed information

about the availability, variety, and price of ten fruits and ten vegetables from

nearly all local supermarkets, greengrocers or convenience stores (other than

small convenience stores attached to petrol stations). All eligible shops which

were identified and operational at this phase of the study were sampled (n=325),

then permission was obtained and data collected for 94% of sampled shops.

Shop-types were identified and defined using a valid and reliable classification

tool described elsewhere (47). The ten fruits and vegetables chosen were the most

commonly consumed according to an Index developed by a major Australian food

retailer (349).

145

5.3.3. Measurement Area-level socioeconomic disadvantage of the CCDs was measured using its

IRSD score (see above). The distribution of IRSD scores was divided into tertiles

which are referred to as low (n=17), medium (n=16) and high socioeconomic

areas (n=17).

Preliminary bivariate analysis indicated that the availability, variety, and price of

fruits and vegetables differed for supermarket, greengrocer, and convenience

stores hence we examine each of the food-outcomes separately for each shop-

type. Availability: the availability of each fruit and vegetable was scored as 0=

‘not available’ or 1= ‘available’. The scores were then summed to derive a

distribution that ranged from 0-10, and for the same shop-type within each

catchment a mean score was calculated as an estimate of the availability of fruit

and vegetables in the area. Variety: variety was originally recorded as ‘not

available’, ‘one variety’, ‘2 or 3 varieties’ or ‘4 varieties or more’. These

categories were subsequently rescored as 0, 1, 2, or 3 respectively, then summed

(theoretical range 0-30 for the ten fruits or vegetables) and a mean score

calculated for each shop-type within the same catchment. Price: Prices were

recorded per kilogram or per item for predesignated sizes and varieties of fruits

and vegetables. Prices that were recorded on a per item basis were divided by

their typical weight (in kilograms) to achieve a common unit (AUS$/kg). Typical

weights were obtained from FoodWorks Professional, a software package based

on comprehensive Australian food composition data (350). We calculated

average prices in each catchment for each item in supermarkets, greengrocers,

convenience stores, and in all local shops. Then, two theoretical food baskets

were derived: one containing ten fruits and the other ten vegetables. The baskets

contained sufficient quantities of fruits and vegetables to meet the minimum

recommendations for a healthy diet for one adult for a fortnight (or two adults for

a week) (51) allowing extra weight for non-edible portions. A similar type of

basket comprising seven fruits was compiled specifically for convenience stores

(due to limited stocking of fruits in these shops). Table 5.1 shows the contents of

the theoretical baskets. The prices of these fruit and vegetable baskets in each

catchment were calculated (separately for each type of shop) using the average

prices ($/kg) of each item.

146

Table 5.1: Hypothetical baskets of widely consumed fruits and vegetables a

Vegetable Basket b

Fruit Basket b 7-Fruit Basket b,c

2 Cauliflower (medium)

½ Rockmelon

1 ½ Rockmelons

2 Broccoli (medium) 1 Pineapple - 2 Cabbage (medium) ¼ Watermelon -

½ Pumpkin ½ Honeydew Melon - 1 kg Carrots 4 Kiwifruit 8 Kiwifruit 1 kg Potatoes ½ kg Apples ¾ kg Apples 1 kg Onions ½ kg Oranges ¾ kg Oranges

1 kg Capsicum ½ kg Bananas ¾ kg Bananas 1 kg Beans ½ kg Nectarines ¾ kg Nectarines

1 kg Tomato ½ kg Pears ¾ kg Pears

Total = 11.9kg

Total = 5.1kg

Total = 5.2 kg a Most common vegetables and fruits according to Coles Fruit and

Vegetable Index 1998 (Coles Supermarkets Pty Ltd., 1998) b Sufficient to meet minimum requirements for 1 adult for 2 weeks, or 2

adults (19yrs+) for 1 week as set out in Australian Guide to Healthy Eating (10.5kg raw edible weight of vegetables and 4.2 kg raw edible weight of fruit) (Kellett et al., 1998).

c A basket containing only seven fruits was created to accommodate the rare availability of pineapple, honeydew melon and watermelon in convenience stores.

5.3.4. Analysis

Statistical analyses were performed using SAS version 8.0 (343). For each shop-

type, we compared the availability, variety, and basket-price of fruits and

vegetables across the tertiles of area-level disadvantage using ANOVA if the

distributions were normal (or could be normalised by transformation) or using the

Kruskall-Wallis test if the distribution could not be transformed. Levene’s test

showed equal variances could be assumed across socioeconomic tertiles for all

outcomes except fruit availability scores in greengrocers, which were therefore

not formally statistically tested. All R2 values reported are maximum-rescaled

(351).

147

5.4. Results

Table 5.2 presents the average availability, variety and prices of fruits and

vegetables in low, medium and high socioeconomic areas.

5.4.1. Availability The areas varied widely in their fruit and vegetable availability when the types of

shops they contained were not considered. Table 5.2 shows that fruit and

vegetable availability in all shops combined were around half to one point greater

in high socioeconomic areas, but differences were not statistically significant.

Socioeconomic differences in the median availability of fruits and vegetables in

greengrocers and supermarkets were minimal and not statistically significant, and

these median scores were all close to the theoretical maximum (ten). However,

the spread of values tended to be smaller in high socioeconomic than other areas,

showing a slight tendency towards lesser availability in low and medium

socioeconomic areas. (This difference in spread was seen as unequal variance by

Levene’s test only for fruit availability scores in greengrocers.) Results for

convenience stores were similar to those for all shops combined. While

differences were not statistically significant, R2 values at times show area

socioeconomic category to explain approximately 10% or more of the variation in

availability.

5.4.2. Variety

Variety scores were generally highest in supermarkets, and lowest in convenience

stores. Vegetable variety scores were highest in high socioeconomic areas and

generally lowest in the most disadvantaged areas. Likewise, fruit variety scores

were generally highest in high socioeconomic areas, but were generally lowest in

medium socioeconomic areas. Differences were generally around half to one

point and did not reach statistical significance. R2 values indicated that only

around 5% of the variation in variety scores could be attributed to area-level

disadvantage.

148

Table 5.2: Availability, variety and price of fruits and vegetables in catchments surrounding areas varying in socioeconomic characteristics a

Areas Average (95% CI) b n= Low SEP Medium SEP High SEP R2 Fruit Availability scores

All Shops 50 6.10 (5.55 to 6.66) 6.06 (5.49 to 6.64) 6.70 (6.14 to 7.26) 0.07 Supermarkets c 47 9.75 (9.25 to 10) 9.82 (8.67 to 10) 9.80 (9.67 to 10) <0.01 Greengrocers c,d 48 9.90 (8.80 to 10) 9.70 (6.67 to 10) 9.88 (9.25 to 10) 0.13 Convenience Stores e 50 3.26 (2.72 to 3.90) 3.24 (1.40 to 3.90) 3.91 (3.26 to 4.66) 0.05

Vegetable Availability Scores

All Shops e 50 6.93 (6.49 to 7.39) 6.96 (6.49 to 7.39) 7.50 (7.02 to 8.00) 0.08 Supermarkets c 47 10 (9.33 to 10) 10 (9.75 to 10) 10 (10 to 10) 0.04 Greengrocers c 48 10 (9.00 to 10) 10 (9.00 to 10) 10 (8.50 to 10) <0.01 Convenience Stores e 50 4.67 (4.10 to 5.37) 4.58 (3.97 to 5.26) 5.52 (4.81 to 6.30) 0.09

Fruit Variety Scores

All Shops 50 8.69 (7.78 – 9.60) 8.78 97.90 – 9.67) 9.55 (8.67 – 10.43) 0.06 Supermarkets 47 16.23 (15.65 – 16.82) 15.62 (15.09 – 16.15) 15.84 (15.32 – 16.37) 0.06 Greengrocers f 48 14.01 (13.31 – 14.67) 13.41 (12.69 – 14.10) 14.17 (13.48 – 14.82) 0.05 Convenience Stores e 50 4.10 (3.36 – 5.00) 3.95 (3.22 – 4.84) 4.85 (3.97 – 5.91) 0.05

Vegetable Variety Scores

All Shops 50 10.92 (9.91 – 11.94) 11.13 (10.09 – 12.18) 12.22 (11.21 – 13.24) 0.04 Supermarkets 47 19.03 (18.53 – 19.53) 19.23 (18.68 – 19.79) 19.49 (18.99 – 19.99) 0.07 Greengrocers c 48 16.67 (14.13 – 18.00) 16.75 (14.00 – 18.67) 17.50 (14.25 – 19.25) - Convenience Stores e 50 5.86 (5.01 – 6.86) 5.76 (4.90 – 6.77) 6.95 (5.94 – 8.13) 0.07

149

Table 5.2 continued: Availability, variety and price of fruits and vegetables in catchments surrounding areas varying in socioeconomic

characteristics a

Areas Average (95% CI) b n= Low SEP Medium SEP High SEP R2 Fruit Basket Prices

All Shops 48 22.44 (21.32 – 23.57) 23.52 (22.40 – 24.65) 23.84 (22.72 – 24.97) 0.07 Supermarkets e 43 24.57 (21.34 – 28.29) 25.78 (21.60 – 30.76) 25.34 (24.64 – 26.09) 0.07 Greengrocers** 45 19.89 (18.63 – 21.14) 19.81 (18.60 – 21.02) 21.07 (19.90 – 22.24) 0.06 Convenience Stores 39 19.68 (17.90 – 21.45) 19.78 (18.13 – 21.42) 19.92 (18.21 – 21.62) 0.00

Vegetable Basket Prices

All Shops 50 28.53 (27.15 – 29.92) 28.86 (27.43 – 30.29) 30.10 (28.71 – 31.49) 0.06 Supermarkets 45 26.01 (24.63 – 27.40) 26.56 (25.31 – 27.81) 27.80 (26.55 – 29.04) 0.09 Greengrocers 48 25.66 (24.44 – 26.88) 26.48 (25.25 – 27.70) 28.78 (27.56 – 30.00) 0.24 Convenience Stores 41 32.67 (29.85 – 35.49) 33.18 (30.46 – 35.89) 33.77 (31.05 – 36.48) 0.01

a Outcomes were measured for shops within a 2.5km radius of 50 sampled CDs. Areas without shops were excluded from analyses, and

areas without all basket items available in at least one shop were excluded from price analyses. Note: Caution should be taken in

interpreting confidence intervals and p-values as shops could be within 2.5km of more than one area.

b Results of ANOVAs reported as mean (95%CI) or median (min – max) where Kruskal Wallis used

c Kruskal-Wallis test used & presents mean (min-max) instead of mean (95% CI)

d Unequal variance (Levene’s test p<0.05) e Backtransformed from natural log

f Backtransformed from square

** significant p<0.05

150

5.4.3. Price

A consistent trend of lower prices with increasing area disadvantage was apparent.

However, socioeconomic differences reached statistical significance only for the

price of a basket of vegetables in greengrocers (F=7.10, df 2,45, p=0.002), which

was $3.12 cheaper in low than high socioeconomic areas. When comparing the

lower with upper socioeconomic areas, price differences were more substantial for a

basket of vegetables ($1.10 to $3.12 cheaper) than for fruits ($0.24 to $1.18 cheaper).

R2 values indicate that socioeconomic category explained virtually no variation in

prices in convenience stores, some variation in supermarket prices and nearly a

quarter of the variation in the prices of a vegetable basket in greengrocers.

5.5. Discussion

In keeping with expectations, living in a socioeconomically disadvantaged area of

Brisbane was not clearly associated with reduced opportunities to purchase fresh fruit

and vegetables locally in terms of price, availability, and variety.

Like the findings of Sooman (346) and Cummins and MacIntyre (193)we found

fruits and vegetables to be similarly priced regardless of areas’ socioeconomic

characteristics. The non-significant differences we noticed were in the same

direction as the study by White et al. (181)which found an increased cost in upper

socioeconomic areas. Retail prices of fruits and vegetables are known to fluctuate

widely as they are affected by seasonality, climate and varying commodity prices.

For example in Australia over the period 1998 to 2003, in each quarter prices were

up to 20% more expensive or 10% cheaper than they were in the corresponding

quarter the year before (208). Lee and colleagues (335)found in 2000, a basket of

fruits and a basket of vegetables and legumes cost around 30% extra in very remote

compared with highly accessible areas in Queensland. In this present study,

conducted within an urban context, fruit and vegetable baskets tended to be 1- 12%

higher in upper socioeconomic areas; a small difference by comparison with typical

price fluctuations and rural-urban disparities. In Australia and internationally, “own-

151

price elasticities” of foods which are considered “staples” (including many fresh

vegetables) show that a 1% reduction in price results in an increase of only 0 to 0.3%

in the quantities of these foods that are purchased by consumers (352). However, for

many foods (including “staple foods”), lower prices result in greater increases in

purchase for lower income- than for higher income households (236). Accordingly,

we would hypothesise that the price differentials observed in our study may

somewhat facilitate purchasing fruits and vegetables for low income residents in low

socioeconomic areas and most likely do not impact on the purchasing behaviours of

high income residents in upper socioeconomic areas. The small observed differences

are more likely to minimise than create inequalities in purchasing fruits and

vegetables.

Within urban Brisbane, supermarkets and greengrocers usually stocked all of the

studied fruits and vegetables, and often in multiple varieties. A non-significant trend

to slightly better relative availability in upper socioeconomic areas was noticeable,

which may have been detectable statistically by a larger study. These differences are

much smaller than the significant and larger rural-urban differences in fruit and

vegetable availability noted at the same time period (335). It remains to be

determined what impact (if any) these small differences in availability would have on

residents’ food purchasing, or on socioeconomic inequalities in diet.

This study did not find food to be more expensive across the socioeconomic areas,

although the findings do suggest that purchasing fruits and vegetables may be

difficult on a low income. There are limitations to how much money low income

households can apportion to foods. For those on lower incomes, fixed expenses like

housing transport and fuel compete with and take priority over the more flexible food

budget. Consequently the food budget is small and is further reduced when

unexpected expenses arise (eg. medical expenses or car repairs) (156). Data from the

1998-1999 Household Expenditure Survey (HES) indicate that on average Australian

households in the lowest income quintile (with an average household size of 1.52

persons) spent $67.15 on all food and beverages per week (353) i.e. just under ninety

dollars per person each fortnight. In our study, sufficient quantities to meet the

fortnightly requirements of one adult (for fruits and vegetables only) cost around

fifty Australian dollars, assuming a roughly even mix of ten vegetables and ten

152

fruits, implying that purchasing sufficient quantities of a wide variety of fruits and

vegetables may be problematic when budgeting on a low income. Considering that

total dietary costs would also include meats, dairy foods and breads and cereals, the

cost of our baskets of fruits and vegetables represents a substantial proportion (14%)

of the fortnightly income of an adult on unemployment benefits when the data were

collected in 2000 (354), which may not be compatible with budgeting on a low

income.

The findings need to be interpreted carefully with consideration of the study’s

methodology. Shops were sometimes located in more than one catchment by virtue

of using a catchment approach to measuring the local shopping environment. Of the

shops in this study, 31.1% were located in one catchment only, 26.5% were located

in two catchments, 23.4% in three, 12.9% in four, 3.4% in five, 2.2% in six and 0.6%

in seven catchments. This lack of full independence of observations affects the

precision rather than the magnitude of the estimates presented, and results in

confidence intervals which are narrower than they would otherwise be. With our

null findings, a greater chance exists (than the overly narrow confidence intervals

would suggest) that true differences could be either smaller or larger than our study

estimates. Secondly, area-level socioeconomic disadvantage was treated as tertiles to

minimise prior assumptions about relationships, however the use of crude categories

means that the variation in prices and availability explained by socioeconomic

disadvantage (R2) is less than what might be found with a more precise continuous

socioeconomic measure.

One limitation is that our study only considered the most common fruits and

vegetables, which may be the least likely to show any variation in availability, and

therefore socioeconomic patterning. Differences in availability of more exotic fruits

and vegetables are possible. Small sample size and the enormity of variation of

prices relative to differences between tertiles of socioeconomic disadvantage means

this study had a limited ability to detect small socioeconomic effects on prices. Also,

quality was not controlled for as our quality measure was unreliable, therefore lower

prices might reflect lower quality. Findings from Newcastle highlighted the

importance of considering shop type and quality in price comparisons as fruits and

vegetables tended to be cheaper in greengrocers and market stalls, but they tended to

153

be of higher quality in supermarkets and department stores (181). While our study

can draw conclusions about prices, it cannot show that fruits and vegetables were of

equal or better value in socioeconomically disadvantaged areas.

Our findings were consistent with earlier BFS analyses which found residents to

have similar tendency to purchase fruits and vegetables despite areas’ socioeconomic

characteristics. However they do not shed much light on why lower income

households are relatively less likely to purchase fruits and vegetables (89)or why

Australian adults in the lowest income quintile consume around 80g less fruit and

just under 20g less vegetables daily compared with those in the highest income

quintile (105). The impact of local availability and prices of fruits and vegetables in

socioeconomically distinct areas on the purchasing behaviours of residents remains

to be determined, and the contribution of local prices and availability to area-level

socioeconomic variation in purchasing remains unknown. However, the findings of

this study do not conflict with the earlier BFS finding that area-level socioeconomic

disadvantage does not independently influence fruit and vegetable purchasing in

Brisbane (116).

5.6. Acknowledgements

We gratefully acknowledge the technical support and advice of Diana Battistutta, and

the contribution of the Brisbane Food Study Team. The second author is supported

by a National Health and Medical Research Council/ National Heart Foundation

Career Development Award (CR 01B 0502). During the writing of this paper Dr

Carla Patterson passed away after a long battle with cancer. Carla was a dear friend,

a valued colleague and an inspiration to all. Her passing has left a deep sadness in

each of us and she will be greatly missed. May she rest in peace.

154

Chapter 6: Local food retailing and the purchase of fruits and vegetables by socioeconomic groups8

8 This manuscript is in preparation for submission to International Journal of Epidemiology as at 20.06.08 and has not been previously rejected by this journal

155

6.1. Summary

Background: As part of a growing examination of the contribution of

neighbourhoods to health behaviours and inequalities, this study examines the extent

to which features of the food retail environment are associated with residents’

purchasing of fruits and vegetables, and mediate socioeconomic differences in fruit

and vegetable purchasing. Methods: This multi-level, cross-sectional study of

1003 individuals living in 50 areas in Brisbane City assessed self-reported fruit and

vegetable purchasing in a household survey and measured distance to shops, the

number of shops, price, or availability of fresh produce within 2.5km of residents’

homes in an environmental audit. Results: In multi-level logistic regression models,

having more shops was associated with significantly less chance of not buying

broccoli (OR: 0.90, 95% CI: 0.82, 0.99) and pineapple (OR: 0.91, 95% CI: 0.87,

0.96). Other relationships were observed but did not reach statistical significance.

The study was under-powered to statistically detect interactions, however price and

distance appeared more strongly associated with purchasing within low- than high-

income households. The relationships between income and fruit and vegetable

purchase were unchanged with adjustment for environmental factors (relative

changes in odds ratios <10 %). Conclusions: Overall, this study found some

evidence the local food retail environment is associated with resident’s fruit and

vegetable purchasing, but found no evidence that it contributes to socioeconomic

differences in fruit and vegetable purchasing in Brisbane, Australia. Keywords:

Fruit, Vegetables, Socioeconomic Factors, Epidemiologic Factors, Behavior and

Behavior Mechanisms

156

6.2. Introduction

Lower fruit and vegetable intake of socioeconomically disadvantaged groups (33, 89,

90, 92-94, 99, 100, 102, 105, 112, 326, 327) may contribute to concomitant

differences in morbidity and mortality (20, 21). There are a number of potential

contributors to this phenomenon, such as taste preferences (141, 142) and nutritional

knowledge (150, 151), and the potential role of the environment in dietary and health

inequalities is increasingly under examination. Observational studies in the United

States and the United Kingdom show lower socioeconomic status of areas, based on

average income, affluence, and composite socioeconomic indices, are associated

with ‘riskier’ behaviours and poorer health, including coronary heart disease, high

body mass index (256, 355, 356) and various dietary outcomes (114) including low

fruit and vegetable intake (108, 111, 115), independently of individual

socioeconomic measures.

Researchers have examined the contribution of food accessibility and affordability

to inequalities in diet by ascertaining whether features of the food supply are

systematically patterned according to the socioeconomic characteristics of areas,

such as the relative abundance of local shops, availability of ‘healthy’ foods, and

food prices (47, 48, 170, 179-181, 187, 188, 333, 337). Socioeconomic

characteristics may reasonably be interconnected with the local food retail

environments through the operation of processes such as redlining (186), less

economic demand (based on ability to pay) leading to lesser supply in poorer areas

(187), price discrimination (202), and cost of living affecting where people choose

to live (338). However, results of findings to date are mixed, with only some of

these studies, mostly from the United States, finding a significant association

between living in a disadvantaged area and having a food supply that is presumably

less health promoting.

Conclusions about the likely effect of socioeconomic differences in food retail

provision are hard to extrapolate from the existing evidence, as relationships appear

complex and have seldom been studied at the neighbourhood level. The availability

157

of fruits and vegetables within stores has been associated with their consumption

(174, 184, 230, 357), mostly based on subjective measures (174, 184, 227) or store-

level sales (230). The relative availability of “healthier choice” items within local

shops has been associated with residents’ dietary practices (222, 224, 225). People

consuming diets higher in fruits and vegetables are more likely to shop at

supermarkets rather than other types of stores (174, 181, 221), which may relate to

either issues of access or consumer choice. Access to shops has at times been

associated with dietary outcomes, although not consistently across locations,

definitions of access or different types of stores (174, 181, 214-216, 219, 221, 358,

359). Price has been associated food purchasing and other diet-related measures in

terms of aggregate price elasticises (230, 234, 238, 239) and subjective measures of

price and affordability (174, 184, 198, 199, 226). Effect modification has been

seldom studied; however the food retail environment may hold stronger relationships

with diet for low-income groups. Price elasticities are stronger for low- than high-

income groups (211, 236-238) and access measures are more often associated with

dietary outcomes within at-risk groups (174, 214, 216, 221, 358) compared groups

less at-risk or general populations (181, 215, 219, 357).

This paper aims to determine whether features of the local food environment (in

terms of the availability of local shops and the price and availability of fruits and

vegetables within shops) are associated with residents’ fruit and vegetable

purchasing behaviours and whether they mediate the socioeconomic differences in

purchasing that have been reported previously (89). The study also examines

whether household income modifies the relationship between aspects of the food

supply and purchasing.

6.3. Material and methods

The multi-level Brisbane Food Study (BFS) was conducted in 2000 in the Brisbane

City Statistical Subdivision and its methods have been reported elsewhere (47, 48,

89): a brief overview follows.

6.3.1. Sampling

158

Census collection districts (CCDs) (n=50) were randomly selected within deciles of

Index of Relative Socioeconomic Disadvantage (360) from the 1,517 out of 1,585

contiguous CCDs within the Brisbane that had pertinent census data available, using

proportional to size sampling. Within CCDs, dwellings were randomly sampled and

the person who reported doing most of the food shopping was invited to participate.

A 64 percent response rate was obtained, yielding a final sample 9 to 36 households

per CCD (20 per CCD on average) and 1,003 households overall. All supermarkets,

greengrocers and convenience stores (excluding those attached to petrol stations)

were sampled that fell within catchments of 2.5 kilometres (km) from the centroids

of the study CCDs. These shopping catchments were created to delineate spatially

uniform neighbourhood boundaries where residents of the sampled CCDs might

reasonably shop.

6.3.2. Data collection

In-home, face-to-face interviews averaging one hour were conducted by trained

interviewers using an instrument that sought information on determinants of food

purchase, socio-demographic characteristics, and household buying habits. Shop

data were obtained through an audit of the shopping catchments conducted between

July and October 2000. Shops were identified from Brisbane City Council licensing

lists and maps, by systematic observation during visits to listed shops and by survey

respondents who reported where they usually shopped. Shops were classified based

on size, primary activity and merchandise according to a reliable tool described

elsewhere (47). All supermarkets (n=58) and greengrocers (n=67) located within

catchments were audited for price and availability, except for two greengrocers

whose managers declined to participate and two greengrocers that were no longer

operating. Due to feasibility constraints, price and availability data were only

collected for the 10 fresh fruits and vegetables most commonly consumed according

to apparent consumption data (349). Trained data collectors observed and recorded

whether or not items were available (yes or no), and their price per kilogram (kg) or

per item, standardized for type and size.

159

6.3.3. Outcome variables

Participants were asked how often they purchased fruits and vegetables (never,

rarely, sometimes, nearly always or always) including those 10 fruits and 10

vegetables in the price and availability audit. Examining purchasing is advantageous

as it is a behaviour that occurs within the food retail context. Price sensitivities show

the price-purchase relationship varies on an item-by-item basis (eg 36, 50-52), so

each fruit and vegetable was analysed separately, to allow for such heterogeneity. To

ensure sufficient responses in each group, responses were collapsed into non-

purchase (never or rarely) and purchase (all remaining categories).

6.3.4. Independent variables

Household-level variables

Sociodemographic characteristics were measured by self report for respondents (age,

gender) and for their households (gross annual income, number and ages of other

household members). Household income was chosen as the socioeconomic indicator

due to its strong relationship with food purchase in a previous analysis (89) and its

obvious theoretical links with material environmental constraints. Respondents were

asked to indicate the total income (including pensions, allowances and investments)

received by all household members as one of 14 categories. To compare extremes

while ensuring sufficient respondents per category, responses were collapsed into

three categories based on current Australian averages: “low” (less than $20,799),

“middle” ($20,800– 51,999) and “high” ($52,000 and above) (361).

160

Area-level variables

The number of shops was defined as the combined number of supermarkets and

greengrocers in the catchment. The distance to nearest shop for each area was

measured in MapInfo Professional (342) as the straight line distance to the nearest

supermarket or greengrocer from the centroid of each CCD. Each area was given a

0-100 availability score for each fruit and vegetable, which was calculated as the

percentage of local shops that had each item available. Average prices in local shops

were calculated for each area, after converting prices to dollars per kilogram using

average weights typical to Australian food databases where necessary (350). Areas

without shops were kept in the analysis by allocating availability scores of zero,

distance values of 2.5 km, and prices from the most nearby shop in which items were

available to avoid biasing results by excluding the areas in which food would be

theoretically hardest to procure.

6.3.5. Analysis

Multi-level logistic regression models with a random intercept were employed to

determine the relative odds of not purchasing each fruit and vegetable according to

socioeconomic and area characteristics, accounting for clustering and adjusting for

confounders. Initially models established the relationship between income and

purchase adjusted for age and gender, as per the initial Brisbane Food Study findings

(Model 0) then adjusted for additional confounders that were identified at the

bivariate level, namely the number of adults and minors in households (Model 1).

Models were then built up to include area disadvantage to separate area-based and

individual socioeconomic effects (Model 2). Next, models included the availablity

and price measures to determine their contribution to purchasing and to the income

effects of earlier models (Models 3-6). Differences between income effects before

and after adjustment for environmental characteristics were calculated as the

difference between the adjusted and original odds ratios (OR) (as a percentage of the

original odds ratios). To account for the correlation between the area- level

characteristics, fully adjusted models with all area level factors were also computed

161

(Model 7). The conditional effects of the environmental factors by income were then

examined by adding interaction terms to the fully-adjusted models, with separate

models being used to examine each interaction (Models 8-11). (All area-level

variables were centred on the mean values to maximise model robustness, especially

in modelling interactions.) Two-tailed hypothesis testing was used. Estimates are

expressed as odds ratios and 95% confidence intervals (CI). Based on the size of

associations reported in studies that used odds ratios or prevalence ratios as outcomes

((171, 214, 257, 358, 362)), the minimum meaningful effect size for reporting was

1.06 (or 0.94) for IRSD, 1.1 (or 0.91) for shops, 1.2 (or 0.83) for availability 1.67 (or

0.6) for distance and price.

6.4. Results

Table 6.1 presents the proportions and the crude relative odds of rarely or never

purchasing common fruits and vegetables according to household income. While all

vegetable items (except cabbage) were purchased by at least two-thirds of

participants, regular fruit purchase was less frequent. Generally, low-income

respondents were the most likely to report not buying fruits and vegetables. Given

the focus on socioeconomic differences in purchasing, further analyses focused on

items that that were commonly not purchased by low-income households (>15%)

and that showed a socioeconomic gradient in their purchase. Only items that low-

income households were substantially more likely not to purchase compared with

high-income households were examined further (OR>2 with adjustment for age and

gender, known suppressors).

162

Table 6.1: Not purchasing common fruits and vegetables by household income among participants of the Brisbane Food Study 2000 †

Rarely or never buy item (%) Crude Results Adjusted results § All

(n=970) Low

Income‡ (n=184)

Middle Income‡ (n=379)

High Income‡ (n=407)

OR (95%CI) Low vs High

Income

OR (95%CI)Middle vs High

Income

OR (95%CI) Low vs High

Income

OR (95%CI)

Middle vs High Income

Kiwifruit 39.1 45.7 42.7 32.7 1.72 (1.20, 2.45)* 1.53 (1.14, 2.04)* 2.01 (1.33, 3.05)* 1.50 (1.11, 2.03)* Orange 17.0 20.1 18.2 14.3 1.49 (0.95, 2.34) 1.32 (0.90, 1.93) 1.42 (0.85, 2.36) 1.30 (0.89, 1.90) Rockmelon 26.6 28.3 28.2 24.3 1.21 (0.82, 1.80) 1.21 (0.88, 1.67) 1.93 (1.24, 3.02)* 1.22 (0.88, 1.69) Banana 4.9 5.4 6.9 3.0 1.90 (0.80, 4.47) 2.43 (1.21, 4.89)* 2.31 (0.92, 5.82) 2.41 (1.19, 4.85)* Pineapple 47.8 58.2 47.2 43.7 1.78 (1.25, 2.53)* 1.15 (0.87, 1.52) 2.16 (1.44, 3.24)* 1.14 (0.86, 1.52) Honeydew Melon 66.2 75 67.3 61.2 1.90 (1.29, 2.80)* 1.30 (0.97, 1.74) 2.53 (1.62, 3.97)* 1.31 (0.97, 1.77) Nectarine 33.4 33.2 40.4 27.0 1.33 (0.91, 1.93) 1.81 (1.34, 2.44)* 2.40 (1.53, 3.75)* 1.89 (1.38, 2.60)* Pear 30.5 30.4 33.8 27.5 1.14 (0.78, 1.67) 1.33 (0.98, 1.80) 1.73 (1.12, 2.66)* 1.36 (1.00, 1.86)* Watermelon 25.1 36.4 23.5 21.4 2.08 (1.42, 3.05)* 1.12 (0.80, 1.56) 2.11 (1.36, 3.26)* 1.08 (0.77, 1.52) Apple 5.0 8.2 5.8 2.7 2.93 (1.34, 6.39) 2.03 (0.99, 4.17) 3.03 (1.27, 7.21)* 1.93 (0.94, 3.98) Broccoli 10.2 17.9 9.5 7.4 2.85 (1.68, 4.81)* 1.33 (0.80, 2.20) 2.42 (1.28, 4.56)* 1.20 (0.70, 2.05) Capsicum 19.6 33.7 19.8 13.0 3.24 (2.14, 4.92)* 1.59 (1.09, 2.32)* 2.51 (1.53, 4.13)* 1.46 (0.98, 2.17) Cabbage 38.8 27.1 35.6 47.2 0.43 (0.29, 0.62)* 0.61 (0.46, 0.82)* 0.69 (0.46, 1.05) 0.63 (0.47, 0.84) Pumpkin 13.2 9.8 15.0 13.4 0.74 (0.42, 1.28) 1.14 (0.76, 1.70) 1.35 (0.74, 2.46) 1.12 (0.74, 1.70) Carrot 3.0 3.3 2.6 3.2 1.02 (0.38, 2.73) 0.83 (0.36, 1.91) 1.44 (0.50, 4.15) 0.79 (0.34, 1.84) Cauliflower 14.2 12.0 13.5 16.0 0.75 (0.45, 1.25) 0.82 (0.55, 1.22) 0.98 (0.55, 1.73) 0.81 (0.54, 1.22) Beans 15.3 14.0 15.0 16.2 0.85 (0.52, 1.39) 0.92 (0.63, 1.35) 1.05 (0.61, 1.81) 0.90 (0.52, 1.55) Tomato 4.1 4.4 6.1 2.2 2.01 (0.76, 5.29) 2.87 (1.31, 6.29)* 1.90 (0.66, 5.47) 2.74 (1.24, 6.02)* Potato 2.5 3.3 3.4 1.2 2.71 (0.82, 8.99) 2.87 (1.01, 8.13)* 3.16 (0.87, 11.45) 2.70 (0.95, 7.69) Onion 3.8 5.4 4.5 2.5 2.28 (0.93, 5.58) 1.87 (0.85, 4.15) 2.52 (0.95, 6.72) 1.82 (0.82, 4.05) † excluding participants missing data on fruit or vegetable purchasing or income. ‡ Annual gross household income: low income (less than $20,799 pa), middle income ($20,800– 51,999 pa) and high income ($52,000 pa +). § adjusted for age and gender (2-level logistic regression with random intercept for area).

163

Multivariable results are presented in Table 6.2. Crude estimates of socioeconomic

differences in fruit and vegetable purchasing were strengthened after adjustment for

age and gender (Model 0). However, further adjustment for the number of adults and

minors living in the households substantially attenuated the estimates of

socioeconomic differences in purchasing produce (Model 1). Consideration of area

socioeconomic disadvantage (Model 2) generally further attenuated the elevated odds

of not purchasing fruits and vegetables attributed to household income, with adjusted

odds ratios being relatively lower by 11-16% for broccoli, capsicum and kiwifruit.

The direction of associations was such that living in areas with less socioeconomic

disadvantage was associated with lower odds of not buying most fruits and

vegetables, independently of household income (Model 2), however there was no

sizeable or significant relationship between IRSD that operated independently of

environmental features (Model 7). Inter-relationships between the different facets of

food access occasionally led to different results being obtained with mutual

adjustment for environmental features (model 7) compared with initial estimates

(model 4).

In mutually adjusted models that also accounted for age, gender, household

composition, household income, and area SES (model 7), price did not appear to

relate to purchase of honeydew melon, kiwifruit or pineapple. Relationships did not

reach statistical significance, however each additional dollar per kilogram above the

average price was associated with greater chances of not buying capsicum and

watermelon but less chance of not buying broccoli. In-store availability was not

associated with the purchase of fruits and vegetables, except for small, borderline-

significant relationship with broccoli. Results for shops were consistent across items

in direction, but were sizeable and significant only for broccoli and pineapple. More

shops were associated with significantly greater purchase of broccoli and pineapple.

Increasing distance was not associated with purchasing fruits and vegetables.

164

Table 6.2: Odds of not buying fruits and vegetables by socioeconomic groups in participants of Brisbane Food Study 2000: influence of local

socioeconomic and built environmental characteristics

Income (low)† Income (middle)† IRSD (/20pts) § Price # Availability †† Number of Shops‡‡ Distance§§

Hon

ey-d

ew M

elon

Model 0 2.53 (1.62, 3.97)* 1.31 (0.97, 1.77) Model 1 ‡ 2.20 (1.35, 3.57)* 1.25 (0.92, 1.69) Model 2 2.05 (1.24, 3.41)* 1.21 (0.88, 1.65) 0.99 (0.96, 1.02) Model 3 2.03 (1.22, 3.38)* 1.20 (0.87, 1.64) 0.99 (0.96, 1.02) 0.89 (0.57, 1.39) Model 4 2.04 (1.23, 3.39)* 1.21 (0.88, 1.66) 0.98 (0.95, 1.02) 1.06 (1.00, 1.12) Model 5 2.10 (1.26, 3.41)* 1.20 (0.87, 1.64) 0.98 (0.95, 1.01) 0.94 (0.90, 0.98)* Model 6 2.10 (1.27, 3.50)* 1.21 (0.88, 1.66) 1.00 (0.96, 1.03) 0.77 (0.60, 0.99)* Model 7 2.08 (1.25, 3.46)* 1.19 (0.87, 1.64) 0.99 (0.95, 1.03) 0.90 (0.57, 1.42) 1.02 (0.95, 1.09) 0.95 (0.90, 1.00) 0.93 (0.66, 1.31)

Kiw

i-fru

it

Model 0 2.01 (1.33, 3.05)* 1.50 (1.11, 2.03)* Model 1 ‡ 1.40 (0.89, 2.19) 1.33 (0.98, 1.81) Model 2 1.24 (0.78, 1.99) 1.26 (0.92, 1.72) 0.97 (0.94, 1.00) Model 3 1.24 (0.78, 1.99) 1.26 (0.92, 1.73) 0.97 (0.94, 1.00) 1.02 (0.75, 1.39) Model 4 1.25 (0.78, 1.99) 1.26 (0.92, 1.72) 0.97 (0.94, 1.00) 1.02 (0.96, 1.09) Model 5 1.25 (0.78, 2.00) 1.26 (0.92, 1.72) 0.97 (0.94, 1.00) 0.98 (0.94, 1.02) Model 6 1.26 (0.79, 2.02) 1.26 (0.92, 1.73) 0.98 (0.95, 1.01) 0.81 (0.62, 1.05) Model 7 1.26 (0.79, 2.02) 1.27 (0.93, 1.73) 0.98 (0.95, 1.02) 0.96 (0.68, 1.34) 0.98 (0.90, 1.07) 0.99 (0.94, 1.04) 0.78 (0.54, 1.12)

Pine

-app

le

Model 0 2.16 (1.44, 3.24)* 1.14 (0.86, 1.52) Model 1‡ 1.73 (1.12, 2.70)* 1.05 (0.78, 1.41) Model 2 1.65 (1.04, 2.59)* 1.02 (0.76, 1.38) 0.99 (0.96, 1.02) Model 3 1.62 (1.03, 2.55)* 1.02 (0.76, 1.38) 0.99 (0.96, 1.02) 0.91 (0.80, 1.04) Model 4 1.64 (1.04, 2.59)* 1.02 (0.76, 1.38) 0.99 (0.96, 1.02) 1.02 (0.96, 1.08) Model 5 1.65 (1.05, 2.60)* 1.02 (0.76, 1.38) 0.99 (0.96, 1.02) 0.95 (0.91, 1.00) Model 6 1.64 (1.04, 2.58)* 1.02 (0.76, 1.38) 0.98 (0.95, 1.02) 1.10 (0.84, 1.43) Model 7 1.60 (1.02, 2.52)* 1.02 (0.75, 1.38) 0.98 (0.95, 1.02) 0.92 (0.81, 1.05) 1.03 (0.96, 1.11) 0.91 (0.87, 0.96)* 1.18 (0.85, 1.65)

165

Table 6.2 continued: Income (low)† Income (middle)† IRSD (/20pts) § Price # Availability †† Number of Shops ‡‡ Distance§§

Wat

er-m

elon

Model 0 2.11 (1.36, 3.26)* 1.08 (0.77, 1.52) Model 1‡ 1.42 (0.89, 2.25) 0.90 (0.63, 1.29) Model 2 1.36 (0.82, 2.26) 0.89 (0.62, 1.28) 0.99 (0.96, 1.02) Model 3 1.36 (0.82, 2.25) 0.89 (0.62, 1.28) 0.99 (0.96, 1.03) 1.60 (0.63, 4.10) Model 4 1.37 (0.83, 2.27) 0.89 (0.62, 1.28) 0.99 (0.96, 1.03) 1.02 (0.96, 1.09) Model 5 1.35 (0.81, 2.24) 0.89 (0.62, 1.28) 0.99 (0.96, 1.03) 1.03 (0.98, 1.08) Model 6 1.38 (0.83, 2.29) 0.89 (0.62, 1.28) 1.00 (0.96, 1.03) 0.91 (0.68, 1.23) Model 7 1.37 (0.82, 2.27) 0.90 (0.62, 1.30) 1.00 (0.97, 1.04) 1.53 (0.57, 4.08) 1.03 (0.95, 1.12) 1.05 (0.99, 1.11) 0.83 (0.57, 1.22)

Cap

sicu

m

Model 0 2.51 (1.53, 4.13)* 1.46 (0.98, 2.17) - Model 1 ‡ 1.86 (1.09, 3.20)* 1.30 (0.87, 1.95) Model 2 1.62 (0.93, 2.82) 1.21 (0.80, 1.83) 0.96 (0.92, 1.00)* Model 3 1.67 (0.96, 2.92) 1.25 (0.83, 1.89) 0.96 (0.93, 1.00) 1.95 (0.92, 4.13) Model 4 1.62 (0.93, 2.82) 1.21 (0.80, 1.83) 0.96 (0.92, 1.00)* 1.00 (0.92, 1.09) Model 5 1.61 (0.92, 2.81) 1.21 (0.80, 1.83) 0.96 (0.92, 1.00)* 0.98 (0.92, 1.04) - Model 6 1.62 (0.92, 2.83) 1.21 (0.80, 1.83) 0.96 (0.92, 1.00) - 1.00 (0.69, 1.43) Model 7 1.67 (0.96, 2.93) 1.21 (0.80, 1.83) 0.97 (0.93, 1.01) 1.97 (0.91, 4.27) 0.99 (0.87, 1.11) 0.98 (0.91, 1.05) 0.91 (0.56, 1.48)

Broc

coli

Model 0 2.42 (1.28, 4.56)* 1.20 (0.70, 2.05) - Model 1 ‡ 1.82 (0.91, 3.64) 1.06 (0.61, 1.82) Model 2 1.53 (0.75, 3.14) 0.97 (0.56, 1.69) 0.95 (0.91, 1.00) Model 3 1.53 (0.75, 3.11) 0.98 (0.56, 1.69) 0.96 (0.91, 1.02) 0.60 (0.29, 1.22) Model 4 1.52 (0.74, 3.11) 0.98 (0.56, 1.70) 0.95 (0.91, 1.00) 0.97 (0.87, 1.07) Model 5 1.49 (0.73, 3.03) 0.98 (0.57, 1.70) 0.96 (0.92, 1.01) 0.92 (0.85, 0.99)* Model 6 1.49 (0.73, 3.04) 0.97 (0.56, 1.68) 0.94 (0.89, 0.99)* 1.35 (0.88, 2.08) Model 7 1.46 (0.72, 2.96) 0.96 (0.56, 1.66) 0.97 (0.92, 1.02) 0.42 (0.17, 1.04) 1.18 (1.00, 1.39) 0.90 (0.82, 0.99)* 1.37 (0.76, 2.49)

* p<0.05 † Low income (less than $20,799 pa) and middle income ($20,800– 51,999 pa) are compared with high income ($52,000 pa +) ‡All models are adjusted for age, gender, presence of 1, 2, 3 or more adults and 0, 1, 2 or more minors in the household (2-level logistic regression with random intercept). § IRSD is the 1996 ABS Index of Relative Socioeconomic Disadvantage (increments of 20 points) # Price is the average price ($/kg) of the item in supermarkets and greengrocers within 2.5km (mean-centred). †† Availabilty is the percentage (0-100) of local supermarkets and greengrocers which sell the item, in increments of 10% (mean-centred) ‡‡ Number of shops is the number of supermarkets and greengrocers within 2.5km (mean-centred). §§ Distance is the straight line distance in (km) to the nearest supermarket or greengrocer from the centroid of the sampled area (mean-centred). Italics indicates evidence of confounding by a shift of >10% in the most crude estimate.

