The Census Area Statistics Myles Gould Understanding area-level inequality & change.

The Census Area Statistics

Myles Gould

Understanding area-level inequality & change

Presentation Content

Nature of Census & CAS data Data Tools Research Uses Analysis Issues Examples: Health Variations

– CAS, SARS & combining with other data

Nature

What is the Census? Cross-sectional snapshot of population on

single date Source of secondary data Can be examined at many geographical

levels Total (nearly!) enumeration (count) &

coverage of national population Coverage is consistent (all households asked

same questions)

Source: Press Association

What is the CHCC?

Collection of Historical & Contemporary Censuses

Census Area StatisticsSample of

Anonymised Records (SARs)• aggregated to zones

• 1971, 1981, 1991, 2001 • representative sample of individuals• 1991 and 2001

Historical Censuses Collection• 1851 & 1881

All datasets are available in UK HE & FE sectors

Census Statistics A range of products are available

Source: Rees et al (2002)

CAS Complex data structure Data aggregated for different geographical units 1991 LBS contain 99 tables for GB, & approx.

20,000 statistical count A relatively simple table…

Census Geography Rich source of local statistics for a range of

hierarchical geographical units

Source: adapted from Martin (1991)

1991 Eng & Wales

2001

2001Outputs

Headcounts: counts males, females, households for unit postcodes upwards (eg LS2 9JT)

Profiles: counts & %s for OAs upwards, but confidentiality protected

Key statistics: 50 variables, mainly %s for OAs upwards, LA data already available

CAS: 7000 counts in cross-tabulations for OAs upwards:

Standard tables: 25,000 counts in more detailed tables for Wards/LAs

Tools

Casweb

Available at http://census.ac.uk/casweb/

CommonGIS

Uses & Examples

Research Uses Describing demographic & socio-economic profiles

of areas– EDA & mapping

Selecting & identifying areas for further study Exploring patterns &/or relationships for

variables/processes– typically generalized linear modelling

Looking at change over time Using as a denominator for calculating other

statistics Combining with other secondary data sources in

multivariate analysis Identifying & classifying areas with similar

characteristics– factor/cluster analysis, composite deprivation indices

Census change over time atlas CommonGIS also used to visualise change over time

Analysis Issues Ecological fallacy MAUP scale aggregation problem &

different results Confounders & ecological analyses Decennial snapshot & out of date quickly Need for standardisation & understanding

underlying composition Dealing with unstable population

denominators (shrunken estimates) Cross-tabulations of a small number of

variables in any one CAS table

Some Research Questions Are there variations between different types of people? Are there variations between places? Are there absolute or relative differences? How does place matter? Is it composition (who’s in a

place) or context that matters? Are variations explained solely by poverty? Do variations vary over time? Is there a widening gap?

Are variations becoming more polarized Are there groups & places we should target with policy

responses? What aspects of place matter?

Health Variations

NB not all these questions can be answered with census data

Health Variations

LTLIHealth

GoodHealth

Fair

30

20

10

Per

cent

age

English Local Authorities, 2001Variations in self-reported heath, amongst

Source: 2001 Census Area Statistics, Crown Copright

Kensington & Chelsea

Easington

Easington

Barnsley

Health Variations

combi.30-5950-5940-4930-39

50

40

30

20

10

0

Per

cent

age

amongst English Local Authorities, 2001Variations in self-reported heath by age,

Source: 2001 Census Area Statistics, Crown Copright

Age category

Liverpool

Easington

EasingtonTow er HamletsManchesterLiverpool

Liverpool

Easington

Easington

Know sleyLiverpool, ManchesterTow er Hamlets

Local Health Variations

Gould & Jones (1996) - Self-reported

Limiting Long-Term Illness Analysis National & sub-regional comparisons Consider compositional vs context debate 2% Individual SARs Use multilevel analysis

– Individual & area variations at same time

419,550 individuals, 42,073 reported illness 278 SAR Areas - combinations of Local

Authority districts (protecting confidentiality of individuals)

Health Variations

Health Variations

Results Marked relationship illness & age but differences between the sexes are not

particularly marked until the older age group there is a 'multiplicative' relationship so that the

worst health of all is experienced by partly skilled/unskilled, local-authority rent, with no car

geographical variation remain after allowing for individual characteristics – area composition (who lives in a place)

Place does make some difference

Health Variations

Health VariationsHealth & Deprivation: Exploratory Survival Analysis Jones, Gould & Duncan (2000) Combine HALS & Census CAS

Health Variations

Results Deprivation little effects on mortality when

in wards where deprivation <0 (mean)– little difference between social classes

in areas of relative affluence Marked differences between classes in

areas of increasing deprivation

Further work SARs, LTLI & Cambridge scores New unpublished work with Kelvyn Jones Looking at absolute & relative variations in morbidity

and social advantage Model interaction quartile Cambridge scores

(individuals), with area means, & with area Gini coefficients

Health Variations

Further work SARs, LTLI & Cambridge scores Min 2.13 ≈driver mate; Ql= 19 ≈security officer Md=36 ≈stores controller Qu=46 ≈ Farmers; Max=94 ≈ General Medical

Practitioner In areas with more equality, individual class effects are

small

Q1Q2

Q1

Q2Q3

Q3Q4 Q4

Low status

High status

Q1Q2

Q1

Q2Q3

Q3Q4 Q4

Health Variations

Self-critique Ethnicity in effect treated chaotic conceptions (Sayer 1992)

– ‘lumping everything together’ MAUP, SAR areas big and crude, what do they mean? Some purchase on modelling complex relationships, but

still only suggesting reasons for variations ML Point us in right direction for other survey work or some

qualitative

Caveats