Understanding differences in life satisfaction between local authority areas: Analysis based on the 2011/12 Annual Population Survey
Post on 01-Nov-2014
818 Views
Preview:
DESCRIPTION
Transcript
1
Understanding differences in life satisfaction between local authority areas
Analysis based on the 2011/12 Annual Population Survey
21 October 2013
Baljit Gill and Rob GreenStatistical and Spatial Analysis Team, Analysis and Innovation Directorate, DCLG
2
Contents
1 Summary 3
2 Introduction 5
3 Background 10
4 Setting the scene: differences in life satisfaction between local authority areas
15
5 Characteristics of places with low or high life satisfaction: correlations
20
6 Understanding differences in life satisfaction between local authority areas: regression modelling
29
7 Annex: Underlying models and the method of modelling 37
3
1. Summary
Nationally, the distribution of people’s life satisfaction ratings on a scale of 0 to 10 (where ‘10’ is completely satisfied) is highly skewed and there is a long tail of low life satisfaction: 7% of the population rate their satisfaction as 0 to 4.
This distribution is relevant in local policy making and delivery to address wellbeing issues: should efforts be focused on those people who are experiencing the lowest levels of wellbeing? Are there relatively cost-effective things that can be done to maintain or improve everyone’s wellbeing?
But this distribution can differ markedly between places. In 2011/12, the proportion rating their satisfaction from 0 to 4 ranged widely between districts, from 1% to 16%, and the proportion who were almost or completely satisfied with their lives (rating their satisfaction as a 9 or a 10) ranged from 11% to 43%.
Differences in self-reported wellbeing between local authority areas can be large. In 2011/12, average ratings of life satisfaction among local authority districts ranged between 7.1 to 8.1 (out of 10). This one-point difference in average life satisfaction is almost as large as that between employed and unemployed people.
ONS have found the biggest drivers of life satisfaction are people’s employment status, their health, and their relationships (in terms of their marital status). Insofar as ONS have been able to explain differences between local authority areas in the life satisfaction levels of their residents, they find the differences between areas reflect the circumstances of the types of people who live there, with an additional positive effect on satisfaction associated with living in rural areas.
Distribution of Life Satisfaction (England)
0
5
10
15
20
25
30
35
0 1 2 3 4 5 6 7 8 9 10
Life Satisfaction
%
Very Low: 7%
Low: 18%
Medium: 50%
High: 26%
Completely satisfied
Not at all satisfied
4
Summary continued
This analysis finds that the characteristics of places where a relatively high proportion of residents report very low life satisfaction correspond well with ONS’s key drivers; these were generally areas with higher levels of poor health, disability or unemployment, and relatively few people living in small market towns.
Actions to increase employment and improve health outcomes should therefore lead to improved wellbeing. Other initiatives to mitigate the detrimental impacts of ill-health, disability or unemployment on wellbeing could be considered. This could include delivering services in ways which meet people’s needs for social contact and building social networks, and building their confidence. Actions could also focus on tackling anti-social behaviour and crime. The findings suggest a key role for Health and Wellbeing boards.
LA areas where a relatively high proportion of residents report high life satisfaction are characterised by having large rural or market town populations, a relatively high proportion of retired people or younger couples without dependent children, and relatively few women working full time in the workforce. In addition to these factors, residents in these districts tended to report a high sense of belonging to their neighbourhood.
It may be possible to maintain or increase life satisfaction from moderate to high levels by focusing on increasing people’s sense of belonging to their neighbourhoods and considering how some of the benefits of rural or market town living can be ‘designed into’ urban areas. This might include actions such as building community spirit and promoting volunteering; reducing fear of crime; supporting wider community wellbeing through the design of housing and the built environment; maintaining a thriving high street and building a sense of distinct identity in areas.
Maintaining or increasing good employment opportunities is important in areas of high life satisfaction too, to meet the needs of all segments of the population, such as young people who might otherwise need to move away from the area.
The analysis presented here explores differences between local authority districts in how residents rate their life satisfaction in terms of the types of people who live there and attributes of the places themselves. The findings suggest that actions to improve life satisfaction should be tailored to populations with very low and high life satisfaction.
%
0
5
10
15
20
25
30
35
0 1 2 3 4 5 6 7 8 9 10
Distribution of Life Satisfaction
4
5
2. Introduction
6
Introduction
The first ONS annual estimates of personal wellbeing were published in July 2012 and showed considerable variation between the 90 county and unitary authorities in the UK.
Our aim:DCLG wish to understand how far differences between places can be explained by the
characteristics of people who live there, and to what extent there are specific features of places which may influence wellbeing.
Approach• We have focused on the life satisfaction
measure of wellbeing from ONS’s 2011/12 survey.
• We explore mean, high and very low levels of life satisfaction by exploring characteristics of areas that are correlated with these, and using linear regression.
• We analyse life satisfaction at local authority district level.
The purpose of this paper is to analyse the factors that are associated with wellbeing, as measured by life satisfaction, at local authority level.
Along with other research on wellbeing by the ONS and Cabinet Office, this exploratory analysis can support discussions with local authorities and Health and Wellbeing boards as they seek to understand wellbeing in their areas and identify actions to improve it.
This analysis aims to cast some light on differences between places alongside ONS’s publication of 23 October 2013 which, for the first time, gives official estimates of wellbeing at LA district level.
We hope it will inspire others to drill more deeply into the data to understand differences in wellbeing between areas and the role of place. ONS have already made a substantial contribution through their ‘What matters most to personal wellbeing’ report and further work is underway to enhance this modelling with an expanded set of place characteristics.
District councils are the main administrative bodies in local government. Where data are sufficiently robust, ONS estimates describe average levels of wellbeing and the distribution in each District. This is a more
granular level of analysis than the county/unitary level reported by ONS in 2012. Districts are more
heterogeneous than counties with respect to wellbeing. For example, ONS estimates from the 2011/12 survey
show West Midlands metropolitan county to have particularly poor wellbeing, but this area includes such varied districts/cities as Solihull, Wolverhampton and
Coventry.6
7
ONS measures of personal wellbeing
As part of the Measuring National Wellbeing Programme, the Office for National Statistics has developed a set of experimental survey-based measures of wellbeing.
The first annual estimates of national wellbeing, published in July 2012, are based on the Annual Population Survey (APS) 2011-2012. This is a very large UK-wide survey of adults living in private households. There were 122,000 respondents In England.
Adults rated their life satisfaction on a scale from 0 to 10 in answer to the question: Overall, how satisfied are you with your life nowadays?
Adults were also asked to rate their wellbeing according to their feeling that the things they do in life are worthwhile, and their levels of happiness and anxiety yesterday.
As with any survey estimate, wellbeing estimates are subject to random sampling error. As such the confidence intervals around wellbeing estimates for local authority areas, particularly at district level, are generally very wide. This means that many of the apparent differences between local authorities are not statistically significant.
Nevertheless, it is possible to explore the data to identify patterns in how respondents rated their wellbeing in different areas and the socio-demographic characteristics of the area or particular attributes of the places themselves.