166

Most interactions were not statistically significant at p<0.10, however some of the

multiplicative factors were substantially different from 1, suggesting modification of

environmental effects by income (Table 3). There was a significant interaction

(p<0.05) between income and distance to the nearest shop in the odds of purchasing

capsicum. Based on main effects and multiplicative terms, each additional kilometre

to the nearest shop was associated with increases in the odds of not buying capsicum

(OR=3.03) and broccoli (OR=2.32) for low-income households rather than the absent

or inverse main effects detected for the referent high-income group (ORs=0.61 and

0.96, respectively). The effect of price also appeared stronger in low-income than

high-income groups, although not always in the expected direction. Increasing price

was associated with increases in the odds of not purchasing watermelon (OR=3.82)

and capsicum (OR=2.30) and reductions in the odds of not buying broccoli

(OR=0.18) that were visibly stronger than the main effects in high-income groups.

There was little evidence of a differential effect by SES for item availability and the

number of shops, with most multiplicative terms being close to 1.

Environmental features did not appear to contribute to socioeconomic differences in

produce purchase. The odds of purchasing items among low- and middle- compared

with high-income households were attenuated after adjustment for neighbourhood

SEP, however very little further reduction (0-6%) occurred with adjustment for

environmental features (Models 3-6 vs Model 2). Effect modification did not appear

to contribute to socioeconomic differences in fruit and vegetable purchasing as the

conditional odds of not buying fruits and vegetables for low- and middle-income

compared to high-income households (with average distances, numbers of shop,

prices and levels of availability) were similar to the odds of not buying most items

without considering interaction effects (0-11% relative differences) (Models 8-11 vs

Model 7, figures not presented).

167

Table 6.3: Odds of not buying fruits and vegetables in participants of Brisbane Food Study 2000: effect modification of local environmental

characteristics by household income †

Honeydew Kiwifruit Pineapple Watermelon Capsicum Broccoli Model 8 Price §

(High-income) ‡ 0.78 (0.38, 1.60) 1.17 (0.75, 1.84) 1.02 (0.84, 1.24) 0.56 (0.12, 2.67) 1.52 (0.46, 5.07) 0.55 (0.18, 1.70) * Middle-income‡ 1.14 (0.45, 2.90) 0.74 (0.39, 1.39) 0.82 (0.63, 1.06) 3.97 (0.47, 33.66) 1.47 (0.29, 7.49) 0.78 (0.17, 3.65) * Low-income‡ 1.71 (0.48, 6.07) 0.56 (0.25, 1.27) 0.97 (0.66, 1.42) 6.82 (0.56, 82.36) 1.51 (0.29, 7.80) 0.33 (0.06, 1.74)

Model 9 Availability #

(High-income) ‡ 1.02 (0.92, 1.12) 1.02 (0.91, 1.14) 0.99 (0.90, 1.10) 0.99 (0.88, 1.11) 1.00 (0.85, 1.17) 1.27 (1.00, 1.62) * Middle-income‡ 1.01 (0.89, 1.15) 0.99 (0.85, 1.14) 1.05 (0.93, 1.20) 1.03 (0.88, 1.20) 1.18 (0.91, 1.52) 0.97 (0.74, 1.27) * Low-income‡ 0.97 (0.83, 1.14) 0.89 (0.76, 1.04) 1.10 (0.95, 1.28) 1.11 (0.94, 1.31) 0.87 (0.73, 1.04) 0.86 (0.67, 1.09)

Model 10 Number of local shops ††

(High-income) ‡ 0.93 (0.86, 1.00) 0.98 (0.91, 1.05) 0.99 (0.93, 1.07) 1.07 (0.98, 1.16) 0.98 (0.88, 1.09) 0.88 (0.76, 1.02) * Middle-income‡ 1.05 (0.95, 1.16) 1.02 (0.92, 1.12) 1.01 (0.92, 1.11) 0.97 (0.86, 1.09) 1.03 (0.90, 1.17) 1.03 (0.86, 1.23)

* Low-income‡ 1.05 (0.93, 1.19) 1.02 (0.91, 1.15) 0.98 (0.87, 1.10) 0.97 (0.86, 1.11) 0.95 (0.82, 1.10) 1.05 (0.87, 1.26) Model 11 Distance to nearest shop ‡‡

(High-income) ‡ 0.88 (0.57, 1.34) 0.77 (0.48, 1.22) 0.71 (0.39, 1.32) 0.65 (0.39, 1.10) 0.61 (0.31, 1.19) 0.96 (0.40, 2.30) * Middle-income‡ 1.08 (0.36, 1.86) 0.95 (0.54, 1.68) 0.93 (0.55, 1.58) 1.53 (0.79, 2.96) 1.31 (0.61, 2.83) 1.21 (0.43, 3.40)

* Low-income‡ 1.20 (0.61, 2.33) 1.16 (0.61, 2.21) 1.29 (0.85, 1.96) 1.44 (0.72, 2.90) 3.03 (1.37, 6.72)* 2.32 (0.87, 6.20) * significant interaction (p<0.05) by Wald test † All models are adjusted for age, gender, presence of 1, 2, 3 or more adults and 0, 1, 2 or more minors in the household, IRSD, other environmental characteristics (price, availability, number of shops, distance) and include a random intercept for area. ‡ Main effect of environmental characteristics among high-income ($52,000 pa +, referent category) household and conditional effects in middle-income ($20,800– 51,999 pa) and low-income (< $20,799 pa) households. § Price is the average price ($/kg) of the item in supermarkets and greengrocers within 2.5km (mean-centred). # Availability is the percentage (0-100) of local supermarkets and greengrocers which sell the item, in increments of 10% (mean-centred) †† Number of shops is the number of supermarkets and greengrocers within 2.5km (mean-centred). ‡‡ Distance is the straight line distance in (km) to the nearest supermarket or greengrocer from the centroid of the sampled area (mean-centred).

168

6.5. Discussion

We found some evidence supporting the hypotheses that access to shops, local prices

and availability of fruits and vegetables are associated with dietary behaviours,

however results were not consistent for all items and did not always reach statistical

significance. Buying fruit and vegetable items was associated with having a greater

number of nearby supermarkets and greengrocers, but not with distance to the nearest

shop. Higher local prices were associated with both increases and decreases in the

odds of buying fruits and vegetables, depending on the item examined. The counter-

intuitive associations may have arisen from confounding by quality or the prices of

other items. For example, broccoli may have been more likely to be purchased at

higher prices if it was also of visibly better quality, or if it remained competitively

priced relative to other possible items that could act as substitutes. Fast-food outlets

were not examined due to the focus on purchasing for at-home consumption. Since

the distribution of different store types might reasonably be related (215), this may

have contributed to some residual confounding if take-away outlets indirectly affect

fruit and vegetable purchasing. For simplicity, we have been describing associations

in terms of effects of environmental features on fruit and vegetable purchasing,

however it is also plausible that residents’ dietary behaviours create a market and

shape the local food environment. Quasi-experimental studies have been of

insufficient quality to support or refute that manipulating the retail environment can

promote dietary change (363).

The relationship between distance and price and vegetable purchase appears to

depend on household income, however the study lacked the statistical power to make

definitive statements regarding interactions. It is unclear why the interactions were

less apparent for fruit purchase, but the results imply that the determinants of fruit

and vegetable consumption may differ, even though these foods are often studied

collectively due to their similar nutritional benefit. The study design permitted

gathering rich data through directly observing all food shops in large catchments,

which covered approximately 80% of the Brisbane SSD, but prohibited having a

large sample size. Hence, the study had limited precision in examining

environmental effects and interactions. Future studies that are not focused on sub-

169

populations may need to consider examining interactions or stratifying results as null

findings overall may be misleading with respect to low-income and other vulnerable

groups.

The measures used may have affected the results, particularly for price and

availability. Area characteristics were measured only once. Consequently,

classification of price and availability, which fluctuate over time, may have been

limited, perhaps more so than numbers of shops and distance, which are

comparatively stable. The associated measurement error was probably non-

systematic and non-differential, and may have attenuated estimates towards the null.

There may also be a small amount of error in the distance measures, being based on

aggregated rather than individual measures, geometric rather than population-

weighted centroids, and straight-line rather than road-network distances (364). A

recent, large study in New Zealand that used more sophisticated assessment of

distances also failed to find large associations between distance and fruit and

vegetable intake (359), so perhaps the null findings were not unduly influenced by

the simple distance measure used.

The outcome measure also has implications for interpreting the findings. Since only

the most common fruits and vegetables were examined, their non-purchase may be

indicative of a diet low in fruit and vegetable variety, however, non-purchase of these

items does not necessarily indicate a lower dietary intake of fruits and vegetables

overall. The purchasing measures did not capture quantity and may have been

relatively insensitive to detecting effects that operate more in terms of quantities

purchased rather than which varieties are chosen. It is also possible that variation in

the purchase, availability and price of more exotic fruits and vegetables may be more

pronounced, and by focusing on very common items, the study missed capturing

associations between price and availability and purchase.

We found little evidence that access to shops, and the price and availability of fruits

and vegetables mediate socioeconomic inequalities in fruit and vegetable purchasing,

consistent with the absence of large differences in access to shops, price and

availability according to area-level socioeconomic disadvantage in the Brisbane Food

Study (47, 48). Similarly, a Melbourne-based, Australian study found store outlet

170

density (per 10,000 residents) did not mediate socioeconomic inequalities in

women’s dietary intake of fruit and vegetables, using education rather than income as

a socioeconomic marker (215).

Socioeconomic differences in fruit and vegetable purchasing were often small and

not statistically significant in this study, unlike previous papers from the BFS (89,

116) and other research underlying the rationale for this study (93-95, 104, 105, 113).

Adjustment for additional confounding factors, in particular household composition,

consistently attenuated the size of differences between low- and high-income

households in fruit and vegetable purchasing. Studies that have partially accounted

for household composition through their choice of income measure have found

relationships between income and vegetable intake as measured by dietary recall

methods (95, 113). Although confounding of the relationship between purchase and

income by household composition may have been especially evident in our study due

to the use of a household-based outcome, household composition may need to be

considered in future studies of income-based differences in diet. Domestic living

arrangements are integrally linked to dietary behaviours and outcomes, even for

individuals (365). Households with more adults (i.e. more potential income earners)

generally have higher gross incomes, especially as the proportion of non-wage-

earning adults declines (366).

Research that has examined food price without a spatial emphasis has been

suggestive that price may contribute to dietary differences across socioeconomic

groups (201, 250, 254, 367) unlike this present study. A framework that ties in more

realistically with how people shop for food may be necessary to adequately assess

the role of food accessibility and affordability in the dietary behaviours of

socioeconomic groups, beyond whether or not low socioeconomic areas have

comparatively fewer shops, lesser food availability or higher prices (47, 170, 180,

188, 333, 337). This study, like others, examined shops near to where people live,

yet people do not typically travel to shops from their homes in isolation, but rather

‘chain’ their trips to shops with those to and from other places, such as work,

maximising convenience and efficiency (368). It is also possible that the shopping

strategies that people adopt enable them to consume items irrespective of average

local prices – for example, shopping at the cheapest nearby shop, shopping at more

171

distant locations or taking advantage of the variability in fruit and vegetable prices.

The spatial scale at which effects may truly operate is unknown, and this study would

not have detected relationships occurring on scales other than a 2.5km radius around

the sampled small neighbourhoods (CCDs).

Findings from this Brisbane-based study may be generalisable to contexts with

similar spatial and socioeconomic patterning, and where residents are similarly

mobile. In Brisbane, as in many cities, population density increases with proximity

to the inner city, however, in Brisbane, the inner city is increasingly becoming

gentrified (345), unlike many cities in the US (186). In Brisbane, residential and

retail space tend to be co-located in neighbourhoods (369) and the population density

was 721.3 persons per square kilometre in 2004 (370). Private transport is common,

with approximately 10% of Brisbane households, mostly low-income households,

reporting not owning a motor vehicle (344).

This study contributes some additional evidence to support that the local food retail

environment is associated with residents’ fruit and vegetable purchasing, particularly

among low-income groups. However, in Brisbane, Australia, the distribution of fruit

and vegetable prices and availability, supermarkets and greengrocers do not appear

to mediate socioeconomic differences in fruit and vegetable purchasing.

172

Chapter 7: Confidence to cook vegetables and the buying habits of Australian households9

9 This manuscript has been to Journal of the American Dietetic Association and is under review following major revisions as at 25.06.08

173

7.1. Abstract

Background: Cooking skills are emphasised in nutrition promotion but their

distribution among population sub-groups and relationship to dietary behaviour is

researched by few population-based studies.

Objectives: This study examined the relationships between confidence to cook,

socio-demographic characteristics and household vegetable purchasing.

Design/ Methods: This cross-sectional study of 426 randomly selected households

in Brisbane, Australia, used a validated questionnaire to assess household vegetable

purchasing habits and the confidence to cook of the person who most often prepares

food for these households.

Analyses: The mutually-adjusted odds ratios of lacking confidence to cook were

assessed across a range of demographic sub-groups using multiple logistic regression

models. Similarly, mutually-adjusted mean vegetable purchasing scores were

calculated using multiple linear regression for different population groups and for

respondents with varying confidence levels.

Results: Lacking confidence to cook using a variety of techniques was more

common among respondents with less education (OR: 3.30, 95% CI: 1.01, 10.75)

and was less common among respondents who lived with minors (OR: 0.22, 95% CI:

0.09, 0.53) and other adults (OR: 0.43, 95% CI: 0.24, 0.78). Lack of confidence to

prepare vegetables was associated with being male (OR: 2.25, 95% CI: 1.24, 4.08),

low education (OR: 6.60, 95% CI: 2.08, 20.91), lower household income (OR: 2.98,

95% CI: 1.02, 8.72) and living with other adults (OR: 0.53, 95% CI: 0.29, 0.98).

Households bought a greater variety of vegetables on a regular basis when the main

chef was confident to prepare them (difference: 18.60, 95% CI: 14.66, 22.54) older

(difference: 8.69, 95% CI: 4.92, 12.47), lived with at least one other adult

(difference: 5.47, 95% CI: 2.82, 8.12) or at least one minor (difference: 2.86, 95%

CI: 0.17 to 5.55).

Conclusions: Cooking skills may contribute to socioeconomic dietary differences

regarding fruits and vegetables.

174

7.2. Introduction

The World Health Organisation strongly emphasises the need for individuals and

populations to increase their intake of fruits and vegetables in their global strategy on

diet, physical activity and health (371). Low intake of fruits and vegetables may be

responsible for approximately 31% of ischaemic heart disease, 19% of

gastrointestinal cancer, 11% of stroke, and 2.7 million deaths (4.9%) each year

worldwide (371). A lack of cooking skills may contribute to the inadequate

consumption of fruits and vegetables in developed nations including Australia (5,

371) and the United States (US)(7), however relationships between cooking skills,

dietary behaviors and health are supported by very few-population based studies

(165, 227, 266, 271, 279, 281, 284).

About 10 percent of people in the UK cite not knowing how to cook as a factor

limiting their food choice (165). Cooking skills interventions have successfully

altered dietary behaviors (153, 154, 272, 372). It is not known whether dietary

improvement is sustained in absence of long-term follow-up, nor whether dietary

improvement occurred as a result of cooking skills, or the more comprehensive

education provided by these programs. In studies from the US, the food preparation

practices of adolescents (276), young adults (279) and women seeking charitable

food assistance (278) have been associated with more health-promoting dietary

measures, however it is likely that factors other than cooking skills contribute to

cooking practices. Among a convenience sample of Australian adults (277),

cooking-related attitudes and practices were associated with vegetable intake, but

sometimes with lower, rather than higher, intakes. Self-rated cooking skills have

been associated with healthy dietary indicators among older men (271), adult

‘household heads’ in the UK (266) and households with children in Vancouver,

Canada (281).

Not all studies have had positive findings. A study of low- to middle-income US

women found associations between fruit and vegetable consumption and food

preparation skills among Hispanic, but not white or African-American women (227).

175

A study of young adults in the US found no associations between dietary measures

and cooking-related skills and resources (279).

A comparative lack of cooking skills may contribute to the lower fruit and vegetable

intakes of low- compared with higher socioeconomic groups (33, 105, 113), as

studies have at times demonstrated a socioeconomic patterning of cooking skills

(284) (266) (276). A study of Irish household heads (266) reported a significant

association between occupation and self-reported cooking skills. A large population-

based study of cooking skills, the UK National Health and Lifestyles Survey (284),

found respondents of higher socioeconomic position (SEP) (according to

occupational status, education and income), were more likely to report being

confident to cook than respondents of lower SEP. Similar socioeconomic differences

in confidence to prepare fresh green vegetables and root vegetables, and confidence

to use many cooking techniques were apparent. Possibly owing to population or

measurement differences, the US research has not replicated these associations. One

study of young adults found parental education was not associated with perceived

adequacy of cooking skills and resources (279) and another study found adolescents

of lower SEP (as defined by parental education) were more likely to help prepare

dinner than their higher socioeconomic counterparts (276).

This present study examines socio-demographic differences in self-reported

confidence to cook and assesses whether confidence is related to household

vegetable purchasing. This study has a particular focus on low socioeconomic

groups, as their comparatively low consumption of fruits and vegetables (33, 93,

105) has been hypothesised to contribute to their poorer health outcomes (20, 373).

7.3. Methods

7.3.1. Study sample and data collection

This cross-sectional study received ethical approval from the Queensland

University of Technology ethics committee and was conducted in Brisbane, Australia

during 2004. It employed a cluster sampling design to recruit participants across the

176

spectrum of socioeconomic position. A total of six areas (collection districts) were

randomly sampled within tertiles of Index of Relative Socioeconomic Advantage/

Disadvantage. Census collection Districts were eligible for inclusion in the study if

they contained at least 180 residential dwellings and had available socioeconomic

data (374). Ineligible areas (n=660) were no different to eligible areas (n=1007), or

Brisbane in general, in terms of available socioeconomic markers: unemployment

rates, ethnic make up, household size, and median figures for age, income, housing

loan repayments, and rent. Within each area, an address was chosen at random from

Local Government BIMAPs, electronically available listings of properties. This

residence was sampled along with the next 164 private, residential dwellings that

were spatially sequential to it in a radial pattern. The person primarily responsible

for most of the food preparation was invited to participate, as presumably his or her

confidence to prepare food would have greatest relevance to household buying

habits.

Data were collected using a self-administered questionnaire that had been reviewed

by a panel of experts, revised and pilot-tested to maximise face validity.

Repeatability was assessed in a random sub-sample of early participants (n=85) who

re-completed the questionnaire after two weeks (n=58, 68% response rate). The

measures of purchasing and confidence to cook showed acceptable repeatability

(intra-class correlations ranged from 0.85 to 0.88). Questionnaires and reply-paid

envelopes were distributed and the sample was followed-up with thank-you cards

and replacement questionnaires. In total, 446 questionnaires were returned, 426 of

these by intended respondent, leaving a response rate of 43%.

7.3.2. Measures

7.3.2.1. Socio-demographic characteristics

Socio-demographic characteristics were measured by self-report: age, gender,

country of birth, socioeconomic position, and household composition. Respondents

were asked whether they self-identified as belonging to the following groups:

Indigenous Australian, person with a disability and person from a non-English-

177

speaking background. SEP was measured using gross household income and

respondents’ education. Income categories were based on the income distribution in

the 2001 census: low income (the bottom ≈20%), middle income (the middle ≈60%)

and high income (the top ≈20%). These categories ensured sufficient respondents

per category while comparing socioeconomic extremes. Respondents’ educational

levels were measured as the highest educational qualification completed, and covered

the broad range from primary school or less through to post-university qualifications.

7.3.2.2. Outcome variables

Domestic cooking skills are complex and have no universal definition or gold-

standard measure. Perceptions vary as to what constitutes cooking, and skilled

cooking, and public perceptions often do not adhere to definitions of researchers or

culinary professionals (45). For some, ‘cooking’ may refer specifically to food

preparation involving heating. To ensure coherence between participants’ responses,

the survey used the term “food preparation” and included the following definition:

“for purposes of this survey, ‘preparing’ food means anything you might do to make

the food suitable to eat (for example, make a salad from it)”.

Confidence to perform an action theoretically increases the likelihood of doing so

(300). Accordingly, we examined confidence to cook rather than quantifying skill

levels by self-report, which is unlikely to accurately capture actual skill levels, or by

observation, which is an intensive approach (45). Respondents indicated how

confident they felt to prepare vegetables, and to use cooking techniques, in categories

ranging from not at all confident (1) to very confident (6). The measure included 21

vegetable items and 10 techniques examined in the Health and Lifestyles Survey

(284). Item scores were summed, forming two scales, which could range from 21 to

126 and 10 to 60, respectively. These scales were skewed towards high levels of

confidence (medians and ranges were 117 (21 to 126) and 53 (14 to 60),

respectively). Both scales had acceptable internal consistency (Cronbach’s

alpha=0.94, and 0.83, respectively).

178

Based on a measure from the Brisbane Food Study (89), respondents indicated how

often they purchased 21 vegetables in categories ranging from never (1) to always

(5). Responses were summed to form an index with a theoretical range of 21 to 105.

Higher scores denote more varied or more regular vegetable purchase. This index

was normally distributed (mean=69.76, standard deviation= 12.29).

7.3.3. Statistical methods

Statistical analyses were performed using SAS 9.1 (SAS Institute Inc., Cary, NC,

USA, 2002) and MLWin 2.01 (Centre for Multilevel Modelling, Bristol, UK, 2004).

Bi-variate relationships with continuous predictors were examined by Spearman’s R,

while bivariate relationships with binomial and other categorical predictors were

examined by t-tests and ANOVAs for purchasing and by Wilcoxon Rank-Sum

(Mann-Whitney) test and Kruskal-Wallis test for the confidence outcomes.

Variables with a significant (two-tailed p<0.05) or substantial relationship (defined

here as R2>=0.3, or difference >3 points in mean or median) with any outcomes were

included in multivariable models.

Multivariable relationships were assessed using multiple linear regression for the

purchasing outcome. The confidence scales were not transformable and lacking

confidence was of interest, considering high level of confidence to cook overall.

Therefore, multiple logistic regression models were used, where the odds of being in

the bottom quartile of confidence (as opposed to any other quartile) were calculated.

To control for any clustering inherent in the design, two-level Penalised Quasi-

Likelihood (PQL) models with a random intercept were employed using MLWin.

Multi-level results were reported only for the technique confidence models, as

clustering was not evident in the other models. All multivariable analyses were

restricted to the participants with full data on all the variables in the full models

(n=401).

7.4. Results

179

7.4.1. Sample characteristics

This study targeted individuals who cooked for their household. Primary

responsibility for cooking was shared with other household members in 21.2% of

households and borne by the respondent in 78.1% of households. There were very

few households in which no-one cooked (0.7%). Respondents were mostly women,

born in Australia, aged 46 on average, and most often categorised their weekly

household income as $1000-1499 before taxes (Table 7.1). About half of the sample

had some post-secondary school training, over two-thirds lived with at least one

other adult in the household, and around one-quarter lived with minors (< 18 years).

Non-response bias could only be estimated based on household-level characteristics

in the absence of identification of persons who mostly cook in the census.

Participating households were very similar to all households in the areas from which

they were sampled, based on 2001 census figures. Proportions of respondents in

each income category and with 1, 2, 3, 4 or 5+ household members were very similar

in this study and the census (with the largest differences in any category being 8.8%

and 6.8%, respectively), although the two samples were statistically different

(p<0.001 for both characteristics). Respondents to this study were much more likely

to fail to report their income than were census respondents (26.8% vs 9.6%,

p<0.001). Excluding respondents with relevant missing data (n=25) is unlikely to

have produced any serious bias. Participants with incomplete (n=158) data scored

similarly on all outcome variables, but were older, had less education, lower incomes

and more often reported having a disability compared with participants with

complete data (n=271) (Table 7.1).

180

Table 7.1: Sample characteristics and descriptive results

All (n=429) a

Partial Completers (n=158) b

Completers (n=271) p c

Gender (males) 100 (23.4%) 35 (22.4%) 65 (24.0%) .813 Age b 46 (18, 99) 57.50 (19, 94) 39.00 (18, 84) .000 Country of Birth .569

Australia 313 (73.6%) 116 (75.3%) 197 (72.7%) Overseas 112 (26.4%) 38 (24.7%) 74 (27.3%)

Aboriginal/ Torres Strait Islander 6 (1.4%) 3 (1.9%) 3 (1.1%) .674 Australian South Sea Islander 1 (0.2%) 1 (0.6%) 0 (0%) .368 Disability 18 (4.2%) 11 (7%) 7 (2.6%) .043 Non-English Speaking Background 47 (11.5%) 15 (9.5%) 32 (11.8%) .523 Highest Education Completed .000

Primary or less 48 (11.5%) 33 (22.6%) 50 (18.5%) Secondary 129 (30.9%) 55 (37.7%) 33 (12.2%) Post Secondary 168 (40.3%) 41 (28.1%) 123 (45.4%)

Post University 72 (17.3%) 17 (11.64%) 65 (24.0%) Household Characteristics Number of adults (18 yrs +) .023

One 132 (32.0%) 57 (40.4%) 75 (27.7%) .007 Two 207 (50.2%) 65 (46.1%) 142 (52.4%) Three or more 73 (17.7%) 19 (13.5%) 54 (19.9%)

Number of teenagers (13-17yrs) .237 None 369 (89.6%) 130 (92.2%) 239 (88.2%) One or more 43 (10.4%) 11 (7.8%) 32 (11.8%)

Number of Children (6-12 years) .614 None 369 (89.6%) 128 (90.8%) 246 (88.9%) One or more 43 (10.4%) 13 (9.2%) 30 (11.1%)

Number of Children (2-5 years) .006 None 384 (93.2%) 138 (97.9%) 246 (90.8%) One or more 28 (6.8%) 3 (2.1%) 25 (9.2%)

Number of Infants (<2 years) 389 (94.4%) 139 (98.58%) 250 (92.3%) .006 None 23 (5.6%) 2 (1.42%) 21 (7.8%) One or more

Minors (under 18 yrs) None 313 (76.0%) 117 (83.0%) 196 (72.3%) One or more 99 (24.0%) 24 (17.0%) 75 (27.7%) .021

Annual Household Income .000 Not reported 95 (22.1%) 64 (40.5%) 50 (18.5%) Less than $20,800 77 (17.9%) 44 (27.8%) 33 (12.2%) $20,800-$77,999 160 (37.3%) 37 (23.4%) 123 (45.4%) $78,000 or more 97 (22.6%) 13 (8.2%) 65 (24.0%)

Descriptive Results Purchasing Index b 69.76 (12.29) 69.79 (13.71) 69.75 (11.87) .977 Vegetable Confidence Index b 117 (21, 126) 116 (21, 126) 117 (25, 126) .432 Technique Confidence Index b 53 (14, 60) 53 (14, 60) 53 (20, 60) .589 a 158 participants partially completed surveys however due to missing data, corresponding figures are based on fewer than 158 responses. b Continuous data are presented as mean (standard deviation) or median (minimum, maximum) for normally and non-normally distributed variables, respectively. c Probability for difference between completers and partial completers by chi-square test (categorical variables), t-test or Mann Whitney U-Test (normal and non normal continuous variables). NB – While there are 429 participants in total, statistics presented for each variable among ‘all participants’ are based on the number of participants who had completed those variables.

181

Table 7.2: Bivariate predictors of respondents’ confidence to prepare vegetables, confidence to use a variety of cooking techniques and vegetable purchasing

* statistically different from referent category by ANOVA at p<0.05 r= Spearman’s R p= probability that at least one category is statistically different to the other according to Mann-Whitney test for binary or Kruskal Wallis test for other categorical variables. For continuous variables p= for Spearman’s R. a Indigenous Australians include anyone who self identified as being Aboriginal or Torres Strait Islander, or Australian South Sea Islander NESB=Non-English Speaking Background

Confidence –

Vegetable

Confidence – Techniques

Buying Vegetables

Confidence – techniques r=0.591,p<.001 r=0.366, p <.001 Confidence – vegetables r=0.591, p<.001 r=0.487, p p<.001 Age (years) r=0.124, p<.05 r=0.264, p<.001 r= 0.232, p<.001 Gender

Male 110 (21, 126) 52 (14, 60) 67.13 (64.30, 69.96) Female 118 (41, 126) 53 (20, 60) 70.65 (69.19, 72.10) p<0.001 p=0.05 p=0.02

Indigenous Australian? a Yes 116 (92, 122) 55 (34, 59) 65.50 (45.10, 85.90) No or missing 117 (21, 126) 53 (14, 60) 69.81 (68.51, 71.11) p=0.54 p=0.79 p=0.49

NESB? Yes 105 (61, 126) 49 (14, 60) 69.21 (65.06, 73.36) No or missing 118 (21, 126) 53 (20, 60) 69.83 (68.46, 71.19) p<0.001 p<0.001 p=0.77

Has a disability? Yes 103.5 (37, 126) 50.5 (35, 60) 67.50 (56.79, 78.21) No or missing 117 (21, 126) 53 (14, 60) 69.84 (68.54, 71.14) p=0.14 p=0.98 p=0.52

Country of Birth Australia 118.5 (21, 126) 53 (20, 60) 70.25 (68.79, 71.70) Overseas 114 (25, 126) 51 (14, 60) 68.30 (65.52, 71.08) p=0.08 p=0.03 p=0.19

Household Income (gross) Prefer not to say 112 (41, 126) 53 (14, 60) 70.99 (68.43, 73.55) < $20,800 p.a. 108 (37, 126) 51 (25, 60) 66.98 (62.94, 71.02) $20,800-77,999 p.a. 119 (21, 126) 52 (20, 60) 70.37 (68.34, 72.40) > $78,000 p.a. (ref) 121 (80, 126) 53 (24, 60) 69.01 (66.37, 71.66) p<0.001 p=0.80 p=0.26

Education Primary 105 (37, 126) 55 (35, 60) 73.79 (67.28, 80.29)* Secondary 116 (42, 126) 53 (27, 60) 70.15 (67.81, 72.49) Post Secondary 117 (21,127) 52 (20, 60) 69.43 (67.56, 71.30) Post University (ref) 121 (93, 126) 52 (24, 60) 68.13 (65.33, 70.93) p=0.006 p=0.056 p=0.31

Adults in Household One 111 (21, 126) 51 (14, 60) 65.40 (62.77, 68.02)* Two 119 (42, 126) 53 (20, 60) 71.52 (69.74, 73.29) Three or more 119 (80, 126) 54 (30, 60) 72.01 (69.74, 74.29) p=0.002 p=0.04 p<0.001

Minors in Household None 116 (21, 126) 53 (14, 60) 72.10 (69.72, 74.49) One or more 119 (41, 126) 53 (27, 60) 68.99 (67.46, 70.52) p=0.03 p=0.21 p=0.04

182

7.4.2. Confidence to cook among socio demographic subgroups

At the bivariate level, household chefs who were women, older, had higher

education, had greater household incomes, came from an English speaking

background, lived with at least one other adult, or lived with at least one minor in

the household had significantly greater confidence to prepare vegetables (Table 7.2).

Confidence scores were not substantially different among other demographic

subgroups. (Too few persons with a disability (n=12) and Indigenous persons (n=5)

were included to draw conclusions about these population subgroups, or include

these variables in further analyses.) Household chefs who are women, older, born in

Australia, come from an English speaking background and lived with other adults

had significantly greater confidence to use a variety of cooking techniques. Greater

confidence was observed among people with higher incomes and less education,

however the differences were not statistically significant.

Table 7.3 summarises multivariable associations between sociodemographic

characteristics and the odds of having very low confidence to cook. In the fully

adjusted model, lacking confidence to prepare vegetables was significantly

associated with being male, having low education, not disclosing income and not

living with other adults. Differences were not statistically significant however,

compared with young respondents (<30 years), those aged 30-45 and 45-60 were less

likely to lack confidence, while those aged 60 and over were more likely to lack

confidence to prepare vegetables. Respondents were less likely to lack confidence to

prepare vegetables using a variety of techniques if they lived with other adults or

lived with minors and were more likely to lack confidence if they had less education.

Unlike the vegetable scale, there was very little difference between men and women,

or older and younger persons, in terms of confidence to use a variety of cooking

techniques. Having a non-English-speaking background was associated with lacking

confidence on both scales, however not significantly.

183

Table 7.3: Multivariable adjusted a sociodemographic differences in the odds of having very low confidence to prepare vegetables and to use a variety of cooking techniques b

Confidence – Vegetables (n=401) c Confidence – techniques (n=401) d Income Education Both Income Education Both Income

Not reported 4.63 (1.74, 12.33) - 3.51 (1.28, 9.62) 2.54 (1.03, 6.28) - 1.99 (0.78, 5.06) <$20,800pa 2.98 (1.02, 8.72) - 2.07 (0.67, 6.33) 2.32 (0.85, 6.36) - 1.72 (0.60, 4.95) $20,800-77,999 p.a. 2.05 (0.78, 5.37) - 1.74 (0.65, 4.66) 1.44 (0.60, 3.46) - 1.20 (0.49, 2.94) ($78,000pa+) 1.00 (referent) - 1.00 (referent) 1.00 (referent) - 1.00 (referent)

Education (Post-university) - 1.00 (referent) 1.00 (referent) - 1.00 (referent) 1.00 (referent) Post secondary - 2.35 (0.95, 5.80) 2.03 (0.81, 5.10) - 2.17 (0.89, 5.31) 2.01 (0.82, 4.95) Secondary - 2.27 (0.87, 5.88) 1.89 (0.71, 5.01) - 3.21 (1.26, 8.14) 2.81 (1.09, 7.23) Primary or less - 8.79 (2.88, 26.84) 6.60 (2.08, 20.91) - 4.16 (1.31, 13.18) 3.30 (1.01, 10.75)

Gender: Male 1.99 (1.12, 3.54) 2.13 (1.19, 3.82) 2.25 (1.24, 4.08) 0.84 (0.46, 1.52) 0.92 (0.50, 1.68) 0.91 (0.49, 1.69) Age

(Under 30 yrs) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) 30-45 0.59 (0.25, 1.37) 0.65 (0.28, 1.49) 0.60 (0.25, 1.42) 0.79 (0.34, 1.85) 0.86 (0.36, 2.03) 0.86 (0.36, 2.06) 45-60 0.69 (0.33, 1.44) 0.61 (0.29, 1.30) 0.59 (0.27, 1.26) 0.99 (0.49, 2.02) 1.00 (0.48, 2.10) 1.02 (0.48, 2.15) 60 years + 2.70 (1.30, 5.58) 2.09 (0.95, 4.57) 1.81 (0.82, 4.01) 1.37 (0.65, 2.89) 1.35 (0.61, 3.02) 1.19 (0.53, 2.69)

Adults in household (One) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) Two 0.57 (0.31, 1.02) 0.50 (0.28, 0.88) 0.53 (0.29, 0.98) 0.46 (0.26, 0.83) 0.39 (0.17, 0.88) 0.43 (0.24, 0.78) Three or more 0.67 (0.30, 1.49) 0.68 (0.31, 1.50) 0.68 (0.30, 1.53) 0.45 (0.20, 1.00) 0.41 (0.23, 0.71) 0.41 (0.18, 0.94)

Minors in household (none) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) 1.00 (referent) At least one 0.98 (0.49, 1.98) 0.82 (0.40, 1.70) 0.82 (0.39, 1.72) 0.24 (0.10, 0.57) 0.22 (0.09, 0.55) 0.22 (0.09, 0.53)

NESB: yes 1.59 (0.70, 3.49) 1.56 (0.68, 3.59) 1.44 (0.62, 3.34) 1.65 (0.73, 3.73) 1.75 (0.77, 3.99) 1.73 (0.75, 3.99) a Mutually adjusted for all other variables in the table b ‘Very low confidence’ is defined as the bottom quintile of the total of respondents’ confidence scores on each confidence scale c Multiple logistic regression estimates and 95% confidence intervals d Multilevel logistic regression estimates and 95% confidence intervals, accounting for design clustering (random intercept model). Household Income – gross annual household income. Education – highest qualification completed. Adults – number of adults aged 18 and older living in the household, including the respondent. Minors - number of minors aged under 18 living in the household. NESB= Non English Speaking Background Referent categories are in parentheses

184

We explored (post-hoc) whether the deliberate selection of the main household chef

in this study could explain the lack of a theoretically expected large gender

difference in confidence to use a variety of techniques. Our study was not powered

to examine interactions, so the size of the gender–confidence relationship was

described separately among one-adult and multiple-adult households (where there is

some choice as to who cooks). In households with only one adult, men were much

more likely than women to lack confidence to prepare vegetables (OR 3.26, 95% CI:

1.16 to 9.15), while smaller gender differences were observed within multiple-adult

households (OR 1.50, 95% CI 0.67 to 3.36). Men living without other adults tended

to be more likely than women to lack confidence to use a variety of cooking

techniques (OR: 1.61, 95% CI 0.61 to 4.25) while men living in multiple-adult

households tended to be less likely than women to lack confidence (OR: 0.56, 95%

CI 0.22 to 1.44).

Age trends were also examined separately for men and women (Table 7.4). Too few

men lived with minors to remove this source of confounding by adjustment, so

stratified analyses were limited to those men and women who did not live with

minors. Due to the small number of males in this study, confidence intervals were

wide and none of the relationships were statistically significant, however age trends

appeared to vary by sex. For women, results were similar to those reported for the

overall sample. However, men in the middle age categories did not have the same

tendency towards greater confidence, unlike women.

185

Table 7.4: Age differences in confidence to cook among men and women who

usually cook for their household, adjusted for other demographic characteristics a

Odds of “very low” confidence b

(OR, 95% CI) Odds of “very low” confidence b

(OR, 95% CI)

Full Sample Adults without minors in the

household Men (n=94) Women (n=307) Men (n=79) Women (n=225)Confidence to prepare vegetables Age (years)

<30 1.00 referent 1.00 referent 1.00 (referent) 1.00 referent

30-45 1.04

(0.16, 6.60) 0.50

(0.17, 1.46) 2.54

(0.28, 22.89) 0.43

(0.104, 1.78)

45-60 0.93

(0.15, 5.68) 0.45

(0.18, 1.13) 1.01

(0.12, 8.44) 0.46

(0.16, 1.28)

60+ 4.80

(0.68, 34.14) 1.45

(0.58, 3.62) 6.60

(0.73, 59.35) 1.42

(0.54, 3.77) Confidence to use a variety of cooking techniques c Age (years)

<30 n/a 1.00 (referent) 1.00 (referent) 1.00 (referent)

30-45 n/a 0.85

(0.30, 2.37) 5.22

(0.39, 68.95) 0.47

(0.17, 1.32)

45-60 n/a 1.19

(0.50, 2.79) 1.71

(0.13, 21.74) 0.65

(0.18, 2.32)

60+ n/a 1.03

(0.40, 2.62) 7.52

(0.60, 97.54) 1.19

(0.47, 3.00) a Logistic regression models also adjust for income, education, adults in household, and minors in household (for the full sample only) b ‘Very low confidence’ is defined as the bottom quintile of the total of respondents’ confidence scores on each confidence scale c also corrected for clustering using a random intercept model n/a not assessed due to model instability

Confidence to cook was comparatively lacking among low socioeconomic groups

(Table 7.3). Not all comparisons attained statistical significance, however,

respondents from low-income (OR 2.98, 95% CI: 1.02 to 8.72) and middle-income

households were more likely to lack confidence to prepare vegetables (OR 2.05, 95%

CI: 0.78 to 5.37) compared with high income households. Compared with high-

income households, low- income (OR: 2.32, 95% CI: 0.85 to 6.36) and middle-

income (OR: 1.44, 95% CI: 0.60 to 3.46) households were also more likely to use a

variety of cooking techniques. Compared with household chefs with a post-graduate

university education, those with only a primary education were much more likely to

lack confidence to prepare vegetables (OR 8.79, 95% CI: 2.88 to 26.84). Similar

results were observed for intermediate educational categories but these did not reach

statistical significance. Respondents in all educational categories had over twice

the odds of lacking confidence to cook vegetables using a variety of techniques

186

compared with respondents who had a post graduate university education, however

the differences were statistically significant for those with a primary or secondary

education only. Educational differences were more pronounced than income-based

differences but the associations between each socioeconomic marker and confidence

were largely independent. Educational and income effects were both attenuated

slightly but not wholly after mutual adjustment, as odds ratios were relatively lower

by 20-30%.

