Improving sub national estimates of disability prevalence

1
Improving sub national estimates of disability
prevalence

• Alan Marshall
• Manchester University

2
Presentation Structure

• Why do we need to improve estimates?
• Concepts of disability and data issues.
• HSE 2001 Locomotor disability prevalence.
• Preliminary Analysis.
• Empirical Bayes techniques.
• Conclusions.

3
Why do we need to improve sub national
estimates?

• Census data lacks detail
• Survey data lacks geography.
• Survey estimates become increasingly unreliable
  at sub national level.
• Data on disability prevalence (by type and
  severity) are important for planning of service
  provision.

4
Disability Concepts and Definitions?

• Medical Model Locates the source of disability
  in the individuals deficiency and his or her
  personal incapacities.
• Social Model - Disability is a form of oppression
  caused by the aspects of society that prevent
  people with physical impairment from
  participating in everyday activities.

(Abberley in Levitas, 1996)
5
Disability official data sources

• Census,
• General Household Survey,
• Health Survey for England
• OPCS disability surveys 1969, 1984 and 1996/7
• Disjointed approach to data collection (Abberely
  in Levitas,
• 1996).
• Estimates vary between sources because surveys
  are not
• consistent in their definitions used or
  population sampled.
• No true disability prevalence figure, depends on
  definitions
• used and motives of collector.

6
HSE 2001

• HSE gives most recent source of detailed data on
  disability.
• 5 disability types are measured
• Locomotor disability is assessed according to
• ability to
• Walk 200m without stopping and without
  discomfort.
• Walk up and down stairs.
• Pick up objects from floor.
• Note - this relates to performance without aids
  e.g. walking stick.

(Bajekal et al, 2002)
7
Analysis HSE 2001

• Evidence of a strong relationship between age and
  disability prevalence.
• Age specific disability schedule similar between
  areas (and by type).
• May be regional differences in the level of
  disability prevalence.
• These findings are important considerations when
  deciding how to improve the precision of
  estimates.

8
Empirical Bayes (EB) Techniques

• EB techniques have been shown to improve
• estimates where rates have strong spatial and age
• patterns (Assuncao et al, 2005).
• Use Bayes theorem to adjust our expectations of
  the
• GOR rate based on knowledge of the national rate.
• Assume the National schedule gives true (or
  most
• reliable) estimates of the Parameters.
• The GOR schedules are variations on this national
• schedule.

9
Empirical Bayes (EB) Techniques

• Variance of the GOR rates are composed of
• The variance of the parameter estimates in
  England.
• The sampling variance in each GOR region (ie the
  variance in the GOR rate given that we know the
  England rate).
• Shrink GOR estimates towards National estimates
• If the majority of variation is due to sub
  national sampling
• then the estimate is moved further towards the
  national
• rate.

10
Evaluation of EB techniques

• HSE 2000 includes data on disability.
• Combine 2000 and 2001 data.
• Compare new estimates to EB estimates.
• Use EB on combined data.
• Expect
• EB on 2001 data to move estimates toward combined
  estimates.
• EB on combined estimates to have less effect.

11
Conclusions

• There is a lack of detailed data on disability at
  sub-national levels.
• No true figure for disability prevalence as
  definitions of disability are contested.
• Disability is strongly related to age, there is
  evidence higher/lower rates in some GOR regions.
• Empirical Bayes techniques enable shrinkage of
  GOR estimates to more reliable national rates.
• Comparison of EB adjusted prevalence rates and
  data from the 2000 and 2001 HSE allows an
  evaluation of the success of EB techniques.

12
Less precise for older ages
Shape similar to mortality curve
13
NE curve above England curve
14
SE curve below England
15
(No Transcript)
16
Large sampling Variance more shrinkage
Smaller sampling variance Less shrinkage