The APS is a large sample survey of adults aged 16 and over in private households in the UK. The survey is particularly focused on measuring economic activity but includes questions on subjective wellbeing. All interviews in England were conducted face-to-face. Wellbeing questions were not asked of another household member by proxy. The survey is boosted in some local authorities to provide greater numbers for analysis, for example, of economic activity. Fieldwork was conducted from mid-April 2011 to mid-April 2012.
8
Analysis presented here is based on the 2011/12 Annual Population Survey (APS)
The analysis presented here is based on the 2011/12 APS and corresponds with the life satisfaction estimates published in July 2012, but uses unpublished data at the more detailed geography of LA districts.
On 23 October 2013, ONS published estimates for LA districts for the first time. As the estimates in both years are based on a sample survey, they will be subject to some degree of random sampling error, reflected in the confidence intervals around the estimates.
Estimates for each local authority area will differ between survey years simply because they are based on different random samples of respondents.
Therefore, in presenting and exploring LA wellbeing levels based on ONS’s 2011-12 APS, DCLG intentionally does not name specific local authorities.
9
Key features of the DCLG analysis
Our analysis focuses on life satisfaction rather than other measures of personal wellbeing. • Life satisfaction is measured by responses to the survey question: Overall, how
satisfied are you with your life nowadays? Respondents rate their satisfaction on a scale from 0-10, where 10 is ‘completely satisfied’, and 0 is ‘not at all satisfied’.
The analysis is conducted at local authority level only• This is not an analysis of individual responses to the Annual Population Survey. • Rather, we summarise wellbeing levels in each local authority area by aggregating the
wellbeing ratings given by APS respondents living in each LA. • There are 324 local authority districts in England (excluding City of London). These
include London Boroughs, metropolitan and non-metropolitan district councils, and unitary councils.
There are three summary measures of life satisfaction used in our analysis to describe the average level of wellbeing in the LA area and the extent to which people experience extreme levels of wellbeing within the LA:
• mean rating of life satisfaction in the LA area. Nationally, this is 7.4. • proportion of adults in the LA area rating their wellbeing as high (9 or 10 out of
10). Nationally, 26 per cent of adults rate their wellbeing at these levels. • proportion of adults in the LA area rating their wellbeing as very low (0 to 4 out of
10). This corresponds to the long tail of low wellbeing, which contains 7 per cent of the population nationally.
10
3. Background
11
The national distribution of life satisfaction
The Annual Population Survey estimates the average (mean) rating for life satisfaction among adults in England is 7.4 on a scale from 0 to 10, where 10 is completely satisfied.
The national distribution shows half of adults rated their life satisfaction 7 or 8, with over a quarter giving a higher rating of 9 or 10.
The distribution is highly skewed: a quarter of adults rated their life satisfaction less than 7, but there is a notably long tail of low life satisfaction with 7% of adults rating it between 0 (meaning not at all satisfied with their life) to 4.
As the following slides show, the average level of life satisfaction and the distribution of life satisfaction varies between local authority areas.
This distribution is relevant when thinking about what, if anything, government should do to raise levels of wellbeing: should government and local authorities focus on people in the low wellbeing tail? Are their relatively cost-effective things that can be done to maintain or improve everyone’s wellbeing?
Distribution of Life Satisfaction (England)
0
5
10
15
20
25
30
35
0 1 2 3 4 5 6 7 8 9 10
Life Satisfaction
%
Very Low: 7%
Low: 18%
Medium: 50%
High: 26%
Completely satisfied
Not at all satisfied
1212
ONS regression analysis identified the key drivers of life satisfaction
ONS have explored the variation in personal wellbeing among APS respondents using regression modelling.
Taking a range of characteristics into account, ONS find the strongest factors associated with the life satisfaction of individuals are:
• Health• Employment status: being employed or retired• Relationship status: being married or a couple• Being content with employment choice/situation
Age has a moderate effect: There is a U shaped relationship between life satisfaction and age even after taking health into account, with the highest satisfaction ratings reported by young adults and older people
There are smaller and negative effects associated with:
• Living alone• Being Black African/Caribbean/Black British• Having no religious affiliation
Among employees: Higher wages help life satisfaction (but not happiness, anxiety or feeling life is worthwhile).
Urban/rural: generally across regions, life satisfaction is higher in rural rather than urban areas, even when other personal characteristics have been taken into account; and Londoners are found to have similar life satisfaction to people in other urban areas in GB.
While there are differences in satisfaction levels between local authority areas, ONS find that most of the variation in satisfaction ratings occurs between individuals, wherever they live, rather than between neighbourhoods or local authorities. ONS plan further to explore the characteristics of places associated with wellbeing.
ONS publications link: http://www.ons.gov.uk/ons/guide-method/user-guidance/well-being/publications/index.html What matters most to personal wellbeing: http://www.ons.gov.uk/ons/dcp171766_312125.pdf
1313
Cabinet Office regression analysis identifies differences in the drivers of low and high satisfaction
Exploratory analysis by Cabinet Office builds on the ONS regression modelling by exploring the factors which are associated with the likelihood of an individual rating their wellbeing as a) very low (0 to 4), or b) high (9,10). Separate individual level models were constructed for these extremes of the wellbeing scale.
This innovative analysis provides useful insight into differences in the characteristics of people who are at the extremes of the wellbeing distribution.
The factors found to be most strongly associated with the likelihood of an individual reporting their life satisfaction as 0 to 4 or 9 to 10 correspond well with those identified by ONS in relation to mean life satisfaction. These were:
• Unemployment – relative to being retired, this was a risk factor for both very low and high satisfaction i.e. being unemployed is associated with a higher likelihood of rating one’s satisfaction as very low and a lower likelihood of rating one’s satisfaction as high. Underemployment (wanting to work more hours) was a stronger risk factor for high satisfaction.
• Marital status – compared with being married, being divorced, separated, widowed or single was a risk factor for levels of very low and high life satisfaction. Being separated was a stronger risk factor for low satisfaction.
• Health – poor health was a risk factor for both levels of very low and high life satisfaction, but it was a stronger risk factor for rating satisfaction as very low.
The term risk factor is used to describe a factor which increases levels of low wellbeing or decreases levels of high wellbeing.
A protective factor is good for levels of low wellbeing (ie it reduces them), or good for levels of high wellbeing (ie it increases them).
1414
Cabinet Office regression analysis identifies differences in the drivers of low and high satisfaction (continued)
There were other shared risk factors for both very low and high life satisfaction. Some shared risk factors were stronger risks for low satisfaction than for high: having poor health or a disability, being in social housing compared with owning outright, being separated, being a smoker, being in receipt of benefits, or being in a Black or minority ethnic group (BME). In contrast, being underemployed was a stronger risk factor for high satisfaction than for very low satisfaction.
Living with dependent children was a protective factor for both low and high satisfaction, but was a stronger protective factor for low satisfaction.