7.4.3. Confidence to cook and household buying habits

Substantially fewer vegetables were purchased regularly in households where the

person doing most of the cooking had less confidence to prepare vegetables, or to use

a variety of cooking techniques, after adjustment for socio-demographic

characteristics (Table 7.5). After mutual adjustment to account for the moderately

strong relationship between the two scales (r=0.59, p<0.001), the effect of

confidence to prepare vegetables remained largely unaffected, but confidence to use

a variety of cooking techniques was diminished to virtually no effect (largest

difference: 2.37, 95% CI: -6.04 to 1.31).

7.4.4. Household buying habits among sociodemographic subgroups

Bivariate results (Table 7.2) show households purchased a greater variety of

vegetables regularly when household chefs were female, older, had greater

confidence to prepare food, lived with other adults or lived with minors. Purchasing

was substantially lower when household chefs had higher levels of education or were

Aboriginal or Torres Strait Islander, although these associations were not statistically

significant. In the fully adjusted demographic model (Table 7.5), greater purchasing

of vegetables occurred when the household chef was older, lived with other adults in

the household, had at least one minor in the household, while education and

household incomes were not substantially or significantly related to purchasing

vegetables. The significant crude difference between the top and bottom educational

categories was no longer significant or sizeable after adjusting for confounding by

other sociodemographic characteristics (Table 7.5).

187

Table 7.5: Socio-demographic characteristics and confidence to cooka as predictors of household vegetable purchasing b among adults who prepare food for their households c Socio-demographic differences in vegetable purchasing Confidence to cook and vegetable purchasing

Income model Education model Mutually adjusted

Model Income model Education model Mutually adjusted

Model Income d not reported 1.91 (-1.89, 5.70) 1.78 (-2.15, 5.71) 4.96 (1.50, 8.42)* 2.33 (-1.49, 6.15) 4.98 (1.52, 8.44)* <$20,800pa 0.22 (-4.38, 4.82) 0.09 (-4.69 4.87) 3.06 (-1.12, 7.24) 0.62 (-4.03, 5.26) 3.14 (-1.05, 7.33) $20,800-77,999 2.05 (-1.27, 5.37) 1.99 (-1.44, 5.42) 3.28 (0.29, 6.27)* 1.96 (-1.37, 5.29) 3.28 ( 0.29, 6.28)* $78,000pa + 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) Education e (post-university) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) post secondary 1.01 (-2.40, 4.42) 0.67 (-2.80, 4.15) 2.55 (-0.48, 5.58) 1.51 (-1.89, 4.90) 2.73 (-0.31, 5.77) secondary 0.56 (-3.14, 4.27) 0.11 (-3.71, 3.94) 1.60 (-1.73, 4.92) 0.50 (-3.23, 4.24) 1.80 (-1.55, 5.14) primary or less 1.22 (-4.26, 6.71) 0.95 (-4.74, 6.64) 6.09 (1.09, 11.10)* 1.48 (-4.07, 7.02) 6.29 (1.25, 11.32)* Gender: male -2.28 (-5.15, 0.59) -2.39 (-5.30, 0.52) -2.27 (-5.19,0.66) 0.04 (-2.52, 2.61) -2.30 (-5.16, 0.56) -0.04 (-2.63, 2.55) Age Years (under 30) 0.00 (referent) 0.00 (referent) 0.00 (referent) 1.00 (referent) 0.00 (referent) 0.00 (referent) 30-45 2.84 (-0.83, 6.51) 2.97 (-0.65 6.60) 2.88 (-0.81, 6.57) 1.63 (-2.29, 4.13) 2.56 (-1.04, 6.16) 1.07 (-2.15, 4.28) 45-60 7.73 (4.46, 11.00)* 7.94 (4.64, 11.24)* 7.76 (4.41, 11.09)* 4.63 (1.67, 7.59)* 6.77 (3.47, 10.06)* 4.72 (1.74, 7.69)* 60 + 9.92 (5.90, 13.94)* 9.80 (5.56 14.05)* 9.84 (5.55, 14.12)* 8.69 (4.93, 12.45)* 8.89 (4.70, 13.08)* 8.69 (4.92, 12.47)* Adults f (one) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent)

two 7.27 (4.27 ,

10.27)* 7.19 (4.32, 10.05)* 7.20 (4.17, 10.23)* 5.61 (2.98, 8.24)* 6.15 (3.18, 9.13)* 5.47 (2.82, 8.12)* three or more 7.56 (3.70, 11.43)* 7.74 (3.96, 11.52)* 7.49 (3.59, 11.39)* 6.10 (2.73, 9.48)* 6.47 (2.65, 10.29)* 5.98 (2.59, 9.38)* Minors g (none) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) 0.00 (referent) at least one 3.31 (0.29, 6.32)* 3.18 (0.12, 6.23)* 3.29 (0.22, 6.35)* 3.14 (0.49, 5.79)* 2.56 (-0.45, 5.56) 2.86 (0.17, 5.55)* NESB h: yes -0.65 (-4.69, 3.39) -0.31 (-4.34, 3.72) -0.65 (-4.71, 3.42) 1.62 (-1.96, 5.21) 0.37 (-3.61, 4.34) 1.72 (-1.87, 5.31)

188

Table 7.5 continued: Socio-demographic characteristics and confidence to cooka as predictors of household vegetable purchasing b among adults

who prepare food for their households c

Socio-demographic differences in vegetable purchasing Confidence to cook and vegetable purchasing

Income model Education model Mutually adjusted

Model Income model Education model Mutually adjusted

Model Confidence (vegetables)

Quartile 1 - - - -19.28

(-22.89,-15.66)* -18.60

(-22.54,-14.66)* Quartile 2 - - - -8.98 (-12.03,-5.93)* -8.78 (-12.11,-5.46)* Quartile 3 - - - -5.46 (-8.37, -2.55)* -5.32 (-8.27, -2.36)* Quartile 4 - - - 0.00 (referent) 0.00 (referent) Confidence (techniques) Quartile 1 - - - -9.09 (-12.88, -5.30)* -2.37 (-6.04, 1.31) Quartile 2 - - - -4.69 (-8.03, -1.34)* 0.15 (-3.12, 3.41) Quartile 3 - - - -4.34 (-7.57, -1.12)* -1.44 (-4.42, 1.54) Quartile 4 - - - 0.00 (referent) 0.00 (referent)

* statistically different at p<0.05 from referent group (indicated in parentheses) a Confidence (veg) is self reported confidence (1-6, not at all to very confident) to prepare 21 vegetables, divided into quintiles and Confidence (tech) is the self reported confidence (1-6, not at all to very confident) to use 10 techniques to prepare vegetables, divided into quintiles b Vegetable purchasing is measured as a 21-105 scale showing frequency (1-5, never to always) of buying 21 vegetables. c multiple linear regression models are based on the 336 respondents with complete data on variables in mutually adjusted models d Income is measured as gross annual household income e Education – highest qualification completed f Adults – number of adults aged 18 and older living in the household, including the respondent g Minors - number of minors aged under 18 living in the household h Non English Speaking Background

189

7.5. Discussion

This study examined confidence rather than actual cooking skills, so it is uncertain to

what extent our findings would apply to cooking skills measured by other means.

However, the confidence measures employed in this study have definite relevance,

being strongly linked to household purchasing habits. Households more regularly

purchased a variety of vegetables when the main household chef had more

confidence to prepare them. Confidence to cook with a variety of techniques was

less relevant to the purchase of vegetables.

Men, lower socioeconomic groups, and younger adults have been shown to have low

intakes of fruits and vegetables compared with their female, higher socioeconomic,

older counterparts (105, 375), and both men and low socioeconomic groups lacked

confidence to prepare vegetables in this study relative to women and higher

socioeconomic groups. However, factors other than confidence seem to be important

in the dietary patterns of different socio-demographic sub-groups. Respondents

living without other adults, without minors, and young respondents still scored lower

on the purchasing index than their counterparts, even after accounting for differing

confidence levels.

Living without other adults was associated with less vegetable purchasing and

lacking confidence to cook on both scales. This finding has at least two plausible

interpretations in view of the study’s focus on household chefs. Confidence to cook

may be one factor considered when adults living together decide who will do the

majority of cooking. If so, one might expect to find more people lacking confidence

in households with only one adult, where the responsibility for cooking falls by

default. Additionally, cooking for oneself has very different meanings and

motivations to cooking for others (365), and people who prepare food to meet the

desires of others as well as themselves may become confident in preparing a wider

range of vegetables. Living with minors was associated with less chance of lacking

confidence to cook on the technique scale. Possibly the need to prepare food for

infants, toddlers and children that they will be willing to eat has contributed to

greater confidence in using a range of techniques.

190

This study did not find a strong or consistent age gradient in confidence to cook,

contrary to the findings of the UK study (165, 284) and the purported secular

‘decline’ in cooking skills alluded to in the literature (266, 267, 376). Changes in the

division of household labour imply age trends might vary between men and women

(377), however, we lacked sufficient numbers to adequately examine age trends for

men, since our sample focused on household chefs and included approximately three

women for every man.

Findings from one of our confidence outcomes were consistent with the gender

differences in self-rated cooking skills (266) and in confidence to cook according to a

wide range of measures reported in other studies (284). Gendered avoidance of

cooking cannot entirely explain the present findings as this study included only men

and women who admitted primary responsibility for household cooking. Although

not statistically conclusive, the post-hoc analysis is suggestive that gender

differences in the general population would be larger than those in this study of

people who cook for their households. The lack of large gender differences on the

techniques scale may also suggest the gender differences are specific to vegetables.

Studies that examine cooking skills more holistically and include both people who do

and do not cook could examine this, along with other complexities associated with

domestic living arrangements and gender differences in cooking skills.

This study demonstrated a socioeconomic patterning in confidence to cook, as did

the U.K. National Health and Lifestyles study (284) . In the UK study, income was

the strongest socioeconomic predictor of confidence, whereas in our study

population, confidence was much more strongly linked to respondents’ education

than to household income. Considering how cooking skills are acquired, several

mechanisms could underpin the relationship between education and confidence to

cook observed in this study. Mothers are the most commonly reported source of

learning to cook, followed by school and cookbooks, at least in the U.K (165, 284).

Cooking skills taught and acquired in the educational system may have contributed

to the greater confidence to cook among people with more education. Cooking skills

are taught as a part of the curriculum in Australian schools at the primary and

secondary levels, although not universally (378, 379), and have been during most of

191

the twentieth century, at least for women (268). Post-school training seldom

includes food preparation, therefore the differences between those with and without

post-school qualifications probably arise by another mechanism. Education provides

generic learning skills, which may facilitate the ability to learn to cook through self-

directed informal learning. Cooking is a common part of adults’ informal learning

(380). The typical dietary patterns documented in various studies (33, 89, 92, 93,

97, 99) may provide limited exposures to varied foods and food preparation

techniques among less-educated groups, which may affect later confidence to prepare

a variety of healthy foods. Experience and familiarity, along with knowledge and

training, increases self-efficacy (confidence) for performing tasks, such as preparing

foods (300). An exposure – confidence – behaviour cycle could transmit inequalities

across generations, as adult socioeconomic position is highly linked to parental SEP

(381, 382). However it is beyond the capacity of this cross-sectional study to

determine which of these mechanisms is most likely producing the results observed.

The extent to which confidence may mediate socioeconomic differences in

purchasing could not be directly examined as this study found no substantial

relationship between SEP and vegetable purchasing. The adjustment for

demographic variables not included in many other studies (33, 89, 91-93, 97, 99,

105) substantially attenuated socioeconomic differences in vegetable purchasing in

this study. Less adjusted results (not reported) were similar to a comparable

previous, Brisbane-based study (89).

Our study had several limitations which should be mentioned to avoid over-

generalising results. The measures showed acceptable face validity and repeatability,

an improvement over many studies to date, however content validity has not been

fully established. Also, validity was not specifically examined within different

ethnic groups, whose typical range of foods and cooking techniques may diverge

from the mainstream. Thus, their apparent differences in cooking skills and

purchasing may be artificial. The response rate was typical for mail-style, health-

related surveys (383), but was not high enough to ensure the sample is representative

of the wider community. Also, households were accessed via occupied, private,

residential dwellings. Although living in non-private residential settings is more

common among low than high socioeconomic groups, associated bias is likely small

192

as over 99% of people enumerated in the 2001 census lived in the types of dwellings

included in our study (344). Further, a small proportion of dwellings may have

contained more than one household (for example, group households in which the

premises are shared but food and finances are not). Our study may have slightly

overestimated confidence and buying habits and underestimated socioeconomic

differences if low socioeconomic households are more likely co-locate in one

dwelling and if the household chefs most interested in food elected to do the survey.

Finally, the precision of the study may have been insufficient to detect some

meaningful differences as statistically significant due a small number of participants

in certain sub-groups, and overall, as evidenced by consistently wide confidence

intervals. Larger studies that over-sample minority groups may be necessary to

overcome this.

This study provides empirical evidence that confidence to prepare vegetables is

associated with their purchase. The contrast between the results for the technique

and vegetable confidence scales suggests that interventions to alter dietary

behaviours by improving cooking skills may need to focus specifically on the healthy

foods they aim to promote. The socioeconomic patterning of confidence to cook

observed in the UK in the 1990s is also evident a decade later in Brisbane, Australia,

and therefore, may be a widespread phenomenon in western nations. Similarities in

the experience of dietary inequalities and inadequate fruit and vegetable consumption

suggest these issues may also have relevance within the United States population.

The associations between confidence to cook, socioeconomic position and vegetable

purchasing were suggestive that confidence to cook may contribute to socioeconomic

differences in vegetable intake, and further consideration of this issue by larger,

better equipped studies appears warranted.

193

Chapter 8: Discussion

8.1. Overall findings As part of the broader understanding of contributors to socioeconomic inequalities in

diet and health outcomes (Figure 1.1), this thesis examined whether cooking skills

and the accessibility and affordability of fruits and vegetables in the local

environment mediate socioeconomic differences in fruit and vegetable purchasing

among Brisbane residents. Findings from the secondary analysis of the Brisbane

Food Study provided little evidence that the food retail environment contributed to

socioeconomic differences in fruit and vegetable purchasing, as in multivariable

models, estimates of the relationship between socioeconomic position and fruit and

vegetable purchasing were not substantially affected by adjusting for features of the

food retail environment. Findings from the cooking skills study generally supported

mediation of socioeconomic differences in fruit and vegetable purchasing by

confidence to cook, however mediation could not be examined in the same manner

as the secondary data analyses, for reasons later discussed. Along with the other

possible mediators of socioeconomic differences in dietary behaviour, such as

nutritional knowledge, attitudes and food preferences, it appears that confidence to

cook may form part of the explanation for socioeconomic differences in fruit and

vegetable purchasing, while neighbourhood shop access, prices and food availability

do not, at least within Brisbane.

The more detailed results of the secondary analyses show why affordability did not

mediate socioeconomic differences in fruit and vegetable purchasing in Brisbane,

when framed in terms of the conceptual links between socioeconomic position,

accessibility, affordability and fruit and vegetable purchasing (Figure 2.1). Having

more supermarkets and greengrocers nearby was associated with small, significant

increases in the odds of purchasing some fruit and vegetable items, however, in

Brisbane, low- and high-socioeconomic areas contained very similar numbers of

these shops. Compared with high socioeconomic areas, low socioeconomic areas

tended to be closer to shops and to have lower prices, but also less in-store

availability of fruits and vegetables. However, socioeconomic differences in fruit and

194

vegetable purchasing were not reduced by distance or increased by in-store

availability because distance to shops and in-store availability were not associated

with purchasing fruit and vegetable items. Similarly, the average differences

between low- and high- socioeconomic areas in the price of a basket of vegetables

($1.10-3.12, i.e. 9c – 26c per kilogram) and fruits ($0.24-$1.18, i.e. 5c – 23c per

kilogram) may be too small to affect socioeconomic differences in purchasing, in

view of the size of the relationships observed between price and purchasing fruit and

vegetable items. (The largest relationships observed were an increase in the odds of

not buying capsicum and a decrease in the odds of not buying broccoli of

approximately two-fold for each additional dollar per kilogram.)

In the cooking skills study, the relationships between socioeconomic position,

confidence to cook and vegetable purchasing were consistent with confidence to

cook mediating socioeconomic differences in vegetable purchasing, as outlined in the

conceptual model (Figure 2.1). Adults with less education or lower incomes

possessed less confidence to prepare vegetables, and less confidence to cook

vegetables using a variety of techniques compared with adults with more education

and higher incomes. According to both confidence scales, lacking confidence to

cook was in turn associated with less vegetable purchase. Mediation of

socioeconomic differences in vegetable purchasing by confidence to cook could not

be examined by changes in estimates before and after adjustment in the manner used

in the secondary data analyses, as socioeconomic differences in vegetable purchasing

were not apparent in this study, for reasons discussed within chapters 6 and 7.

8.2. Findings in context with other research

8.2.1. Socioeconomic differences in shop access, in-store availability and price

Some of these findings of this thesis are consistent with the broader international

research, while there are also some marked differences. The shop access findings

differ from those of many studies, mostly from the United States, that have

documented a relative lack of supermarkets, larger or chain stores and an over-

195

abundance of smaller, non-chain stores in lower socioeconomic areas (166-170)

(171-173, 175, 179). The lower purchasing power in poor areas (187), and the

specific avoidance of investing in low-income areas (redlining) (186), were

hypothesised to contribute to the lack of large, well-equipped shops in poor areas.

However, as in this present study, research conducted outside of the United States

has mostly failed to find the phenomenon of worse shop access in poor areas (180,

181). Literature that has emerged subsequent to these thesis findings has provided

further evidence against poorer food access in low-socioeconomic areas outside of

the US. A national study of access across New Zealand found that travel times to

food shops generally and supermarkets specifically were shorter in the most-deprived

compared with the least-deprived areas (384). Likewise, a study in Montreal,

Canada, found comparatively better access to shops in lower socioeconomic census

tracts (385). Possibly, these processes or others operate to a greater extent in the

United States compared with elsewhere.

The tendency of poor areas to cluster towards the inner city may have driven the

socioeconomic differences in shop access noted in the U.S. research (166-173, 175),

as some of these studies have shown that inner city areas tend to be served by smaller

rather than larger shops. The U.S. findings may be a by-product of the “white flight”

that has occurred throughout the U.S., in which many inner-city areas have been left

populated by ethnic minority and low-income groups after higher income, white

families have moved to the outer suburbs (386). Subsequent disinvestment in the

inner-city areas, due to real and imagined declines in the profitability of these areas

may have left some cities in the US with a disproportion of low-income, poorly

provisioned areas around the inner-city(176, 186, 189). There is little evidence this

process has occurred in many of the study settings in the United Kingdom (182, 218)

and Canada (178, 385), and the present, Brisbane-based study did not show evidence

of this type of spatial patterning of disadvantage. Low- and high-socioeconomic

areas were similarly distant from the central business district, and researchers have

noted that Brisbane City has experienced gentrification and an influx of supermarkets

towards the inner city (387). Furthermore the combination of racial and social

polarisation that has occurred within the US may also have contributed to the unique

findings there, as the Detroit study provided some evidence that the effects of

poverty were more notable as racial segregation increased and vice versa (176).

196

Non-spatial contextual differences between countries may also be important. The

focus on income- or poverty-based socioeconomic disadvantage in the U.S. research

and the focus on multiple deprivation or disadvantage in the U.K. and Australia may

serve to highlight that the experience and outworking of disadvantage may also vary

between contexts that have different welfare policies and levels of social polarisation

(388).

In part, the different findings from Brisbane compared with the United States may

also be methodological, as this present study eliminated some of the probable biases

in previous U.S. studies by measuring shop access on a scale that standardizes for

geographical size (catchments of 2.5 km radius). The U.S studies that noted low-

income areas have fewer shops per administrative boundary may have found no

difference between high- and low- socioeconomic areas if access had been measured

per square kilometre. The location of low-income areas towards the inner-city may

mean these low-income zip-codes, census tracts and census blocks are smaller than

corresponding high-income areas, in view of high population density occurring near

the inner city. Figures presented in the ARIC study (170) indicated the low-income

zip codes were approximately three times smaller than the high-income zip codes,

and also found a three-fold difference in the prevalence of shops in low- compared

with high-income areas. The difference in findings of the present, Brisbane-based

study and the bulk of the U.S. research is unlikely to be entirely methodological, as a

recent, large, U.S. study found socioeconomic differences in shop access in the U.S,

having adjusted for geographic size (173).

In Brisbane, there were small, non-significant differences in the availability of fruits

and vegetables, with the measures of availability being lower in low- compared with

high-socioeconomic areas for convenience stores (-10 to -17%), but not for

supermarkets and greengrocers (-5% to +3%). Studies are not directly comparable,

as the items examined have differed and previous studies have not standardized for

shop type (171, 181, 191-193), however the present findings fit within the range of

findings elsewhere. Case studies have reported small socioeconomic differences in

fruit and vegetable availability within Glasgow (192) and in St Louis (171) and New

York (191). Larger studies have reported no significant socioeconomic differences

197

in the in-store availability of fruits and vegetables in Newcastle (181) and Glasgow

(193).

The small, non-significant price differences for low- compared with high-

socioeconomic areas (1-11%) fits within the range of mixed findings in the broader

food price research. Studies have reported finding no differences (188, 204), higher

prices (166, 192, 202, 203) or lower prices (181, 193, 197, 201) of food items in low-

socioeconomic areas, or paid by low-socioeconomic households (202), with price

differences being small, as in the present study. Notably, the only studies to report

higher prices for fruits and vegetables specifically have used a non-representative

sample of areas (192) or done so on the basis of only one food item (197), providing

very little evidence for widespread price disparities. As with in-store availability,

findings are not directly comparable as there has been a general failure to standardize

for shop type in the existing literature. Methodological issues aside, true variation in

findings across contexts is also possible, if the various processes that could lead to

higher, or lower, prices occurring in low-socioeconomic areas are more likely to

operate within some contexts than others (see Figure 2.1). Poor areas could

experience lower prices if marketers in these areas are more likely to reduce their

prices to increase sales in their low-income market, based on their comparatively

greater price sensitivity (210). Price discrimination could lead to higher prices in

poor areas, as could greater operating costs (for example from higher crime rates)

(197). Higher prices in low-income areas could be created by the reduced presence

of middle-income households, as it has been argued that competition occurs mostly

for the patronage of middle-income consumers, who can most afford both the time

and money to search around for the cheapest store from which to purchase foods

(203).

8.2.2. Relationships between shop access, in-store availability and price and fruit and vegetable purchasing, intake, or other dietary outcomes

In addition to an absence of substantial socioeconomic differences in shop access, in-

store prices and availability of fruits and vegetables, there was also only minimal

evidence that these environmental characteristics were associated with fruit and

198

vegetable purchasing. The findings of this study are not directly comparable to other

research, having examined whether or not respondents reported buying specific fruit

and vegetable items rather than taking a quantitative measure of fruit and vegetable

purchasing or intake like most studies. However, the results provide further evidence

to the limited research on this topic and are largely consistent with emerging research

from New Zealand, which has detected only small or no relationships between access

to shops and fruit and vegetable intake (359).

In this study, each additional supermarket or greengrocer within 2.5 kilometres was

associated with significantly lower odds of residents not purchasing pineapple (OR:

0.91, 95% CI: 0.87, 0.96) and broccoli (OR: 0.90, 95% CI: 0.82, 0.99), but showed

only very small, non-significant relationships with the odds of not purchasing other

fruit and vegetable items (odds ratios ranged between 0.98 and 1.03). Similarly, for

white Americans (214) and women living in Melbourne, Australia (215),

relationships observed between access to supermarkets or greengrocers and fruit and

vegetable intake have been small and non-significant. A more sizeable and

significant association between supermarket access and fruit and vegetable intake

was observed among African Americans (214), and an increase in fruit and vegetable

intake occurred following the construction of a supermarket in a low-income area of

the U.K., which was substantial among residents with the lowest baseline intakes

(216). Therefore, the relationship between shop access and purchasing was expected

to be modified by household income, however the findings did not support this

expectation, at least in terms of access as measured by the number of nearby shops.

Consistent with findings from other studies, each additional kilometre to the nearest

supermarket or greengrocer showed no large or significant relationship with

purchasing fruit and vegetable items overall. A recent study in New Orleans found

no association between distance to supermarkets and fruit and vegetable intake(220);

greater distance to the nearest supermarket was associated with little or no increase in

fruit intake and vegetable intake among residents of Yorkshire in the UK; and the

Newcastle study (181) reported that distance to the nearest shop (of any type) was

not significantly associated with participants’ fruit and vegetable intakes. In contrast

with the findings for socioeconomically varied populations, a study of Food Stamp

recipients across the US (221) found living further from the store mostly used for

199

shopping (mostly supermarkets) was associated with lower fruit usage and

significantly lower vegetable usage within this low-income population. Similarly, in

this present study, the size of the relationship between distance and purchasing was

more substantial within low- income households compared with middle- and high-

income households. Unfortunately, statistical evidence regarding interactions of the

associations between environmental characteristics and purchasing was limited, as

the BFS was not powered to detect interactions.

Shopping at a supermarket rather than other shop types (174), and having ‘easy’

access to supermarkets as partly defined by shopping at a supermarket (221), have

been associated with fruit and vegetable intakes. The difference between these

studies and the present study, and other studies of supermarket access, may indicate

that the type of store patronised may be an indicator of choice rather than access, as

the foods people wish to consume may influence their decision regarding where to

shop.

The diverse methods employed in examining the relationship between in-store

availability and dietary measures make findings difficult to compare directly,

however the absence of a relationship between in-store availability and purchasing

fruit items differs slightly from most other studies that have noted relationships

between dietary measures and in-store access measured objectively (220, 222, 224,

225) and subjectively (174, 184, 226). Only one recent study, conducted in New

Orleans, has supported an association between in-store availability and fruit and

vegetable intakes (220), and this study examined grocery stores within 100m rather

than supermarkets and greengrocers within 2.5 kilometres. In this thesis, having

examined very common fruit and vegetable items in supermarkets and greengrocers

which tend to be well stocked with fruits and vegetables, there may not have been

enough variation in availability to find associations between availability and

purchase, which may exist for less common items. Being the major purchasing

source for fruits and vegetables, supermarkets and greengrocers were examined in

this thesis. In view of the findings that subsequently emerged (220), the next step

would be to re-examine in-store access within smaller, convenience stores.

200

The absence of substantial or significant relationships between neighbourhood price

and purchase of most items resemble the findings from a study in Yorkshire (219), in

which there was no substantial or significant relationship between Yorkshire

residents’ fruit and vegetable intakes and the prices of fruits and vegetables in the

shops they mostly patronised. The present research found some evidence for a

relationship between price and purchase of some fruit and vegetable items, and in

line with the economic literature that shows price sensitivity varies with income

(210, 211, 236-238), there was a marked tendency for stronger relationships between

price and purchase to occur within low- compared with high- income households.

Overall, there have been too few epidemiological studies of this issue to be confident

about the relationship between neighbourhood variation in price and purchasing or

intake of fruits and vegetables, especially in view of the limitations of this present

study, discussed later, and those of the Yorkshire study, described in the literature

review (Chapter 2).

Other studies examining price, but not in terms of neighbourhood variation, have

been generally supportive of a relationship between price and purchase or other

outcomes, including fruit and vegetable purchase and intake (230, 234, 238, 239,

257). Studies examining facets of affordability other than price, such as perceptions

of the affordability of fruits and vegetables (174, 184, 259), and the relative

importance of price in buying food (42, 198), have sometimes found these to be

associated with fruit and vegetable intake (184, 198), although not in all studies (174,

259). One study reported insufficient detail to be certain of its findings (42).

8.2.3. Confidence to cook

Having found little evidence for the food retail environment in mediating

socioeconomic differences in fruit and vegetable purchasing, this thesis examined

relationships between confidence to cook, socioeconomic position, and vegetable

purchasing. In general, the findings were consistent with a contributory role of

confidence to cook in socioeconomic differences in fruit and vegetable purchasing.

People who purchased food for their households were more likely to lack confidence

to prepare vegetables and to prepare vegetables using a variety of techniques when

201

they had less education or lower household incomes. Lacking confidence to prepare

vegetables, or to use a variety of techniques to cook vegetables, were associated with

buying fewer vegetables on a regular basis for the household. Educational

differences in confidence were more pronounced than income differences, and the

vegetable confidence scale was more strongly associated with vegetable purchasing

than the techniques scale. However, it could not be properly assessed whether

different levels of confidence to cook mediated socioeconomic differences in

purchasing vegetables, as there were no substantial significant socioeconomic

differences in purchasing vegetables in this study, after adjusting for further

confounding factors not examined in many previous studies of income-based

socioeconomic differences in fruit and vegetable purchasing (89, 116, 311) or intake

(33, 91-93, 97, 99, 105). Households with more adults had higher incomes, as might

be expected from the increase in dual-income earning households (366). Households

with more adults purchased a higher variety of vegetables, which may be partly due

to the non-standardised household-level outcome measure employed in the study, or

may be due to true associations between living alone and lower fruit and vegetable

purchasing. Associations between fruit and vegetable intakes and living alone (271),

or marital status (375), have been observed in studies that have used individual-level

outcome measures.

The relationship between confidence to cook and vegetable purchase contributes to

the small body of research that generally supports a relationship between various

cooking-related measures and diet-related outcomes, mostly among non-probabilistic

samples of populations, and sub-populations. Dietary measures have been associated

with the food preparation practices of U.S. adolescents (276) and women seeking

charitable food assistance in the United States (278), with self-rated cooking skills

among older men (271), adult ‘household heads’ in the United Kingdom (266)and

households with children in Vancouver, Canada (281), and have been associated with

cooking-related attitudes and practices in a convenience sample of Australian adults

(277). While the literature has generally supported associations between cooking

skills and diet, some studies have reported null associations, especially among young

adults (279, 280).

202

The socioeconomic patterning of cooking skills in this study is consistent with

findings from the UK National Health and Lifestyles Survey (HLS) (284), reported

in various studies (46, 165, 267, 285), where confidence to cook was highest among

respondents with more education and higher incomes. Occupation was also

significantly associated with confidence to cook in the HLS (284) and in a population

based study conducted in Ireland (266). Findings differed from those of a study

among young adults in the United States, reporting no relationship between parental

educational status and perceived adequacy of cooking skills and resources (279),

which might be accounted for by the different population and measures in this study.

A U.S. study found adolescents of lower socioeconomic position (as defined by

parental education) were more likely to help prepare food for dinner than their higher

socioeconomic counterparts (276), although the effect this may have on adult

cooking skills and eating behaviour is unknown.

8.3. Strengths and limitations

8.3.1. Secondary analysis of Brisbane Food Study

Design and analysis

The research in this thesis had several key strengths in examining whether the food

retail environment varies systematically across socioeconomically diverse areas or is

associated with residents’ dietary behaviours. The issue was examined using a study

that took representative random samples of areas, and an entire population of shops,

which represents an improvement over the earlier studies that have compared non-

probabilistic samples. The study quantified access to shops surrounding, rather than

within, small areas of varied socioeconomic characteristics (census collection

districts). This allowed for the possibility that residents, especially those living

toward collection district boundaries, shop outside their CCD and may represent an

improvement over previous studies that have quantified access and socioeconomic

disadvantage at the same spatial level. Smaller boundaries capture more

homogeneous socioeconomic populations, which arguably provides the best measure

of socioeconomic context (120, 389). However, the scale at which socioeconomic

203

effects may operate is a matter of debate (120, 389-392) as is the spatial scale at

which the food environment might affect behaviours (56, 163)).

Having utilised a multi-level study, it was possible to employ multi-level analysis,

thereby separating individual and area-level variation, and accounting for the non-

independence of observations (118, 393-400). Multilevel analysis has been

described as the “gold standard” for contextual analysis (401), which is the way in

which the relationships between environmental features such as shops and dietary

outcomes have been treated in most of the literature reviewed. However, both the

cross-sectional study and to a lesser extent the multi-level analysis may still have had

some non-independence of observations. The non-independence is created by the

extra similarities that may be shared by areas that have overlapping catchments, or

are spatially closer compared with areas that are more distant (402).

Spatial autocorrelation was not accounted for in the ecological analyses of

socioeconomic differences in food supply features and is not fully accounted for in

multilevel analyses such as the one employed (401, 402), which assume that spatial

correlation can be reduced to within-neighbourhood correlation and the distribution

of “neighbourhoods at risk in space” is random (401). Based on the spatial

distributions of the study areas, socioeconomic disadvantage and shops (Figures

A5.1-2, Appendix 5) there was no reason to suspect the type of smooth spatial

variability that would necessitate a fully spatial approach. Semi-variograms (403)

based on the residuals from the early analyses of shop location (Study 1) indicated a

small degree of spatial autocorrelation may have been present (see Appendix 6).

Thus, standard errors may have been underestimated, confidence intervals appear

more narrow than they ought and some ‘significant’ findings may still be attributable

to chance. With the largely null results, the net result of any bias has been in a

conservative direction, i.e. the true differences between socioeconomic areas in food

supply variables may be larger (in either direction) than indicated by the confidence

intervals, and that food supply variables may increase, or decrease the chances of

purchasing fruits and vegetables to a greater extent than indicated as likely by the

overly narrow confidence intervals. Alternate approaches could be utilised in the

future to overcome some of the limitations. Cross-classified models can somewhat

account for non-independence created by shops being located in multiple catchments

204

(404), however this approach was not taken as it would have required reframing the

research hypotheses to consider shops as units of analyses rather than areas. Spatial-

based approaches that identify and remove spatial auto-correlation, such as those

utilised in ecology (402), and modifications to multi-level models to accommodate

the spatial component of area-based data, such as distance decay parameters (405),

could be utilised in the future.

A key issue in examining fruits and vegetables is their seasonality. In the BFS,

seasonality was controlled by the short study time frames and the wording of the

questions regarding purchase of items when in season, but was not explored in any

depth. The purchasing data were gathered between September and December and

the availability and price survey was conducted in November and December (i.e. in

spring and summer, when fruits and vegetables are highly available and highly

consumed). Greater examination of seasonality, such as through multiple

assessments of dietary behaviours and availability across seasons of greater and

lesser availability of fruit and vegetable items, may provide further information, as

price and availability may be associated with the types and amounts of fruits and

vegetables purchased at some certain times of the year and not others.

Finally, the study was cross-sectional and observational, meaning temporal

associations could not be firmly established, and plausible counter-arguments can be

made. For example, associations between the food retail environment and

purchasing could ensue from effects of purchasing on the food retail environment.

While shops do not contribute to personal socioeconomic position, the presence of

shops (and other associated amenities) may form types of neighbourhoods that attract

a certain type of resident, thereby influencing the socioeconomic composition of

areas. Most known confounders were considered, however residual confounding

could have contributed to results. Quality of fresh produce was not assessed, and

quality is a contributor to produce prices (205). This could have contributed to the

null and varied findings in the relationship between price and purchase, and might

change the interpretation of the absence of price differences across socioeconomic

areas if the fruits and vegetable items were not of equal quality. Having considered

fruit and vegetable items in isolation may have precluded detecting cross-price

effects or any relationships between total costs and purchase. Fast-food outlets were

205

not examined within this thesis, however, due to the interconnectedness of dietary

behaviours, it is plausible that a connection between these shops and fruit and

vegetable purchasing may have been overlooked. Hence, there may have been

residual confounding due to the lack of information on this factor (215).

Measures

The measures available for use in this thesis were both a strength and limitation. The

dietary outcome measure available in this thesis was limited, being semi-quantitative

and not comprehensive of all fruits and vegetables purchased. Associations between

the food retail environment and the quantities of each item purchased or total

household purchasing of vegetables could not be assessed.

Measures of access to shops were an improvement over many previous studies.

Using multiple indicators to capture several facets of access appeared important as

findings were not the same for shop density and distance. Access measures were

based on shop location data obtained via a comprehensive process that rectified the

inaccuracies contained in business and council listings. The measure of shop density

(shops per catchment) is not open to differential misclassification unlike measures of

shop density per zipcode or census tract, as it is based on a consistent geographic size

that is independent of socioeconomic position and population density. This

approach, as opposed to a shops per capita approach to shop density, enabled the

effect of population density to be modelled separately, without having to consider a

greater number of shops servicing a large population to be equal with a small number

of shops servicing a small population. In absence of street addresses of participants,

distance to the nearest shop was approximated by straight line distance from the

centroid of CCDs to shops rather than distances from participants’ homes or road

distances. Straight-line distances will underestimate distance to the nearest shop

relative to more accurate road distances, but are unlikely to systematically vary with

socioeconomic position or dietary measures, and therefore are unlikely to have

introduced any differential bias in results.

206

The measures of price and availability employed in this thesis were a limitation.

Price measures were taken at one time point only, and are known to fluctuate. The

measurement error in this study is likely to be non-differential with respect to

socioeconomic position, though may be differential with respect to diet. The chances

of finding extreme high or low prices might be dependent on consumption patterns,

since retailers adjust prices to prevent spoilage of fresh produce that which have not

sold (205). However, supply-related factors are the key driver of fluctuations in

fresh produce prices (205)so misclassification is more likely to be non-differential.

Studies based on routinely collected business data have had less control over the way

in which price data are collected, but have the advantage of more accurate price

measures, since these have been collected repeatedly over a wider time span. The

ideal design would involve the quality control of deliberate research and the accuracy

of the repeated price measures approach utilised within routine data collection.

In this study, availability was measured in an absolute sense (yes/no) and relative

availability in a semi-quantitative fashion as the number of varieties available in

stores. A shelf-space measure may have been more sensitive in detecting

socioeconomic differences in availability than the measure employed. However,

shelf space may not have been a better overall choice, as it might be more likely to

detect spurious relationships occurring via the practice of food retailers to allocate

shelf space according to sales (224) whereas the absolute absence of an item in shops

could more plausibly influence whether people buy that item. Items that were

usually available were considered as being available, even if they were out of stock

on the day of the audit, which minimises the chance of differential misclassification

that could arise if items appear artificially unavailable, or less available, in shops

where they have been quickly consumed. More obviously, the focus on very

common fruits and vegetables meant that items were almost universally available in

supermarkets and greengrocers, which may have precluded detecting associations

between availability and dietary patterns, or detecting socioeconomic differences in

availability that possibly might exist for more exotic fruits and vegetables.

207

Generalisabilty

The data for the secondary analyses were collected in Brisbane in the year 2000. A

number of differences across cities and changes over time in food retailing and

socioeconomic disadvantage may influence whether or not similar results are likely

to be obtained in other Australian cities, or at later time points.

The spatial patterning of socioeconomic disadvantage is probably important in terms

of whether or not the food retail environment contributes to socioeconomic

inequalities in diet-related behaviours. The spatial patterning of poverty and social

segregation varies across cities (388), and therefore the findings from Brisbane may

not be replicated in dissimilar Australian cities. Appendix 5 presents thematic maps,

produced in MapInfo-CDATA, of the distribution of socioeconomic disadvantage

(IRSD) across the Brisbane Statistical Subdivision and the Statistical Subdivisions of

other Australian Capital cities in the year 2000. In Brisbane, socioeconomically

disadvantaged CCDs are mostly interspersed with more advantaged CCDs, although

small pockets of disadvantage can be seen in the South, and a large pocket of

advantage exists in the West of Brisbane (Figure A5.2). Socioeconomic

disadvantage is also quite interspersed in inner city Melbourne, and Darwin, although

Darwin has greater levels of socioeconomic disadvantage overall than Brisbane.

Perth, Hobart, Canberra, Sydney, and Adelaide all tend to show more clustering of

disadvantaged areas together compared with Brisbane (Figures A5.3-9). If the

spatial patterning of socioeconomic disadvantage is a driver of socioeconomic

differences in shop provision, then the Brisbane findings may not apply to Perth,

Hobart, Canberra, Sydney and Adelaide. Also, if there are substantial changes to the

socioeconomic patterning of Brisbane City over time, then the conclusions of this

research regarding the socioeconomic patterning of accessibility and affordability

may no longer apply. The distribution of socioeconomic disadvantage across

Brisbane has changed little between the 1996 and 2000 census periods, as indicated

by Figure A5.1 in Appendix 5. Unfortunately, further changes since this time cannot

be properly assessed, as Socioeconomic Indexes for Areas are not due for release

until March 2008 (406).