The modelling identified protective place-related and community-based factors which were associated with rating one’s satisfaction as high, but were not associated with rating one’s satisfaction as very low. These were:
• The proportion of people in the respondent’s local authority who feel a sense of belonging to their area
• The proportion of people in the respondent’s local authority who feel they can influence local decisions
• Rurality (living in a hamlet or village, rather than in an urban area)• Having moved to a new address in the past year
15
4. Setting the scene: differences in life satisfaction between local authority areas
16
Average life satisfaction ratings vary among unitary and county councils
6.2
6.4
6.6
6.8
7.0
7.2
7.4
7.6
7.8
8.0
8.2
Rut
land
UA
Cor
nwal
l UA
Cum
bria
Bat
h an
d N
orth
Eas
t Som
erse
t UA
Wes
t Sus
sex
Che
shire
Eas
t UA
Dor
set
Wilt
shire
UA
Shr
opsh
ire U
AN
orth
Yor
kshi
re
Win
dsor
and
Mai
denh
ead
UA
Brig
hton
and
Hov
e U
AE
ast R
idin
g of
Yor
kshi
re U
AH
amps
hire
Dev
onS
uffo
lkY
ork
UA
Her
tford
shire
Wok
ingh
am U
AG
louc
este
rshi
reD
arlin
gton
UA
Cen
tral
Bed
fors
hire
UA
Sur
rey
Poo
le U
AW
arrin
gton
UA
Sto
ckto
n-on
-Tee
s U
AN
orth
ampt
onsh
ireW
orce
ster
shire
Ken
tIs
le o
f Wig
ht U
AN
orth
Som
erse
t UA
Ply
mou
th U
A
Che
shire
Wes
t and
Che
ster
UA
Nor
folk
Som
erse
tS
win
don
Cou
nty
Dur
ham
UA
Der
bysh
ireLe
ices
ters
hire
Her
efor
dshi
re, C
ount
y of
UA
Buc
king
ham
shire
Oxf
ords
hire
Nor
thum
berla
nd U
ALa
ncas
hire
Nor
th L
inco
lnsh
ire U
AE
ssex
Red
car
and
Cle
vela
nd U
ALi
ncol
nshi
reE
ast S
usse
xH
artle
pool
UA
Sta
fford
shire
Tel
ford
and
Wre
kin
Sou
tham
pton
UA
Hal
ton
UA
Leic
este
r U
AN
ottin
gham
UA
Wes
t Ber
kshi
re U
A
Kin
gsto
n U
pon
Hul
l, C
ity o
f UA
Nor
th E
ast L
inco
lnsh
ire U
AB
rack
nell
For
est U
AB
ourn
emou
th U
AM
iddl
esbr
ough
UA
Por
tsm
outh
UA
Tyn
e an
d W
ear
Met
Cou
nty
Not
tingh
amsh
ireC
ambr
idge
shire
Wes
t Yor
kshi
reS
outh
end-
on-S
ea U
AM
ilton
Key
nes
UA
Rea
ding
UA
Bris
tol,
City
of U
AS
outh
Yor
kshi
reS
toke
-on-
Tre
nt U
AS
outh
Glo
uces
ters
hire
UA
Pet
erbo
roug
h U
AM
erse
ysid
e M
et C
ount
yB
edfo
rd U
AO
uter
Lon
don
Slo
ugh
UA
Gre
ater
Man
ches
ter
Der
by U
AW
arw
icks
hire
Points show the mean rating. The bars show the 95% confidence intervals around this average, indicating a margin of error. The confidence intervals are wider where survey sample sizes were smaller. Where the confidence intervals do not overlap, the differences between areas are statistically significant. (This is an approximate test: technically: even where the bars overlap a little, the differences between places may also be statistically significant).
Decreasing wellbeing
Source: ONS Annual Population Survey 2011-12, England
Average life satisfactionEngland: 7.40
Mean rating given by adults to the question: 'Overall, how satisfied are you with your life nowadays?' on a scale of 0-10 where 10 is completely satisfied
17
Local authority area distributions of life satisfaction can differ markedly
On the following slide, we present examples of places with different distributions which can broadly be segmented into the following groups based on their life satisfaction levels:
“Polarised”
Higher proportions of both very low and high wellbeing
“Struggling”
Large proportions of low wellbeing, little high
“Moderate”
Mainly ‘middle’ wellbeing, few at the extremes
“Satisfied”
Large proportions of high wellbeing, little low
The proportion of adults rating their life satisfaction as high (9,10) ranges across LA districts from 11% to 43% (average = 27%)
The proportion of adults rating their life satisfaction as very low (0-4) ranges across LA districts from 1% to 16% (average = 6.3% ).
18
Examples of LA areas with different life satisfaction distributions
Satisfied: West Sussex, largely rural population
Struggling: West Midlands, entirely urban population Moderate: Warwickshire, largely urban population
Polarised: Staffordshire, largely urban population
0
5
10
15
20
25
30
35
40
45
0 1 2 3 4 5 6 7 8 9 10
%
0
5
10
15
20
25
30
35
40
45
0 1 2 3 4 5 6 7 8 9 10
%
0
5
10
15
20
25
30
35
40
45
0 1 2 3 4 5 6 7 8 9 10
%
0
5
10
15
20
25
30
35
40
45
0 1 2 3 4 5 6 7 8 9 10
%
Each chart shows the distribution of life satisfaction ratings among survey respondents in the LA.
Respondents rated life satisfaction from 0 to 10 (0 = 'not at all satisfied’; 10 = ‘completely satisfied’).
19
Segmenting local authority areas according to their levels of very low and high life satisfaction
In this scatter diagram, we plot the levels of high and very low life satisfaction for each local authority district.
•Districts higher up the vertical axis have a greater proportion of their residents rating life satisfaction as high (9,10).
•Districts to the right of the chart have a smaller proportion of their residents rating the life satisfaction as low (0-4). (The horizontal axis, representing the proportion of adults with very low life satisfaction, is reversed.)
• Districts with average levels of satisfaction on both dimensions are plotted around the centre of the diagram.
•All estimates are subject to fairly wide confidence intervals on both dimensions.
Observations:
Districts with the greatest proportions of residents rating their satisfaction as high are almost exclusively non-metropolitan districts.
Few places have truly polarised populations, with large proportions of people giving ratings at both extremes (0-4 and 9-10).
-16
0
16
-16016
% of residents with very low life satisfaction
(mean centred; axis is reversed)
Incr
ea
sin
g %
of
resi
de
nts
with
hig
h li
fe s
atis
fact
ion
Met district/London borough
Unitary council
Non-met district
Struggling places: low wellbeing, little high wellbeing
Polarised places: low wellbeing and also high wellbeing
Satisfied places: high wellbeing with little low
wellbeing
Moderate places: neither low nor high wellbeing
Decreasing % of residents with very low life satisfaction
% of residents with very high life satisfaction
(mean centred)
19
20
5. Characteristics of places with low or high life satisfaction: correlations
21
Introduction to this section
This section begins by identifying the factors which are most strongly correlated with life satisfaction at the LA district level as measured by:
• the mean rating of life satisfaction in the LA area. • proportion of adults in the LA rating their life satisfaction as high (9 or 10 out of 10). • proportion of adults in the LA rating their life satisfaction as very low (0 to 4 out of 10)
These correlates can inform the potential levers for improving wellbeing.
A key statistic is the correlation coefficient, ρ (rho), which describes the degree to which two factors are associated. Because the work presented here is area-based analysis, DCLG consider a correlation of 0.4 or higher as large, and a correlation of 0.5 or higher as very large.
DCLG explored a wide range of factors as described in the following two slides.
We begin by examining the strongest correlates with average levels of life satisfaction at LA level. We then contrast the factors most strongly correlated with levels of a) very low and with levels of b) high wellbeing among LAs.
The analysis in this section is descriptive, and many of the factors are also correlated with each other. For example, in areas where unemployment rates are high, there will generally also be higher levels of ill-health or disability, and higher levels of crime. In the next section, we present the findings of regression modelling which asks: of the many characteristics that are correlated with life satisfaction at LA level, which ones – when taken together – explain the differences between LA areas?
All findings indicate association rather than causation. Causality is particularly difficult to establish in relation to attributes of places: while certain attributes such as the amount of green space or having coastline may be conducive to high wellbeing, where family and other ties permit, people who value quality of life/high wellbeing and have the means to afford it, will elect to live in areas with particular attributes. Further, places where wellbeing levels are high may meet the needs of some segments of the population better than others, and this may cause some people – with lower wellbeing – to leave the area. For example, young people may move out of areas with high average wellbeing for employment, better pay or career opportunities.
21
22
The association between life satisfaction levels across LA districts and a wide range of compositional and contextual factors
The following slide presents the wide range of factors that have been considered in this work as potential drivers of levels of life satisfaction in LA areas.
We have endeavoured to collect a range of both compositional and contextual indicators for each local authority district, across a range of sources. Much of the data is aggregated up from individual respondents to the Annual Population Survey within each LA. Other sources include Census 2011, NOMIS for employment data, Place Survey 2008 for residents’ perceptions of their area, and a range of administrative or GIS data.
The factors considered cover the domains of wellbeing identified by ONS in their framework for measuring national wellbeing (left), and incorporate local authority measures from the Public Health Outcomes Framework.
In gathering data for this analysis, particular attention has been given to attributes of places, in terms of people’s perceptions of the place and social capital/behaviours, and also characteristics of the living or natural environment. The most notable omissions are local indicators of mental health or the strength of social networks/quality of relationships.
A compositional indicator is one which reflects the personal circumstances of people living in each LA area. A contextual factor reflects the characteristics or attributes of the area itself.
22
23
Potential drivers of life satisfaction in local authority areas
Unless stated otherwise factors are the proportion of adults in the LA with the particular characteristic
Broad group Factor
1 - Self reported health Self-reported health (5 point scale)% of 16-64 year olds that are economically active/inactive (NOMIS)Economic activity by type: e.g. Retired, Employee, Unemployed% of employed or self-employed that work part time, full time, long hours% of employed or self-employed that work part time, full time, long hours - split by male/female% of employed or self-employed that are male/femaleWant new, additional job or longer hoursPermanent/non-permanent jobCouple/not couple; with/without dependent children - split by age groupWith dependent children - by numberMarital statusLives alone
5 - Age Age in bands, mean age, median age6 - Ethnicity Ethnicity split into 9 groups
Tenure (owned outright, with a mortgage, private renter, social renter)% of dwellings that are flats% of dwellings that are HMOs (Houses of Multiple Occupation)% of households with overcrowding in rooms/bedrooms
8 - Religion Religion (8 groups)9 - Socio-economic status NS-SEC Socio Economic Classification (8 levels - main or last main job)10 - Highest qualification Highest qualification achieved (7 levels)11 - Gender Female
Day to day activities limited by health problem or disabilityLegally and work limiting disabledDisability Living Allowance claimants% of households with no/some/all adults speaking English as main languageRecent immigrant or born in UKResident in UK - by number of years (4 bands)Volume of total migration per 1,000 populationVolume of international migration per 1,000 population
14 - Unpaid carers Informal carers by hours per week (4 bands)Perceive that anti-social behaviour is a problem in their local areaPerceive that drug dealing is a problem in their local areaPerceive that drunk or rowdy behaviour is a problem in their local areaPerceive that respect is a problem in their local areaRobbery offences recorded per 1,000 population Burglary dwelling offences recorded per 1,000 population Violence against the person offences per 1,000 population
12 - Disability
13 - Migration
15 - Crime/anti-social behaviour
2 - Economic activity
3 - Employment details
4 - Living arrangements/relationships
7 - Tenure and dwelling
Broad group Factor
School appeals heard as % of admissions - primary/secondarySchools net imports as % of school populationHousing rent - median, average, lower quartileHouse prices - median, lower quartile, ratio to earnings% of pupils achieving at 5 or more GCSEs at grades A*-C
17 - Time at same address Length of time at address - by number of years (6 bands)Life expectancy and disability free life expectancySlope Index of Inequality (SII) for life expectancy at birthAge standardised mortality ratesAverage weekly net equivalised household income before/after housing costs% of pupils eligible for free school mealsIndices of DeprivationGross disposable household income per head (by counties/groups of unitary authorities)Gross Value Added per head (by counties/groups of unitary authorities)% of 16-18 year olds Not in Education, Employment or Training% household spaces with no usual residentArea sizeDwelling densityPopulation densityArea of domestic gardens - as a % of area of domestic buildingsArea of greenspace - as a % of total areaArea of water - as a % of total areaAuthorities with coastline% Population by urban/rural area typePopulation sizeUrban-Rural classification (6 groups)
22 - Commuting distance Commuting distance (9 bands) - for 16-74 year olds employed in the LA23 - Governance Type of Local Authority (4 types)
Civic participationPerceive that parents take responsibility for behaviour of childrenRegular formal volunteeringParticipate in sport (1 x 30 mins per week; 3 x 30 mins per week)Perceive that people from different backgrounds get on well (cohesion)Feel that they belong to immediate neighbourhoodSatisfied with local areaPerceive fair treatment by local servicesPerceive ability to influence local decisions
21 - Natural environment classifications
24 - Area perceptions/behaviours
16 - Area desirability/prices
18 - Health inequality and life expectancy
19 - Area deprivation/income
20 - Natural environment attributes
24
At local authority level, the factors most strongly correlated with mean life satisfaction are as follows
Very high correlations (coefficient of 0.5 up to 0.6, positive or negative)
• Unemployment (-0.50); and households with dependent children and no adult in employment (-0.57)
• Indicators of income and area deprivation. IMD crime deprivation (-0.53) and income deprivation (-0.50)
• Women working full time (-0.50) and people working from home (0.50)
• Couples with no dependent children (0.51) and lone parent households (-0.55)
• Perceptions of anti-social behaviour (-0.56) and that respect is a problem (-0.54)
• Satisfaction with local area (0.55), sense of belonging to the neighbourhood (0.50), and perceptions that local services are fair (0.54)
• Population living in rural areas (0.51)
• Mean and median age (both 0.51). (Also the proportion of 60-64 year olds (0.51)
Highly correlated: correlation coefficient of 0.4 up to 0.5• % retired (0.42) or working full time (-0.45), % women working
full-time (-0.48). % working as small employers (0.42)• Living as a couple (0.46), single (-0.44), and % having 0-4
dependent children in household (-0.47)• Age groups: correlations were positive for age-groups from 45
and over (ranging between 0.42 and 0.46) and • negative for 25-29 (-0.47) and 30-34 (-0.40)• Tenure/housing: people own outright (0.49), social tenants (-
0.43), overcrowded bedrooms (-0.46)• Ethnicity/religion: % White (0.47), Mixed (-0.42), Asian (-0.42),
Black (-0.40); and Muslim (-0.43)• Migration: % resident in UK 5-10 years (-0.41); having some
or no adults in household with English as their main language (-0.41, -0.43)
• % providing 1 to 19 hours unpaid care per week (0.49). (Note: correlation for % providing no weekly unpaid care is negative (-0.38))
• % participating in regular formal volunteering (0.45)• Crime/ASB: burglary, robbery, violent offences (ranging
between -0.47 and -0.43); perceptions of drunk/rowdy behaviour or of drug use/dealing (both -0.49)
• Other perceptions: feeling that people from different backgrounds get on well i.e. cohesion (0.42) and that parents take enough responsibility for their children (0.42).