208

Changes to the price of food since the Brisbane Food Study was conducted will

affect the absolute costs of fruit and vegetables calculated in the study. According

to data from the Australian Bureau of Statistics Consumer Price Index (407), the

price of food has risen by an average of by 4% each year from 2000 to 2007 (see

Table A5.1, Appendix 5). The prices of fruits and vegetables have risen since the

BFS was conducted, by an average of 12% and 6% per year, respectively. The

drought and disease have contributed to the particular increases in fruit and vegetable

prices (205, 407) and the destruction of the Queensland banana crop from a tropical

cyclone in March 2006 partly contributed to the rise in fruit prices (408). The

changes in fruit and vegetable prices have been large, and have exceeded the price

changes of food in general. The effect these changes may have on the applicability

of the findings to the present cannot be determined for certain, however, if the price-

purchase relationship operates at a threshold, stronger relationships between price

and purchase might operate currently than those estimated in this thesis.

Similarly, there have been a number of trends in the food industry that may have

affected the distribution of retail outlets, opening hours and prices. The absence of

socioeconomic differences in shop opening hours found in the analysis of the 2000

Brisbane Food Study may no longer be the case, as unrestricted trading hours were

implemented in Queensland in August 2001. Also, since the time of the study, there

has been general expansion of food retail stores, including an expansion of discount

retailers (e.g. Aldi), increased growth of independent retail groups, and an increased

effectiveness of specialty retailers including fruit and vegetable retailers (408). A

growth in the number of shops would be expected to mean improvements in shop

access since the time of the study, and depending on where the retail growth has

occurred, socioeconomic differences in shop access may be increased, reduced, or

unchanged. The growth of discount retailers may produce changes to the types of

supermarkets located near low- and high-socioeconomic areas if the socioeconomic

patterning of discount supermarkets that has occurred elsewhere (175) also occurs

here. There has also been a trend towards retail consolidation (408). The exact

impact of retail consolidation is unknown, and a matter of some controversy,

especially regarding whether it may cause higher prices through reduced

competition, a known determinant of food price (205). If more aggregated

209

ownership means less variability in prices and availability, socioeconomic

differences in price and availability may be less now than at the time the BFS was

conducted.

8.3.2. Cooking Skills In interpreting the overall findings as to the role of accessibility, affordability and

cooking skills in socioeconomic differences in fruit and vegetable purchasing, the

strengths and limitations of the cooking skills study must also be considered. The

cooking skills study was designed to be compatible with the secondary analyses, and

thus shared the same limitations previously outlined, especially with respect to the

purchasing outcome measure. The study also had some unique issues pertaining to

its design and implementation, discussed below.

Study design

The design of the cooking skills study had several strengths, including its

quantitative nature, probabilistic sampling of households and specific focus on

persons who most often prepare foods for households. However, despite the

probabilistic sampling method, a low response rate was achieved (43%), which is

comparable to many similar mail-style health surveys (383) but may have resulted in

an unrepresentative sample. Analyses indicated slight, but statistically significant,

differences between census figures and participating households on comparable

measures of income, and indicated that respondents without full, useable data tended

to be older, have less education, and lower incomes. There is no reason to suspect

differential biases that would affect estimates of socioeconomic differences in

cooking skills measures or purchasing, however the descriptive figures of cooking

skills may not represent the wider Brisbane population. The low survey response

resulted in a smaller sample size (n=401 in multivariable models) than a priori

calculations indicated were necessary to achieve 80% power. However, these

calculations appear to have been too conservative, as the standard deviation of the

outcome variables (12.3) and level of clustering (ICC=0.005) were less than initially

assumed (15.2 and 0.009, respectively). Repeating the initial sample size calculation

210

using observed rather than assumed values, the necessary sample size to achieve 80%

power was actually achieved by the study (n=315) (Appendix 7).

The cross-sectional design did not enable determination of whether skills preceded

purchasing or vice versa. The vegetables typically purchased may influence the

acquisition of skills to prepare them. Confounding was minimised by adjustment for

other characteristics known to relate to cooking skills or purchase, however residual

confounding from unmeasured variables may have occurred. On further reflection,

working hours may have been one such confounding factor. Working longer hours

may be associated with higher household income, and perhaps with less confidence

to cook, since time is often reported as a barrier to healthy eating (409). Thus,

income differences in confidence to cook may have been underestimated.

Furthermore, this study examined cooking skills in isolation, and did not consider

possible mediating influence of psychosocial constructs that may relate to cooking

skills and vegetable purchase, such as general self-efficacy, interest in food and

nutrition, or food and nutrition knowledge.

Measures The measures of cooking skills employed in this study represent an improvement

over previous studies, which have generally not reported the psychometric properties

of the measures they employ. In this thesis, the measures developed were shown to

have acceptable face validity among a diverse convenience sample, and acceptable

internal consistency and test-retest reliability among the general Brisbane population.

However, many issues surrounding the measurement of cooking skills have not been

overcome, such as the lack of validation against actual objectively measured skill

levels or actual performance, and other food-related skills were not measured, such

as planning ahead, using and modifying recipes, or improvising.

Despite these limitations, based on the limited existing knowledge base, there is not

sufficient reason to believe that actual, objectively measured skill levels are

genuinely the gold standard in measuring cooking skills (45). Just as self-efficacy is

a predictor of the performance of many health-related behaviours (300), confidence

211

may be a more important predictor of food-related behaviours than degree of skill.

Conceptual differences between confidence to cook and actual skill levels may mean

confidence and actual skill are not necessarily proxies of varying quality for the same

general cooking skills construct. It is likely that actual skill level contributes to a

person’s perception of his or her own skill level, and confidence to cook is formed on

the basis of self-perceived skill and other attitudes and beliefs. While this thesis has

demonstrated that levels of confidence to cook are lower among low socioeconomic

groups, and are in turn associated with vegetable purchasing, this does not mean

actual skill levels necessarily vary across socioeconomic groups or relate to

purchasing.

8.4. Nutritional implications of findings The limitations of the secondary analyses and cooking skills study arising from the

outcome measure have been discussed from a technical perspective. From a nutrition

perspective, the purchasing outcome largely determines the implications the findings

have for the diets of socioeconomic groups and subsequent health inequalities.

Higher intakes of fruits and vegetables have been associated with reduced risks of

stroke (63-65), heart disease (65-69) and some cancers (70-74). This study assessed

fruit and vegetable purchasing, based on the expected relationships of accessibility,

affordability and cooking skills and purchasing for at-home consumption (Figure

2.1). A greater variety of vegetables being purchased may be indicative of a higher

total quantity being purchased and subsequently consumed, but this is not certain as

quantities and intakes were not assessed. Further, fruit and vegetables purchased for

food shopping are not equivalent to total fruit and vegetable intake. Total intake

would also include fruits and vegetables grown or received for free, or contained in

foods consumed in away-from-home settings, and would not include fruits and

vegetables purchased but not consumed due to inedible portions, spoilage, wastage or

items consumed by people other than household members. A conceptual schema of

the contribution of fruits and vegetables purchased to total fruit and vegetable intake

is pictured in Figure 3.1.

212

While purchasing and intake are not equivalent, the findings still have nutritional

relevance. The availability of fruits and vegetables at home is an enabler of healthy

dietary choices, being associated with higher fruit and vegetable intakes among

adolescents (410, 411) and children (412, 413). Thus, the findings have relevance in

terms of understanding ways in which to promote the types of behaviours (such as

purchasing a variety of fruits and vegetables) that would contribute to a diet high in

fruits and vegetables. Also, while the measure could not be standardised to

individuals, a greater variety of fruits and vegetables being purchased at the

household level may be indicative of greater dietary variety for household members.

Variety, as well as quantity, of fruits and vegetables may be important for reducing

the risk of chronic diseases. Greater dietary variety of fruits, vegetables and

wholegrains has been associated with reduced risk of all-cause mortality, cancer,

coronary heart disease and stroke (414). Since fruits and vegetables are

heterogeneous in their composition, a varied intake of these foods is promoted as a

means of achieving adequate intakes of nutrients and other substances, such as

phytochemicals, that may reduce the risk of chronic disease when consumed in

combination (415).

Further exploration of the nutritional relevance of the shop access findings could be

achieved by examining overall fruit and vegetable purchase, or purchase of sub-

groups of fruits and vegetables, such as green leafy-, cruciferous-, or orange-yellow

vegetables. Possible heterogeneity of the price-purchase relationship was the reason

for examining items individually, and based on the item-by-item results, examination

of indices or sub-groupings may be appropriate. Similarly, the nutritional relevance

of the relationship between confidence to cook and vegetable purchase could be

explored in more depth by examining nutritional groupings of vegetables. Measures

of nutrient availability at the household level are not achievable since quantities were

not examined, however the different composition of these classes of vegetables

would provide an indication of whether access to shops, or confidence, were

associated with purchasing types of vegetables known to be rich sources of folate

(green leafy vegetables), glucosinolates (cruciferous vegetables), and vitamin A

(orange-yellow vegetables). The example above is only one possible grouping, with

alternates including the “dark green” and “deep-yellow vegetables” specifically

promoted in the U.S dietary guidelines (415).

213

The relationships between shop access, confidence and purchasing therefore indicate

that having greater access to shops locally and greater confidence to cook are

associated with a greater likelihood of performing the types of behaviours consistent

with consuming a varied diet that may also be high in fruits and vegetables overall.

Conversely, the lack of relationships between other accessibility and affordability

measures and purchasing do not necessarily rule out that distance to shops, in-store

availability and price may be associated with the quantity of fruits and vegetables

purchased, or consumed. Small relationships might have been missed, such as those

documented elsewhere by studies using quantitative measures of fruit and vegetable

purchase or intake. In summary, the nutritional and health consequences of the

findings cannot be empirically determined. However, this thesis provides insight

into the connections between accessibility, affordability and some of the behaviours

(purchasing fruit items and purchasing vegetable items) that might ultimately

contribute to the lower fruit and vegetable intakes (33, 89-105), and higher burden of

diet-related diseases (19, 21-30) among low socioeconomic groups observed

elsewhere.

8.5. Future directions for health promotion and research

Overall, this thesis contributes to the body of knowledge related to contributors to

socioeconomic differences in purchasing and possibly the subsequent consumption

of fruits and vegetables, by demonstrating the potential importance of cooking skills

and the relative unimportance of unequally-distributed food retail environment, at

least in Brisbane, Australia. Based on the findings of this thesis, there may be little

value in public policy and intervention strategies to reduce socioeconomic

inequalities in nutrition by encouraging the development of shops in low-

socioeconomic areas or by targeting the fruit and vegetable supply of supermarkets

and grocery stores. Low- socioeconomic groups may benefit from promotion of

cooking skills, as lower levels of confidence to cook among adults with less

education and lower incomes might contribute to socioeconomic differences in

vegetable purchasing. This is not to suggest that cooking-related interventions

214

necessarily need to be targeted at low- income groups, however promotion of the

skills to prepare vegetables, for example through education, may benefit population

sub-groups among whom confidence to prepare vegetables is comparatively lacking.

The estimates are based on very few individuals with a primary education or less,

however the particular lack of confidence to cook among this group suggests that

incorporating cooking skills earlier than currently occurs, which is mostly in early

high school (416-419), may benefit low socioeconomic groups.

Further research is warranted before recommending teaching cooking skills per se as

a means to improving population nutrition or reducing inequalities. Future research

could examine the relationships between cooking skills and nutritional knowledge,

attitudes and beliefs, in their relationship with dietary behaviours. This approach

could be further extended to ascertain whether the association between cooking skills

and behaviours is a better explained by super-ordinate psychosocial constructs (such

as general self-efficacy). It would be informative to know whether cooking skills

affect behaviours directly, or via intermediaries such as nutritional knowledge, or

alternatively whether other factors such as interest in food and nutrition or nutrition

knowledge promote both cooking skills and healthy dietary behaviours. Future

studies could examine the inter-relationships between confidence to cook and other

psychosocial constructs using a structural equation modelling approach. While this

approach does not demonstrate causality, it can determine whether direct or indirect

relationships between the constructs are more likely.

From a health promotion perspective, the effectiveness of cooking skills as an

intervention to achieve dietary change is possibly more immediately relevant than the

manner in which cooking skills may influence behaviours such as fruit and vegetable

purchasing. In Australia, the teaching of cooking skills has been treated as a

‘vehicle’ for providing more comprehensive nutrition education (416, 418). The

evidence that teaching cooking skills, with or without other concurrent nutrition

education, achieves dietary change is limited. Stitt (164) described similarities

between health promoting dietary profiles and educational policies of countries that

do and do not promote teaching of cooking skills. He concluded that educational

curricula that include cooking skills may promote positive dietary behaviours in the

population. Food-skills based interventions have successfully altered dietary

215

behaviours (153, 154, 272, 372), at least in the short term. Higher quality evidence is

available from a recent, community-based, food skills intervention in a low-

socioeconomic setting in Scotland, which included a control group and a six-month

follow-up period (420). Relative to controls, the intervention resulted in increases in

participants’ confidence to cook, but achieved only small increases in fruit intake

(one portion per week), and lesser, non-significant increases in vegetable intake

following the intervention, and dietary changes did not appear to be sustained at

follow-up six months later.

Additional research is necessary to further explore the impact of the food retail

environment on the dietary patterns of low-income groups, and whether this explains

socioeconomic variation in health. It should be emphasised that this thesis focused

on a contributory role of an unequally distributed food retail environment and did not

examine other ways in which accessibility and affordability in the broader sense may

relate to socioeconomic differences in fruit and vegetable purchase. This study, plus

other studies to date, have found only small relationships, or no relationship, between

neighbourhood features and the purchase or intake of fruits and vegetables.

However, qualitative studies, and studies using subjective measures of accessibility

and affordability, have tended to find more substantial associations between

accessibility, affordability and diet-related measures. Perhaps a broad-brush

approach to locating and counting shops, and taking average measurements of price

and availability, is not sufficiently coherent with the way people shop to make it a

useful way to examine accessibility and affordability issues.

Examining additional ways in which accessibility and affordability are associated

with dietary behaviours or dietary intake may yield useful insight into how to assess

socioeconomic differences in accessibility and affordability in ways that have clear

consequences for dietary inequalities. Future studies could examine price variation

(for example, minimum prices), as purchasing decisions based on prices may be

influenced more by the minimum prices available than the average prices. Similarly,

access to shops near both home and work may also reflect a more realistic appraisal

of the relationships, in view of the way people ‘chain’ their shopping trips with other

travel (421). Furthermore, the spatial scale at which the food environment may

potentially influence behaviour is not necessarily the arbitrary 2.5km distance used in

216

this thesis. Future studies should compare catchments of various sizes to determine

whether there is any particular spatial scale at which access-related measures most

relate to dietary behaviours. Additionally, subjective approaches to delineating

neighbourhoods meaningfully (56, 163), which factor in natural boundaries such as

rivers, could be explored, as access to shops within areas that people use as their

neighbourhood may have greater relevance to dietary behaviours. Additionally,

findings of this thesis were suggestive that there might be a differential impact of the

food environment according to household income, but were not definitive as the

Brisbane Food Study was not originally powered to detect interactions. Future

studies could be better equipped by including larger samples of individuals and areas,

or alternatively, by examining the presence/ absence of a relationship between the

food environment and purchasing in a larger sample of low-income households.

The difference between Brisbane and other locations outside of the United States in

the presence or absence of socioeconomic patterning in shop location may have

implications for research and health promotion. Further examination by future

studies of the conditions under which socioeconomic differences in access to shops

occur may highlight the processes that operate to produce (or prevent) an inequitably

distributed food supply, such as “white flight” (386) and socio-spatial polarisation

more generally (388). The spatial patterning of poverty and social segregation is

argued to differ from city to city because the patterning partly results from past land

and housing practices (388). A greater understanding of the underlying processes

may be necessary to identify locations in which inequitable shop provision is likely

to occur, and to identify the urban planning policies and practices that are likely to be

beneficial for health promotion. In summary, this thesis has provided important

findings for social epidemiology and health geography by contributing to growing

body of evidence that neighbourhoods may contribute comparatively less to

inequalities outside of the US and articulating some of the complexities of studying

this issue rigorously within the international literature.

217

8.6. Concluding remarks

Within a socio-ecological framework of how socioeconomic differences in diet-

related behaviours and outcomes may occur, disparities in the availability of shops

and the price and availability of fruits and vegetables do not provide an explanation

for socioeconomic differences in fruit and vegetable purchasing, at least within

Brisbane. Therefore, these price and availability disparities probably provide a poor

explanation for subsequent socioeconomic differences in fruit and vegetable intake

and diet-related health outcomes. This conclusion may not pertain to places where

disparities in the availability of shops have been observed, as this thesis did find that

greater access to shops was associated with fruit and vegetable purchasing, albeit to a

minor extent. Findings of this thesis would tend to suggest that purely environmental

strategies, such as building more shops in low-socioeconomic areas, are not likely to

substantially improve fruit and vegetable intake or reduce inequalities. Indeed,

caution has been expressed that such strategies may worsen food accessibility over

time by reducing the presence of small shops, especially in absence of concurrent

residential expansion to provide additional demand (182). The limitations of this

study and the literature to date have highlighted some important areas in which future

research into environmental determinants of inequalities in diet can benefit: namely,

in expecting and understanding differences across contexts in the socioeconomic

patterning of environmental characteristics, in examining differential effects on

socioeconomic groups of environmental characteristics and in quantifying

environmental characteristics. While the environmental factors studied did not

contribute to inequalities, at least in terms of being inequitably distributed within

Brisbane, there was evidence that lower confidence to cook contributes to the less

varied vegetable purchasing of low socioeconomic groups. Further research in this

area is warranted to determine the potential of a focus on cooking skills to reduce

socioeconomic differences in fruit and vegetable intakes and subsequent health

outcomes that may ensue.

i

Appendices

ii

I. Appendix I: Details of major studies included in the literature review

iii

Table A1.1: Studies examining the relationship between socioeconomic position and shop availability (excluding fast food shops), prices and food availability (United States) Study Design Location Sample SES measure Outcome Results/ Conclusions & Limitations Macdonald and Nelson 1991 (166)

cross-sectional

USA (Atlanta, Boston, Denver, Detroit, Houston, LA, NY, Philadelphia, Pittsburgh, St Louis)

random sample of n=332 supermarkets (chain and non-chain) from subsample of n=10 metropolitan statistical areas from larger stratified random sample

1980 Census (Zip code) Low income areas by two cutoffs (>10% and >20% of households below the poverty line)

Price of basket of foods based on National Food Consumption Survey (average of price at three survey time points) Store size (square feet) Independent vs chain status of supermarket

- Compared with other areas, low income areas had smaller stores within the central city (11600 vs 19500 sq feet) and suburbs (18200 vs 19500 sq feet) (not statistically tested). Median Neighbourhood income was significantly positively related to store size (B=0.680, p<0.01 for log median income and log store size (square feet)). - Compared with other areas, low income areas had a higher proportion of independent stores in both the central city (36% vs 12.5%) and the suburbs (32.5% vs 20.7%). - Mean price of NFCS baskets higher in low income than high income areas by both the 10% poverty ($101.43 vs $99.35, p<0.05) and 20% poverty definitions for low income ($101.66, $99.88, p<0.05). Same did not hold true in a sub-sample of inner city stores ($102.49, $102.11, n.s.). (Higher proportion of low income areas in inner city, and mean prices were higher for the central city ($102.38 than elsewhere ($98.30) p<0.05.) -No large or significant association between median income and price (B=0.019) after adjustment for for other neighbourhood characteristics (average household size, car ownership) and store characteristics (labour & insurance costs, store size, services offered, whole sale practices, other supermarkets in zip code, warehouse stores within 5 miles, and central city location). Limitations – no generalisablity beyond study area, insensitive socioeconomic measure

Alwitt and Donley (1997) (187)

cross-sectional ecological

Chicago, USA

all residential zipcode in Chicago

Sourcebook of Zipcode demographics & US Census 1990

Census of retail Trade 1990 & CD-rom telephone

- There were significantly fewer supermarkets and large grocery stores, but more small grocery stores in poor compared with non-poor zip-codes, and nearby (within 2, and 3 miles) to poor compared with non-poor zip-codes. - Per million dollars of purchasing power there were more small

iv

poor= bottom quartile of poverty, high school graduation, labour force participation, top quartile of unemployment non-poor= rest (zip-codes)

database number of retail stores (zip code) supermarkets grocery stores (1-9, 10+ employees)

grocery stores in poor areas than non-poor areas (0.048 vs 0.016 p<0.01), but similar numbers of large grocery stores (0.011 vs 0.010) and supermarkets (0.003 vs 0.003). Mean number of food stores in poor and non-poor zip-codes

Poor Non-Poor

Within 2 miles Within 3 miles Poor Non

Poor Poor Non-

Poor all 20.9 16.4 15.9 17.9 15.6 18.0 small 16.9** 10.2 11.8 10.9 10.8 11.3 large 4.0** 6.2 4.2 6.4** 4.7 6.7* s’mkts 1.1** 2.4 1.6 2.4* 1.6 2.7*

*p<0.05 **p<0.01 NB: poor & non poor zip-codes had similar population density Limitations - the poor zip-codes were disproportionately located in the inner city, size & urbanicity not adjusted

Finke, Chern, and Fox (1997) (202)

cross-sectional

USA From Household Nationwide Food Consumption Survey (1987-88) n=12522 individuals in 4495 housholds. n=10427 in the analysis

Household Nationwide Food Consumption Survey (1987-88) Low income = bottom quartile 1987 census (assume household) High income = all other households

Household Nationwide Food Consumption Survey (1987-88) ( Average of normalised prices paid for 9 items (i.e. whole milk, white flour, white sugar, large eggs, regular ground beef,

- Overall, low income consumers pay slightly higher prices for the same foods than high income consumers, to a borderline significant level (normalised prices 1.024 vs 1.000, p=0.052). - This difference in normalised price is significant for some subgroups, namely urban residents (1.045 vs 0.997), black urban residents (1.078 vs 1.012) and similar sized differences (though not significant) were seen for African Americans (1.012 vs 0.983), and urban whites (1.012 vs 0.983). No differences were observed for suburban residents (1.004 vs 1.002) or White residents (0.998 vs 0.990). -Overall, the urban location of low income people may drive the higher prices paid for food. Limitations – socioeconomic cut-off points, limited range of foods examined, no accounting for clustering, (exclusions and sampling not

v

pork chops, whole chicken, white potatoes and bananas)

stated)

Chung and Myers Jr., 1999 (189)

cross-sectional, ecological

USA Hennepin and Ramsey counties, Minneapolis

all n=526 grocery (chain and non-chain) and convenience stores in study area (from listing), subsample of n=55 of these (stratified by revenue) for pricing component zip codes

1990 Current Population Survey Poverty rates (zip code) (<10%, 10-20%, >20%

Price of 49 items from USDA Thifty Food Plan (TFP), individually and grouped into 6 groups, Group A= fresh fruit and vegetables Group C= tinned vegetables/ legumes white rice and tuna Price of a basket based on the TFP Availability of each item Availability Index (based on all items)

- 89% of chain grocery stores but 60% of non-chain grocery stores were located in non-poor zip-codes. - Exact figures were not presented, however a similar distribution by income was reported for stores with revenue >$10 and <$10 million. - Prices in chains are consistently lower in chain than non-chain stores by 10-40%, and in suburban compared with inner city areas (by approximately 2% on average). - Basket prices were non-significantly higher for poor vs non-poor zip-codes (110.36 vs 105.21), similar for inner city vs suburban (107084 vs 106.66) and significantly higher for non-chain vs non-chain stores (109.90 vs 93.28). Regression analysis found no significant relationship between poverty and basket price (B=-1.5223) controlling for the chain status of stores, and availability. - Availability was lower in the inner city compared with suburban stores for all items, on average by 25% but by up to 50% for some items particularly for fruits and vegetables. Availability was higher in chain compared with non-chain stores for all items, by even larger amounts, often exceeding 100%. - Regression analysis found relationship between poverty and availability (B=-0.2129) controlling for the chain status of stores. Limitations – no generalisablity beyond study area, limited socioeconomic measure

Fisher and cross- New York n=503 food stores Median Percentage of - Each additional $10, 000 pa median household income was

vi

Strogatz, 1999 (222)

sectional City, USA (randomly sampled) in n=53 zip-codes (randomly selected from n=7 counties ranging from from large metropolitan to rural)

household income (zip code) 1990 US census data

milk available that is low fat (1% or less) in stores (zip code level)

associated with an increased percentage of low-fat milk in stores at the zip code level (B= 5.5, SE=2.1, p=0.01), adjusted for region (urbanicity) and ethnic make-up -Areas with higher median incomes have a greater proportion of low-fat milk in store shelves. Limitations: ecological fallacy, no account for clustering, store types not specified – different proportion of store types in lower socioeconomic areas

Kaufman (1999) (169)

cross-sectional

Lower Mississippi Delta, USA

n=36 low-income counties in the Lower Mississippi Delta

Low-income zip codes (so classed based on proportion of low-income households) within study counties vs Lower Delta core counties as a whole

Supermarkets per square mile Accessibility ratios (accessible food stamp redemption as a proportion of food stamp sales)

- Low-income counties had fewer supermarkets per square mile than the average for rural counties in Arkansas, Louisiana and Mississippi (1 per 190.5 vs 1 per 153.5 square miles). - Accessibility ratios > 1, which indicated poor accessibility, were found for 30.8% of the study sample compared with 22.5% of all zip-codes in the Lower Delta core counties overall. Limitations: ecological fallacy, no account for clustering, non-exclusive comparison groups

Hayes (2000) (197)

cross-sectional

USA, national

Stores in Bureau of Labor Statistics data set with relevant data (sampling not specified but authors claim “representative national sample” based on region, population size income, heating type (electric/ fuel), ethnicity, and retirees.)

US Census Proportion of the population in poverty Poor (>20% in poverty) Non-poor (<20% in poverty) (Other cut-offs were explored, similar results)

Data from US Bureau of Labour Statistics and Private data (SPECTRA) Price of 5 item ‘basket’ (sum of items, weighted by low income expenditure)

- Compared with non-poor zip-codes, in poor zip-codes there are comparatively more stores per square mile (2.935 vs 1.559) and fewer people population per store (6.304 vs 6.834), however these stores are smaller 26592 vs 32000 sq feet) and are less likely to be chain stores (59% vs 85%). (Statistical tests not performed). - Compared with non-poor zip-codes, in poor zip-codes lettuce was slightly more expensive (0.805 vs 0.758 $/pd, p<0.05), oranges were slightly cheaper (0.680 vs 0.832 $/pd, p>0.05) and an overall food basket was significantly cheaper (1.235 vs 1.321 $/pd, p<0.05) Overall, based on the price dispersions, for poor neighbourhoods each additional increase in median household income of $10,000 was associated with overall increase in prices of the 42.7% (p<0.05) (after

vii

n=19386 Stores in SPECTRA study (comparable to BLS sample) n=2181

Median Household Income ($10000 pa increments)

($ per pound & price dispersion statistics) & price of Iceberg lettuce & navel oranges. Shop access: Number of stores/ square mile & population/ store, & store size (sq. feet)

adjustment for the neighbourhood demographics, crime statistics, store format, region and local area fixed effects). No relationship between income and price without these adjustments, and for affluent neighbourhoods a non-significant tendency in the opposite direction occurred, an increase in prices (p<0.05). - Demographics including population density, unoccupied housing and inner-city location can confound the relationship between neighbourhood income and price. Limitations – basket was weighted by low income group expenditures, which may bias comparisons to favour low income areas (if price is related to purchase), quality not controlled

Frankel and Gould (2001) (203)

ecological cross-sectional, and panel

USA Data from American Chamber of Commerce Research Association 184 cities 1979/80 and 1989/90 n=49 cities in sample (based on data availability) sampling unknown, types of shops unknown Data from Bureau of Labour Statistics

US Census Poverty (3 measures) Percentages of people with incomes below the poverty line, 1-2x the poverty line Income (three measures) proportion of families falling into bottom, second and top three quintiles of family

Data from American Chamber of Commerce Research Association Price of ‘basket’ of grocery items (100% pork, sausage, light tuna, whle milk, sugar, eggs (branded or graded for quality)) Calculated

- Having greater proportions of poor, or upper-middle income, rather than lower-middle income populations is associated with higher retail prices at the city level. Correlations between socioeconomic measures, and grocery prices (both on log scale): change over time (∆) in both IV and DV, and cross-sectional data

∆ grocery price grocery price % with income < poverty 0.38 0.62 % with income 1-2x poverty 0.35 0.48 % with income > 2x poverty n/a n/a % in bottom income quintile 0.33 0.62 % in 2nd income quintile n/a n/a % in top 3 income quintiles 0.40 0.59

adjusted for region, tax, percentage of population who are over 65 year, Hispanic, and female, proportion of non-family households, population density, and population growth rate (p values not reported)

viii

used to ‘confirm’ models

income (City level) Both absolute and change over time

the bi-annual averages, and also the change over time (1979/80 to 1989/90)

-used alternate data source to ‘confirm’ findings Limitations – representativeness of sample not established, however both datasets would need to share same bias to explain findings

Morland et al 2002 (170)

cross sectional ecological

USA (Mississippi, North Carolina, Maryland & Minnesota)

all residential census tracts in study locations

1990 census (census tracts) Neighbourhood wealth

Business listings Number of grocery food stores (supermarkets = chain stores & grocery = non-chain stores) and convenience stores (census tract)

-More wealthy neighbourhoods had significantly more supermarkets & fewer grocery stores, but contained similar numbers of convenience stores. Prevalence Ratio (95% CI) of food stores by neighbourhood wealth

Supermarkets Grocery Convenience low income 1.0 1.0 1.0 low-middle 2.8 (1.2, 6.7) 0.9 (0.6, 1.3) 1.1 (0.7, 1.9) middle 2.6 (1.1, 6.4) 0.6 (0.3, 0.9) 1.4 (0.9, 2.4) high-middle 3.6 (1.5, 8.7) 0.6 (0.3, 0.9) 1.0 (0.6, 1.7) high income 3.3 (1.4, 7.9) 0.6 (0.3, 0.9) 1.0 (0.6, 1.8)

adjusted for racial segregation and population density NB - Mean area of zip-codes increases with neighbourhood wealth (Low =8km2, low mid=7km2, mid=19km2, high mid=15km2, high=20km2) Limitations – unequal sizes of zip-codes not accounted for (may be alternate explanation), urbanicity not considered, site not considered (only pooled analyses), limited classification of shop types.

Topolski, Boyd-Bowman, & Ferguson (2003) (209)

cross-sectional

USA, “a mid-sized southern city”

Two grocery storesd (matched for chain) each from low, middle and high income zip-codes random selection of strawberries, bananas and green

US Census 2000 Median household income (low: $36236 & $14232pa, middle: $38947

standardized procedure, 18 independent, blinded raters Fruit Quality (Strawberries, bananas & grapes)

- Fruit from the high SES stores appeared fresher and tasted better than fruit from the low SES stores. - The difference between fruit appearance in stores in high and low SES areas is not consistent across store chains. Appearance and taste of fruit by store zipcode SES, mean scores (SE) and rank order within Supermarket Chains Y and Z

Appearance Taste (T) Chain Y Chain Z

ix

grapes not on sale sampling of areas and shops not stated

& $54222, high $54222 and $70809 (NB median housing prices for these areas also classed as low, middle & high)

Freshness appearance 9-point scale, and Relative Ranking (most to least fresh) Taste (9 point scale, and relative ranking)

SES (A) A T A T Low 2.46 (1.22) 2.53 (0.96) 4.28 4.04 4.80 3.71 Mid 3.30 (1.29) 3.47 (1.19) 3.73 3.61 3.68 2.79

Hi 4.74 (0.88) 4.50 (0.65) 2.87 2.94 1.65 2.27 For the 9 point scale, higher scores show more positive ratings. For the rankings, lower scores show more positive ratings p<0.01 for all SES differences in appearance and taste, p<0.01 for interaction between SES and chain type (greater SES discrepancy in one chain) for appearance but not taste. Limitations: Sample not established to be random, and unlikely to be representative given n=3 areas and n=6 stores.

Horowitz, Colson, Herbert & Lancaster 2004 (171)

case-study Harlem (New York City), USA

All general food stores (not restaurants) on licensing list that were small (1 register), medium (2-4 registers) or large (>4 registers) mapped to census blocks, and residents of census blocks in 2000 census

East Harlem (predomiantly non-white, low median income) vs Upper East Side (predominantly white, high median income)

number of stores (per capita & by shop type) in-store availability (% shops) of low-carbohydrate or high fibre bread, low/non-fat milk, fresh fruit, fresh green vegetables, diet/ club soda desirable stores

- Compared with residents of the low SES area, residents of the high SES area had a greater proportion of all available stores being classed s medium (RR: 3.0, 95% CI: 1.5, 6.1) and large (RR: 2.8, 95% CI: 1.4, 5.8), and fewer classed as small (RR 0.7 (0.7, 0.9)). However, residents of the high SES area had fewer stores overall (RR:0.4) and were more likely to have no stores (RR:1.3, 95% CI:1.3, 1.3). -Compared with residents of the low SES area, residents of the higher SES area had a greater proportion of stores being classed as desirable overall (3.2 (2.2, 4.6)) and for small stores 5.3 (3.1, 9.1) but not medium 1.0 (0.9, 1.2) or large stores (0.9 (0.9, 1.0)). Residents of the high SES area were more likely to have a desirable store in their area (RR: 1.2 (1.2, 1.2)) have only desirable stores (RR: 2.5, 95% CI: 2.5, 2.6) - Compared with residents of the low SES area, residents of the higher SES area had a greater availability of stores selling bread (RR:2.3, 95% CI:1.7, 3.2), low/non fat milk (RR:1.9, 95% CI; 1.6, 2.3)), fresh fruit (RR:1.2, 95% CI: 1.1, 1.4) , fresh green vegetables (RR:1.3, 95% CI: 1.1, 1.5), and a similar availability of shops selling diet or club soda (RR:1.0, 95% CI: 0.9, 1.1). Limitations: only two non-randomly sampled areas that varied on

x

(carried >1 of above items)

more dimensions than socioeconomic position, no removal of clustering, spatial size and population density not considered, non-stratified results may be due to store size

Gallagher (2005) (175)


Chicago, USA

n=75 of 77 “community areas” in Chicago, excluding the airport area, and the “downtown loop” all “major player” grocers (Jewel, Domicck’s, Aldi, Cub Foods)

US Census data Annual per capita income (Low: < $9,999, low-mid: $10-19999, mid-high: $20-29999, high: $30000+)

Number of “major palye all “major player” grocers (based on size and range) (Jewel, Domick’s, Aldi, Cub Foods)

-Overall, more grocers per capita are located in areas with higher incomes. This pattern of store location was most obvious for Jewel (“Chicago’s #1 chain”) and opposite for the discount grocer, Aldi. Number of stores per 100,000 residents in “community areas”

All Jewel Dominick Aldi Low income 1.7 0.0 0.6 1.1 Low-middle income 2.6 0.8 0.5 1.0 Middle- high income 3.5 1.6 1.2 0.6 High income 4.5 2.4 1.5 0.6

-Trend more obvious in majority >50% Black than majority white communities (figures not presented), possibly due to the greater spread of low incomes. Limitations: descriptive only, standardized per capita not per unit area, size differences of communities and inner city location not considered

Zenk et al., 2005 (176)

cross-sectional

Detriot, USA all n=869 census tracts in the tri-county Detroit metropolitan area.

U.S. Census data Percentage of residents below the poverty line (tertiles): Low (<5.03%), Medium (5.03-17.20%), high (17.23-81.96%)

Government and phone directory listings. Travel distance to the nearest supermarket (Mannhattan Block method) from geographic centroid to

-Distance to shops increased with poverty. Compared with low poverty areas, high-poverty areas had significantly greater distances to shops (coefficient=0.703 SE=0.092, p<0.001). Results were adjusted for poverty and for spatial autocorrelation by moving average spatial regression. - There was a significant interaction between poverty and percentage of African American residents, such that the extra distance associated with poverty increased as the percentage of African Americans in areas also increased. Limitations: correction spatial autocorrelation between adjacent areas does not correct for proximity to the inner city

xi

street address Baker et al ., 2006 (191)

cross sectional

St Louis, Mo, USA

n=8 “high 5 – low Fat” intervention sites all supermarkets (chain and non-chain) in study sites based on business census geocoded to site by address

2000 Census Median Household Income (from all zip codes within site)

78 item audit checklist Number of fruits vegetables (fresh/ frozen/ canned) available, averaged for all stores in site No. of shops

- Supermarkets in higher income areas had a greater selection of fruits and vegetables than those in lower income areas, regardless of area racial composition. “The selection of fruits and vegetables was highest in sites with the highest income and tended to decrease as area level income decreased”. (Figures not presented by authors.) -Sub-group analysis showed “mixed racial sites had less selection when they had lower area-level income and more selection when they had higher area-level income (data not shown)”. Limitations – non random selection of sites (generalisablity), no valid statistical comparisons, magnitude of differences not presented

Jetter and Cassady (2006) (201)

cross sectional

Sacramento & Los Angeles (USA)

Zip codes chain and non-chain supermarkets and small independent grocery stores within 5 miles of core study areas (from lists) (n=6 within Zip code and n=6 outside but within 5 miles per area) sampling method for shops and areas not stated

2000 Census Median Annual Household Income Very Low (VL)=<$27,000 Low (L)= $30-34000 Middle (M)=$42-46000 High (H)= $57-64000 Income (The “very low” cut off represents food stamp receipt criteria: 135%

Audit of stores, repeated over 12 month period Average price of two baskets (Thrifty Food Plan (TFP) & modified Thrifty Food Plan (TFP+) with “healthier” alternatives Availability of 19

- With the exception of very low income areas, prices of both the regular and healthier basket of groceries tended to increase with income (without standardizing for shop type). (Statistical comparisons by income not performed.) - The “healthier” food basket was consistently more expensive than the less healthy basket across areas of all income levels in Sacramento and Los Angeles. - Some socioeconomic difference in availabitly. Authors report “Except for the frozen fish filets, all items that were never available were recorded for stores located in very low or low income neighbourhoods.” Price of TFP and healthier TFP+ baskets

Sacramento LA Sacramento LA TFP TFP+ TFP TFP+ TFP TFP+ TFP TFP+ VL 205 238 196 225 200 236 185 224 L 188 220 159 192 M 203 239 218 255 204 238 211 250 H 205 238 204 245

xii

of the poverty line for a family of 4)

healthier alternative items (no. of shops missing each item)

Limitations: non-probabilistic sample, no valid statistical comparison by income, lack of standardization for shop type

Powell et al., 2007 (173)

cross sectional (ecological)

USA n=2805675874 people in 28050 zip-codes with available grocery store and census data, >300 residents. (representativeness not established) subsample n=4404 urban zip codes

2000 US Census Median household income ( low=bottom quintile, middle= middle three quintiles, high=upper quintile) (zip-codes)

Business listings number chain supermarkets (ChS), non-chain supermarkets (NChS), grocery stores (GS), & convenience stores (CS) per zipcode

- Most differences were small, however, relative to middle income areas, there were significantly fewer chain supermarkets, but more non-chain supermarkets and grocery stores in low income areas, overall and within urban areas. There were significantly fewer and more convenience stores in low income areas, overall and in the urban sub-sample respectively. - Most differences were small, however, relative to middle income areas, there were fewer shops of all types in high income areas, overall and within urban areas (mostly p<0.05). Relative rates of shops in low- and high- vs middle income zip-codes

All (n=28050) Urban (n=4404) Low High Low High Chain supermarkets 0.75*** 0.84*** 0.74*** 0.89** Non-chain s’mkts 1.10*** 0.79*** 1.10* 0.90 Grocery stores 1.18*** 0.73*** 1.40*** 0.83*** Convenience stores 0.96** 0.62*** 1.26*** 0.58***

* p<0.05, ** p<0.01 *** p<0.001, controlled for population size, degree of urbanisation and region Limitations: controls for region, but not in a way that accounts for clustering, did not account for zip code size

xiii

Table A1.2: Studies examining the relationship between socioeconomic position and shop availability (excluding fast food shops), prices and food availability (Canada) Study Design Location Sample Measures IV Measures DV Results/ Conclusions & Limitations Travers et al 1997 (204)

cross sectional

Nova Scotia (Canada)

8 counties (random selection, stratified by urbanicity), grouped into Health Department counties systematic sample of supermarkets and grocery stores (business listings) serving 4 population types (rural/ urban, low/ mixed income)

Shops classed as serving low-income population, serving predominantly mixed income population according to data from store regional managers

Store audit Price of four food baskets (Nutritious food basket (NFB), Thrifty nutritious food basket (TNFB) both based on nutritional adequacy, alternate food basket (AFB) which includes low-fat, high-fibre substitutes etc, and Consumption food Basket (CFB) which is based on usual consumption patterns and diets of a sample of survey respondents

-No significant differences between costs of baskets priced in stores servicing low vs mixed income populations. Thrifty Nutritious Food basket cheapest, followed by nutritious food basket. AFB $140, CFB $132, $ NFB, $119 Limitations: socioeconomic measure not defined in magnitude, nor to specified spatial level

Smoyer-Tomic , Spence & Amhein 2005


Edmonton, Canada

All n=212 residential neighbourhoods (postal areas) in Edomonton, Canada

Percentage of households classed as low income (definition from Statistics Canada for urban areas based on average income and household sizes)

Number of supermarkets within a 1-km radius by road of neighbourhoods (defined by postal areas) Distance to the nearest

There was a significant weak-to-moderate positive association between percentage of low income households and the number of supermarkets within 1km of neighbourhoods (Spearman’s R=0.350, p<0.001). Similarly there was a significant, weak-to-moderate inverse correlation between the percentage of low-income households and distance to the nearest supermarket. - The rate of low-income households was 1.9 times higher in the highest compared with lowest quintile of access to supermarkets within 1-km (p<0.001) and was 2.9 times higher for the highest compared with lowest

xiv

supermarket (by road)

quintile of distance to the nearest supermarket (p<0.001). Limitations: Results are unadjusted. Low-income areas tended to cluster in the inner-city, which had significantly higher numbers of supermarkets and lesser distances to supermarkets.