• Other indicators of income and area deprivation: proportion receiving free school meals (-0.45). IMD employment deprivation (-0.40)
• Natural environment: green space (0.49), population density (-0.42)
• Urban/rural classifications: % of population in rural or large market towns (0.47), in villages (0.48), in dispersed rural areas (0.45), or major urban authorities (-0.40) 24
25
% in LA area rating life satisfaction as very low or high by LA-level unemployment rate
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20%
Unemployment rate (NOMIS 2011-12, modelled)
Per
cen
tag
e o
f ad
ult
s ra
tin
g li
fe s
atis
fact
ion
as
very
lo
w o
r h
igh
% very low life satisfaction % high life satisfaction
25
At LA level, unemployment is the factor most strongly correlated with ratings of very low life satisfaction – but is less strongly correlated with high satisfaction
For very low satisfaction, other explanatory factors include: disability, economic inactivity due to ill health/disability; poor health; ASB.
ρ = 0.44
ρ = -0.25
ρ (rho) indicates the correlation coefficient, a statistic that describes the degree to which two factors are associated (in this case it describes the correlation between the level of high – or very low – life satisfaction with the unemployment rate). In the work presented here, DCLG consider a correlation of 0.4 or higher as large, and a correlation of 0.5 or higher as very large.
26
% in LA area rating life satisfaction as very low or high by LA-level self reported poor health
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
0% 2% 4% 6% 8% 10% 12% 14%
% of adults rating their health as bad or very bad (Census 2011)
Per
cen
tag
e o
f ad
ult
s ra
tin
g li
fe s
atis
fact
ion
as
ver
y lo
w o
r h
igh
% very low life satisfaction % high life satisfaction
26
Poor health is also highly correlated with low life satisfaction at LA level – but there is no apparent relationship with high life satisfaction
For high satisfaction, the most strongly correlated factors include: whether the area is urban/rural; % retired; and the extent of green space. But the factor with the strongest association with high wellbeing levels is shown overleaf.
ρ = -0.02
ρ = 0.41
ρ (rho) = correlation coefficient
27
% in LA area rating life satisfaction as very low or high by LA-level sense of belonging
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
20% 30% 40% 50% 60% 70% 80%
Sense of belonging (2008 Place Survey)
Per
cen
tag
e o
f ad
ult
s ra
tin
g li
fe s
atis
fact
ion
as
ver
y lo
w o
r h
igh
% very low life satisfaction % high life satisfaction
27
The extent to which people feel they belong to their neighbourhood is the factor most strongly correlated with LA level ratings of high life satisfaction
ρ = 0.52
ρ = -0.19
Sense of belonging was measured by the Place Survey 2008. Citizenship Survey analysis at the national level and qualitative case studies show sense of belonging is related to having strong social networks (such as having 3 or more close friends, or having family and friends in the area); being older, having lived in the area for a long time; a sense of community, cohesion, and local pride; and – particularly in deprived areas – feeling safe in the area.
ρ (rho) = correlation coefficient
28
Different factors are most strongly correlated to very low and high life satisfaction
For very low life satisfaction at LA level the strongest correlates are:
Large correlations:• Unemployment (0.44); households with dependent
children and no adult in employment (0.41)• Very bad or bad self reported health (0.42, 0.41)• Disability: % who are DDA and work disabled (0.44)• % giving 20 to 49 hours unpaid care per week (0.41). • Income deprivation (0.43) and employment deprivation
(0.44) • Perceptions of ASB (0.41), of drug use/dealing (0.42),
and that respect is a problem (0.40)• Satisfaction with local area (-0.41)
For high life satisfaction at LA level the strongest correlates are:
Very large correlations:• Sense of belonging to the immediate neighbourhood (0.52)• % 60-64 year olds (0.52)• Urban/rural classifications: proportion of population living in
rural areas (0.50)
Large correlations:• Economic activity and hours worked: % Retired (0.47);
Women working full time (-0.47); Small employers (0.42)• Relationships and living arrangements: couples with no
dependent children (0.47), singles (-0.42), 0-4 dependent children in household (-0.43), widows (0.42), lone parents in part time employment (0.42)
• Property owned outright (0.43)• Mean/median age (0.49/0.50). Correlations were positive for
age-groups from age 45 to 84 (range: 0.43 and 0.49) but negative for age-groups 25-29 (-0.43) and 30-44 (-0.44)
• Ethnicity: Mixed (-0.44), White (0.41) • Resident in UK 5-10 years (-0.41)• 1 to 19 hours unpaid caring per week (0.43). • Burglary offences (-0.41) and IMD Crime deprivation (-0.40)• Green space (0.45)• Urban/rural classifications: proportion of population living in
rural towns or large market towns (0.48), village population (0.45), dispersed rural population (0.46)
28
29
6. Understanding differences in life satisfaction between local authority areas: regression modelling
30
Introduction to this section
The previous section identified the factors which were most strongly correlated with life satisfaction at the LA level.
In this section, we present the findings of regression modelling. These seek to answer the question:
Of the many characteristics that are correlated with life satisfaction at LA level, which ones – when taken together – best explain differences between LA areas?
This work is undertaken as a series of three regression models looking at differences between LA areas according to:
• their mean life satisfaction,• the percentage of people rating life satisfaction as very low (0-4), and • the percentage of people rating life satisfaction as high (9,10).
DCLG tested a wide range of factors to explore the main compositional and place-related factors that are associated with life satisfaction. In building these models, we were strongly guided by ONS findings on what matters most to individual life satisfaction. This was to ensure the findings at local authority level were grounded in evidence on what drives individual life satisfaction.
Our findings suggest that, at the LA level, there are different factors associated with the extremes of the life satisfaction distribution. This is consistent with the findings from the Cabinet Office exploratory analysis presented on slides 13 and 14.
These findings are summarised here, but detailed models are presented in the annex, with a fuller discussion of the findings and the methods used.
As in the presentation of correlations, these findings from regression models indicate association rather than causation.
See slide 32 for an explanation for notes on the model and terms used in table, and slide 43 for a description of the method of modelling. 30
31
Mean life satisfaction: regression results
DCLG conducted regression modelling, testing a wide range of factors, guided by ONS findings on what matters most to individual life satisfaction. The regression model below shows that when the following factors about each LA area are considered together, they explain almost half of the variation observed in mean life satisfaction between districts (R square = 0.493). Factors with coefficients shown in red are negatively associated with mean life satisfaction i.e. they are associated with reduced average wellbeing.
Most of the factors associated with mean life satisfaction in LA areas are related to the personal circumstances of people living there, with the proportion of adults living in couples without dependent children being the most highly associated factor. Self-reported health, age, employment and hours worked, educational qualifications and ethnicity are also highly associated.
We also find the proportion of land covered by the LA district which is green space is a positive contextual factor, although this is in itself highly correlated with predominantly rural populations.
Alternative models find the proportion of people living in rural areas is a positive and significant factor when other factors are taken into account.
Economic activity as measured by Gross Value Added per head for the county/group of unitary councils is also positively associated with mean life satisfaction.
Regression model for mean satisfaction in each LA district
FactorUnstandardized
CoefficientsStandardized Coefficients
(Constant) 7.68 **% Living as a couple with no dependent children 0.75 ** 0.24% Aged 20-24 0.93 ** 0.20% of female employees that work 31-48 hours per week -0.98 ** -0.19% Greenspace in area 0.17 ** 0.17% Mixed ethnic group -3.37 ** -0.14Gross Value Added per head1 0.00 ** 0.14% Unemployed -1.08 * -0.13% Aged 80 & over 0.94 * 0.11% Self-reported health: bad -1.23 * -0.11% No qualifications -0.57 * -0.10% Semi-routine occupations -0.59 * -0.10% Self-reported health: very bad -2.04 * -0.10% would work more hours at same basic rate -0.52 * -0.10
(Model R-Squared = 0.493)* significant at 5% level; ** significant at 1% level
32
Notes on the model and terms used in the table
All findings indicate association rather than causation.
The model shows a high R-Sq (0.493) which indicates that almost half of the total variance in mean life satisfaction between local authority areas can be explained by factors shown in the table. This is consistent with other research at this high geographical level (e.g. US cities).
The factors identified as significant in the model are broadly consistent with factors identified by ONS in their modelling at individual level.
Where population characteristics do not vary substantially at local authority level, they will not influence differences in wellbeing at this level of analysis. This may explain why some factors which one might expect to be influential are not identified in this modelling.
While the model has identified specific factors as significantly associated with life satisfaction, this does not mean other factors or combinations of factors were not significant or that the factors selected are the underlying ones which drive satisfaction. Many of the factors entered into the models are inter-correlated. The combination of factors identified by the models, when taken together, offer the most explanatory power in understanding life satisfaction levels in places but they also reflect the priority they were given when factors were entered in steps during model building. We therefore built models giving priority to factors found to be important at the individual level, and entered place-level attributes only after compositional characteristics were taken into account. A number of alternative models were also considered where the underlying drivers were less clear.
Factors are ordered by their standardised coefficients to indicate the factors that are most strongly associated with life satisfaction.
Unstandardised coefficient: the size of this coefficient indicates the extent to which mean life satisfaction is expected to increase or decrease given a one-point increase in the independent factor, holding all other factors constant.
Standardised coefficient: this measure tells us the number of standard deviations that the outcome will change as a result of one standard deviation change in the predictor. It takes into account that each factor is distributed differently between local authority areas and may be measured on a different scale to other factors in the model. This is a useful statistic as the size of these coefficients tells us which factors are most strongly associated with the outcome.
33
Discussion of factors associated with average life satisfaction ratings
Age: Being younger (aged 20-24) or older (aged 80+) is positively associated with high average life satisfaction (after controlling for other factors).
Unemployment is negatively associated with mean life satisfaction, as is the proportion of people who would work more hours (at the same rate of pay) if they could. There are parallels with ONS findings on unemployment and underemployment being negative drivers of life satisfaction at the individual level.
We also find a negative association with the proportion of working (or self-employed) women who work 31-48 hours per week. This is consistent with findings which show that the proportion of people rating their satisfaction as high (9-10) in each LA area is negatively associated with the proportion of the workforce made up of women working such hours. These findings may indicate that satisfaction is highest in areas where women do not need to work full-time, for example, because the cost of living is lower, or where household wages tend to be higher.
Self-reported health: As one might expect, the proportion of people with bad or very bad self-reported health is negatively associated with mean life satisfaction.
• The coefficients show that a ten point increase in the percentage of people in the LA who report very bad health is associated with a reduction in mean life satisfaction of about 0.2, all other things being constant.
• There is some debate over the relationship between self-reported health and personal wellbeing. As ONS points out, Dolan et. al (2008)1 note some of the association may be caused by the impact that wellbeing has on health. Other research has shown that subjective evaluations of health matter more than objective measures in terms of the relationship with personal wellbeing (Brown et. al., 20102; Diener et al., 19993).
Ethnicity, qualifications and socio-economic status: Local authority composition in terms of the proportion in the Mixed ethnic group was significantly and negatively associated with mean life satisfaction. LA areas with a higher proportion of people without qualifications or working in semi-routine occupations also tend to have lower mean life satisfaction.
1. Dolan, P., Peasgood, T. and White, M. (2008). ‘Do we really know what makes us happy/? A review of the economic literature on the factors associated with subjective well-being’, Journal of Economic Psychology, 29, 94-1222. Diener, E., Suh, E. M., Lucas, R. E. and Smith, H. L. (1999). ‘Subjective well-being: Three decades of progress’, Psychological Review, 125, 276–3023. Brown, D., Smith, C. and Woolf J. (2010), ‘The Determinants of Subjective Wellbeing in New Zealand: An Empirical Look at New Zealand’s Social Welfare Function’, New Zealand Association of Economists Conference 2010, Statistics New Zealand
34
Discussion of factors associated with average life satisfaction ratings (continued)
Contextual effects: The proportion of green space is a contextual factor associated with higher average life satisfaction, but alternative models find the proportion of people living in rural areas is also a positive and significant factor when other factors are taken into account. This factor has a large association with the proportion of green space in the LA, which was selected in the model shown above.
Economic activity as measured by Gross Value Added per head for the county/group of unitary councils is positively associated with mean life satisfaction. This measure reflects the income generated by resident individuals or corporations in the production of goods and services. The coefficient is very small (rounded to zero) as the range is extremely wide.
35
Factors associated with very low and high life satisfaction scores
LA levels of very low life satisfaction are, broadly speaking, positively associated with compositional characteristics of: -% with poor health or a limiting disability-% unemployed-% of working men who work 16 to 30 hours per week-% separated or divorced/lone parents
Very low satisfaction levels are negatively associated with: -% living as a couple, or % retired-% living in large market towns-% living close to work.