Latham and Moffat 2007 (177)

mixed methods

Hamilton, Canada

n=2 areas, deliberately selected for socioeconomic disparity all supermarkets, grocery stores and variety stores identified from

“Uptown” high SES area (residents typically higher income & more education, lower unemployment, fewer lone parent families, fewer immigrants) vs “Downtown” NB: As well as socioeconomic differences, Downtown is a more inner city area with higher population density

number of shops (per capita, per km2) by shop type (variety (V), Grocery (G), Supermaket (S), Speciatly (Spc) Price of 20 common items from Ontario Nutritious Food Basket ($) in and variety stores, and all ONFB items in supermarkets Availability of n=18 fresh produce items

-Compared with the high SES area, the lower SES area had a comparatively greater proportion of convenience stores (64% vs 44%), a lesser proportion of supermarkets (3% vs 11%) and specialty stores (23% vs 33%) and a similar proportion of grocery stores (9% vs 11%). - Per capita, the low SES area had comparatively 2.8 times as many variety stores (1.62 vs 0.58), 1.7 times as many grocery stores (0.24 vs 0.14), 1.4 times as many specialty stores (0.59 vs 0.43) per capita, but 75% fewer supermarkets (0.08 vs 0.14). However, per square kilometre, the low socioeconomic area had approximately six times more of all types of shops than the high socioeconomic area (10.5 vs 1.62 convenience stores, 1.54 vs 0.24 grocery stores, 0.51 vs 0.08 supermarkets, 3.85 vs 0.59 specialty stores). -The low and high socioeconomic area had similar median produce availability in convenience stores (0 (0 to 1), n=4 vs 1.0 (0 to 14), n=39), grocery stores (11, n=1 vs 15 (7, 17), n=5) and supermarkets (18, n=1 vs 17.5 (17 to 18), n=2). -Median prices in the few uptown shops were generally within the price ranges of shops in the downtown area within convenience stores ($43.08, min 45.31 to max 40.84) vs $39.57, 37.36 to 46.19) and grocery stores ($119.64 (n=1) vs $122.38, 111.38 to 128.35) (n=3)). Prices were higher in the Uptown supermarket ($160.22) than the two Downtown supermarkets ($132.17 and $132.35). Limitations: design prevents valid statistical comparisons and generalisation (small and non-random sample)

xv

Table A1.3: Studies examining the relationship between socioeconomic position and shop availability (excluding fast food shops), prices and food availability (United Kingdom) Study Design Location Sample Measures IV Measures DV Results/ Conclusions & Limitations Sooman et al., 1993 (192)

case -comparison

Glasgow, UK

n=2 regions Sample: selected for socio-economic contrast, deliberate sample 7 localities and 10 shops per region. (shops representative of each region, not standardised for type)

North West region (High SEP) vs South West region (Low SEP)

Price - ‘healthy basket’ & ‘less healthy basket’ ‘fruit & veg basket’ (used smallest packet sizes); Availability - number shops in which basket items available Quality – (scale 1 to 5 for fruit and veg, subjective assessment of 2 field workers)

- A basket of nine fruit and vegetable items was similarly priced in low and high socioeconomic areas (3.54 vs 3.59 £ / pound), but were of a lower average quality in the low socioeconomic area (2.6 vs 3.3). - Compared with the high socioeconomic area, the low socioeconomic area had a slightly higher mean price of a healthy food basket (10.48, 9.94 £ / pound) but not of a less healthy food basket (9.02 vs 8.99 £ / pound). - Compared with the high socioeconomic area, the lower socioeconomic area had comparatively lesser average availability of fruits and vegetables (7.4 vs 8.5), healthy food items (4.8 vs 6.5), and less healthy items (8.6 vs 9.0). -The ‘healthy basket’ was more expensive than ‘less healthy basket’ only in the low SEP area. Limitations: two non-randomly selected areas only

Ellaway and MacIntyre 1998 (115)

cross-sectional

Glasgow, Scotland

n=4 areas deliberately sampled for socioeconomic contrast n=318 40 year olds, n=373 60 year olds (probabilistic within age groups)

Census data West End [most advantaged], Garscaddan, Mosspark, Greater Pollock [most disadvantaged] face to face interviews social class

face to face interviews Use of the local area for food shopping (Y/N)

- Residents of Garscaddan, Mosspark, Greater Pollock were less likely to shop locally than residents of the most advantaged area (West End) (Odds ratios 0.50*, 0.39*, 0.64, respectively), adjusted for age, gender and personal social class. *=p<0.05. In part this may have had to do with residents of these areas being more likely to shop at all compared with residents of West End (64.9%, 67.4% and 66.0% vs 54.9%). - No significant relationship between personal social class and the chances of shopping locally. Compared with people of the highest social classes (1-2), people in the intermediate non-manual class (3) were more likely to shop locally (OR: 1.47) and people in the manual social class category were equally likely to shop locally (OR: 0.94),

xvi

(occupation) of household head

adjusted for age, gender and neighbourhood. Limitations: findings may have been due to a greater chance of shopping at all rather than shopping locally per se. Non-probabilistic sample of areas

Cummins and MacIntyre, 1999 (180)

cross –sectional ecological

Scotland, UK

all Greater Glasgow Health Board district all multiple stores & random sample (1 in 9) of non-multiple food stores

Census data Carstairs-Morris DEPCAT of postcode where store located (7 point scale, based on over-crowding, unemployment, social class, & car ownership)

Business listings Number of stores (fruit and veg, independent grocer, affiliated independent grocer, freezer stores, discount supermarkets & multiples)

Fewest stores of all types in the most deprived areas (DEPCAT 1, 2 and 3) and most in advantaged areas (DEPCAT 4, 5, 6, 7) (Actual figures not presented, significance not tested). Limitations: significance not tested, no consideration of urbanicity, population density, size

Furey et al., 2002 (196)

case -study Ireland, UK

n=4 deliberately sampled areas, varied socioeconomic and access characteristics random sample of food shops in areas (not stratified but mostly chain franchises (symbol), and multiples)

Area (unemployment rates and ‘deprivation’) Ballymena (affluent, some deprivation, low unemployment, rural and urban, high car ownership) Coleraine (moderate unemployment, low car ownership) Londonderry (high

Shop audit Availability (y/n) Price (food basket) (£) Price of food items/ groups relative to mean as (z- score) of MAFF low cost healthy diet basket (includes fresh green vegetables, fresh vegetables, frozen vegetables

- Fruit and vegetable items tended to be more expensive than average only in one of the low socioeconomic areas, and cheaper than average in the other areas. - In symbol stores, prices tended to be slightly cheaper in the poorer areas. In multiples (the cheaper stores overall), lower prices were observed in one of the lower, and one of the higher socioeconomic areas, and these same two areas tended to show most items to be more expensive than the average price in all shops. Mean basket prices, and z-scores for price of fruit and vegetable items relative to average

Bally. Coler. Lon. Strab. Basket Price (symbols) £ 7.57 £ 7.23 £ 7.09 £ 6.97 Basket Price (multiples) £ 6.34 £ 5.00 £ 5.46 £ 6.52 potato (-z scores) -0.02 -0.04 -0.03 0.09 Fresh Green (-z scores) -0.01 -0.08 0.53 -0.23 Fresh other (-z scores) -0.04 -0.08 0.39 -0.09 Frozen veg (-z scores) -0.25 0.03 0.72 0.07

xvii

unemployment, very low car ownership) Strabane (high ‘deprivation’, very high unemployment)

and fresh fruit) Process veg (-z scores) 0.21 -0.19 0.27 -0.02 fresh fruit (-z scores) -0.15 -0.01 0.20 -0.17 n /26 cheaper than average 8 20 6 21

n /26 cheap = number of items cheaper than average (/26) Limitations: descriptive only, areas were not randomly selected and varied across more dimensions than SES.

Dibsdall et al., 2003 (184)

cross-sectional

East Anglia, UK

n=680 low income adults residing in public housing in East Anglia who mostly purchased food for their households. Random sample, 23% RR

Questionnaire Employment status (Full time, Part time, Jobseeker, On sick leave, Looking after family, Retired) NB all are on low incomes

Questionnaire scales derived from items measured on 1-7 Likert scale Perceived Affordability mean attitude rating (higher mean values of indicate better affordability) Choice (perceived availability) mean attitude rating (direction of scale not reported) Transport difficulties (mean attitude rating) (direction

- Employment groups had significantly different attitudes towards affordability (difficulty buying more fruit / vegetables than already do (2 items), cannot afford to buy organic fruit/ vegetables (2 times) , lack of money prevents me from eating healthily), with full time employees reporting the most positive attitudes and job-seekers the least positive attitudes. - Employment groups had significantly different attitudes towards transport difficulties (often use taxis, often use bus, often get shopping delivered, bus service is affordable). Attitude ratings mean (+SD) among different employment groups (portions per day)

FT job PT job

Job-seeker

Sick Leave

Family care

Retired

Afford- ability

3.6 (1.6)

3.3 (1.7)

2.5 (1.5)

3.0 (1.5)

3.1 (1.5)

3.4 (1.7)

Trans-port

5.7 (1.2)

5.6 (1.2)

5.3 (1.4)

5.7 (1.3)

5.5 (1.2)

4.9 (1.5)

- Employment groups had no significant difference in perceived availability of fruits and vegetables (Where I shop has a wide choice of fresh vegetables/ fresh fruit/ frozen veg / tinned feg / tinned fruit (5 items), I am satisfied with the shop where I buy most of my food, I think veg are affordable to me where I buy most of my food, visiting a supermarket is easy for met to do, there is a wide choice of food shops in my local area), however results were only reported only as p>0.05.

xviii

of scale not reported) Car ownership (Y/N)

- Car ownership rates were highest among people employed part time (68%), full time (68%), or on sick leave (60%), and were lower among people looking after families (46%), retirees (44%) and were lowest among jobseekers (26%) Limitations – low response (23%) but similar to UK national figures, exclusively low- income population limits generalisabilty

Cummins & Macintyre 2002 (193)


Glasgow, UK

all postcode sectors in Greater Glasgow Health Board District n=325 food retail outlets (all multiple retailers, 1 in 8 independent food retailers), data obtained for n=250 (77%) of shops

Census Carstairs-Morris DEPCAT (7 point scale, based on over-crowding, unemployment, social class, & car ownership) (postcode sector)

Price (cheapest, branded) Availability (yes/ no) [overall including grocery stores, bakers, butchers fruit and vegetable etc and by shop type (multiple/ independent)] 57 food items from London Family Budget Unit’s “modest but adequate diet” new potatoes, old potatoes, frozen chips, cabbage (per pound), lettuce (each), 450g Baked beans, 400g

-Average, cheapest-method prices (£) were similar for most areas (DEPCAT 1-6) compared with the most deprived (DEPCAT 7) for fruit (1.18 vs 1.14 (-3%)), vegetables (1.01 vs 0.96 (-5%)), fish and meats (p>0.05 for all comparisons). -In multiple stores, 4/57 items (teacakes, sausages, burgers, chocolate) were significantly cheaper in the more deprived areas, 1/57 was more expensive (apple) while 52/57 were similarly priced, using the “cheapest price” method. -In multiple stores, 3/57 items (burgers, fish fingers, orange juice) were significantly cheaper in the more deprived areas, 7/57 were more expensive (margarine, vegetable oil, sultanas, wholemeal bread, digestive biscuits, tea, coca cola) while 47/57 were similarly priced, using the “branded price” method. -In independent stores 7/57 and 5/57 items were cheaper in more deprived areas while 5/57 and 2/57 items were cheaper in less deprived areas, using the cheapest and branded price methods respectively. -In all stores, availability of 46 /57 items was similar across deprivation categories, however 10/57 items were less available in the more deprived areas, while only 1 item (coca cola) was comparatively more available in more deprived areas). NB multiples had lower prices and greater availability than other shop types

xix

tinned tomatoes, 2 lb frozen peas, apples, bananas, pears (per pound) OJ (/L), orange (each)

NB repeated analysis at postcode district and health board locality levels – results unaffected Limitations: availability conducted on an all stores basis and may reflect varying proportions of shop types

Guy et al., 2004 (179)


Cardiff, UK

all n=28 electoral divisions in Cardiff

Census Welsh Index of Multiple Deprivation (income, employment, health & disability, education, skills & training, housing, access to services)

Council listings Store openings & closures (1990s-2001) Presence/ absence of food stores 1990 and 2001 Effective delivery scores [level of shop provision per household] (based on expenditure, perceived attractiveness of stores, distance to the stores)

- More deprived areas experienced the same number of store openings (7 vs 7) , but more store closings (8 vs 3) compared with less deprived areas. (Difference not formally tested) - At each time period, similar numbers of more and less deprived areas lacked shops (5 vs 4). (Not formally tested) - Shop accessibility, as measured by effective delivery scores, was worse for the 50 most deprived areas compared with the 50 least deprived areas in 1990 (61 vs 91) and 2001 (73 vs 118). The most deprived areas experienced the least increase in accessibility from he increase from 1990 to 2001 (12 vs 27 points, 19.7% vs 29.7%). Limitations – no formal statistical testing, access to services (including food shops) forms a small part of the IV

White et al ., 2004 (181)

multilevel cross sectional

Newcastle Upon Tyne, UK

n=5044 individuals (83% RR) n=3153 households (18% RR)

Household socioeconomic index (based on Standard of living index, household income per adult equivalent,

Distance to shop selling 10 or 14 F&V, 10 good quality fruit and vegetables, shop mostly patronised

- Socioeconomic deprivation is associated with lesser distance to the shop mainly utilised and shops selling a wide range of fruit and vegetables, healthy food items and unhealthy food items, and quality fruit and vegetables. - Average TDS of the locations of different shop types varied substantially, but not significantly: supermarkets (9.02), discount

xx

n=560 food shops (85% response)

household tenure, house size, state benefit receipt), in quintles Household TDS Townsend Deprviation Score (TDS) (TDS of enumeration district household lives in, or where shop is located depending on analysis)

Type of store patronised Have difficulty carrying shopping home (yes/ no) Travel mode to main store Average deprivation (TDS) of enumeration district in which each shop type is located Weekly opening hours Percentage of fruits and vegetables being “high quality” (by observation)

supermarkets (7.85), department stores (2.52), convenience stores (10.55), freezer centres (7.27) and local discounters (8.63) (p>0.05). - Odds of shopping at a supermarket were significantly higher for less deprived households (OR = 2.4, 1.8, 1.9, 1.4 for quintiles 1 to for compared with 5 (most deprived)) and low socioeconomic groups more often shopped at discount rather than multiple supermarkets. - Low socioeconomic groups were less likely to travel by car to shops compared with high socioeconomic groups and were more likely to travel by all other means (bus, metro, taxi, bicycle, on foot). Low socioeconomic groups also were more likely to report having difficulties carrying shopping home (lowest vs highest quintiles of SES by household measure 32.1% vs 3.0%, or TDS 26.3% vs 7.7%). - Townsend Deprivation Score was not associated with opening hours (linear regression) (figures not reported) - Among shops which stocked all relevant items, most of which were supermarkets. the Cost of 33 food items was not associated with TDS (r=0.14, p=0.5), but the cost of 10 fresh fruit and vegetables associated with TDS (r=-0.42, p=0.002) such that they were more expensive in more affluent areas. - Quality was not associated with TDS (r=0.049 p=0.5) in all shops, or in only those shops selling all fruits and vegetables (mostly supermarkets). - The median number of items available in the shops (mostly used by respondent) in areas of varied socioeconomic characteristics were equal in for low and high socioeconomic respondents, by the household SES measure and TDS (14 vs 14 for 14 fruits and vegetables, 21 vs 21 for 21 healthy items and 10 vs 10 for 10 unhealthy items). - Lower income was associated with spending less on food per week

xxi

in absolute terms, and more as a proportion of income. Top vs bottom quintile comparisons of accessibility and affordability by household socioeconomic index and TDS

Household SES SES (TDS) Low High Low High Median distance to the nearest shop selling…

10 F&V 638 677 608 903* 10 F&V quality 709 814* 706 1059*

14 F& V 982 1018 981 1275* 21 healthy 1075 1116 1058 1289*

10 unhealthy 628 835* 616 983* main shop 1325 2419* n/a n/a

Usually shop at (%)… multiple 57.3 92.7 60.4 88.9 discount 31.6 1.7 28.8 3.2

dept store 2.7 1.7 3.2 2.6 other 3.4 0.4 2.3 0.9

missing 4.9 3.4 5.3 4.4 p<0.05 p<0.05 Expenditure…

£ / wk 22.0 £ 31.5 £* 23.5 £ 29.4 £* % income 27.7 13.2* 14.9 25.0*

Mode of transport from shops (%)… car 23.0 94.5 36.8 88.4 bus 29.4 0.7 25.2 3.7

metro 1.5 0.2 0.8 0.9 taxi 11.8 0.9 11.0 3.3

bicycle 0.8 0.2 0.9 0.0 foot 33.5 3.5 25.3 3.7

p <0.05 p <0.05 Limitations – population density & urbanicity not considered

xxii

Table A1.4: Studies examining the relationship between socioeconomic position and shop availability (excluding fast food shops), prices and food availability (Australia) Study Design Location Sample Measures IV Measures DV Results/ Conclusions & Limitations Burns and Inglis, 2007 (183)


Casey (S.E. Melbourne, VIC AUS,)

all city of Casey all supermarkets and all major fast food chains (>10 franchises in Australia)

2000 Census Socioeconomic Indexes for Areas (Advantage/ Disadvantage) (Census Collection District level)

City data for shop locations, road networks & boundaries, slope Travel time (minutes) (calculated in ARCView Software) for walking, driving and public transport to supermarkets and fast food stores (shorter to supermarket, equidistant, shorter to fast food)

- Areas in which travel was quicker to a supermarket than a fast food outlet had greater socioeconomic disadvantage (mean (SE) 1016.2 (81.6)) a than areas in which travel was equal to both (988.0 (54.2)), or shorter to a supermarket (957.9 (75.9)) p<0.01 overall, each group different at p<0.05. - Over 50% of fast food chains were co-located with supermarkets Limitations: no adjustments for population density.

xxiii

Table A2.1: Studies examining the relationship between shop availability (excluding fast food shops) (abundance and or distance) and diet, or diet-related outcomes (U.S studies) Study Design Location Sample Measures IV Measures DV Results/ Conclusions & Limitations Shankar and Klassen 2001 (135)

mixed methods (focus group + cross –sectional)

US African American urban residents, n=10 (30-65 y.o) for focus groups, n=230 for questionnaire, convenience samples

Questionnaire Distance to supermarket (1-5 blocks, >5 blocks) Car ownership (Y/N) Focus group: barriers to fruit and vegetable purchase, preparation and consumption

Questionnaire Food behaviours (Y/N): Plan before buying food ; Shop at least weekly; Use free food; Dinner is main meal; meals made by self only; fast food 1+ times per week

- From the focus groups, cost was cited as the primary structural barrier to fruit and vegetable consumption, with fruits and vegetables being believed to be expensive relative to other foods by volume or satiety provision. - from the questionnaire, there were no significant relationships between distance and behavioural outcomes. The small tendencies were as follows: Compared with women living within 5 blocks of supermarkets, women who lived further from supermarkets more often planned before buying foods (49 vs 44%) had dinner for a main meal (76 vs 65%), made meals alone (82 vs 74%) and less often shopped weekly (23 vs 30%, used free food (53 vs 59%) and consumed fast food (54 vs 53%). - No significant relationships between car ownership and behavioural outcomes. The tendencies were as follows: Compared with women who owned a car, women did not own a car more often made meals by themselves (83 vs 75%), shopped at least weekly (28 vs 23%) and used free food (59 vs 51%), and less often plan before buying food (45 vs 49%), had dinner as a main meal (70 vs 73%) and consumed fast food at least weekly (51 vs 58%). Limitations: non-probabilistic sampling, limited distance measure

Morland et al., 2002 (214)

cross-sectional

US, 4 counties

n=2392 Black & n=8231 White ARIC participants with valid data living in n=221 census tracts

Presence or absence in census tract of supermarket (S) grocery store (G)

Semi-quantitative FFQ (10 fruit items, 16 vegetable items)

- Presence of a supermarket (but not other shop types) associated with greater odds of meeting dietary guidelines for fruits and vegetables (fat, and saturated fats) for black, but not white Americans. RR and 95% CI of meeting dietary guidelines fruit & vegetable intake S’mkts (white Americans): 1.11 (0.93, 1.32) , adj. 1.08 (0.89, 1.30) S’mkts (African Americans): 1.42 (1.06, 1.91), adj. 1.54 (1.11, 2.12) Grocery stores (White Am.): 0.92 (0.78, 1.09), adj. 0.93 (0.78, 1.10) Grocery stores (African Am.): 0.99 (0.80, 1.23), adj. 1.07 (0.83, 1.38) adj. adjusted for other types of food stores, food service places, income,

xxiv

education (all not confounded by age) NB – results similar for total fat, saturated fat, no associations for cholesterol & car access 3x lower among African Americans, Limitations: not adjusted for gender, site differences not examined

Laraia et al., 2004 (358)

cross-sectional

US (Wake County, NC) (mid-sized metropolitan area)

n=973 Participants of PIN cohort living in Wake County, lower-middle income women recruited 24-29 weeks gestation from 4 prenatal clinics

USDA 2000 inspection registry supermarkets, grocers, convenience stores (excl. stores attached to petrol stations, ethnic grocers, specialty stores) Straight line distance to nearest shop Number of shops in census block Number shops within 0.5km

self administered FFQ (120 item modified Block) Dietary Quality Index for pregnancy (range 0-80) (based on intake of grains, vegetables, fruits, folate, iron, calcium, fat and meal patterns), collapsed in tertiles

- Living further from shops (supermarkets and convenience stores) is associated with poorer diet among pregnant women, adj. age, race, income, education, marital status. Adjusted odds of being in lowest (vs highest) tertile of DQI-P for distance (4 miles + vs <2 miles) to nearest: convenience store 1.17 (1.02, 1.35) grocery store 1.08 (0.97, 1.19) supermarket 2.46 (1.4, 4.3) (supermarket 2.16 (1.2, 4.0) when further adjusted for distance to other shop types) -No association shop numbers and DQI-P (figures not reported) adj. age, race, income, education, marital status NB – wide range of distance exposures women lived on average 2 miles from each shop type, but lived up to 8.3 miles from supermarkets, 6.3 miles from convenience stores and 11.0 miles from grocery stores

Chou et al., 2004 (256)

ecological time-series

US (nation-wide)

n=6 years, states per year (min. n=15, max n=51)

Census of retail trade 1982, 1987, 1992, 1997 (interpolating for missing years) per capita

Behavioural Risk Factor Surveillance System (1984-1999) (annual telephone surveys, adults 18+)

- Increasing availability of restaurants (including fast food) per capita at the state level is associated with increasing average BMI and obesity at the state level. -Elasticities BMI with per capita restaurants (0.17) -increase in obesity with 10% increase in per capita restaurants (1.39%)

xxv

restaurants (including fast food) (State level)

BMI & % obesity (State level)

Rose and Richards, 2004 (221)

cross-sectional

US n=963 (nationally representative) Food Stamp Program Participants random sample of Households in Food Stamp Program from random sample n=43 counties (proportional to size) (67% RR)

1996-97 National Food Stamp Program Survey + road network data Distance to main store used for shopping Travel to principal store (>30 minutes) Car ownership Self-reported store access (based on whether respondent shopped at supermarket, distance & travel time)

Survey weekly household food use (foods used from the home food supply – bought or home grown for preparing and/or consuming at home) Household fruit and vegetable use (grams per adult male equivalent per day)

- Compared with participants living less than one mile from the shop they mostly utilised, those living 1-5 miles away had similar or lower fruit intake (Mean difference (95%CI) -15 (-64, 34)) and those living 5 or more miles away had significantly lower fruit intake (-62 (-117, -7)). Compared with living less than 1 mile from shops, living 1-5 miles or 5 or more miles from shops was not substantially associated with vegetable intake (mean differences (95% CI): -20 (-101, 61) and -36 (-108, 35), respectively.. - Shorter travel time (<30 mins vs 30 mins+) and car ownership showed a no substantial associations with fruit intake (23 (-41, 88) and -13 (-63, 38), respectively) or vegetable intake (30 (-22, 81) and -30 (-78, 19), respectively). - Compared with respondents who did not shop at supermarkets (who were classed as having little access), respondents who had moderate access (i.e. they buy mostly from supermarkets, but have no car and the round trip takes 30 minutes or more) had higher intakes of fruits (64 (-39, 166)) but not vegetables (-7 (-106, 92)), and respondents with the greatest access (i.e. they shop at a supermarket and do not meet the ‘moderate’ criteria) have significantly higher fruit intake (86 (7, 164)) and substantially, but not significantly higher vegetable intake 51 (-55, 156). -All results adjusted for urbanisation, household income, size, race/ ethnicity, schooling, single parent status, employment status of respondent. (NB: virtually no change in the access measure results when adjusted for dietary attitudes and awareness of the food guide pyramids)

Sturm and Datar, 1995

longitudinal, 3 year follow up

US multistage cluster n=1000 schools

US Bureau 1999 Zip Code Business Patterns files

Anthropometric measurements BMI

Figures not reported. No significant relationship between availability of shops and BMI gain in kindergarten children over a three year period.

xxvi

(255) n=13282 kindergarten children postcode (school & residence)

Number of shops / 1000 residents - grocery - convenience

None significantly related at p<0.05

Zenk et al., 2005 (174)

cross-sectional

US, eastside Detroit 2001

n=266 African American women recontacted (80% RR) women still living in Detriot from women in 1996 ESVWHP study (who were a probability sample, 81%RR).

Per capita family income Shop most often utilised (supermarket, specialty store vs independent and suburban vs city)

Behavioural Risk Factor Surveillance System Survey fruit and vegetable intake (times consumed daily)

-Shopping at a supermarket rather than independent store is associated with greater fruit and vegetable intakes among African American women. - Higher income is not directly, but indirectly associated with fruit and vegetable intake (via shopping at supermarkets) Direct effects: B (SE) Income: -0.10 (0.15) p>.05 supermarket: 1.22 (0.33) p<.0001 specialty: 2.37 (1.06) p<.01 suburban -0.54 (0.36) p>.05 Indirect effects: B (SE) Income: -supermarket 0.10 (0.05) p<.05 -specialty -0.01 (0.03) -suburban-0.01 (0.01) Limitations: only 7 women shopping at specialty stores, less than ideal model fit (RMSEA= 0.26, CFI= 0.28)

Mobley et al., 2006 (355)

cross-sectional

US (five states)

n=2692 women enroled in WISE-WOMAN program (2001-02) low income, uninsured, many Hispanic

Census data Neighbourhood (zip code) -grocery stores -“minimarts” /1000 population

WISEWOMAN data BMI 10 year CHD risk (gender, age, TC, HDL, SBP, smoking, diabetes)

- No association between the abundance of grocery stores, stores, or minimarts per capita (at the ZIP code level) and BMI or Coronary Risk Scores among low income women in the US. Regression estimates (SE) for BMI: grocery stores: -0.37 (0.42) ns, minimarts -0.25 (0.54) ns Regression estimates (SE) for 10 year CHD risk (log) grocery stores: -0.01 (0.05) ns

xxvii

and African American 40-60y (not random, nor representative)

minimarts -0.09 (0.08) ns mutually adjusted, and also adjusted for age, smoking ,education, race/ ethnicity, land use mix, fitness facilities per 1000 residents, Index of Racial Segregation, robbery arrests per 100000 residents, Income inequality, income (neighbourhood)

Morland et al., 2006 (362)

cross-sectional

US (4 states)

population based sample n=207 census tracts n=10,763 men & women

Government listings (1999) presence/ absence in tract of shop: - supermarket (S) - grocery (G) - convenience (C)

Medical examinations Overweight (BMI >25-30) & obesity (BMI >30). Diabetes Hypertension (systolic or diastolic) High Cholesterol (total)

- Prevalences of overweight and obesity were significantly lower in tracts which had a supermarket, significantly higher in tracts with a convenience store and non-significantly higher in tracts with a grocery store. All shops showed similar patterns of association with hypertension (though non-significant), and supermarkets and grocery stores showed similar trends with diabetes. Shops showed nearly null associations with the prevalence of high cholesterol. - Examination of combinations of shop types in areas showed the prevalence of overweight and obesity was higher when areas lacked a supermarket (even with other shop types present), and was lowest when only supermarkets (or supermarkets and grocery stores) were present. Prevalence Ratios & 95% CI of diet related outcomes according to the presence / absence of each shop type in the local census tract

Overweight Obesity Supermarkets 0.94 (0.90, 0.98) 0.83 (0.75, 0.92) Grocery stores 1.03 (1.00, 1.07) 1.07 (0.99, 1.16) Convenience stores 1.06 (1.02, 1.10) 1.16 (1.05, 1.27)

diabetes ↑ cholesterol hypertension S’mkts 0.96 (0.84, 1.10) 0.98 (0.94, 1.03) 0.92 (0.85, 1.01) Grocery 1.11 (0.99, 1.24) 0.99 (0.95, 1.03) 1.08 (1.00, 1.17) Conv. 0.98 (0.86, 1.12) 1.00 (0.96, 1.04) 1.08 (0.99, 1.18)

overweight obese S’mkt only 1.00 (ref) 10.0 (ref) S’mkt & Grocery 1.04 (0.71, 1.53) 0.95 (0.81, 1.12)

xxviii

S’mkt & Conv. 1.35 (1.05, 1.73) 1.11 (1.00, 1.23) S’mkt, grocery & conv 1.28 (1.00, 1.64) 1.09 (0.98, 1.21) Grocery only 1.48 (1.12, 1.94) 1.14 (1.01, 1.27) Convenience only 1.45 (1.16, 1.82) 1.12 (1.02, 1.23) Grocery & Convenience 1.60 (1.28, 2.00) 1.18 (1.08, 1.30) No shops 1.28 (1.00, 1.63) 1.08 (0.97, 1.19)

adjusted for other food service places, gender, race/ ethnicity, age, income, education, physical activity - Limitations: population density, size, area SEP not considered.

Bodor et al.., 2007 (220)

cross sectional

New Orleans, USA

Adult primary food shoppers from n=102 households (random sample from phone directory, 53% RR) from 4 contiguous census tracts (conveniently sampled for high socioeconomic and racial diversity)

Presence or absence of a supermarket within 1000m, and small food stores within 100m. Car ownership (yes/ no) definition small food store= sales <1 million per annum definition supermarket = sales >5 million per annum

24 hour recall by telephone (asked for fist sized serves of 4 fruit, 10 veg items listed plus up to two unlisted fruit and vegetables, excluding mixed dishes ) Fruit intake, vegetable intake (serves/ day)

- Access to small food store within 100m is associated with greater intake of vegetables of around 1 serve/ day (mean (SE) 3.3 (2.3) vs 2.4 (1.6), p<0.05) and around ½ serve per day of fruit (2.4 (2.1) vs 1.8 (1.4), p>0.05). - Access to a supermarket within 1 km was not associated with intake of fruits (2.0 (1.4) vs 2.1 (1.9), p>0.05) or vegetables (2.5 (1.5) vs 2.9 (2.1) , p>0.05). - Slight but non-significant tendency to greater vegetable intake with car ownership (3.0 (2.1) vs 2.3 (1.5), p=0.09), but no difference in fruit intake (2.1 (2.2) vs 2.0 (2.2), p>0.05) (all results adjusted for age, gender, income (poverty income ratio which is specific to household size), and race/ ethnicity) -Limitations – questionable representativeness: a) used other adults where main food shopper not available b) phone directory sampling frame for households d) convenience sample of census tracts, low power, validity of normality assumption not examined, amount and handling of missing data unclear, clustering not considered. No stratified analyses.

xxix

Table A2.2: Studies examining the relationship between shop availability (excluding fast food shops) (abundance and or distance) and diet, or diet-related outcomes (U.K studies) Study Design Location Sample Measures IV Measures DV Results/ Conclusions & Limitations Cade et al., 1999 (258)

cross-sectional, telephone survey

UK Random sample of 104 women within the UK Women’s Cohort study with low and high dietary scores

Telephone Interview Time to main shop (<10 mins vs >10mins) Use car to shop Shopping + travel time (< 1-1.5 hrs or > 1-1.5 hrs) More expensive to eat healthy (Y/)

Dietary score unhealthy group (0) or healthy group (8) on a 0-8 indicator based on based on intakes (as a percentage of energy) of saturated fats, polyunsaturated fats, protein, complex carbohydrates and free sugars, and total intake of dietary fibre, fruit and vegetables, pulses nuts and seeds.

- Women displaying very healthy were similarly likely to use a car when shopping (85% vs 90%, p>0.05) and similarly likely or less likely to travel less than ten minutes to shops (73% vs 83%, p>0.05), compared with women displaying very unhealthy diets, however the healthy diet follwoers were less likely to have a round travel and shopping time longer than 1-1.5 hours (60% vs 91%, p<0.05). - Women displaying very healthy diets were less likely to report perceiving an extra expense to following a healthy diet (29% vs 40%) but were more likely to perceive availability to be a problem (38% vs 27%), compared with women following the least healthy diets (all comparisons p>0.05). Limitations: small sample, limited power.

Wrigley et al., 2002 (216)

case study UK (Leeds - deprivedarea)

n=1009 (Wave 1, 34% RR) n=615 (wave 2) respondents mostly responsible for food in household

Interviewer administered questionnaire Pre vs post construction of large superstore (Tesco)

self-completed 7 day food diary Intake (portions per day) fruits and juices vegetables both

- Mean fruit and vegetable consumption increased slightly after construction of a superstore in a deprived area, overall (Pre: 2.88, Post 2.92 portions/ day). -Mean fruit and vegetable intake increased substantially among participants with low baseline fruit and vegetable intakes (<7 portion/week): (4.13 to 9.83 portions/ week F&V & 0.77 to 3.92 portions/ week of fruit & juices) - Mean fruit and vegetable intake increased substantially among participants with low baseline fruit and vegetable intakes (<14 portion/week): (9.17 to 12.25 portions/ week F&V and 2.82 to 4.59 portions/ week fruit and juices. (data presented are for participants with both wave)

xxx

Limitations: Sample non-representativeness, no statistical comparisons, concurrent changes

White et al., 2004 (181)

multilevel, cross-sectional

UK (New-castle Upon Tyne)

n=5044 individuals (83% RR) n=3153 households (18% RR)

distance to nearest shop (km) type of store mostly used (multiple supermarket, discount supermarket, department stores, other, unkown) mode of travel to store (foot/bicycle, taxi, public transport, private car)

FFQ (modified from EPIC study) fruit & vegetable consumption (g/day) Fat consumption (% E from fat) Non-Starch-Polysaccharide intake (g/ day) all standardised to Z scores Overall dietary index (composite of indices for F&V, NSP & fat)

- At a bivariate level, measures of accessibility were significantly associated with all dietary outcomes. However, after adjustment, no measures were significantly associated with fruit and vegetable intake though type of shop used remained associated with fat and NSP intake, and mode of travel remained related to NSP intake Dietary indices by accessibility (bivariate, ANOVA results)

Bivariate, ANOVA results Multiple linear regression estimates (B)

F &V fat NSP F&V

fat NSP

Shop at p <0.05 p<.01 p<.001 discount 0.005 -0.03 -0.02 ns ref ref multiple -0.09 0.18 0.12 ns -0.13** -0.08 ns dept 0.07 0.17 -0.13 ns 0.07 ns -0.32 ** other -0.10 0.12 -0.21 ns -0.11 ns -0.06 ns unkown 0.17 -0.10 0.07 ns -0.19* 0.03 ns Travel p<.01 p<.01 p<.001 ns ref ref foot/ bike

-0.11 / 0.24

0.12 / 0.49

-0.08/ 0.11

ns ns ref

taxi 0.16 -0.06 0.13 ns ns 0.14 ns public 0.05 0.08 0.09 ns ns 0.14 * car -0.004 -0.03 -0.02 ns ns 0.04 ns ns -0.12 ** ns

NB: Bivariate results for distance not reported * p<0.05 ** p<0.01 ***p<0.01 Multivariable results are independent of the other accessibility variables and: age, gender, BMI, dietary knowledge, alcohol consumption, ethnicity, marital status, household composition, gross household income, state benefits, health benefits, work status, standard of living index, car ownership, house ownership, number of rooms, education, illness,

xxxi

smoking, physical activity, cost of weekly food - Limitation: cannot determine from bivariate and final models only whether SEP differences in diet are mediated by differences in accessibility, Interaction between SEP and accessibility not examined, multivariable results reported only for predictors with “significant” associations (magnitude of association not assessable).

Pearson et al., 2005 (219)

Cross-sectional

UK (Barnsley, South Yorkshire)

Main food shoppers from random sample of electoral roll addresses n=426 (42% RR) from purposive sample of 4 wards

Questionnaire, shopping basket survey, road network data: Road distance to nearest supermarket (km) Potential difficulties shopping (Y/N) (car ownership, mobility)

Single item 24 hour recall Portions consumed over the last day (defined size)

- fruits - vegetables

- No significant or large relationship between distance to nearest supermarket, or potential difficulties shopping and fruit and vegetable consumption. Fruit consumption – Beta coefficient (95% CI) Distance 0.05 (-0.02, 0.12) Has difficulties 0.02 (-0.36, 0.39) Vegetable consumption – Beta coefficient (95% CI) Distance 0.01 (-0.05, 0.07) Has difficulties -0.26 (-0.69, 0.16) adjusted for sex, age, area SES, fruit and vegetable price (ward as a random effect) Limitations: Personal income, education not considered (could affect area SES effect estimation), mutual adjustment for effects of area SES & distance, difficulties etc. could cancel out (if indirect relationship occurring), limited geographical range (4 wards) however may have sufficient range of exposures regarding distance to supermarkets (median distance (road)=1.9 km, from 0.1 to 9.4 km)

Cummins et al., 2005 (218)

Quasi-exeprimental

UK, Glasgow

Intervention and Control study site. Cluster sample of n=3975

Intervention site (I) experienced opening of superstore two months after the

Questionnaire: Pre and post intervention self reported consumption of

- Dietary changes that occurred tended to be in a positive direction, however there was no evidence the change in the intervention group exceeded that of the control group - ANCOVA results showed the intervention change net of control to be

xxxii

households within postcode sectors (stratified by DEPCAT)

baseline survey while control site did not (C). Follow up period 12 months after baseline survey.

fruits and vegetables (portions/day) (2-items)

close to zero for fruit (-0.03, 95% CI -0.25, 0.30), vegetables (-0.11, 95% CI: -0.44, 0.22) and combined fruit and vegetable intake (-0.10, 95% CI -0.59, 0.40), having adjusted for baseline consumption, sex age education, employment and education. - Similarly, among residents of intervention area, switching to the new store was not associated with large or substantial improvements in fruit, vegetable, or combined fruit and vegetable intake. (Differences ranged from 0.09 portions per day for vegetables to 0.35 portions per day for fruits and vegetables combined.

Stafford et al., 2007 (422)

pooled two national cross sectional studies (Health Survey of England – HSE and Scottish Health Survey - SHS)

UK HSE 1995 & SHS 1998 (Probabilistic sample, response rates 69% to 81%, representative of British and Scottish populations) n=438 (HSE) n=81 (SHS) postcode sectors chosen (stratified by population density & deprivation) n=6848 participants with complete data (similar to excluded sample)

Face to face interviews Supermarkets per postcode district

Face to face interviews BMI (height and weight measured by trained nurse)

- No substantial or significant relationship between the number of supermarkets per postcode and residents’ BMI or WHR in England and Scotland. (the corresponding coefficients from Structural Equation models were -0.001 and -0.005, respectively.) Independent of: age, sex, social class, Neighbourhood disorder, Sports participation rate, swimming pools, population density, high street facilities (financial and health related stores) and distance to nearest post office for the WHR model.) Population density (a common correlate of shop location) was inversely associated with BMI and WHR (correlations -0.024 ns and -0.041 respectively). -Limitations –deprived postcodes underrepresented in England, and over-represented in Scotland. SES was included in the models but an SES-supermarket – BMI specification was not examined (nor SES – other factors – BMI). .

xxxiii

Table A2.3: Studies examining the relationship between shop availability (excluding fast food shops) (abundance and or distance) and diet, or diet-related outcomes (Australian) Study Design Location Sample Measures IV Measures DV Results/ Conclusions & Limitations Burns and Inglis., 2007 (215)

multilevel cross-sectional

Melbourne, AUS

n=45 suburbs (stratified random sample) n=1347 women aged 18-65 yrs (oversampled low & mid SES) (50% RR)

Council registries & phone directories Supermarkets, fruit and vegetable stores per 10, 000 capita - suburb level

Questionnaire (1-item, validated against NNS) Fruit intake, vegetable intake (serves/ day)

- Access to supermarkets and fruit and vegetable stores not significantly associated with fruit or vegetable intakes. (Small, non-significant associations only) Correlations (Pearson’s R for bivariate and B for multivariable) between accessibility and fruit and vegetable intake

Fruit Vegetable S’markets Bivariate -0.03 NS 0.04 NS Adjusted not tested not tested F &V -0.04 NS 0.06 p<.05 Adjusted 1 not tested 0.027 NS Adjusted 2 not tested 0.016 NS

1 Adjusted for age, marital status and education 2. Adjusted as 1 and for psychosocial constructs Limitations: possibility of suppression, no account for income, population density, suburb size, insufficient detail provided to justify parametric modelling choices

xxxiv

Table A3.1: Studies examining in-store availability or price and consumption of fruits and vegetables (U.S., U.K., Canada, Australia) Study Design Location Sample Measures IV Measures DV Results / Conclusions & Limitations Fisher and Strogatz, 1999 (222)

ecological cross-sectional

New York City, USA

n=250 households per zipcode selected randomly from telephone listings (RR > 75%), any adult in household sampled from n=19 zip-codes randomly selected from larger, random sample of n=503 stores (within 7 counties from rural to major metropolitan in urbanicity) Exclusion: households not buying milk

Shop audit Percentage of milk in stores that is low fat (<%1)

Telephone interview Household low-fat milk consumption (y/n whether <1% fat milk was usually present in the household milk usually available)

- Proportion of low-fat milk in stores associated with proportion of households with low-fat milk available in-home (B=0.81, 95% CI: 058, 1.07) (for n=19 zip-codes). - Sub-group analyses showed this was stronger for respondents who shopped within their residential zipcode (B=1.02) compared with those who shopped outside their residential zipcode (B=0.68). Limitations: ecological fallacy, confounding bias (did not adjust for sociodemographics), possible misclassification of exposure (failed to specify store types), possible inaccuracy of household availability measures from using any adult rather than main food shopper or similar, failure to account for clustering

Edmonds et al., 2001 (223)

Cross-sectional

Texas, USA

172 African American boy scouts aged 11-14 yrs (sample=all within selected troops, 83%RR) from n=12 troops ( >80% African American, meet >3

Availability of fruits (10), juices (3), vegetables (12) in -grocery stores (may include supermarkets) (observed availability Y/N & shelf space)

24 hour dietary recall (trained nutritionist & food models) Intake of fruits, juices, vegetables (including French fries)

- No relationship between in store availability and African American boys’ intakes of fruits, vegetables and juices. - Restaurant availability strongly correlated with boys intakes of vegetables and fruit juices. Correlation between availability and intake of fruits, vegetables and fruit juice

Fruit Veg Fruit Juice home 0.11 ns 0.33 ns -0.32 ns

xxxv

times per month) within Sam Houston Area Council of Boy Scouts (sample type & RR unknown)

-restaurants (Y/N on menu) (one grocery store per tract selected, all restaraunts selected) census tract level (or contiguous tracts if live near border)

home a ns ns ns home b ns ns ns store -0.17 ns 0.02 ns 0.17 ns store a ns ns ns store b ns ns ns restaurant -0.29 ns 0.53 ns 0.61 ns restaurant a -0.33 0.72 * 0.70 * restaurant b ns ns ns

* p<0.05 ns p > 0.05 a adjusted for income (median African American family income – census tract) b adjusted as a plus for other availability measures (NB – only p values provided for some multivariable models) - Limitaitons: misclassification of exposure – census tract, 1 store only, unclear how treated areas without grocery stores. Mentions supermarkets & convenience stores but presents no results – unsure if these were included.

Zenk et al., 2005 (174)

cross-sectional

US, eastside Detroit 2001

n=266 African American women recontacted (80% RR) women still living in Detriot from women in 1996 ESVWHP study (who were a probability sample, 81%RR).

selection/ quality (1 (poor ) to 4 (excellent)) & affordability (1 (very affordable) to 4 (not at all affordable)) of fresh fruit and vegetables of store usually shopped at) )

Behavioural Risk Factor Surveillance System Survey fruit and vegetable intake (times consumed daily)

- Self rated selection/ quality but not affordability of fresh produce is significantly associated with fruit and vegetable intake - Better self reported selection/ quality mediated the relationships between shopping at a specialty store (vs independent grocer) and shopping at a suburban (vs city) shop. Effect on Fruit and Vegetable intake B (SE) Direct effects: B (SE) Selection/ Quality 0.43 (0.20) p<0.05 Affordability -0.05 (0.26) ns (p>0.1) Indirect effects: B (SE) Supermarket -selection/quality: -0.07 (0.05) ns -affordabilty: 0.01 (0.07) ns Specialty: -selection/quality 0.33 (0.16)p<.05

xxxvi

-affordabilty -0.03 (0.17) ns Suburban: -selection/quality 0.23(0.11)p<.05 -affordabilty: <-0.01 (0.01) ns ns= p>0.05 NB – model fit not ideal (RMSEA=0.26, CFI=0.28) - Limitations: self reported selection / quality & affordability, poor model fit, findings for specialty stores based on 7 women.

Bodor et al., 2008 (220)

cross sectional

New Orleans, USA

Adult primary food shoppers from n=102 households (random sample from phone directory, 53% RR) from 4 contiguous census tracts (conveniently sampled for high socioeconomic and racial diversity)

Total in-store availability (shelf space (m) of fresh, frozen, canned) of fruit and vegetables in all small food stores within 100m of respondents. definition =small food store sales <1 million per annum

24 hour recall by telephone (asked for fist sized serves of 4 fruit, 10 veg items listed plus up to two unlisted fruit and vegetables, excluding mixed dishes ) Fruit intake, vegetable intake (serves/ day)

- Greater in-store access to vegetables, but not fruit in small stores within 100m is significantly related to intake. Daily intake (mean serves + SD) by total metres of shelf space devoted to fruits and vegetables in small food stores within 100m

Fruit Veg None 2.0 (1.7) 2.4 (1.6) 0-3 m 1.8 (1.3) p>0.1 3.3 (2.4) * >3 m 2.3 (2.2) p>0.1 4.5 (2.4) *

adjusted for age, gender, income (poverty income ratio which is specific to household size), and race/ ethnicity Daily intake of fruits (mean serves + SD) No food assistance 1.9 (1.7) / Food assistance 2.8 (2.1)* Daily intake of vegetables (mean serves + SD) No food assistance 2.8 (2.0) / Food assistance 3.2 (2.1) -Limitations: as per table A2.1. Also, results could reflect effect of shop presence, not shelf space), greater total shelf space may indicate more rather than better stocked shops

Ard et al., 2007 (257)

cross-sectional

Birmingham, Alabama, US

Participants in the Hi5+ intervention n=1355 parents of 4th graders

Expenditure data (1999 AC Neilson Homescan) Price of fruit juice

Questionnaire (at baseline of Hi5+ intervention) availability in the

- Each additional 10c per serve associated with less chance having item at home OR=0.77 (95% CI: 0.75, 0.79) (p<0.001) adjusted for race, income, BMI, parent’s BMI, gender - Using three cost cut-offs, a threshold effect might operate.

xxxvii

cluster sample

and vegetable items (average cost per serve) based on all shops (supermarket, grocery, convenience store etc.)

home in previous 2 weeks of: 3x 100% juice items, 13 fruits, 18 vegetables (fresh, frozen or canned)

Compared with low cost (<20c/serve), Medium cost (20-29c/serve) OR=0.99 (0.94, 1.03) & High cost (30c +/serve) OR=0.67 (0.63, 0.71). -Relationships were slightly stronger for whites (OR=0.72 (95% CI: 0.7, 0.75)) compared with African Americans OR=0.89 (95% CI: 0.82, 0.96). (Based on discrete choice analysis, accounts for the availability of multiple items) Limitations: doesn’t examine quantity

Watters, 2007 (259)

cross-sectional

North Carolina, US

random sample 658 African- Americans, aged 18–70 years six counties of North Carolina (3 urban, 3 rural) from Department of Motor Vehicles roster 17.5% response rate

Questionnaire - “Do you feel that you can afford to purchase healthy foods, such as fruits and vegetables?” Yes/Sometimes/ No

7-item screening tool (National Cancer Institute) Fruit and vegetable intake during the past 3 months (serves/ day) Fruit intake (fruit and juice) Vegetable (green or lettuce salad, potatoes (boiled, baked or mashed), other vegetables, beans and peas, and vegetables in mixed dishes)

Mean intake (serves per day) of fruits and vegetables according to perceptions of ability to afford healthy foods, including fruits and vegetables

F &V F V Yes 2.52 0.90 1.63 Some-times

2.49 0.85 1.63

No 2.39 0.73 1.66 p 0.88 0.41 0.99

NB: SD/SE not presented. adjusted for body mass index, education, age and gender and other ‘enabling’ factors (belief it takes time and trouble to prepare healthy foods, belief it is easy to order healthy foods at restaurants and believing have need to know how to prepare healthy foods)

Cade et al., 1999 (258)

(see table A2.2)

U.K.

Dibsda cross- East n=680 low income Questionnaire Questionnaire - Low income respondents who consumed 5 or more fruits and

xxxviii

ll et al., 2003 (184)

sectional Anglia, UK

adults residing in public housing in East Anglia who mostly purchased food for their households. Random sample, 23% RR

scales derived from items measured on 1-7 likert scale Perceived Affordability Perceived availability (choice) (for detail, see table )

Number of ‘helpings’ of fruits and vegetables consumed daily (as perceived by respondents)

vegetables daily indicated greater availability of fruits and vegetables to them (in-store, and in terms of choice of shops) and indicated greater affordability of fruits and vegetables to them, compared with those consuming 0-2 portions of fruits and vegetables per day. Attitude ratings mean (+SD) among groups according to fruit and vegetable intake (portions per day)

0-2 (n=317)

3-4 (n=207)

5+ (n=114)

Afford 3.0 +0.6 3.2 +1.5 3.8 +1.8* Choice 2.0 +0.9 1.9 +0.9 1.7 + 0.7*

* different to 0-2 group at p<0.01 NB – scaled so that higher mean values of choice indicate less choice, and higher mean vaulues of affordability indicate better affordabiltiy) Limitations –low response (23%), however response bias and generalisabilty may not be an issue as sample was representative according to national figures.

Pearson et al., , 2005 (219)

Cross-Sectional

UK (Barnsley, South Yorkshire)

Main food shoppers from random sample of electoral roll addresses n=426 (42% RR) from purposive sample of 4 wards

Shopping basket survey (14 day period) Lowest price for carrots, onions, cauliflower, and potatoes, lettuce, tomatoes, apples, bananas, oranges (£ per standardized weight unit)

Portions consumed over the last day (defined size)

- fruits vegetables

- No substantial or significant association between price and fruit or vegetable intake, but slightly greater tendency for vegetables than fruits. Fruit (portions / day) F&V price Price (£) β= -0.01 (-0.52, 0.50) Vegetables (serves/ day) F&V price Price (£) β= -0.26 (-0.69, 0.16) adjusted for sex, age, area SES, distance, difficulties shopping (ward as a random effect) Limitations: Prices measured for fruits, especially, may not have been typical of the prices of the fruits actually consumed which may contribute to null associations.

Giskes et al..,

cross-sectional,

Brisbane, AUS

Main adult food purchaser from

Survey perceived

usual purchase of regular vs

-Nutritionally ‘superior’ versions were always more expensive. Price differences ranged from as little as 1% more for low

xxxix

2007 (226)

multilevel n=1003 households randomly selected from 50 CCDs (random sample, stratified by SES) who shopped at a supermarket which was included in the audit and had available data.

availability of “nutritionally superior” versions of 14 common grocery items (Yes/ No/ Not Sure) Perceived comparative price of nutritionally superior versions (more expensive / cheaper / the same / unsure) Audit of supermarkets usually patronised by respondents Actual comparative price of superior versions (as % of regular) in shop audit (lowest prices, sizes standardized) supermarket NB Examined actual availability of items (yes/ no) but not further analysed as almost universally available.

nutritionally superior versions of 14 common grocery items. (NB – buying both kinds collapsed with nutritionally superior category) face to face survey

saturated fat solid cooking fat to 64.4% more for reduced fat margarine. For fruit and vegetable items, differences were substantial - Salt reduced legumes (+47.0%), tinned fruit in natural juice (55.6%) 100% fruit juice (19.1%). -Actual price differences not significantly or substantially related to the odds of purchasing item. Largest associations were for reduced fat milk (0.81 (0.63-1.05) and reduced fat margarine (1.15 (0.96, 1.32)). -Respondents were less likely to purchase nutritionally superior items they perceived to be more expensive – significantly so for 6/14 items (bread, milk, cheese, yoghurt, chicken and tinned fish). Largest association for chicken (0.35, 0.23-0.53) and smallest for cheese (0.64 (0.47 – 0.87). -Nutritionally superior items were all more likely to be purchased when respondents perceived them to be available, significantly so for 12 / 14 items and with most OR >2. However confidence intervals were wide as estimates were based on only a small number of respondents who did not perceive items as available. -Substantial attenuation (> 10%) of difference between low and high income households in the odds of purchasing the nutritionally superior varieties occurred with adjustment for perceived availability (for bread, tinned fruit, cheese, tinned fish and solid cooking fat), but not perceived or actual price. Limitations – referent categories for outcome variables, and perceived availability and price unclear, making interpretation difficult. This leaves alternate explanations which could explain results (people who don’t buy any version of an item are more likely to be unsure about comparative price and availability). The dietary behaviours examined my have little relevance to health outcomes as there was little actual nutritional difference between many of the “nutritionally superior” and regular grocery items (eg salt reduced vs regular butter).

xl

II. Appendix II: Relevant materials used in the BFS data collection

This Appendix contains materials from the Brisbane Food study relevant to the

secondary analyses. Pages from the household questionnaire pertaining to fruit and

vegetable purchasing and demographic characteristics are included, followed by

instructions and tools used for identifying and classifying shops and measuring fruit

and vegetable prices and availability. (The pricing and availability audit tool for

supermarkets and convenience stores is identical regarding fruit and vegetable items

to the tool for fruiterers / greengrocers that has been included in this Appendix).

xli

II.1. Household survey – relevant sections

xlii

xliii

xliv

xlv

xlvi

xlvii

xlviii

xlix

l

li

lii

liii

liv

II.2. Audit tools and instructions

lv

lvi

lvii

lviii

lix

lx

lxi

III. Appendix III: Relevant materials used in the cooking skills survey

III.1. Ethical clearance

From: Wendy Heffernan [mailto:[email protected]] Sent: Tuesday, 22 June 2004 3:31 PM To: [email protected] Cc: [email protected] Subject: Confirmation of Level 1 ethical clearance - 3559H

Dear Elisabeth I write further to the Level 1 (Low Risk) application received for your project, "Looking and buying fruits and vegetables" (QUT Ref No 3559H). The Chair, University Human Research Ethics Committee, has considered your application and requested I contact you on her behalf. The Chair has confirmed that the project qualifies for Level 1 (Low Risk) ethical clearance. This approval is subject to:

• confirmation regarding the size of the participant pool; and • the questionnaire cover sheet being provided in accordance with Booklet 11

of the University Human Research Ethics Manual (http://www.research.qut.edu.au/oresearch/policyandpro/ethics/humanmanual.jsp).

However, you are authorised to immediately commence your project. This authorisation is provided on the strict understanding that the above information is provided as soon as possible. Please do not hesitate to contact me further if you have any queries regarding this matter. Regards Wendy

lxii

University Human Research Ethics Committee Information in relation to ethical clearance

What is the duration of my ethical clearance? The ethical clearance awarded to your project is valid for three years commencing from 22 June 2004. Recruitment, consent and data collection / experimentation cannot be conducted outside the duration of the ethical clearance for your project. Please note that a progress report is required annually on 22 June or on completion of your project (whichever is earlier). You will be issued a reminder around the time this report is due. The progress report proforma can be located on the University Research Ethics Webpage http://www.research.qut.edu.au/oresearch/policyandpro/ethics/index.jsp. Extensions to the duration of your ethical clearance within the 3-5 year limit must be made in writing and will be considered by the Chair under executive powers. Extensions beyond 5 years must be sought under a renewal application (usually involving the completion or a checklist for Researchers seeking expedited ethical review). Standard conditions of approval The University’s standard conditions of approval require the research team to: 1. conduct the project in accordance with University policy, NHMRC / AVCC guidelines and regulations, and the provisions of any relevant State / Territory or Commonwealth regulations or legislation; 2. respond to the requests and instructions of the University Human Research Ethics Committee (UHREC) 3. advise the Research Ethics Officer immediately if any complaints are made, or expressions of concern are raised, in relation to the project; 4. suspend or modify the project if the risks to participants are found to be disproportionate to the benefits, and immediately advise the Research Ethics Officer of this action; 5. stop any involvement of any participant if continuation of the research may be harmful to that person, and immediately advise the Research Ethics Officer of this action; 6. advise the Research Ethics Officer of any unforeseen development or events that might affect the continued ethical acceptability of the project; 7. report on the progress of the approved project at least annually, or at intervals determined by the Committee;

lxiii

8. (where the research is publicly or privately funded) publish the results of the project is such a way to permit scrutiny and contribute to public knowledge; and 9. ensure that the results of the research are made available to the participants. Modifying your ethical clearance The University has an expedited mechanism for the approval of minor modifications to an ethical clearance (this includes changes to the research team, subject pool, testing instruments, etc). In practice this mechanism enables researchers to conduct a number of projects under the same ethical clearance. Any proposed modification to the project or variation to the ethical clearance must be reported immediately to the Committee (via the Research Ethics Officer), and cannot be implemented until the Chief Investigator has been notified of the Committeeï¿½s approval for the change / variation. Requests for changes / variations should be made in writing to the Research Ethics Officer. Minor changes (changes to the subject pool, the use of an additional instrument, etc) will be assessed on a case by case basis and interim approval may be granted subject to ratification at the subsequent meeting of the Committee. It generally takes 5 -10 days to process and notify the Chief Investigator of the outcome of a request for a minor change / variation. Major changes to your project must also be made in writing and will be considered by the UHREC. Depending upon the nature of your request, you may be asked to submit a new application form for your project. Audits All active ethical clearances are subject to random audit by the UHREC, which will include the review of the signed consent forms for participants, whether any modifications / variations to the project have been approved, and the data storage arrangements. Wendy Heffernan Research Ethics Officer Office of Research O Block Podium Tel: 07 3864 2340 Fax: 07 3864 1304

lxiv

III.2. Cooking Skills Study questionnaire

lxv

lxvi

lxvii

lxviii

lxix

lxx

III.3. Invitation letter to potential participants

lxxi

III.4. Invitation letter for repeatability sub-study

lxxii

III.5. Thank-you cards

lxxiii

IV. Appendix IV: Questionnaire development, validity and reliability

A summary of the questionnaire development is provided in Chapter 3. More detail

about the process, and more detailed results are reported here.

IV.1. Methods

IV.1.1. Face validity To gauge and maximise face validity of the new constructs in the questionnaire, two

different draft versions of the questionnaire were sent to members of the community

who were recruited through personal contacts. Participants were chosen to have

disparate salient demographic characteristics (age, gender, single/cohabiting,

with/without dependent children). These people completed the questionnaire, and

gave written feedback on which version they found better, and any difficulties they

had in completing the questionnaire. In addition, some respondents, when possible,

were asked to describe what they thought the confidence questions were asking them

(in their own words).

IV.1.1.1. Content validity

To ensure content validity, the draft questionnaire was passed for comment to

relevant ‘experts’, two of whom agreed to comment on the questionnaire. One is a

Professor whose previous work includes measurement of cooking skills in the United

Kingdom. The other is a Dietician who has worked in programs teaching cooking

skills including among low income groups. A panel of peers (fellow PhD students)

was also asked for feedback in a Measurement Development Forum held within the

Centre for Health Research, Public Health.

IV.1.1.2. Repeatability: test-retest

To assess the repeatability of the questionnaire, a sub-sample of 85 respondents were

chosen at random to complete a second questionnaire two weeks after initial delivery

of the first questionnaire. This time delay was chosen to minimise recall of previous

lxxiv

responses, and real change in respondents’ typical purchasing patterns and

confidence levels. Participants were given a gratuity (a $1 scrath-it) whether they

completed the re-test questionnaire or not. Fifty-eight (68%) of households

completed the questionnaire. Two surveys were excluded, as the same person did

not complete the questionnaire (according to mis-match on key demographic

characteristics including gender and age). A substantial number of surveys had at

least one item missing, and complete data on all variables were obtained for only 26

participants. To avoid loss of power analyses were conducted using all surveys with

relevant data.

Most statistical analyses were performed in SPSS 13.0. Test-retest reliability was

assessed using kappas for categorical variables and intraclass correlation coefficients

with Bland-Altman plots for continuous variables. Kappas calculated for ordinal

variables were weighted for the level of disagreement using Allison – Cichetti based

weights (319) in SAS 8.1. Where missing levels on one questionnaire required

calculation of weighted kappa by the method of Armitage and Berry (p446)(320).

Intra-class correlations were calculated using a two-way mixed effects model, with

participants treated as random effects, and test/ retest treated as a fixed effect (321).

Agreement was assessed according to the levels described by Landis and Koch (0-0.2

poor, 0.2-0.4 fair, 0.4-0.6 moderate, 0.6-0.8 substantial, and 0.8-<1.0 almost perfect)

(317)

IV.1.1.3. Internal consistency

The vegetable purchasing measure was treated as an index rather than a scale,

therefore internal consistency was not expected or required. Participants’ responses

items composing the vegetable confidence and technique confidence scales were

checked for internal consistency as indicated by Cronbach’s alpha. Calculations

were performed in SPSS 13.0. Adequacy was assessed using the rating system

described by George and Mallery (316).

lxxv

IV.2. Results

IV.2.1. Face validity Respondents’ feedback is presented in Table A4.1. Comments on specific questions

and the subsequent changes made to the questionnaire are presented in Table A4.2.

The majority preferred the Likert-Style format to the 0-100% format, and

importantly, none of those who preferred the numeric format had difficulty with the

Likert- format. The preference for verbal rather than numeric labelling fits with

general advice in psychometric survey research (423) Some of the comments raised

issues concerning the face validity of the purchasing measurement, as respondents

pointed out that purchasing does not equate to consumption and purchasing in bulk

means a low frequency of purchase but not necessarily a low frequency of

consumption. Since the index aims to capture purchasing (not consumption), and

this is what respondents were answering, this issue does not affect the face validity of

the measure.

Participants own words generally matched the intended meaning of the questions,

and they clearly communicated that ‘confidence’ related to whether they thought

they could do whatever was being asked about, and the scaling referred to how

certain or uncertain were that they could do it.

lxxvi

Table A4.1: Respondents’ comments on the survey Comment Who Their demographic

characteristics Numeric vs Likert

“probably too many options found the numbering system too broad at first. Why not 1 to 10. “

R6 Female ?age, born in Australia, cohabiting,

I feel that this survey (Likert-style survey)was more easier to understand, and more quicker to get through, so I would say that this one is better than the other (Likert survey)

R30 32 y F, born Australia, single, primary school education).

Question 4 – needed a box between confident and a little confident. Preferred the wording over choosing between 0-100

R6 Female ?age, born in Australia, cohabiting,

prefer the not numbered one R46 Female 17 y, born in Australia, lives with parents, junior high school education

I would prefer the scale of the first from two filled in. Just briefly I did not consider the 0-100 scale of any help

0047 –Female ? age, single, no children, primary school education

I prefer v2 [the numeric scale]. R48 F 60y, Born Australia, TAFE education, cohabiting, no children

I prefer this one (numeric) R45 Male 56 y, cohabiting, with children, TAFE education

Prefer format in version 2 [Likert-type]– easier for respondent to tick a box rather than deriving a %. Don’t have to read scale first.

R36 M 36y, Single, no children University education)

Indicators of comprehension and easy of completion Comment on seasonality therefore purchasing was hard to answer. Wanted to know the difference between boiling and poaching.

(0033 – 62 y F born overseas, cohabiting, no children, TAFE education).

Rate the survey at 80%. Flowed nicely, easy to do R31 28y F, born Australia, cohabiting with children, senior high school

I find it easier to cook Asian food than Australian food R49 48y female, born overseas, NESB, University education, cohabiting with children

‘Buy’ does not give automatic indicator of frequency of ‘use’. eg grow, barter, be given, buy in bulk.

R48 F 60y, Born Australia, TAFE education, cohabiting, no children

Some of questions tricky or a bit deceptive eg potatoes onions etc buy may buy in bulk – so you don’t buy often but eat often? so in Qs it should differentiate between buying often and eating often. I barter a lot especially with vegetables so a lot of vegetables I eat but don’t buy – some I buy and give to others.

Misccellaneous “Age question? – why not 20-30 or 30-3. Choosing an age group rather than just putting one’s age down.”

R6 Female ?age, born in Australia, cohabiting, with children, senior high school education).

Might want to include in cover letter or front page of this survey a disclaimer that this info used will not be relayed or utilized for marketing or sales in the future.

R36 Male, 36y, Single, no children University education)

lxxvii

IV.2.2. Content validity

Text boxes A4.1 and A4.2 present the comments from experts and text box

A4.3 presents the issues raised by peers at the measurements forum. There was

a fair degree of overlap between issues raised by participants and by experts.

No concern was expressed by experts over whether the questionnaire was

adequately measuring the constructs contained in it, but a fair level of caution

was expressed as to the extent which the constructs measured in the

questionnaire may or may not relate to other commonly used constructs in this

field of research.

Table A4.2: Comments on specific questions and refinements

Comment Q Solution Income before or after tax Q13 Now specified before tax Difference between poaching and boiling? Q6 Knowldedge? Overlapping categories (Q12 2-6 yrs and 6-12 yrs) Typographical error

fixed - now reads 2-5 yrs, 6-12 yrs

Weekends, dinnerparties vs weeknights? Q1 Not focus of survey how do you incorporate people who grow their own veges?”

Q2 Validity concern for purchasing scale

“This [prelude] is good as it clarifies – does washing constitute preparing (eg a celery stick)?”

Q3 Validity concern– participants interpretations of food preparation

“Are you a member of ANY OF the following groups?”

Q10 Changed

“May be misconstrued if educated outside qld (primary school)”

Q11 Coding scheme for other category

“Income question: disclaimer why this measured; suggest phrasing as broad income categories”

Q13 “These are a few very important questions about yourself and your household that will be used to make sure we’re considering people from all walks of life, and household circumstances”

lxxviii

Text Box A4.1: Email communication from expert panel – Expert 1 From: "Lang, Timothy" <[email protected]> To: "'Elisabeth Winkler'" <[email protected]> Cc: "'[email protected]'" <[email protected]> Subject: RE: measuring cooking skills Date: Tue, 18 May 2004 15:56:27 +0100 MIME-Version: 1.0 X-Mailer: Internet Mail Service (5.5.2657.72) Content-Type: multipart/alternative; boundary="----_=_NextPart_001_01C43CE7.7E647592" X-MailScanner-Information: Please contact Computing Services for more information X-City-MailScanner: Found to be clean X-City-MailScanner-SpamCheck: not spam (whitelisted), SpamAssassin (score=0.8, required 5, HTML_20_30 1.16, HTML_MESSAGE 0.10, ORIGINAL_MESSAGE -0.50) X-Junkmail-Status: score=20/50, host=mail-router01.qut.edu.au Dear Elisabeth, thanks for this. It looks good. My colleague Martin Caraher and I are really interested in this. But it is a minefield, as we know. We went down this route 10 years ago and now we have our PhD students doing close studies, eg observing and spending time with people. Tapping into the Depth and complexities of the skills-eating relationship is what you need to explore rather than just overt attitudes. Inevitably there are limitations from such an approach. For a PhD you will need to explore those and cover yourself. For instance, our concerns here would be things like: - how can you validate this? ie check that it brings out differences between those who really are lacking in confidence or not - it assumes attitudes drive behaviour. This may not be the case. - how truthful are people being? Can you 'triangulate' with known behaviour in any way? But we are very keen for you to do this. Did you see our various articles from our attitudinal research in the mide 1990s? I have forwarded yr email on to Martin: [email protected] Do contact him, too. I am copying this reply to him, as well. best wishes and good luck with this. good stuff. Lots of enthusiasm from us. tim lang see the weblink for the new ATLAS OF FOOD, eds Erik Millstone & Tim Lang http://www.earthscan.co.uk/asp/bookdetails.asp?key=3848 Tim Lang PhD FFPH Professor of Food Policy Dept Health Management & Food Policy Institute of Health Sciences City University Northampton Square London EC1V 0HB UK email: [email protected] our general website: www.city.ac.uk/ihs/hmfp/foodpolicy MSc in Food Policy: www.city.ac.uk/ihs/hmfp/foodpolicy/msc

lxxix

Text Box A4.2: Email communication from expert panel – Expert 2 X-Mailer: Novell GroupWise Internet Agent 5.5.6.1 Date: Fri, 21 May 2004 08:55:44 +1000 From: "Barbara Radcliffe" <[email protected]> To: <[email protected]> Subject: Re: measuring cooking skills/confidence Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="=_D8F9386A.395A353B" X-Junkmail-Status: score=37/50, host=mail-router01.qut.edu.au Hi Elisabeth The questionnaire is looking good and well formatted. Have just made a couple of suggested changes (attached). I also have a few other comments: 1. What about people who grow their own veges or have them given to them by families/neighbours? You could ask this as an additional question or could you broaden your question on purchase of veges to include these? OR do you want to find out how they access veges ie internet, food co-op, F&V shop, supermarket, weekend markets, their garden, friends/neighbours (or a combination) - and are they organic? I think this could provide you with some interesting associations. 2. Are you interested in the participant's role in the household eg mother, father, other adult guardian, etc? 3. I was wondering also if you want to know how many times per week they eat vegetables with their meal at night (ie is confidence associated with frequency of intake?) AND how many nights a week a meal is prepared at home. 4. do you have a separate confidentiality statement or should it be at the beginning of the questionnaire along with the purpose of the survey? 5. I was also wondering if there are too many income groups. It may make people suspicious that you have links with the tax or welfare departments. You will probably have to merge them anyway if you want to look for significant associations. 6. Finally in question 3 you ask about whether they are the main person preparing food by this is also on the title page. Is this just a cross check? Hope these comments are useful. If you have any questions, just give me a call. Good luck with your studies Barbara Barbara Radcliffe Community Nutrition Unit QEII Hospital Health Service District Annerley Road Community Health Service PO Box 3077 South Brisbane BC Qld 4101 phone: (07) 3010 3550 fax: (07) 3010 3552

lxxx

Expert 1 offered the opinion that the questionnaire looked adequate however it is

uncertain to what extent any psychometric measure would correlate with actual skill

level, or behaviours. This is a major issue, particularly if this questionnaire is to be

used purposes other than that of this thesis which is to determine whether confidence

(not actual skill) relates to purchasing behaviours. The comments of Expert 2

showed a few examples of how purchasing may not equate to intake. From the

measurements forum, it was suggested that the term food preparation may carry

different meanings for participants which may undermine the validity of their

responses. To avoid this concern a simple definition was added to the survey:

“For this survey, ‘preparing’ food means anything you might do to make the food

suitable to eat (for example, make a salad from it)”.

One validity issue that was not raised by any of the panel that is also worthy of

consideration is the lack of a definable time period on the purchasing measures.

Dietary patterns are not fixed, and though it questions respondent’s usual purchase,

dietary responses are often biased toward current behaviour. Furthermore the

categories were qualitative between always and never and did not refer to fixed

amounts, or frequencies. Thus the measure should probably be best interpreted as

respondents’ reported perceptions of their current typical purchasing patterns.

Overall from the face and content validity, there are no obvious concerns with how

the questionnaire measures self-reported purchasing and confidence levels of

participants, however there are numerous concerns with the validity of trying to

equate these measures to dietary intake or skill levels.

Text Box A4.3: Comments from peers at measurements forum • possibility of respondents having various interpretations of the term food

preparation o SOLUTION: definition added to the survey

• short questionnaire o length determined by feasibility constraints and concern for

respondent burden • readability is good • suggestion to compare against gold standard

o SOLUTION: no validated measures known to be available • suggestion to compare against actual behaviour

o cannot be performed in this thesis due to feasibility constraints. However this warrants investigation for future endeavours, especially as confidence and performance are expected to relate, but not perfectly.

lxxxi

IV.2.3. Internal consistency

Reported in Chapter 3.

IV.2.4. Test-retest reliability

IV.2.4.1. Representativeness and Missing Data

Table A4.3 presents a comparison between survey participants and the reliability

sub-study participants. Participants who responded to the second questionnaire were

largely representative of the respondents to the original survey; however there was a

slight overrepresentation of females, people with less education and on lower

incomes, and older respondents.

Table A4.3: Characteristics of people in repeatability study compared with total

study respondents

Repeatability sample

Entire sample

Gender (male) 9 (16.07%) n=55 100 (23.4%) Age 53 (18-89) n=54 46 (16-94) n=419 Adults 2.00 (1 – 6) n=52 2.00 (1-6) n=412 Teenagers 0.00 (0-3) n=52 0.00 (0-3) n=412 Children 0.00 (0 – 2) n=52 0.00 (0-3) n=412 Young Children 0.00 (0-1) n=52 0.00 (0-2) n=412 Infants 0.00 (0-1) n=52 0.00 (0-2) n=412 Total Household 2.00 (1-6) n=52 2.00 (1 – 8) n=412 Purchasing Index 73.00 (34.00-102)n=45 70 (28-105) n=349 Vegetable Confidence Index

118 (84-126) n=44 117 (21 – 126) n=362

Technique Confidence Index

53 (24 – 60.00) n=43 53 (14 - 60) n=360

ATSI (yes) 1 (1.89%) n=53 6 (1.5%) ASthSI (yes 0 (0%) n=52 1 (0.2%) NESB (yes) 3 (5.77%) n=52 18 (4.4%) Disability (yes) 1 (1.92%) n=52 47 (11.5%) Post school education? (yes)

26 (49.06%) n=53 240 (57.6%)

Income prefer not to say bottom tertile middle tertile upper tertile

11 (20.37%) 16 (29.63%) 11 (20.37%) 14 (25.93%)

95 (23.2%) 100 (24.4%) 82 (20.0%) 133 (32.5%)

lxxxii

While there was a substantial degree of missing data, there were no obvious

differences between partial and full completers (Table A4.4). Compared to

completers, partial completers have similar numbers of household members of all

age-groups, have similar scores on the purchasing scores and confidence indices, are

equally likely to be male, from non-English speaking backgrounds, Aboriginal or

Torres Strait Islanders, Aboriginal or South Sea Islanders. However they tended to

be older (median 66.50 vs 45 years), were less likely to report having post-school

qualifications (34.62% vs 62.96%), were more likely to ‘prefer not to say’ what their

household income is (33.33% vs 10.00%), were less likely to report high incomes

(8.33 vs 30.00%) but were equally likely to report low incomes.

Table A4.4: Characteristics of participants with complete and incomplete data

Incomplete data

Completers (n=26)

Gender 4 (13.33%) 5 (19.23%) Age 66.5 (19 – 89) n=28 45 (18 – 76) Adults 1.50 (1 – 6) n=26 2.00 (1 – 4) Teenagers 0.00 (0 – 1) n=26 0.00 (0 – 3) Children 0.00 (0 – 2) n=26 0.00 (0 – 2) Young Children 0.00 (0 – 1) n=26 0.00 (0 – 1) Infants 0.00 (0 - 0) n=26 0.00 (0 – 1) Total Household 2.00 (1 – 6) n=26 2.00 (1 – 6) Purchasing Index 72 (34-90) 73.5 (41-102) Vegetable Confidence Index 119 (96-126) 118 (84-126) Technique Confidence Index

53 (24-60) 53.5 (39-60)

ATSI (yes) 1 (3.70%) n=27 0 (0%) ASthSI (yes 0 (0%) n=26 0 (0%) NESB (yes) 2 (7.69%) n=26 1 (3.85%) Disability 1 (3.85%) n=26 0 (0%) Post school education? (yes) 9 (34.62%) n=26 17 (62.96%) Income

prefer not to say bottom tertile middle tertile upper tertile

8 (33.33%) 10 (41.67%) 2 (8.33%) 2 (8.33%)

3 (10.00%) 6 (40.00%) 9 (20.00%) 12 (30.00%)

lxxxiii

The commonality across the missing items may be age. Older persons tend to be on

lower incomes and are less likely to have post-school qualifications, particularly as

the sample is predominantly female. Importantly, the ranges indicate that

respondents across the full spectrum of age ranges and purchasing scores did

complete the survey. According to data entry, much of the missing data came from

older respondents ticking more than one box on the same line and skipping the next

on lists, indicating that visual acuity may account for much of the missing data. This

is important because as opposed to an unwillingness to answer certain items, this

type of missing data is unlikely to bias estimates.

IV.2.4.2. Categorical data

Chapter three presents the kappa statistics in tabular form and discusses the results

for the confidence and purchasing items. Demographic characteristics all showed

almost perfect agreement (317) between tested and retested measures, except where

agreement could not be tested due to insufficient variation in responses.

IV.2.4.3. Continuous data

Chapter 3 presents the tabular results for all ICCs and discusses results for the

purchasing index and confidence scales are presented in Chapter 3. Mean

differences (test – retest) and their limits of agreement shown on the Bland-Altman

plots (Figures to ) indicate that the difference between values of interval/ continuous

variables reported at testing and retesting were very similar for the age (mean

difference, limits of agreement =0.00(-0.67 to 0.67)), the number of household

teenagers (0.02 (-0.27 to 0.31), children (-0.05 (-0.64 to 0.56)), young children (0.02

(0.46 to 0.52)), infants (0.02 (-0.31 to 0.27) and overall number of household

members (-0.08 (-0.77 to 0.61)). The largest discrepancy was that the number of

adults in the household was reported as being slightly greater in the retest

questionnaire (mean difference (test- retest) = -0.61 (-0.55 to 0.42)). Generally, the

points of difference were symmetrically distributed above and below the line, and to

the left and right of the graphs, indicating that the differences between tested and

retested values did not appear to be biased towards the either questionnaire, or occur

to a greater extent for lesser or greater values of the measured variables. There was a

lxxxiv

slight tendency for differences to be negative, indicating higher values on the retested

questionnaire, but the scarcity of discrepant values means this does not show a

definitive bias.

Variation between tested and retested values was small compared with overall

variation for these measures which were collected continuously. Age, number of

household members who were adults, teenagers and children all had excellent

ICC>0.9. The number of young children and infants in the household had substantial

repeatability (ICCs 0.65 and 0.79 at the individual level). The total number of

household members still had excellent repeatability (ICC>0.9), as the average

number of infants and young children was low.