LA levels of high life satisfaction are, broadly speaking, positively associated with:-a good sense of belonging-% in rural areas or market-towns-% retired, or younger couples without dependent children
High satisfaction levels are negatively associated with: -% women working full-time in the workforce-% unemployed
Of the many characteristics that are correlated with the percentage of people rating life satisfaction as very low (0 to 4) in each of the 324 LADs, which ones – when taken together – best explain differences between LA areas? And which ones best explain differences in the percentage of people rating life satisfaction as high (9,10)? To answer this, DCLG conducted regression modelling, testing a wide range of factors, guided by ONS findings on what matters most to individual life satisfaction. Our findings suggest that, at the LA level, there are different factors associated with the extremes of the distribution of life satisfaction. (Detailed findings are presented in the annex.)
%
0
5
10
15
20
25
30
35
0 1 2 3 4 5 6 7 8 9 10
Distribution of Life Satisfaction
35
3636
Implications for action to improve life satisfaction
To uplift very low life satisfaction, actions could include: -increasing employment; -early health interventions-mitigating the detrimental impacts of ill-health or disability on wellbeing e.g. by building social networks, or by supporting people in employment. -Other actions include tackling ASB and crime. This suggests a key role for Health & Wellbeing Boards.
To maintain or increase life satisfaction from moderate to high levels, actions could focus on:-increasing ‘belonging’ e.g. through volunteering and building community spirit, and reducing fear of crime. -addressing issues for full-time women workers, or other challenged groups.Can some aspects of rural living be created in urban areas for wellbeing benefits?
These regression findings, and the findings on correlations, suggest actions to improve life satisfaction should be tailored to populations with very low and high life satisfaction.
%
0
5
10
15
20
25
30
35
0 1 2 3 4 5 6 7 8 9 10
Distribution of Life Satisfaction
36
37
Annex: Underlying models and the method of
modelling
38
Modelling very low and high life satisfaction
On slide 31 we identified the main factors associated with mean life satisfaction. We now seek to understand the quite marked differences between local authorities in the proportions of their populations experiencing very low or high life satisfaction.
The purpose of this is to see how factors associated with life satisfaction at the extremes of the distribution differ. This may, in turn, suggest different sets of policy actions to improve wellbeing.
The research question is to identify the main factors associated with levels of very low life satisfaction and levels of high satisfaction, and to observe differences between the two sets of factors. The dependent variables used in regression modelling are:
• the proportion of people with very low life satisfaction – defined as those rating their life satisfaction from 0 to 4; and
• the proportion of people with high satisfaction – defined as those rating their life satisfaction as 9 or 10.
In some local authorities, our estimates for both of these dependent variables are based on small numbers of survey respondents and are therefore subject to relatively high levels of uncertainty. We have therefore adapted our approach to regression modelling from Ordinary Least Squares to Weighted Least Squares regression. There is no definitive way to assign weights but we have sought to ensure all types of council are included in our analysis while giving lower weight to those with less reliable estimates on the dependent variable (see slide 44).
As such, we recognise that the ensuing models are less robust than the one presented earlier for mean wellbeing. But in justification, the findings do indicate interesting differences in the main factors associated with very low and high life satisfaction.
39
Very low life satisfaction: regression results
Regression model for the percentage reporting very low satisfaction (0-4) in each LA district
FactorUnstandardized
CoefficientsStandardized
Coefficients(Constant) 0.03 *% Self-reported health: very bad 0.57 ** 0.25% Self-reported health: fair 0.14 ** 0.22% Unemployed 0.19 ** 0.21% Commuting distance 0-2km -0.09 ** -0.21% Self-reported health: bad 0.23 ** 0.19% Lone parents aged 45 or over 0.36 ** 0.14% Commuting distance 10-20km -0.07 * -0.12% of male employees that work 16-30 hours per week 0.12 * 0.10
(Model R-Squared = 0.379)* significant at 5% level; ** significant at 1% level
The model gives a good indication of the main factors associated with levels very low life satisfaction at LA level. But other factors could also be important, as described on the next slide.
40
Very low life satisfaction: findings from alternative models
Health: very low life satisfaction is positively associated with the proportions of adults reporting their health as very bad, bad or fair. Alternative models find that disability has an equally strong association with levels of very low life satisfaction as self reported health. The proportion of people with disabilities has a large association with the proportion reporting health as very bad, bad or fair. Self reported health factors were entered into the model before disability so had a greater chance of being selected in the model. If disability factors had been entered into the model before disability it is likely that they would have come out as significant factors themselves.
In addition to unemployment rates, we find the proportion of working men who work between 16 to 30 hours per week to be positively associated with low satisfaction levels in the LA area. Both of these may be indications of weak local employment conditions.
Areas with a high proportion of retired people are likely to have fewer people with low life satisfaction: the proportion of people who are retired has come up as a significant negative factor in alternative models. (A negative factor implies a reduction in the number of people with low life satisfaction.)
The proportion of people that are in couples and the proportion of people that are separated or divorced can also be considered to be associated with very low life satisfaction. These were found to be significant factors in alternative models.
Commuting distance and urban/rural indicators: The model finds that the proportions of people who have a short commute (of under 2 km) or a moderate commute of 10 to 20 km are negatively associated with very low life satisfaction (i.e. this factor reduces the number of people with low wellbeing). This finding may reflect the trade-offs people make between being close to work, and also living in areas where work remains accessible but where other amenities can be enjoyed such as larger houses and more green space.
Alternative models find that having a high proportion of residents living in large market towns is a significant negative factor associated with the level of low wellbeing in an area. Had urban/rural factors been entered into the above model before commuting factors, the proportion of people living in large market towns would have been identified as a significant factor.
41
The strongest factor – sense of belonging – relates to place attachment. This is the proportion of adults who feel they belong to the immediate neighbourhood, as measured by the Place Survey 2008.
Citizenship Survey analysis and qualitative case studies show sense of belonging is related to having strong social networks (such as having 3 or more close friends, and having family and friends in the area); being older, having lived in the area for a long time; a sense of community, cohesion, and local pride; and – particularly in deprived areas – feeling safe in the area.
The inclusion of ‘sense of belonging’ in the model did not add substantially to its explanatory power (adding only one percentage point to the proportion of variation in levels of high life satisfaction that were explained by the model). But the ‘sense of belonging’ factor was highly effective in drawing together a range of alternative models containing disparate factors. This reflects that levels of high satisfaction are associated with a diverse set of characteristics: some characteristics are influential in some LAs, but in other LAs, other characteristics are influential. The overall sense that people belong to the area emerges as a unifying factor among these areas.