Figure A4.1. Bland Altman Plot of Test – Retest Age

lxxxv

Figure A4.2. Bland Altman Plot of Test – Retest Number of Adults in Household

Figure A4.3. Bland Altman Plot of Test – Retest Number of children (aged 6-12) in

Household

lxxxvi

Figure A4.4. Bland Altman Plot of Test – Retest Number of Young Children (aged

2-5) in Household

Figure A4.5. Bland Altman Plot of Test – Retest Number of Infants (aged 0 to <2) in

Household

lxxxvii

Figure A4.6. Bland Altman Plot of Test – Retest Number of Persons in

Household

Figure A4.7. Bland Altman Plot of Test – Retest Vegetable Purchasing Index

lxxxviii

Figure A4.8. Bland Altman Plot of Test – Retest Confidence to Cook Vegetables

Scale

Figure A4.9. Bland Altman Plot of Test – Retest Confidence to Using Cooking

Techniques Scale

lxxxix

V. Appendix V: Other relevant data

V.1. Maps of Brisbane and other capital cities

Figure A5.1a: Distribution of study catchments, supermarkets and greengrocers in

the study areas around Brisbane

Figure A5.1b: Distribution of study areas and shops in Brisbane: distribution of

convenience stores and study catchments around Brisbane

xc

Figure A5.2a: Distribution of Index of Relative Socioeconomic Disadvantage (IRSD)

(deciles) in Brisbane Statistical Subdivision (SSD), 199610

10 Higher numbers (darker colours) indicate less socioeconomic disadvantage

Legend: IRSD

xci

Figure A5.2a: Distribution of IRSD (deciles) in Brisbane SSD, 2000

Legend: IRSD

xcii

V.2. Spatial patterning of socioeconomic disadvantage (IRSD) across census collection districts in the Statistical Subdivisions of Australian Capital Cities in 2000

Figure A5.3: Distribution of IRSD (deciles) in Darwin SSD, 2000

Legend: IRSD

xciii

Figure A5.4: Distribution of IRSD (deciles) in Hobart SSD, 2000

Legend: IRSD

xciv

Figure A5.5: Distribution of IRSD (deciles) in Inner Melbourne SSD, 2000

Legend: IRSD

xcv

Figure A5.6a: Distribution of IRSD (deciles) in North Canberra SSD, 2000

Figure A5.6b: Distribution of IRSD in South Canberra SSD, 2000

Legend: IRSD

Legend: IRSD

xcvi

Figure A5.7a: Distribution of IRSD in Western Adelaide SSD, 2000

Legend: IRSD

xcvii

Figure A5.7b: Distribution of IRSD (deciles) in Eastern Adelaide SSD, 2000

Legend: IRSD

xcviii

Figure A5.7c: Distribution of IRSD (deciles) in Northern Adelaide SSD, 2000

Legend: IRSD

xcix

Figure A5.8: Distribution of IRSD (declies) in Perth (Central Metropolitan WA

SSD), 2000

Legend: IRSD

c

Figure A5.9a: Distribution of IRSD (deciles) in Inner Sydney SSD, 2000

Legend: IRSD

ci

Figure A5.9b: Distribution of IRSD (deciles) in Inner Western Sydney SSD, 2000

Legend: IRSD

cii

Figure A5.9c: Distribution of IRSD (delices) in Central Northern Sydney SSD, 2000

Legend: IRSD

ciii

Figure A5.9d: Distribution of IRSD in Central Western Sydney SSD, 2000

Legend: IRSD

civ

Figure A5.9e: Distribution of IRSD (deciles) in Lower Northern Sydney SSD, 2000

Legend: IRSD

cv

V.3. Evidence of change in food prices from the Consumer Price Index (CPI)

Table A5.1: Change (%) in prices of food, fruits and vegetables in the CPI from

corresponding quarter of previous year in Brisbane 2000-01 to 2006-07

All Food Fruit & Vegetables Fruit Vegetables Mar-01 7.5 % 26.3 % 23.9 % 28 % Jun-01 7.3 % 11.4 % 18.2 % 6.5 % Sep-01 6.3 % 12.3 % 19.1 % 6.4 % Dec-01 9.2 % 22.6 % 27.5 % 17.9 %

Mean for 2001 7.6 % 18.2 % 22.2 % 14.7 %Mar-02 6 % 0.1 % 9.1 % -7.3 % Jun-02 4.8 % 0.7 % 13.3 % -9.4 % Sep-02 4.6 % 6.8 % 0.7 % 12.6 % Dec-02 2.1 % -3.3 % -15.9 % 9.9 %

Mean for 2002 4.4 % 1.1 % 1.8 % 1.5 %Mar-03 4.5 % 10.1 % -9.4 % 29.4 % Jun-03 4.1 % 9 % -8.6 % 26.9 % Sep-03 3 % -3 % -5 % -1.3 % Dec-03 4 % 7 % 8.6 % 5.7 %

Mean for 2003 3.9 % 5.8 % -3.6 % 15.2 %Mar-04 2.5 % 5.4 % 9.4 % 2.6 % Jun-04 1.5 % 2.6 % 10.2 % -3 % Sep-04 1.3 % 1.7 % 7.6 % -3.2 % Dec-04 1.6 % 0.7 % 11.9 % -8.5 %

Mean for 2004 1.7 % 2.6 % 9.8 % -3.0 %Mar-05 1.2 % -7.8 % 1.4 % -14.6 % Jun-05 2.7 % -2.6 % -4.9 % -0.7 % Sep-05 3.1 % 1.4 % -2.1 % 4.2 % Dec-05 3.4 % 1.8 % -8.6 % 12 %

Mean for 2005 2.6 % -1.8 % -3.6 % 0.2 %Mar-06 4.6 % 9 % -3.3 % 20.2 % Jun-06 8.7 % 38.7 % 65.5 % 15.9 % Sep-06 11.4 % 58 % 103.8 % 14.8 % Dec-06 9.5 % 41.8 % 82.7 % 6.1 %

Mean for 2006 8.6 % 36.9 % 62.2 % 14.3 %Mar-07 4.1 % 3.1 % 14.5 % -5.6 % Jun-07 2.2 % -7 % -18.2 % 6.7 % Sep-07 1.8 % -7.3 % -26.1 % 24 %

Mean for 2007 2.7 % -3.7 % -9.9 % 8.4 %Average Change

2000-2007 +4.50 % +8.44 % +11.27 % +7.33 %Source: Australian Bureau of Statistics (2007) 6401.0 Table 14. CPI: Groups, Sub-groups and Expenditure Class, Percentage change from corresponding quarter of previous year by Capital City. Bold figures are calculated mean of changes in each quarter (March, June, September and December) from the same quarter of the corresponding year

cvi

VI. Appendix VI: Tests for spatial autocorrelation

An estimation of the degree of spatial autocorrelation in the ecological analysis

(Study 1, Chapter 4) was made using semivariograms (403) of the model residuals

against the distances between each area. Experimental semivariograms were

produced in MapInfo, which provide best-fit estimates of the relationship between

semivariance (half the average squared difference between residuals) and the

distances between each area. Distances were examined in two directions (North-

South, and East-West). Figures A6.1 and A6.2 present the semivariograms based on

the residuals from the Poisson models examining the density of supermarkets and

greengrocers, and convenience stores, respectively.

A small degree of spatial autocorrelation was present, as the semivariance (half the

average squared difference between residuals) was low for particularly for areas very

close together (less than approximately 2 km), and higher for areas further apart.

Based on the semivariograms, there may have been a small to moderate degree of

spatial autocorrelation present, however, the semivariogram provides little evidence

for a smooth spatial process occurring. For distances further than 2 kilometres, the

spread of semivariance around the line of best fit was large, irrespective of distance.

The increase in semivariance for areas 20km apart compared with 2 km apart was

small relative to the maximum semivariance observed in the data (approximately

10% for supermarkets and greengrocers and approximately 25% for convenience

stores). Thus the likely impact of spatial autocorrelation was addressed in the first

two studies, and a multilevel analysis was used for study 3 without the more complex

modifications of a fully spatial approach.

cvii

Figure A6.1: Semivariogram of residuals from Poisson model of number of supermarkets and greengrocers by area socioeconomic disadvantage

Sem

ivar

ianc

e (A

vera

ge sq

uare

d di

ffer

ence

bet

wee

n re

sidu

als /

2)

Distance between paired observations (metres)

cviii

Figure A6.2: Semivariogram of residuals from Poisson model of number of convenience stores by area socioeconomic disadvantage.

Sem

ivar

ianc

e (A

vera

ge sq

uare

d di

ffer

ence

bet

wee

n re

sidu

als /

2)

Distance between paired observations (metres)

cix

VII. Appendix VII: Checking assumptions of a priori sample size calculations.

The cooking skills study did not recruit sufficient participants to meet the sample size

needed based on a priori calculations. However, the a priori calculations

overestimated the required sample size as the actual spread of the outcome variable

and the degree of clustering were less than what was anticipated from the literature

(standard deviation 12.29 vs 15.2 and ICC 0.005 vs 0.009). Based on a posteriori

calculations (below), only 315 participants were needed to detect a minimum

difference of 6.3 points on the purchasing scale. Accordingly, the study may not

have underpowered as would have been expected based on a priori assumptions

(statistical models based on n=401 participants).

Figure A7.1: Sample size requirements: actual and a priori estimations Sample size calculation Where ‘I’ is 2.8 for two-tailed α =0.05 and β=0.2, ‘s’ is the population standard error, and ‘diff’ is the minimum difference of interest =6.3, actual s=12.29. Number per group = (2 * I2 *s2)/ diff2

= (2 * 2.82 * 12.292)/6.32 = 59.6717 (+ statistical adjustments (15%)) (& measurement error (20%)) = 80.5568 (x 1.3 for cluster sampling (Design Effect)) = 104.724 Overall number (for 3 groups) = 314.172 Design effect Where ICC is intra-cluster correlation and k is number of individuals in each cluster (average) Actual design effect - ICC: = cluster variance / total variance

= 0.722/150.354 = 0.004802

(Cluster variance and total variance estimated from VARCOMP in SPSS version 16.0 using type 3 sums of squares.) Design Effect (DEFF) = 1 + (ICC* (k-1)) = 1 + 0.004802*(66.83-1) ≈ 1.3

cx

VIII. References 1. Kirkwood BR, Sterne JAC. Essential medical statistics. Oxford: Blackwell Science; 2003. 2. Carlin JB, Hocking J. Design of cross-sectional surveys using cluster sampling: An overview with Australian case studies. Australian and New Zealand Journal of Public Health 1999;23(5):546-551. 3. Australian Institute of Health and Welfare. Australia's health 2006. Canberra: Australian Institute of Health and Welfare; 2006. 4. Mathers C, Vos T, Stevenson C. The burden of disease and injury in Australia. In: Australian Institute of Health and Welfare; 1999. 5. Magarey A, McKean S, Daniels L. Evaluation of fruit and vegetable intakes of Australian adults: the National Nutrition Survey 1995. Australian and New Zealand Journal of Public Health 2006;30(1):32-37. 6. Australian Institute of Health and Welfare. Australia's health 2004. Canberra: Australian Institute of Health and Welfare; 2004. 7. Casagrande SS, Wang Y, Anderson C, Gary TL. Have Americans increased their fruit and vegetable intake? The trends between 1988 and 2002. American Journal of Preventive Medicine 2007;32(4):257-263. 8. Hoare J, Henderson L, Bates CJ, Prentice A, Birch M, Swan G, et al. The National Diet & Nutrition Survey: adults aged 19 to 64 years, volume 5 2004. London: Food Standards Agency, Departments of Health, Office for National Statistics and Medical Research Council Human Nutrition Research; 2004. 9. Pradhan A. Obesity, metabolic syndrome, and type 2 diabetes: inflammatory basis of glucose metabolic disorders. Nutrition Reviews 2007;62(12):s152-156. 10. Lobstein T, Millstone E. Context for the PorGrow study: Europe's obesity crisis. Obesity Reviews 2007;8(s2):7-16. 11. World Health Organisation. The world health report 2002 - Reducing Risks, Promoting Healthy Life. Geneva: World Health Organisation,; 2002. 12. Baghurst P, Beaumont-Smith N, Baghurst K, Cox D. The relationship between the consumption of fruits and vegetables and health status: Report to Department of Health and Aged Care and the Strategic Intergovernmental Nutrition Alliance. Campbell, Australia: CSIRO Health Sciences & Nutrition; 1999. 13. Fiscella K, Franks P. Individual income, income inequality, health, and mortality: What are the relationships? Health Services Research 2000;35(1):307-318. 14. Hayward MD, Crimmins EM, Miles TP, Yang Y. The significance of socioeconomic status in explaining the racial gap in chronic health conditions. American Sociological Review 2000;65(6):910-930. 15. Berkman L, Lochner K. Social determinants of health: meeting at the crossroads. Health Affairs 2002;21(2):291-293. 16. Adler N. Socioeconomic disparities in health: Pathways and policies. Health Affairs 2002;21(2):60-76. 17. Chandola T, Bartley M, Wiggins R, Schofield P. Social inequalities in health by individual and household measures of social position in a cohort of healthy people. (Theory and Methods). Journal of Epidemiology and Community Health 2003;57(1):56-63.

cxi

18. Draper G, Turrell G, Oldenburg B. Health inequalities in Australia: mortality. Canberra: Quuensland University of Technology & The Australian Institute of Health and Welfare; 2004. Report No.: AIHW Cat No. PHE55. 19. Siskind V, Najman J, Veitch C. Socioeconomic status and mortality revisited: an extension of the Brisbane area analysis. Australian Journal of Public Health 1992;16(3):315-320. 20. Wilkinson R, Marmot M. The solid facts. London: World Health Organisation and International Centre for Health and Society; 1998 1998. 21. James WPT, Nelson M, Ralph A, Leather S. The contribution of nutrition to inequalities in health (Socioeconomic determinants of health, part 8). British Medical Journal 1997;314(7093):1545-1549. 22. Van Rossum CT, van de Mheen H, Breteler MMB, al. e. Socioeconomic differences in stroke among Dutch elderly women: the Rotterdam study. Stroke 1999;30:357-362. 23. Maheswaran R, Elliot P, Strachan DP. Socioeconomic deprivation, ethnicity and stroke mortality in Greater London and south east England. Journal of Epidemiology and Community Health 1997;51:127-131. 24. Kunst AE, del Rios M, Groenhof F. Socioeconomic inequalities in stroke mortality among middle-aged men: an international overview, European Union Working Group on Socioeconomic Inequalities in Health Stroke. Stroke 1998;29:2285-2291. 25. Bennett S. Socioeconomic inequalities in coronary heart disease and stroke mortality among Australian men, 1979-1993. International Journal of Epidemiology 1996;25:266-275. 26. Davey Smith G, Hart C, Blane D, Hole D. Adverse socioeconomic conditions in childhood and cause specific adult mortality: prospective observational study. British Medical Journal 1998;316(7145):1631-1635. 27. Wannamethee SG, Whincup PH, Shaper G, Walker M. Influence of fathers' social class on cardiovascular disease in middle-aged men. The Lancet 1996;348(9037):1259-1263. 28. Paeratakul S, Lovejoy JC, Ryan DH, Bray GA. The relation of gender, race and socioeconomic status to obesity and obesity comorbidities in a sample of US adults. International Journal of Obesity 2002;26(9):1205-1210. 29. O'Dea JA, Caputi P. Association between socioeconomic status, weight, age and gender, and the body image and weight control practices of 6-to 19-year-old children and adolescents. Health Education Research 2001;16(5):521-532. 30. Lahmann PH, Lissner L, Gullberg B, Berglund G. Sociodemographic factors associated with long term weight gain, current body fatness and central adiposity in Swedish women. International Journal of Obesity 2000;24(6):685-694. 31. Taylor R, Qunie S. Differences in life expectancy: social factors in mortality rates. Ausralian Journal on Ageing 1992;11(1):21-35. 32. House S. Understanding social factors and inequalities in health: 20th century progress and 21st century prospects. Journal of Health and Social Behaviour 2002;43(2):125-142. 33. De Irala-Estévez J, Groth M, Lohansson L, Oltersdorf U, R. P, Martínez-González MA. A systematic review of socio-economic differences in food habits in Europe: consumption of fruits and vegetables. European Journal of Clinical Nutrition 2000;54(9):706-714. 34. Kant A, Graubard BA. Secular Trends in the association of socio-economic position with self-reported dietary attributes and biomarkers in the US population:

cxii

National Health and Nutrition Examination Survey (NHANES) 1971 - 1975 to NHANES 1999-2002. Public Health Nutrition 2005;10(2):158-167. 35. Strategic Intergovernmental Nutrition Alliance. Eat Well Australia: a strategic framework for public health nutrition. In: Strategic Intergovernmental Nutrition Alliance; 2001. p. 1-37. 36. Giles-Corti B, Donovan RJ. The relative influence of individual, social and physical environment determinants of physical activity. Social Science and Medicine 2002;54(12):1793-1812. 37. Stokols D. Establishing and maintaining healthy environments: toward a social ecology of health promotion. American Psychologist 1992;41(1):6-22. 38. Hagdrup NA, Simoes EJ, Brownson RC. Fruit and vegetable consumption in Missouri: knowledge, barriers and benefits. American Journal of Health Behavior 1998;22(2):90-100. 39. Strecher V, Wang C, Derry H, Wildenhaus K, Johnson C. Tailored interventions for multiple risk behaviors. Health Education Research 2002;17(5):619–626. 40. Bruijn D, Gert-Jan, Kremers SPJ, De Vet E, De Nooijer J, Van Mechelen W, et al. Does habit strength moderate the intention-behaviour relationship in the Theory of Planned Behaviour? The case of fruit consumption. Psychology & Health 2007;22(8):899-916. 41. Bogers RP. Explaining fruit and vegetable consumption: the theory of planned behaviour and misconception of personal intake levels. Appetite 2004; 42(2):157-166. 42. Glanz K, Mullis RM. Environmental interventions to promote healthy eating: a review of models, programs, and evidence. Health Education Quarterly 1998;15(4):395-415. 43. Lechner L, Brug J. Consumption of fruit and vegetables: how to motivate the population to change their behavior. Cancer Letters 1997;114(1-2):335-336. 44. Baranowski T, Cullen KW, Baranowski J. Psychosocial correlates of dietary intake: Advancing dietary intervention. Annual Review of Nutrition 1999;19(1):17-40. 45. Short F. Domestic cooking skills: what are they? Journal of the HEIA 2003;10(3):13-22. 46. Lang C. Is there a culinary skills transition? Data and debate from th UK about changes in cooking culture. Journal of the Home Economics Institute of Australia 2001;8(2):2-14. 47. Winkler E, Turrell G, Patterson C. Does living in a disadvantaged area mean fewer opportunities to purchase fresh fruit and vegetables in the area? Findings from the Brisbane food study. Health & Place 2006;12(3):306-319. 48. Winkler E, Turrell G, Patterson C. Does living in a disadvantaged area entail limited opportunities to purchase fresh fruit and vegetables in terms of price, availability, and variety? Findings from the Brisbane Food Study. Health & Place 2006;12(4):741-748. 49. Commonwealth Department of Health and Ageing. National Health Priority Areas. Better Health Outcomes Newsletter 1999:3-5. 50. National Health and Medical Research Council. Dietary guidelines for Australian Adults: National Health and Medical Research Council; 2003 10/04/2003. 51. Kellett E, Smith A, Schmerlaib Y. The Australian guide to healthy eating: background information for consumers. In. Canberra: Public Health Division Commonwealth Department of Health and Family Services; 1998.

cxiii

52. SIGNAL. Eat Well: the benefits of vegetables and fruit. In: National Public Health Partnership; 2000. 53. Queensland Public Health Forum. Eat Well Queensland: 2002-2012; 2002. 54. SIGNAL. National action plan to increase consumption of vegetables and fruit. In: National Public Health Partnership; 2000. 55. World Health Organisation. Ottawa charter for health promotion. Copenhagen: World Health Organisation; 1986. 56. MacIntyre S, Ellaway A, Cummins S. Place effects on health: how can we conceptualise, operationalise and measure them? Social Science and Medicine 2002;55:125-139. 57. Nutrition – Health Development CASN, Nutrition – Health Development TESN, Services) NaPAUDoHaC. Market basket survey of remote community stores in the Northern Territory: April – June 2001. Casuarina, Northern Territory: Northern Territory Government Department of Health and Community Services; 2001. 58. Queensland Health. Healthy Food Access Basket (HFAB) survey: Queensland Health; 2004. 59. Queensland Health. Summary: the 2000 Healthy Food Access Basket: Queensland Health; 2000. 60. Lee AJ. Price and availability of healthy food: healthy food access basket surveys. Food Chain 2003:14-16. 61. Rissanen TH, Voutilainien S, Virtanen JK. Fruits, berries, vegetables and mortality. Nutrition Research Newsletter 2003;22(2):14-16. 62. Khaw K-T, Bingham S, Welch A, Luben R, Wareham N, Oakes S, et al. Relation between plasma ascorbic acid and mortality in men and women in EPIC-Norfolk prospective study. Lancet 2001;357(9257):567-664. 63. Gillman MW, Cupples LA. Protective effects of fruits and vegetables on the development of stroke in men. Journal of the American Medical Association 1995;273(14):1113-1118. 64. Bradley S, Shinton R. Why is there an association between eating fruit and vegetables and a lower risk of stroke? Journal of Human Nutrition and Dietetics 1998;11(5):10. 65. Ness AR, Powles JW. Fruit and vegetables, and cardiovascular disease: a review. International Journal of Epidemiology 1997;26(1):1-13. 66. Noakes m, Clifton P. The role of diet in cardiovascular health. Australian Journal of Nutrition and Dietetics 1999;56(s3):s3-s23. 67. Bradbury J. Lutein may help reduce atherosclerosis. Lancet 2001;357(9273):2030. 68. Martinez-Gonzalez MA, Fernandez-Jarne E, Serrano-Martinez M, Marti A, Martinez JA, Martin-Moreno JM. Mediterranean diet and reduction in the risk of a first acute myocardial infarction: an operational healthy dietary score. European Journal of Nutrition 2002;41(4):153-161. 69. John JH, Ziebland S, Yudkin P, Roe LS, Neil HAW. Effects of fruit and vegetable consumption on plasma antioxidant concentrations and blood pressure: a randomised controlled trial. Lancet 2002;359(9322):1969-1975. 70. Silalhahi J. Anticancer and health protective properties of citrus fruit components. Asia Pacific Journal of Clinical Nutrition 2002;11(1):79-85. 71. Choi S-W, Mason JB. Folate and carcinogenesis: an integrated scheme. Journal of Nutrition 2002;132(12):129-133.

cxiv

72. Weisburger JH. Prevention of cancer and other chronic diseases worldwide based on sound mechanisms. Biofactors 2000;12(1-4):73-82. 73. Levi F, Pasche C, La Vecchia C, Lucchini F, Franceschi S. Food groups and colorectal cancer risk. British Journal of Cancer 1999;97(7-8):1283-1288. 74. Potter JD. Vegetables, fruit, and cancer. The Lancet 2005;366:527-530. 75. Higdon JV, Delage B, Williams DE, Dashwood RH. Cruiciferous vegetables and human cancer risk: epidemiological evidence and mechanistic basis. Pharmacological Research 2007;55:224-236. 76. Temple NJ, Gladwin KK. Fruit, vegetables and the prevention of cancer: research challenges. Nutrition 2003;19(5):467-470. 77. Omenn GS. What accounts for the association of vegetables and fruits with lower incidence of cancers and coronary heart disease? Annals of Epidemiology 1995;5(4):333-335. 78. Russo GL. Ins and outs of dietary phytochemicals in cancer chemoprevention. Biochemical Pharmacology 2007;74:533-544. 79. Feeney M. Fruits and the prevention of lifestyle-related diseases. Clinical and Experimental Pharmacology and Physiology. 2004;31(s1):s11-s13. 80. Heinecke JW. Oxidants and antioxidants in the pathogensis of atherosclerosis: implications for the oxidised low density lipoprotein hypothesis. Atherosclerosis 1998;141(1):1-15. 81. Bazzano LA, Serdula MK, Liu S. Dietary intake of fruits and vegetables and risk of cardiovascular disease. Current Atherosclerosis Reports 2003;5(3):492-499. 82. Conlin PR, Chow D, Miller ER, Svetkey LP, Lin P-H, Harsha DW, et al. The effect of dietary patterns on blood pressure control in hypertensive patients: results from the Dietary Approaches to Stop Hypertension (DASH) trial. American Journal of Hypertension 2000;13:949-955. 83. John JH, Ziebland S, Yudkin P, Roe LS, Neil HAW. Effects of fruit and vegetable consumption on plasma antioxidant concentrations and blood pressure: a randomised controlled trial. Lancet 2002;359(9322):1969-1974. 84. Ganji V, Kafai MR. Frequent consumption of milk, yoghurt, cold breakfast cereals, peppers and cruciferous vegetables and intakes of dietary folate and riboflavin but not vitamins b-12 and b-6 are inversely associated with serum total homocysteine concentrations in the US population. American Journal of Clinical Nutrition 2004;80(6):1500-1507. 85. Brown AA, Hu FB. Dietary modulation of endothelial function: implications for cardiovascular disease. American Journal of Clinical Nutrition 2001;73:673-686. 86. Nayak DU, Karmen C, Frishman WH, Vakili BA. Antioxidant vitamins and enzymatic and synthetic oxygen– derived free radical scavengers in the prevention and treatment of cardiovascular disease. Heart Disease 2001;3(1):28-45. 87. Houston MC. Nutraceuticals, vitamins, antioxidants, and minerals in the prevention and treatment of hypertension. Progress in Cardiovascular Diseases 2005;47(6):396-449. 88. Ferruzzi MG, Blakeslee J. Digestion, absorption, and cancer preventative activity of dietary chlorophyll derivatives. Nutrition Research 2007;27(1):1-12. 89. Turrell G, Hewitt B, Patterson C, Oldenburg B, Gould T. Socioeconomic differences in food purchasing behaviour and suggested implications for diet-related health promotion. Journal of Human Nutrition and Dietetics 2002;15(5):355-364. 90. Woodward DR, Shaw KN, Rathborne MC, Cumming FJ, Ball PJ, Thomson A, et al. Socio-demographic correlates of the consumption of specific foods in a non-

cxv

metropolitan area of Tasmania. Australian Journal of Nutrition and Dietetics 1999;56(1):31-38. 91. Popkin BM, Zizza C, Siega-Riz AM. Who is leading the change? U.S. dietary quality comparison between 1965 and 1996. American Journal of Preventive Medicine 2003;25(1):1-8. 92. Trudeau E, Kristal AR, Li S, Patterson RE. Demographic and psychosocial predictors of fruit and vegetable. Journal of the Amercian Dietetic Association 1998;98(12):1412-1417. 93. Xie B, Gilliland FD, Li YF, Rockett HRH. Effects of ethnicity, family income, and education on dietary intake among adolescents. Preventive Medicine 2003;36(1):30-40. 94. Leiyu S. Sociodemographic characteristics and indvidual health behaviors. Southern Medical Journal 1998;91(10):933-941. 95. Kant A, Block G, Schatzkin A, Ziegler RG, Nestle M. Dietary diversity in the US population, NHANES 11, 1976-1980. Journal of the Amercian Dietetic Association 1991;91(12):1526-1531. 96. Shimakawa T, Sorlie P, Carpenter MA, Dennis B, Tell GS, Watson R, et al. Dietary intake patterns and sociodemographic factors in the atherosclerosis risk in communities study. Preventive Medicine 1994;23(6):769-780. 97. Galobardes B, Morabia A, Bernstein M. Diet and socioeconomic position: does the use of different indicators matter? International Journal of Epidemiology 2001;30:334-340. 98. Braddon FEM, Wadsworth MEJ, Davies JMC, Cripps HA. Social and regional differences in food and alcohol consumption and their measurement in a national birth cohort. Journal of Epidemiology and Community Health 1988;42:341-349. 99. Johansson L, Thelle DS, Solvoll K, Bjorneboe G-EA, Drevon CA. Healthy dietary habits in relation to social determinants and lifestyle factors. British Journal of Nutrition 1999;81(3):211-220. 100. Mishra G, Ball K, Arbuckle J, Crawford D. Dietary patterns of Australian adults and their association with socioeconomic status: results from the 1995 National Nutrition Survey. European Journal of Clinical Nutrition 2002;56(7):687-693. 101. Steele P, Dobson A, Alexander H, Russell A. Who eats what? A comparison of dietary patterns among men and women in different occupational groups. Australian Journal of Public Health 1991;15(4):286-295. 102. Thompson RL, Margetts BM, Speller VM, McVey D. The Health Education Authority's health and lifestyle survey 1993: who are the low fruit and vegetable consumers? Journal of Epidemiology and Community Health 1999;53:294-299. 103. Billson H, Pryer JA, Nichols R. Variation in fruit and vegetable consumption among adults in Britain. An analysis from the dietary and nutritional survey of British adults. European Journal of Clinical Nutrition 1999;53:946-952. 104. Auld GW, Bruhn CM, McNulty J, Bock MA, Gable K, Lauritzen G, et al. Reported adoption of dietary fat and fiber recommendations among consumers. Journal of the Amercian Dietetic Association 2000;100(1):52-58. 105. Giskes K, Turrell G, Patterson C, Newman B. Socioeconomic differences among Australian adults in consumption of fruit and vegetables and intakes of vitamins A, C and folate. Journal of Human Nutrition and Dietetics 2002;15:375-385.

cxvi

106. Deshmukh-Taskar P, Nicklas TA, Yang S-J, Berenson GS. Does food group consumption vary by differences in socioeconomic, demographic, and lifestyle factors in young adults? The Bogalusa Heart Study. Journal of the American Dietetic Association 2007;107(2):223-234. 107. Inglis V, Ball K, Crawford D. Why do women of low socioeconomic status have poorer dietary behaviours than women of higher socioeconomic status? A qualitative exploration. Appetite 2005;45(3):334-343. 108. Forsyth A, MacIntyre S, Anderson A. Diets for disease? Intraurban variation in reported food consumption in Glasgow. Appetite 1994;22(3):259-274. 109. Lindstrom M, Hanson B, Wirfalt E, Ostergren P. Socioeconomic differences in the consumption of vegetables, fruit and fruit juices. European Journal of Public Health 2001;11:51-59. 110. Hulshorf KFAM, Lowik MRH, Kok FJ, Wedel M, Brants HAM, Hermus RJJ, et al. Diet and other life-style factors in high and low socioe-economic groups (Dutch Nutrition Surveillance System). European Journal of Clinical Nutrition 1991;45:441-450. 111. Diez-Roux AV, Nieto FJ, Caulfield L, Tyroler HA, Watson RL, Szklo M. Neighbourhood differences in diet: the Atherosclerosis Risk in Communities (ARIC) Study. Journal of Epidemiology and Community Health 1999;53(1):55-63. 112. Rankin JW, Winett RA, Anderson ES, Buckley PG, Moore JF, Leahy M, et al. Food purchase patterns at the supermarket and their relationship to family characteristics. Journal of Nutrition Education 1998;30(2):81-88. 113. Popkin B, Siega-Riz A, Hanes P. A comparison of dietary trends among racial and socioeconomic groups in the United States. New England Journal of Medicine 1996;335(10):716-721. 114. Ecob R, MacIntyre S. Small area variations in health related behaviours; do these depend on the behaviour itself, its measurement, or on personal characteristics? Health and Place 2000;6:261-274. 115. Ellaway A, MacIntyre S. Does where you live predict health related behaviours?: A case study in Glasgow. Health Bulletin 1996;54(6):443-446. 116. Turrell G, Blakely TA, Patterson C, Oldenburg B. A multilevel analysis of socioeconomic (small area) differences in household food purchasing behaviour. Journal of Epidemiology and Community Health 2004;58(3):208-215. 117. Forsyth A, Macintyre S, Anderson A. Diets for disease? Intraurban variation in reported food consumption in Glasgow. Appetite 1994;22:259-274. 118. Kreft I, De Leeuw J. Introducing multi-level modelling. London: SAGE Pubications; 1998. 119. Dutton T, Turrell G, Oldenburg B. Measuring socieoeconomic position in population health monitoring and health research. Brisbane, Australia: Queensland University of Technology; 2005. 120. Boyle MH, Willms JD. Place effects for areas defined by administrative boundaries. American Journal of Epidemiology 1999;149(6):577-585. 121. Roache C. Researcher tries to accurately measure socioeconomic differences in health disparities. Around the School: News & Notices of the Harvard School of Public Health 2001;Sect. 2. 122. Davis CG. Linkages between socioeconomic characteristics, food expenditure patterns, and nutritional status of low income households: a critical review. American Journal of Agricultural Economics 1982(December 1982):1017-1025.

cxvii

123. Turrell G. Structural, material and economic influences on the food-purchasing choices of socioeconomic groups. Australian and New Zealand Journal fo Public Health 1996;20(6):611-617. 124. Donkin AJM, Johnson AE, Lilley JM, Morgan K, Neale RJ, Page RM, et al. Gender and living alone as determinants of fruit and vegetable consumption among the elderly living at home in urban Nottingham. Appetite 1998;30(1):39-51. 125. Morgan KJ. Socioeconomic factors affecting dietary status: an appraisal. American Journal of Agricultural Economics 1986(December):1240-1246. 126. Turrell G. Determinants of gender differences in dietary behaviour. Nutrition Research 1997;17(7):1105-1120. 127. Turrell G. Determinants of healthy food choice in a population-based sample. American Journal of Health Behaviour 1998;22(5):342-357. 128. Lutz SM, Blaylock JR, Smallwood DM. Household characteristics affect food choices. Food Review 1993;May-August:12-17. 129. Murphy SP, Rose D, Davis MA, Neuhaus JM, Lein D. Living arrangements over an 8-13 year period and food group consumption by older adults. Nutrition Research 1993;13:1239-1252. 130. Charles N, Kerr M. Eating properly, the family and state benefit. Sociology 1986;20(3):412-429. 131. Harnack L, Story M, Martinson B, Neumark-Sztainer D, Stang J. Guess who's cooking? The role of men in meal planning, shopping, and preparation in US families. Journal of the American Dietetic Association 1998;98(9):995-1000. 132. Calnan M, Cant S. The social organisation of food consumption: a comparison of middle class and working class households. International Journal of Sociology and Social Policy 1990;10(2):53-79. 133. Crawford DA, Baghurst KI. Diet and health: a national survey of beliefs, behaviours and barriers to change in the community. Australian Journal of Nutrition and Dietetics 1990;47(4):97-104. 134. Reicks M, Randall JL, Haynes BJ. Factors affecting consumption of fruits and vegetables by low-income families. Journal of the American Dietetic Association 1994;94(11):1309-1311. 135. Shankar S, Klassen A. Influences on fruit and vegetable procurement and consumption among urban African-American public housing residents, and potential strategies for intervention. Family Economics and Nutrition Review 2001;13(2):34-46. 136. Hupkens CLH, Knibbe RA, Van Otterloo AH, Drop MJ. Class differences in the food rules mothers impose on their children: a cross-national study. Social Science & Medicine 1998;47(9):1331-1339. 137. Randall E, Sanjur D. Food preferences - their conceptualization and relationship to consumption. Ecology of Food and Nutrition 1981;11:151-161. 138. Drewnowski A, Hann C. Food preferences and reported frequencies of food consumption as predictors of current diet in young women. American Journal of Clinical Nutrition 1999;70:28-36. 139. Drewnowski A. Energy intake and sensory properties of food. American Journal of Clinical Nutrition 1995;62(suppl):1081S-1088S. 140. Lupton D. The heart of the meal: food preferences and habits among rural Australian couples. Sociology of Health and Illness 2000;22(1):94-109. 141. Tomlinson M. Do distinct class preferences for foods exist? An analysis of class-based tastes. British Food Journal 1994;96(7):11-17.

cxviii

142. Turrell G. Socioeconomic differences in food preference and their influence on healthy food purchasing choices. Journal of Human Nutrition and Dietetics 1998;11:135-149. 143. Pliner P. The effects of mere exposure on liking for edible substances. Appetite 1982;3:283-290. 144. Trudeau E, Kristal AR, Li S, Patterson RE. Demographic and psychosocial predictors of fruit and vegetable intakes differ: Implications for dietary interventions. Journal of the American Dietetic Association 1998;98(12):1412-1417. 145. Cotugna N, Subar AF, Heimendinger J, Kahle L. Nutrition and cancer prevention knowledge, beliefs, attitudes, and practices: the 1987 National Health Interview Survey. Journal of the American Dietetic Association 1992;92(8). 146. Haralson MK, Sargent RG, Schluchter M. The relationship between knowledge of cardiovascular dietary risk and food shopping behaviors. American Journal of Preventive Medicine 1990;6(6):318-322. 147. Contento IR, Murphy BM. Psycho-social factors differentiating people who reported making desireable changes to their diets from those who did not. Journal of Nutrition Education 1990;22(1):6-14. 148. Charny M, Lewis PA. Does health knowledge affect eating habits? Health Education Journal 1987;46(4):172-176. 149. Shepard R, Stockley L. Nutrition knowledge, attitudes, and fat consumption. Journal of the American Dietetic Association 1987;87(5):615-619. 150. Turrell G. Educational differences in dietary guideline food practices: Are they associated with educational differences in food and nutrition knowledge? Australian Journal of Nutrition and Dietetics 1997;54:25-33. 151. Variyam JN, Blaylock J, Smallwood DM. Modelling nutrition knowledge, attitudes, and diet-disease awareness: the case of dietary fibre. Statistics in Medicine 1996;15:23-35. 152. Foley RM, Pollard CM, McGuinness DJ. Food Cent$-achieving a balanced diet on a limited food budget. Australian Journal of Nutrition and Dietics 1997;54(4):167-172. 153. Ranson D. 'Real men do cook'. A positive program for men. Australian Journal of Nutrition and Dietetics 1995;52(4):201-202. 154. Hermann J, Brown B, Heintz S. Impact of a nutrition promotion program on dietary behaviours, dietary intake and health measures in adults over fifty-five years of age. Journal of Nutrition for the Elderly 2000;19(3):1-14. 155. Forman C. Better buying skills help families save. Food and Nutrition 1988;18(3):18-19. 156. Crowley S. Affordability of a health diet for low income families: re-addressing the issues. Australian Journal of Nutrition and Dietetics 1997;54(4):165-166. 157. Jones E. An analysis of consumer food shopping behavior using supermarket scanner data: Differences by income and location. American Journal of Agricultural Economics 1997;79(5):1437-1443. 158. Strategic Intergovernmental Nutrition Alliance. Eat Well Australia: an agenda for action for public health nutrition: National Public Health Partnership; 2001. 159. Donkin AJM, Dowler EA, Stevenson SJ, Turner SA. Mapping access to food at a local level. British Food Journal 1999;101(7):554-564. 160. Bostock L. Pathways of disadvantage? Walking as a mode of transport among low-income mothers. Health and Social Care in the Community 2001;9(1):11-18.

cxix

161. Turrell G. Structural, material and economic influences on the food purchasing choices of socioeconomic group. Australian and New Zealand Journal of Public Health 1996;20(6):611-617. 162. Australian Bureau of Statistics. 4102.0 - Australian social trends, 1996: Special features: Registered cars. 1996. 163. Cummins S, Macintyre S, Davidson S, Ellaway A. Measuring neighbourhood social and material context: generation and interpretation of ecological data from routine and non-routine sources. Health and Place 2005;11(3):249-260. 164. Stitt S. Taking the cooking out of food: nutrition and the national curriculum. Nutrition and Health 1995;10(2):155-164. 165. Caraher M. The state of cooking in England: the relationship of cooking skills to food choice. British Food Journal 1999;101(8):509-609. 166. MacDonald JM, Nelson PE. Do the poor still pay more? Food price variations in large metropolitan areas. Journal of Urban Economics 1991;30:344-359. 167. Alwitt LF, Donley TD. Retail stores in poor urban neighbourhoods. the Journal of Consumer Affairs 1997;31(1):139-164. 168. Chung C, Myers S. Do the poor pay more for food? An analysis of grocery store availability and food price disparities. Journal of Consumer Affairs 1999;33(2):276-296. 169. Kaufman JS. Rural poor have less access to supermarkets, large grocery stores. Rural Development Perspectives 1999;13(3):19-26. 170. Morland K, Wing S, Diez Roux A, Poole C. Neighborhood characteristics associated with the location of food stores and food service places. American Journal of Preventive Medicine 2002;22(1):23-29. 171. Horowitz CR, Colson KA, Herbert PL, Lancaster K. Barriers to buying healthy foods for people with diabetes: evidence of environmental disparities. American Journal of Public Health 2004;94(9):1549-1553. 172. Moore LV, Diez Roux A. Associations of neighbourhood characteristics with the location and type of food stores. American Journal of Public Health 2006;96(2):325-331. 173. Powell LM, Slater S, Mirtcheva D, Bao Y, Chaloupka FJ. Food store availability and neighbourhood characteristics in the United States. Preventive Medicine 2007;44:189-195. 174. Zenk S, Schulz AJ, Hollis-Neely T, Campbell RT, Holmes N, Watkins G, et al. Fruit and vegetable intake in African Americans: income and store characteristics. American Journal of Preventive Medicine 2005;29(1):1-9. 175. Gallagher M. Chain reaction: income, race and access to Chicago's major player grocers. Chicago: Metro Chicago Information Centre; 2005. 176. Zenk S, Schultz AJ, Israel BA, James SA, Bao S, Wilson ML. Neighbourhood racial composition, neighbourhood poverty, and the spatial accessibiltiy of supermarkets in metropolitan Detroit. American Journal of Public Health 2005;95(4):660-667. 177. Latham J, Moffat T. Determinants of variation in food cost and availablity in two socioeocnomically contasting neighbourhoods of Hamilton, Ontario, Canada. Health & Place 2007;13:273-287. 178. Smoyer-Tomic KE, Spence JC, Amrhein C. Food deserts in the praries? Supermarket accessibility and neighbourhood need in Edmonton, Canada. The Professional Geographer 2006;58(3):307-326.

cxx

179. Guy C, Clarke G, Eyre H. Food retail change and the growth of food deserts: a case study of Cardiff. International Journal of Retail and Distribution Management 2004;32(2):72-88. 180. Cummins S, MacIntyre S. The location of food stores in urban areas: a case study in Glasgow. British Food Journal 1999;101(7):545-553. 181. White M, Bunting J, Williams E, Raybould S, Adamson A, Mathers JC. Do 'food deserts' exist? A multi-level, geographical analysis of the relationship between retail food access, socio-economic position and dietary intake. London: Food Standards Agency; 2004. 182. Clarke G, Eyre H, Guy C. Deriving indicators of access to food retail provision in British cities: Studies of Cardiff, Leeds and Bradford. Urban Studies 2002;39(11):2041. 183. Burns C, Inglis AD. Measuring food access in Melbourne: Access to healthy and fast foods by car, bus and foot in an urban municipality in Melbourne. Health & Place 2007;13:877-885. 184. Dibsdall LA, Lambert N, Bobbin RF, Frewer LJ. Low-income consumers' attitudes and behaviour towards access, availability and motivation to eat fruit and vegetables. Public Health Nutrition 2003;6(2):159-168. 185. Ingene C. Structural determinants of market potential. Journal of Retailing 1984;60(1):37-64. 186. Eisenhauer E. In poor health: Supermarket redlining and urban nutrition. GeoJournal 2001;53(2):125-33. 187. Alwitt LF, Donley TD. Retail stores in poor urban neighbourhoods. The Journal of Consumer Affairs 1997;31(1):139-163. 188. Chung C, Meyers Jr SL. Do the poor pay more for food? An analysis of Grocery Store Availability and Food Price Disparities. The Journal of Consumer Affairs 1999;33(2):276-296. 189. Chung C, Myers Jr. SL. Do the poor pay more for food? An analysis of grocery store availability and food price disparities. The Journal of Consumer Affairs 1999;33(2):276-296. 190. McLennan W. Information paper: 1996 Census of population and housing Socio-economic Indexes for Areas. In. Canberra: Australian Bureau of Statistics; 1998. p. 46. 191. Baker EA, Kelly C, Barnidge E, Strayhorn J, Schootman M, Struthers J, et al. The garden of eden: Acknowledging the impact of race and class in efforts to decrease obestiy rates. American Journal of Public Health 2006;96(7):1170-1174. 192. Sooman A, MacIntyre S, Anderson A. Scotland's Health - a more difficult challenge for some? The price and availability of healthy foods in socially contrasting localities in the West of Scotland. Health Bulletin 1993;51(5):276-284. 193. Cummins S, Macintyre S. A systematic study of an urban foodscape: The price and availability of food in greater Glasgow. Urban Studies 2002;39(11):2115-2130. 194. U.K. Competition Commission. Supermarkets: a report on the supply of groceries from multiple stores in the United Kingdom: Department of Trade; 2000. 195. Cheadle A, Psaty BM, Curry S, Wagner E, Diehr P, Koepsell T, et al. Community level comparisons between the grocery store environment and individual dietary practices. Preventive Medicine 1991;20:251-261. 196. Furey S, Farley H, Strugnell C. An investigation into the availability and economic accessibility of food items in rural and urban areas of Northern Ireland. International Journal of Consumer Studies 2002;26(4):313-321.

cxxi

197. Hayes LR. Do the poor pay more? An empirical investigation of price dispersion in food retailing. Princeton: Department of Economics - Industrial Relations Section; 2000. 198. Bowman SA. A comparison of the socioeconomic characteristics, dietary practices, and health status of women food shoppers with different food price attitudes. Nutrition Research 2006;26:318-324. 199. Glanz K, Basil M, Miaibach E, Goldberg J, Snyder D. Why Americans eat what they do: taste, nutrition, cost, convenience and weight control concerns as influencers of food consumption. Journal of the Amercian Dietetic Association 1998;98(10):1118-1126. 200. Travers KD, Cogdon A, McDonald W, Wright C, Anderson B, Maclean DR. Availability and cost of heathy dietary changes in Nova Scotia. Journal of the Canadian Dietetic Association 1997;58(4):176-183. 201. Jetter KM, Cassady DL. The availability and cost of healthier food alternatives. American Journal of Preventive Medicine 2006;30(1):38-44. 202. Finke MS, Chern WS, Fox JJ. Do the urban poor pay more for food? Issues in measurement. Advancing the Consumer Interest 1997;9(1):13-17. 203. Frankel DM, Gould ED. The retail price of inequality. Journal of Urban Economics 2001;49(2):219-239. 204. Travers KD, Cogdon A, McDonald W, Wright C, MacLean DR. Availability and cost of heart healthy dietary changes in Nova Scotia. Journal of the Canadian Dietetic Association 1997;58(4):176-183. 205. Spencer S. Price detrmination in the Australian food industry. Canberra: Department of Agriculture Fishery and Forestry,; 2004 2004. Report No.: ISBN: 0 642 539073. 206. Department of Agriculture Fisheries and Forestry. Australian Food Statistics 2005. Canberra: Department of Agriculture, Fisheries and Forestry; 2005. 207. Donkin AJM, Dowler EA, Stevenson SJ, Turner SA. Mapping access to food in a deprived area: the development of price and availability indices. Public Health Nutrition 1999;3(1):31-38. 208. Australian Food and Grocery Council. Food prices. In. Barton (ACT): Australian Food and Grocery Council; 2003. 209. Topolski R, Boyd-Bowman K, Ferguson H. Grapes of wrath: discrimination in the produce aisle. Analyses of Social Issues and Public Policy 2003;3(1):111-119. 210. Hoch SJ, Byung-Do K, Montgomery AL, Rossi PE. Determinants of store-level price elasticity. Journal of Marketing Research 1995;32:17-29. 211. Wakefield KL, Inman JJ. Situational price sensitivity: the role of consumption occasion, social context and income. Journal of Retailing 2003;97:199-212. 212. Seale J. Monitoring the price and availabilty of healthy food - time for a national approach? Nutrition & Dietetics 2004;61(4):197-198. 213. Kamphuis C, Van Lenthe FJ, Giskes K, Brug J, Mackenbach JP. Perceived environmental determinants of physical activity and fruit and vegetable consumption among high and low socioeconomic groups in the Nedtherlands. Health & Place 2007;13:493-503. 214. Morland K, Wing S, Diez Roux A. The contextual effect of the local food environment on residents' diets: The Atherosclerosis Risk in Communities Study. American Journal of Public Health 2002;92(11):1761-1767.

cxxii

215. Ball K, Crawford D, Mishra G. Socio-Economic inequalites in women's fruit and vegetable intakes: a multilevel study of individual, social and environmental mediators. Public Health Nutrition 2006;9(5):623-630. 216. Wrigley N, Warm D, Margetts BM, Whelan A. Assessing the impact of improved retail access on diet in a 'food desert': a preliminary report. Urban Studies 2002;39(11):2061-2082. 217. Wrigley N, Warm D, Margetts B. Deprivation, diet and food-retail access: findings from the Leeds 'food deserts' study. Environment and Planning A 2003;35(1):151-188. 218. Cummins S, Petticrew M, Higgins C, Findlay A, Sparks L. Large scale food retailing as an intervention for diet and health: quasi-experimental evaluation of a natural experiment. Journal of Epidemiology and Community Health 2005;59:1035-1040. 219. Pearson T, Russell J, Campbell MJ, Barker ME. Do 'food deserts' influence fruit and vegetable consumption? - a cross-sectional study. Appetite 2005;45:195-197. 220. Bodor JN, Rose D, Farley TA, Swalm C, Scott SK. Neighbourhood fruit and vegetable availability and consumption: the role of small food stores in an urban environment. Public Health Nutrition 2008;in press:1-8. 221. Rose D, Richards R. Food store access and household fruit and vegetable use among participants in the US Food Stamp Program. Public Health Nutrition 2004;7(8):1081-1088. 222. Fisher BD, Strogatz DS. Community measures of low-fat milk consumption: comparing store shelves with households. American Journal of Public Health 1999;89(2):235-237. 223. Edmonds J, Baranowski T, Baranowski J, Cullen KW, Myres D. Ecological and socioeconomic correlates of fruit, juice, and vegetable consumption among African-American boys. Preventive Medicine 2001;32(6):476-481. 224. Cheadle A, Psaty BM, Curry S, Wagner E, Diehr P, Koepsell T, et al. Community-level comparisons between the grocery store environment and individual dietary practices. Preventive Medicine 1991;20:250-261. 225. Cheadle A, Psaty BM, Curry S, Wagner E, Diehr P, Koepsell T, et al. Can measures of the grocery store environment be used to track community-level dietary changes? Preventive Medicine 1993;22:361-372. 226. Giskes K, Van Lenthe FJ, Brug J, Mackenbach JP, Turrell G. Socioeconomic inequalities in food purchasing: The contribution of respondent-perceived and actual (objectively measured) price and availabilty of foods. Preventive Medicine 2007;45(1):41-48. 227. Devine CM, Wolfe WS, Frongillo EA, Bisogni CA. Life-course events and experiences: Association with fruit and vegetable consumption in 3 ethnic groups. Journal of the American Dietetic Association 1999;99:309-314. 228. French SA, Stables GJ. Environmental interentions to promotie vegetable and fruit consumption among youth in school settings. Preventive Medicine 2003;31(37):593-610. 229. Sorensen G, Stoddard A, Peterson K, Cohen N, et al. Increasing fruit and vegetable consumption through worksites and families in the Treatwell 5-a-day study. American Journal of Public Health 1999;89(1):54-60. 230. Curhan RC. The effects of merchandising and temporary promotional activities on the sales of fresh fruits and vegetables in supermarkets. Journal of Marketing Research 1974;11(3):286-294.

cxxiii

231. Frazao E, Allhouse J. Strategies for intervention: commentary and debate. The Journal of Nutrition 2003;133(3):s844-847. 232. Furst T, Connors MM, Bisogni CA, Sobal J, Winter Falk L. Food choice: a conceptual model of the process. Appetite 1996;26:247-266. 233. Owen KM, Griffith GR, Wright VE. One little lebanese cucumber is not going to break the bank: price in the choice of fresh fruits and vegetables. Australian Journal of Agricultural and Resource Economics 2002;2002(46):2. 234. Lechene V. National Food Survey, 2000: Income and price elasticites of demand for foods consumed in the home. London: National Food Survey Committee, Department for Environment Food and Rural Affairs; 2000. 235. Richards TJ. A discrete / continuous model of fruit promotion, advertising, and response segmentation. Agribusiness 2000;16(2):179-196. 236. Jones E, Akbay C, Roe B, Chern WS. Analyses of consumers' dietary behaviour: an application of the AIDS model to supermarket scanner data. Agribusiness 2003;19(2):203-221. 237. Akbay C, Jones E. Food consumption behavior of socioeconomic groups for private labels and national brands. Food Quality and Preference 2005;16(7):621-631. 238. Regmi A, Deepak MS, Seale Jr. JL, Bernstein J. Changing structure of global food consumption and trade. Washington DC: Market and Trade Economics Division, Economic Research Service, U.S. Department of Agriculture, Agriculture and Trade; 2001. Report No.: WRS-01-1. 239. Kinsey J, Bowland B. How can the US food system deliver food products consistent with the dietary guidelines? Food marketing and retailing: an economist's view. Food Policy 1999;24:237-253. 240. Murchu CN, Blakely T, Wall J, Rodgers A, Jiang Y. Strategies to promote healthier food purchases: a pilot supermarket intervention study. Public Health Nutrition 2007;10(6):608-615. 241. Seymour JD, Yaroch AL, Serdula M, Blanck HM, Khan LK. Impact of nutrition environment inventions on point-of-purchase behavior in adults: a review. Preventive Medicine 2004;39:s108-s136. 242. Kristal AR, Goldenhar L, Muldoon J, Morton RF. Evaluation of a supermarket intervention to increase consumption of fruits and vegetables. American Journal of Health Promotion 1997;11:422-425. 243. French SA, Story M, Jeffery RW, Snyder P, Eisenberg M, Sidebottom A, et al. Pricing strategy to promote fruit and vegetable purchase in high school cafeterias. Journal of the Amercian Dietetic Association 1997;97(9):1008-1010. 244. Fiske A, Cullen KW. Effects of promotional materials on vending sales of low-fat items in teacher's lounges. Journal of the Amercian Dietetic Association 2004;104(1):90-93. 245. Jeffery RW, French SA, Raether C, Baxter JE. An environmental intervention to increase fruit and salad purchases in a cafeteria. Preventive Medicine 1994;23(6):788-792. 246. Horgen KB, Browning KD. Comparison of price change and health message interventions in promoting healthy food choices. Health Psychology 2002;51(5):505-512. 247. French SA, Jeffery RW, Story M, Hannan P, Snyder P. A pricing strategy to promote low-fat snack choice through vending machines. American Journal of Public Health 1997;87(5):849-851.

cxxiv

248. French SA, Jeffery RW, Story M, Brieitlow KA, Baxter JS, Hannan P, et al. Pricing and promotion efforts on low-fat vending snack purchases: The CHIPS study. American Journal of Public Health 2001;91(1):112-117. 249. Darmon N, Ferguson EL, Briend A. Do economic constaints encourage the selection of energy dense diets? Appetite 2003;41:315-322. 250. Darmon N, Darmon M, Maillot M, Drewnowski A. A nutrient density standard for vegetables and fruits: nutrients per calorie and nutrients per unit cost. Journal of the Amercian Dietetic Association 2005;105(12):1881-1887. 251. Drewnowski A, Darmon N. Food choices and diet costs: an economic analysis. Journal of Nutrition 2005;135:900-904. 252. Drewnowski A, Darmon N, Briend A. Replacing fats and sweets with vegetables and fruits - a question of cost. Research Practice 2004;94(9):1555-1559. 253. Andrieu E, Darmon N, Drewnowski A. Low-cost diets: more energy, fewer nutrients. European Journal of Clinical Nutrition 2006;60(4):434–436. 254. Darmon N, Ferguson EL, Briend A. A cost constraint alone has adverse effects on food selection and nutrient density: An analysis of human diets by linear programming. The Journal of Nutrition 2002;132(12):3764-3771. 255. Sturm R, Datar A. Body mass index in elementary school children, metropolitan area food prices and food outlet density. Public Health 2005;119:1059-1068. 256. Chou S-Y, Grossman M, Saffer H. An economic analysis of adult obesity: results from the Behavioural Risk Factor Surveillance System. Journal of Health Economics 2004;23:565-587. 257. Ard JD, Fitzpatrick S, Desmond RA, Sutton BS, Pisu M, Allison DG, et al. The Impact of Cost on the Availabilty of Fruits and Vegetables in the homes of School children in Brimingham, Alabama. American Journal of Public Health 2007;97(2):367-371. 258. Cade J, Upmeier H, Calvert C, Greenwood D. Costs of a healthy diet: analysis from the UK Women's Cohort Study. Public Health Nutrition 1999;2(4):505-512. 259. Watters J. Associations of psychosocial factors with fruit and vegetable intake among African-Americans. Public Health Nutrition 2007;10(7):701-711. 260. Vickner SS, Davies SP. Estimating strategic price response in a product-differentiated oligopoly: The case of a domestic canned fruit industry. Agribusiness 2000;16(2):125-140. 261. Walden MJ. Testing implications of spatial economics models: some evidence from food retailing. The Journal of Consumer Affairs 1990;24(1):24-43. 262. Thomadsen R. The effect of ownership structure on prices in geographically differentiated industries. The Journal of Economics 2005;36(4):908-929. 263. Whitehead K, Tretheway L. Edith Devitt, domestic science and feminist ideas in state schools, 1910-1925. Journal of the Historical Society of South Australia 2001;29:85-93. 264. McGraw-Hunter M, Faw GD, Davis PK. The use of video self-modelling and feedback to teach cooking skills to individuals with traumatic brain injury: a pilot study. Brain Injury 2006;20(10):1061-1068. 265. Schuster JW. Cooking instruction with persons labeled mentally retarded: A review of literature. Education & Training in Mental Retardation 1988;23(1):43-50. 266. Furey S, McIlven H, Strugnell C, Armstrong G. Cooking skills: a diminishing art? Nutrition and Food Science 2000;30(5):263-266.

cxxv

267. Porter JA, Patterson CM. Are we contributing to the development of a generation who cannot prepare food? Journal of the HEIA 1994;1(4):37-42. 268. Kingston B. When did we teach our girls to cook? Australian Cultural History 1996;15:89-101. 269. Stead M, Caraher M, Wrieden W, Longbottom P, Valentine K, Anderson A. Confident, fearful and hopeless cooks: Findings form teh development of a food-skills initiative. British Food Journal 2004;106(4):274-286. 270. Bisogni CA, Jastran BS, Shen L, Devine CM. A biographical study of food choice capacity: standards, circumstances and food management skills. Journal of Nutrition Education and Behaviour 2006;37(6):284-291. 271. Hughes G, Bennet KM, Hetherington MM. Old and alone: barriers to healthy eating in older men living on their own. Appetite 2004;43:267-276. 272. Foley RM, Pollard CM. Food Cent$ - implementing and evaluating a nutrition education project focusing on value for money. Australian and New Zealand Journal of Public Health 1998;22(4):494-501. 273. Richins RA, Christensen N, Israelsen M, Albrecht C, Washburn C. Development, evaluation, and efficiency of a heart healthy curriculum in two different education setiings: emphasising food portioning and cooking skills, increased fruits, vegetables, whole grains, low-fat dairy and exercise. Journal of the American Dietetic Association 2007;107(8(s3)):A-94. 274. Lang J. Advocacy: a call to action. Community College Journal 2006;76(5):12-16. 275. Spence F. A qualitative evaluation of community-based cooking classes in Northeast Scotland. International Journal of Health Promotion and Education 2005;43(2):59-63. 276. Larson NI, Story M, Eisenberg ME, Neumark-Sztainer D. Food preparation and purchasing roles among adolescents: associations with sociodemographic characteristics and diet quality. Journal of the American Dietetic Association 2006;106(2):211-218. 277. Nijmeijer M, Worsley A, Astill B. An exploration of the relationships between food lifestyle and vegetable consumption. British Food Journal 2004;106(7):520-533. 278. Mclaughlin C, Tarasuk V, Kreiger N. An examination of at-home food preparation activity among low-income, food-insecure women. Journal of the American Dietetic Association 2003;103(11):1506-1512. 279. Larson NI, Perry CL, Story M, Neumark-Sztainer D. Food preparation by young adults is associated with better diet quality. Journal of the American Dietetic Association 2006;106(12):2001-2007. 280. Williams J. A study to determine actual consumption, awareness of health benefits and factors influencing fruit and vegetable consumption in first-year undergraduate students. Journal of Human Nutrition and Dietetics 2000;13:363-371. 281. Broughton MA, Janssen PS, Hertzmann C, Innis SM, Frankish CJ. Predictors and outcomes of household food insecurity among inner city families with preschool children in Vancouver. Canadian Journal of Public Health 2006;97(3):214-216. 282. Koszewski WM, Yokum G, Jones G, Boeckner L. A study of college athletes dietary habits, food preparation skills, and source of nutrition information. Journal of the American Dietetic Association 2007;107(8(s1)):A-107. 283. Strugnell C, Furey S, Farley H. Food deserts - an example of social exclusion? In: Fairchild R, editor. Empowering the Consumer in the 21st Century,;

cxxvi

2003; UWIC, Llandaff Centre, Cardiff: International Journal of Consumer Studies; 2003. p. 218-251. 284. Health Education Authority. HEA health and lifestyle survey: a report on the secondary analysis of a national dataset of health-related knowledge, attitudes and behaviour. In. version two ed. London: Health Education Authority; 1998. 285. Lang T, Caraher M. Cooking skills and health. London: Health Education Authority; 1999. 286. Rissel CE, Gano CA. Nutrition and low socioeconomic status. Journal of the Home Economics Institute of Australia 1990;22(3):76-79. 287. Lynch J, Kaplan G. Socioeocnomic position. In: Berkman L, Kawachi I, editors. Social Epidemiology. New York: Oxford University Press; 2000. 288. Krieger N, Williams DR, Moss NE. Measuring social class in US public health research: concepts, methodologies and guidelines. Annual Review of Public Health 1997;18:341-378. 289. Berkman L, Macintyre S. The Measurement of social class in health studies: Old measures and new formulations. In: Kogevinas M, Pearce N, Susser M, Boffeta P, editors. Social Inequalities and Cancer. Lyon, France: International Agency for Research on Cancer; 1997. p. 51-64. 290. Vågerö D. Health inequalities in women and men: studies of specific causes of death should use household criteria. British Medical Journal 2000;320(7245):1286-1287. 291. Bartley M, Sacker A, Firth D, Fitzpatrick R. Understanding social variation in cardiovascular risk factors in women and men: the advantages of theoretically based measures. Social Science & Medicine 1999;49(6):831-845. 292. BenShlomo Y, Smith GD. Commentary: Socioeconomic position should be measured accurately (Postcodes as useful markers of social class: population based study in 26 000 British households). British Medical Journal 1999;318(7817):844-845. 293. Donkin Aea. Mapping access to food in a deprived area: the development of price and availability indicies. Public Health Nutrition 1999;3(1):31-38. 294. Merriam-Webster I. Merriam-Webster online dictionary. In: Merriam-Webster, Incorporated; 2004. 295. Anderson AS. A questionnaire assessment of nutrition knoweldge - validity and reliabilty issues. Public Health Nutrition 2002;5(3):497-503. 296. Reichler G, Dalton S. Chefs' attitudes toward healthful food preparation are more positive than their food science knowledge and practices. Journal of the American Dietetic Association 1998;98(2):165-169. 297. Solin MS, Dalton S. Knowledge of food identification, food preparation techniques, and attitudes toward the professional need for food skills of registered dietitians and dietetic students. Journal of the American Dietetic Association 1997;97(9, Supplement 1):A104. 298. Violan N. American cooks fail cooking quiz but want to learn more. Environmental Nutrition 1992:1-2. 299. Anderson AS, Cox DN, McKellar S, Reynolds J, Lean ME, Mela DJ. Take Five, a nutrition education intervention to increase fruit and vegetable intakes: impact on attitudes towards dietary change. [see comments]. British Journal of Nutrition 1998;80(2):133-40. 300. AbuShaba R. Review of self-efficacy and locus of control for nutrition - and health- related behaviour. Journal of the American Dietetic Association 1997;97(10):1122-1132.

cxxvii

301. Seaman C. How are dietary surveys planned? British Food Journal 1995;97(2):18-21. 302. Natarajan L, Flatt SW, Sun X, Gamst AC, Major JM, Rock CL, et al. Validity and systematic error in measuring carotenoid consumption with dietary self-report instruments. American Journal of Epidemiology 2006;163(8):770-778. 303. Michels KB, Welch AA, Luben R, Bingham SA, Day NE. Measurement of fruit and vegetable consumption with diet questionnaires and implications for analyses and interpretation. American Journal of Epidemiology 2005;161(10):987-994. 304. Day N, McKeown N, Wong M, Welch A, Bingham S. Epidemiological assessment of diet: a comparison of a 7-day diary with a food frequency questionnaire using urinary markers of nitrogen, potassium and sodium. International Journal of Epidemiology 2001;30(2):309-317. 305. Australian Bureau of Statistics. 4102.0 - Australian social trends, 1998: Health related actions: Food and energy intake. In: Australian Bureau of Statistics, editor. Australian Social Trends. Canberra: AGPS; 1998. 306. Ransley JK, Donnelly JK, Botham H, Khara TN, Greenwood DC, E CJ. Use of supermarket receipts to estimate energy and fat content of food purchased by lean and overweight families. Appetite 2003;41:141-148. 307. Hewitt B, Turrell G, Patterson C, Gould T, Roy M. The Brisbane Food Study: A multi-level and spatial investigation of socioeconomic differences in food purchasing behaviour. Brisbane: Queensland University of Technology; 2003 November 2003. 308. Turrell G, Patterson C, Oldenburg B, Gould T. The socioeconomic patterning of survey participation and non-response error in a multilevel study of food purchasing behaviour: area and individual-level characteristics. Public Health Nutrition 2003;6(2):181-189. 309. Australian Bureau of Statistics. 1996-1997 Apparent consumption of foodstuffs. In. Canberra: Australian Government Publishing Service; 1998. 310. Australian Bureau of Statistics. National Nutrition Survey: Foods eaten in Australia. In: Australian Government Publishing Service; 1995. 311. Turrell G, Hewitt B, Patterson C, Oldenburg B. Measuring socio-economic position in dietary research: is choice of socio-economic indicator important? Public Health Nutrition 2003;6(2):191-200. 312. Turrell G, Najman JM. Collecting food-related data from low socioeconomic groups: how adequate are our current research designs? Australian and New Zealand Journal of Public Health 1995;19:410-416. 313. Dillman DA. The tailored design method. 2 ed. New York: John Wiley & Sons, Inc.; 2000. 314. Tseng M, DeVellis RF. Fundamental dietary patterns and their correlates among US whites. Journal of the Amercian Dietetic Association 2001;10(8):929-932. 315. Gleim JA, Gleim RR. Calculating, interpreting and reporting Cronbach's Alpha reliability coefficient for Likert-type scales. In: Midwest Research-to-Practice Conference in Adult, Continuing, and Community Education; 2003 8/10/2003; Ohio State University; 2003. p. 82-88. 316. George D, Mallery P. SPSS for Windows step by step: a simple guide and reference. 11.0 update. 4 ed. Boston: Allyn & Bacon; 2003. 317. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159-174.

cxxviii

318. Bowker AH. A test for symmetry in contingency tables. Journal of the American Statistical Association 1948;43(244):572-574. 319. Cicchetti DV, Allison T. A new procedure for assessing reliabiltiy of scoring EEG sleep recordings. American Journal of EEG Technology 1971;11:101-9. 320. Armitage P, Berry G. Statistical methods in medical research (3rd Edition). London: Blackwell; 1994. 321. Shrout PE, Fleiss JL. Intra-class correlations: uses in assessing rater reliabiltiy. Psychological Bulletin 1979;86(2):420-428. 322. John JH, Ziebland S, Yudkin P, Roe LS, Neil HAW, The Oxford Fruit and Vegetable Study Group. Effects of fruit and vegetable consumption on plasma antioxidant concentrations and blood pressure: a randomised controlled trial. The Lancet 2002;359(9322):1969-1974. 323. Rogers L, Sharp I, National Forum for Coronary Heart Disease Prevention. Preventing coronary heart disease: The role of antioxidants, vegetables and fruit: report of an expert meeting. London: National Forum for Coronary Heart Disease Prevention; 1997. 324. Kaumudi JJ, Ascherio A, Manson JE, Stampfer MJ, al e. Fruit and vegetable intake in relation to risk of ischemic stroke. Journal of the American Medical Association 1999;282(13):1233-1239. 325. Lahmann P, Lissner L, Berglund G. Sociodemographic factors associated with long-term weight gain, current body fatness and central adiposity in Swedish women. International Journal of Obesity 2000;24:685-694. 326. Magarey A, Boulton J. The Adelaide Nutrition Study 5. Differences in energy, nutrient and food intake at ages 11, 13, and 15 years according to fathers' occupation and parents' educational level. Australian Journal of Nutrition and Dietetics 1997;54(1):15-23. 327. Rutishauser I, Wheeler C, Conn J, O'Dea K. Food and nutrient intake in a randomly selected sample of adults: demographic and temporal influences on energy and nutrient intake. Australian Journal of Nutrition and Dietetics 1994;51(4):157-166. 328. Turrell G. Educational differences in dietary guideline food practices: are they associated with educational differences in food and nutrition knowledge? Australian Journal of Nutrition and Dietetics 1997;54(1):25-33. 329. De Irala-Estevez J, Groth M, Johansson L, Oltersdorf U, Prattala R, Martinez-Gonzalez MA. A systematic review of socio-economic differences in food habits in Europe: consumption of fruit and vegetables. European Journal of Clinical Nutrition 2000;54:706-714. 330. Duncan C, Jones K, Moon G. Do places matter? A multi-level analysis of regional variation in health related behaviour in Britain. Social Science and Medicine 1993;37(6):725-733. 331. Curtis S, Jones IR. Is there a place for geography in the analysis of health inequality? Sociology of Health and Illness 1998;20(5):645-672. 332. Whichelow MJ, Erzincliglu SW, Cox BD. Some regional variations in dietary patterns in a random sample of British adults. European Journal of Clinical Nutrition 1991;45:253-262. 333. Graddy K. Do fast-food chains price discriminate on the race and income characteristics of an area? Journal of Business & Economic Statistics 1997;15(4):391-401. 334. Cummins S, Macintyre S. "Food deserts"--evidence and assumption in health policy making. British Medical Journal 2002;325(7361):436.

cxxix

335. Lee AJ, Darcy AM, Leonard D, Groos AD, Stubbs CO, Lowson S, K, et al. Food availability, cost disparity and improvement in relation to accessibility and remoteness in Queensland. Australian and New Zealand Journal of Public Health 2002;26(3):266-272. 336. Australian Bureau of Statistics. Year book Australia 2002: Population distribution. In: Australian Bureau of Statistics; 2002. 337. Reidpath DD, Burns C, Garrand J, Mahoney M, Townsend M. An ecological study of the relationship between social and environmental determinants of obesity. Health and Place 2002;8:141-145. 338. Frankel DM, Gould ED. The Retail Price of Inequality. Journal of Urban Economics 2000;49(2):219-239. 339. Australian Bureau of Statistics. Population growth: The trend is city living or seachange - ABS. In: Australian Bureau of Statistics; 2002. 340. Townsend P. Deprivation. Journal of Social Policy 1987;16(2):125-146. 341. Morris R, Carstairs V. Which deprivation? a comparison of selected deprivation indexes. Journal of Public Health Medicine 1991;13(4):318-326. 342. MapInfo Corporation. MapInfo Professional. In. 6.5 ed: MapInfo Corporation; 1985. 343. SAS Institue Inc. SAS. In. 8.2 ed. Cary NC, USA: SAS Institute Inc.; 1999. 344. Australian Bureau of Statistics, MapInfo Pty Ltd. CDATA2001. In. 2nd Release version 7 ed: Commonwealth of Australia; 2002. 345. Pritchard W. Beyond the modern supermarket: geographical approaches to the analysis of contemporary Australian retail restructuring. Australian Geographical Studies 2000;38(2):204-218. 346. Sooman A, Macintyre S, Anderson A. Scotland's health - a more difficult challenge for some? The price and availability of healthy foods in socially contrasting localities in the west of Scotland. Health Bulletin 1993;51(5):276-284. 347. Levy PS, Lemeshow S. Sampling of populations: methods and applications. 3 ed. New York: John Wiley & Sons, Inc.; 1999. 348. Cohen MP. Determining sample sizes for surveys with data analyzed by hierarchical linear models. Journal of Official Statistics 1998;14(3):267-275. 349. Coles Supermarkets Pty Ltd. Coles Fruit and Vegetable Index. In; 1998. 350. Xyris Software. FoodWorks professional. In. 3.02 ed. Brisbane: Xyris Software. 351. Nagelkerke N. A note on a general definition of the coefficient of determination. Biometrika 1991;78(3):691-692. 352. Lester IH. The selling and buying of food. In: Australia's Food and Nutrition. Canberra: Australian Government Publishing Service; 1994. p. 97-136. 353. Australian Bureau of Statistics. 6530.0 Household Expenditure Survey, Australia: Summary of results. In. Canberra: Australian Bureau of Statistics,; 2005. 354. Commonwealth of Australia. Centrelink. In; 2005. 355. Mobley LR, Root DE, Finkelstein EA, Khavjou MA, Farris RP, Will JC. Environment, obesity, and cardiovascular disease risk in low-income women. American Journal of Preventive Medicine 2006;30(4):327-332. 356. King T, Kavanagh AM, Jolley D, Turrell G, Crawford D. Weight and place: a multilevel cross-sectional survey of area-level social disadvantage and overweight/ obesity in Australia. International Journal of Obesity 2006;30:281-287. 357. Bodor JN, Rose D, Farley TA, Swalm C, Scott SK. Neighbourhood fruit and vegetable availability and consumption: the role of small food stores in an urban environment. Public Health Nutrition 2007;in press:1-8.

cxxx

358. Laraia BA, Siega-Riz AM, Kaufman JS, Jones SJ. Proximity of supermarkets is positively associated with diet quality index for pregnancy. Preventive Medicine 2004;39(5):869-875. 359. Pearce J, Hiscock R, Blakely T, Witten K. The contextual effects of neighbourhood access to supermarkets and convenience stoers on individual fruit and vegeatable consumption. Journal of Epidemiology and Community Health 2008;62:198-201. 360. MacLenan DL, Gottschall-Pass K, Larsen R. Fruit and Vegetable Consumption: Benefits and Barriers. Canadian Journal of Dietetic Practice and Research 2004;65(3):101-105. 361. Australian Bureau of Statistics. Household income and income distribution, 2000-01. Canberra: Australian Bureau of Statistics; 2003. Report No.: 6523.0. 362. Morland K, Diez Roux A, Wing S. Supermarkets, other food stores, and obesity. American Journal of Preventive Medicine 2006;30(4):333-339. 363. Cummins S, Findlay A, Higgins C, Sparks L, Thompson H. Reducing inequalities in health and diet: Findings from a study of the impact of a food retail development. Environment and Planning 2008;40(2):402-422. 364. Apparicio P, Abdelmajid M, Riva M, Shearmur R. Comparing alternative approaches to measuring the geographical accessibilty of urban health services: Distance types and aggregation-error issues. International Journal of Health Geographics 2007;7(7):14. 365. Charles N, Kerr M. Women, food and families. Manchester and New York: Manchester University Press; 1988. 366. Broomhill R, Sharp R. The changing male breadwinner model in Australia. Labour and Industry 2004;15(2):1-23. 367. Darmon N, Briend A, Drewnowski A. Energy-dense diets are associated with lower diet costs: a community study of French adults. Public Health Nutrition 2004;7(1):21-27. 368. Anas A. A unified theory of consumption, travel and trip chaining. Journal of Urban Economics 2006;in press. 369. Brisbane City Council. Brisbane city plan 2000. Brisbane: Brisbane City Council; 2000. 370. Australian Bureau of Statisitcs. Regional profile Brisbane City 2002 to 2004: selected characteristics. Canberra: Australian Bureau of Statisitcs,; 2006. Report No.: 1379.0.55.001. 371. World Health Organisation. World Health Report 2002: reducing risks, promoting healthy life. Geneva: WHO; 2002. Report No.: 1O20-3311. 372. Doeleman BH, Kunkel ME, Cason KL. Knowledge, skills, and behavior changes of expanded Food and Nutrition Education Program graduates in South Carolina. Journal of the American Dietetic Association 1998;98(9, Supplement 1):A48. 373. James W, Nelson M. The contribution of nutrition to inequalities in health. British Medical Journal 1997;314(7093):1545-1550. 374. Trewin D. Technical Paper - Census of population and housing: Socio-Economic Indexes for Area's (SEIFA). Canberra: Australian Bureau of Statistics; 2004. Report No.: 2039.0.55.001. 375. Magarey A. Evaluation of fruit and vegetable intakes of Australian adults: the National Nutrition Survey 1995. Australian & New Zealand Journal of Public Health 2006;30(1):32-7. 376. Kramer L, Pollack J. On the run. Advertising Age 1998 24/0898:1-2.

cxxxi

377. McMahon A. Blokus domesticus: the sensitive new age guy in Australia. Journal of Australian Studies 1998;56:147-157. 378. Williams P. Home economics: between a rock and a hard place. Journal of the HEIA 1994;1(3):8-17. 379. Varley P. Evaluation of the National Nutrition Education in Schools project. Journal of the HEIA 1995;2(4):48-54. 380. Livingstone DW. Exploring the icebergs of adult learning: findings of the first Canadian survey of informal learning practices. Ottawa, Canada: Social Sciences and Humanities Research Council of Canada; 1999. 381. Monden CWS, Van Lenthe FJ, Mackenbach JP. A simultaneous analysis of neighbourhood and childhood socio-economic environment with self assessed health and health-related behaviours. Health & Place in press:10. 382. Lynch J, Kaplan GA, Salonen JT. Why do poor people behave poorly? Variation in adult health behaviours and psychosocial characteristics by stages of the socioeconomic lifecourse. Social Science & Medicine 1997;44(6):809-819. 383. Picavet HJL. National health surveys by mail or home interview: Effects on response. Journal of Epidemiology and Community Health 2001;55(6):408-413. 384. Pearce J, Witten K, Hiscock R, Blakely T. Are socially disadvantaged neighbourhoods deprived of health-related community resources? International Journal of Epidemiology 2007;36:348-355. 385. Apparicio P, Cloutier M-S, Shearmur R. The case of Montral's missing food deserts: Evaluation of accessibility to food suprmerkets. International Journal of Health Geographics 2007;6(1):4. 386. Frey WH. Status selective white flight and central city population change: a comparative analysis. Journal of Regional Science 1980;20(1):71–88. 387. Pritchard WN. Beyond the modern supermarket: geographical approaches to the analysis of contemporary Australian retail restructuring. Australian Geographical Studies 2000;38(2):204-218. 388. Badcock B. Restructuring and spatial polarization in cities. Progress in Human Geography 1997;21(2):251-262. 389. Carstairs V. Small area analysis and health service research. Community Medicine 1981;3:131-139. 390. Krieger N, Chen JT, Waterman PD, Soobader MJ, Subramanian SV, Carson R. Geocoding and monitoring of US socioeconomic inequalities in mortality and cancer incidence: does the choice of area-based measure and geographic level matter? American Journal of Epidemiology 2002;156(5):471-482. 391. Geronimus AT, J. B. Use of census-based aggregate variables to proxy for socioeconomic group: evidence from national samples. American Journal of Epidemiology 1998;148(5):475-86. 392. Davey-Smith G, Ben-Shlomo Y, Hart C. Use of census-based aggregate variables to proxy for socioeconomic groups: evidence from national samples (letter to the editor). American Journal of Epidemiology 1999;150(8):996-997. 393. Diez-Roux AV. Multilevel analysis in public health research. Annual Review of Public Health 2000;21:171-192. 394. Diez-Roux AV. A glossary for multilevel analysis. Journal of Epidemiology and Community Health 2002;56:588-594. 395. Duncan C, Jones K, Moon G. Context, composition, and heterogeneity: using multilevel models in health research. Social Science and Medicine 1998;46(1):97-117.

cxxxii

396. Blakely TA, Woodward AJ. Ecological effects in multilevel studies. Journal of Epidemiology and Community Health 2000;54:367-374. 397. Greenland S. A review of multi-level theory for ecological analysis. Statistics in Medicine 2002;21:389-395. 398. Hox J, Kreft I. Multilevel analysis methods. Sociological Methods and Research 1994;22:283-299. 399. Susser M. The logic in the ecological: I. The logic of analysis. American Journal of Public Health 1994;84(825-829). 400. Susser M. The logic in ecological: II. The logic of design. American Journal of Public Health 1994;84(5):830-835. 401. Chaix B, Merlo J, Subramanian SV, Lynch JW, Chauvin P. Comparison of a spatial perspective with a mutlilevel analytical approach in neighbourhood studies: the case of mental and behavioural disorders due to psychoactive substance use in Malmo, Sweeden, 2001. American Journal of Epidemiology 2005;162(2):171-182. 402. Legendre P. Spatial autocorrelation: Trouble or new paradigm? Ecology 1993;74(6):1659-1673. 403. Cliff AD, Ord JK. Spatial processes : models & applications. London: Pion; 1981. 404. Goldstein H. Multilevel cross-classified models. Sociological Methods and Research 1994;223(3):364-375. 405. Langford IH. Multilevel modellijng of the geographical distributions of disease. Applied Statistics 1999;48(2):253-268. 406. Australian Bureau of Statistics. 2006 Census: SEIFA 2006. In. Canberra: Australian Bureau of Statistics; 2007. 407. Australian Bureau of Statistics. 6401.0 TABLE 14. CPI: Groups, sub-groups and expenditure class, percentage change from corresponding quarter of previous year by capital city. In. Canberra: Australian Bureau of statistics; 2007. 408. Berry P, Hogan J, Peat A, Ryan A, Vincent J, Connely M, et al. Australian Food Statitstics 2006. Canberra: Department of Agriculture Fishereis adn Forestry; 2006. Report No.: 1444-0458. 409. Eikenberry N, Smith C. Healthful eating: perceptions, motivations, barriers, and promoters in low-income Minnesota communities. Journal of the Amercian Dietetic Association 2004;104(7):1158-1161. 410. MacFarlane A, Crawford D, Ball K, Savige G, Worsley A. Adolescent home food environments and socioeconomic position. Asia Pacific Journal of Clinical Nutrition 2007;16(4):748-756. 411. Arcan C. Parental eating behaviours, home food environment and adolescent intakes of fruits, vegetables and dairy foods: longitudinal findings from Project EAT. Public Health Nutrition 2007;10(11):1257-1265. 412. Gallaway MS. Psychosocial and demographic predictors of fruit, juice and vegetable consumption among 11-14-year-old Boy Scouts. Public Health Nutrition 2007;10(12):1508-14. 413. Klepp K-I, Rodrigo CP, Thorsdottir I. Promoting and sustaining health through increased vegetable and fruit consumption among European schoolchildren: The Pro Children Project. Journal of Public Health 2005;13(2):97-101. 414. Tucker K, Mayer J. Eat a variety of healthful foods: old advice with new support. Nutrition Reviews 2001;59(5):159-158. 415. Krebs-Smith SM, Scott Kantor L. Choose a variety of fruits and vegetables daily: understanding the complexity. The Journal of Nutrition 2001;131(2s):s487-s501.

cxxxiii

416. Reynolds J. Nutrition education and the technology learning area. Journal of the HEIA 1998;5(3):18-36. 417. Brennan M. Debates on schooling: the futures we have to have or the futures we might stand up for? Journal of the HEIA 2001;8(1):37-44. 418. Dommers E. The health promotive school - a supportive environment for food and nutrition education. Journal of the HEIA 1995;2(4):39-40. 419. McGregor SLT. The impact of different paradigms on home economics practice. Journal of the HEIA 1997;4(3):23-33. 420. Wrieden W, Anderson AS, Longbottom P, Valentine K, Stead M, Caraher M, et al. The impact of a community-based food skills intervention on cooking confidence, food preparation methods and dietary choices - an exploratory trial. Public Health Nutrition 2007;10(2):203-211. 421. Anas A. A unified theory of consumption, travel and trip chaining. Journal of Urban Economics 2007;62(2):162-186. 422. Stafford M, Cummins S, Ellaway A, Sacker A, Wiggins RD, Macintyre S. Pathways to obesity: Identifying local, modifiable determinants of physical actiivty and diet. Social Science & Medicine 2007;65:1882-1897. 423. Krosnick JA. Survey research. Annual Review of Psychology 1999;50(1):537-567.

Food accessibility, affordability, cooking skills and ... · United Kingdom ..... 41 Overall quality and extent of the evidence for socioeconomic differences in-store ... purchase

Documents