High life satisfaction: regression results
Regression model for the percentage reporting high satisfaction (9-10) in each LA district
FactorUnstandardized
CoefficientsStandardized
Coefficients(Constant) 0.20 **% who feel they belong to their immediate neighbourhood 0.19 ** 0.24% Living as a couple, aged less than 45, with no dependent children 0.24 ** 0.19% Aged 30-34 -0.21 * -0.15At least 80% population in rural settlements and large market towns 0.02 ** 0.14% Retired 0.10 * 0.14% of employees that are female and work full time 31 to 48 hours -0.33 * -0.13
(Model R-Squared = 0.351)* significant at 5% level; ** significant at 1% level
42
High life satisfaction: further discussion of regression results
High life satisfaction levels are also associated with the most rural local authorities, where at least 80% of their population in rural settlements and large market towns (as opposed to living in more urban areas). This is the most rural of a six-band typology, with ‘major urban’ areas at the opposite extreme.
As with the model for average life satisfaction, we find the proportion of people living as a couple without dependent children in the household to be significant and positively associated with levels of high satisfaction – but here, it is based on those aged under 45 years.
As one might expect, having a high proportion of retirees is positively associated with levels of high satisfaction.
The proportion of people who are unemployed does not come up as a factor in the final models, but it only becomes non-significant when the proportion living in rural areas is taken into account. The proportion of people in the Mixed ethnic group has come up as a significant negative factor in alternative models.
‘Sense of belonging’, along with most of the other perceptions measures, were entered during the last step in modelling. This was to give compositional characteristics and attributes of areas priority in explaining differences in levels of high life satisfaction between areas.
Department of Communities and Local Government (2009). 2007-08 Citizenship Survey, Community Cohesion Topic Report http://resources.cohesioninstitute.org.uk/Publications/Documents/Document/DownloadDocumentsFile.aspx?recordId=149&file=PDFversion Department of Communities and Local Government (2011) Community Spirit in England: A report on the 2009-10 Citizenship Survey http://webarchive.nationalarchives.gov.uk/20120919132719/http:/www.communities.gov.uk/publications/corporate/statistics/citizenshipsurvey200910spirit Livingston, M., Bailey, N., and Kearns, A. (2008) People’s attachment to place – the influence of neighbourhood deprivation, Joseph Rowntree Foundation http://www.jrf.org.uk/system/files/2200-neighbourhoods-attachment-deprivation.pdf
43
Method of Modelling
We have used the Ordinary Least Squares (OLS) technique for modelling mean life satisfaction. A variation (Weighted Least Squares) was used in modelling very low and high life satisfaction. This is to account for different sample sizes between LAs.
Explanatory variables were entered in the models in groups according to the order specified in the list of potential drivers given in slide 23). Each group was entered into the model using the stepwise regression procedure. The significant variables are taken, and kept in subsequent models, where additional factor groups are added to the model using the stepwise procedure. If variables become non-significant at any stage they are removed.
Factor groups were entered in the model in a specified order, according to their importance in influencing wellbeing at the individual level. This order of importance is taken from the ONS publication ‘Measuring National Well-being – What matters most to Personal Well-being?’ This gives the factors highlighted as having a large effect on individual life satisfaction the best opportunity to emerge as LA level predictors. The approach also helps to separate individual and ‘place’ effects, giving priority to the former.
A key reason for taking a stepwise approach was to avoid the problems arising from multi-collinearity among the factors being considered. Where two or more predictor variables in the regression model are highly correlated with each other, the individual regression coefficients cannot be estimated precisely. In this event, the model may not give valid results about individual predictors or about which predictors are most powerful in explaining variance in the outcome variable, and which are redundant.
In effect, if variables chosen in the stepwise procedures appeared to be multi-collinear (as indicated by the VIF, below) then they were removed manually. Although this occurred minimally, it is possible that variables chosen by the model might have been replaced by other very similar ones that had been removed from the procedure.
The VIF (Variance Inflation Factor) is an indication of multi-collinearity in the model. A value greater than 5 for a single variable (or an average of the VIFs of all the factors in the model much higher than 1) would indicate that multi-collinearity was causing problems in the model. In the models presented here, most factors were under 2, and the highest VIF was 2.7 for ‘a sense of belonging’.
44
Weighting models where estimates of very low and high life satisfaction are less robust
Because our estimates for both of these dependent variables are based on small numbers of survey respondents in some local authorities, they can be subject to relatively high levels of uncertainty.
We therefore adapted our approach to regression modelling from OLS to weighted least squares regression. Most local authorities are given equal weight (a weight of 1), but those with the greatest uncertainty around the estimates, as measured by the Relative Standard Error (RSE), are given a reduced weight. RSE is the ratio of the standard error around an estimate to the estimate itself. ONS have used RSE < 20% as their benchmark in assessing the robustness of local authority estimates.
An alternative would have been to drop LAs from analysis where estimates were considered less robust. We did not favour this option because LAs with the highest RSE tend to be non-metropolitan district councils (where sample sizes are less than 300), and these tend to also have the highest levels of life satisfaction. Excluding these LAs from analysis from modelling would be unacceptably biasing.
• In modelling high life satisfaction, 29 LAs (9%) had estimates where RSE > 20%, and were given weights in inverse proportion to their RSE. In effect the lowest weight given was 0.65.
If the same benchmark were used in weighting estimates of very low wellbeing, the majority (68%) of local authorities would have been down-weighted. Rather than do this, we have taken a more relaxed threshold (of RSE > 40% in determining robustness.
• In modelling very low life satisfaction, 90 LAs (28%) had estimates where RSE > 40%, and were given weights in inverse proportion to their RSE. In effect the lowest weight given was 0.38.
The method of correction is ad-hoc, and attempts to avoid introducing bias into modelling.
45
Sample sizes by type of LA district
Annual Population Survey sample sizes vary considerably according to the type of local authority with the largest samples in unitary authorities and Metropolitan district councils. This reflects both the larger population sizes and also any over-sampling conducted on behalf of the LA.
Sample sizes were lowest for non-metropolitan district councils (ranging from 56 to 268, average = 148). Five unitary authorities had sample sizes under or around 500 – notably new unitaries: Bedford (272), Rutland (276), Cheshire West and Chester (434), Central Bedfordshire(475) and Cheshire East (515).
Sample size by type of local authority district
0
100
200
300
400
500
600
700
800
900
1,000
1,100
1,200
London Boroughs
Metropolitan Districts
Unitary Authorities
Non-metropolitan Districts
46
The reliability of local authority district level estimates of very low life satisfaction
The scatter diagram shows the levels of high and very low life satisfaction for each LA district.
The colour of the data points indicates the reliability (Relative Standard Error) of the estimates of very low satisfaction.
Blue: RSE < 40% (ie more reliable) Pink: RSE >= 40%
We have not depicted the reliability of estimates of high life satisfaction, as these are more reliable than those for very low life satisfaction in all but one of the 324 LA districts.
-16
0
16
-16016
% of residents with very low life satisfaction
(mean centred; axis is reversed)
Incr
ea
sin
g %
of
resi
de
nts
with
hig
h li
fe s
atis
fact
ion
Struggling places: low wellbeing, little high wellbeing
Polarised places: low wellbeing and also high wellbeing
Satisfied places: high wellbeing with little low wellbeing
Moderate places: neither low nor high wellbeing
Decreasing % of residents with very low life satisfaction
% of residents with high life satisfaction
(mean centred)
top related