Analyzing Health Equity Using Household Survey Data

1
Analyzing Health Equity Using Household Survey
Data

• Lecture 2
• Data for Health Equity Analysis Requirements,
  Sources and Sample Design

2
Data requirements Health outcomes
Murray and Chen (1992) classification of
morbidity measures
Self-Perceived Self-Perceived
Symptoms and impairments Occurrence of illness or specific symptoms during a defined time period
Functional disability Assessment of ability to carry out specific functions and tasks, or restrictions on normal activities (activities of daily living, e.g., dressing, preparing meals, or performing physical movement)
Handicap Self-perceived functional disability within a specifically defined context
Observed Observed
Physical and vital signs Aspects of disease or pathology that can be detected by physical examination (e.g., blood pressure and lung capacity)
Physiological and pathophysiological indicators Measures based on laboratory examinations (e.g., blood, urine, feces, and other bodily fluids), body measurements (anthropometry)
Physical tests Demonstrated ability to perform specific functions, both physical and mental (e.g., running, squatting, blowing up a balloon, or performing an intellectual task)
Clinical diagnosis Assessment of health status by a trained health professional based on an examination and possibly specific tests
3
Data requirements Health-related behavior

• Health care utilization
• Payments for health care
• Smoking, drinking, diet
• Sexual practices
• Household-level behavior (cooking, sanititation,
  etc.)

4
Data requirements Living standards or
socioeconomic status

• Living standards
• Direct approaches e.g., income, expenditure
• Cardinal can compare magnitudes of differences
• Proxy measures e.g., assets index
• Ordinal provide rankings
• Socioeconomic status
• Education (level or years)
• Occupational class

5
Data requirements for health equity analysis
Health Utilization Living standards (ordinal) Living standards (cardinal) Unit subsidies User payments Back-ground vbls
Health inequality ? ?
Equity in utilization ? ?
Multivariate analysis ? Or ? ? ?
Benefit-incidence analysis ? ? ? (?)
Health financing Progressivity Catastrophic payments Poverty impact ? ? ? ? ? ?
6
Data sources

• Household surveys and non-routine data
• Large-scale, multi-purpose surveys e.g., LSMS
  (World Bank), MICS (UNICEF)
• Health / demographic surveys e.g., DHS (ORC
  Macro), WHS (WHO)
• Household budget surveys
• Facility-based surveys (exit polls)
• Routine data
• Administrative data from HIS, vital registration,
  etc.
• Census data

7
Pros and cons of household survey data
Examples Advantages Disadvantages
Living Standards Measurement Study (LSMS), Demographic and Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS), World Health Surveys (WHS) Data are representative for a specific population (often nationally), as well as for subpopulations Many surveys have rich data on health, living standards, and other complementary variables Surveys are often conducted on a regular basis, sometimes following households over time Sampling and non-sampling errors can be important Survey may not be representative of small subpopulations of interest
8
Pros and cons of user exit poll data
Examples Advantages Disadvantages
Ad hoc surveys, often linked to facility surveys Cost of implementation is relatively low Detailed information that can be related to provider characteristics is provided about users of health services Data on payments and other characteristics of visit are more likely to be accurate Exit polls provide no information about nonusers Data often contain limited information about household and socioeconomic characteristics Survey responses may be biased from courtesy to providers or fear of repercussions
9
Pros and cons of administrative data
Examples Advantages Disadvantages
HIS, vital registration, national surveillance system, sentinel site surveillance Data are readily available Data may be of poor quality Data may not be representative for the population as a whole Data contain limited complementary information, e.g., about living standards
10
Pros and cons of census data
Examples Advantages Disadvantages
Implemented on a national scale in many countries Data cover the entire target population (or nearly so) Data contain only limited data on health Data collection is irregular Data contain limited complementary information, e.g., about living standards
11
Sample design and the analysis of survey data

• Multi-purpose and health surveys often have a
  complex design
• Stratification separate sampling from
  population sub-groups e.g., urban / rural
• Cluster sampling clusters of observations not
  sampled independently e.g., villages
• Unequal selection probabilities e.g.
  oversampling of the poor, uninsured

12
Standard stratified sampling

• Population categorised by relatively few strata
  e.g. urban/rural, regions
• Separate random sample of pre-defined size
  selected from each strata
• Sample strata proportions need not correspond to
  population proportions ? sample weights (separate
  issue)

13
Stratification and descriptive analysis

• If pop. mean differs by strata, stratification
  reduces sample variance of its estimator
• Standard errors for means and other descriptive
  stats. should be adjusted down
• If regression used to estimate conditional means,
  then adjust the standard errors

14
Cluster sampling

• Two (or more) stage sampling process
• Clusters sampled from pop./strata
• Households sampled from clusters
• Observations are not independent within clusters
  and likely correlated through unobservables
• Standard errors of parameter estimates should be
  adjusted to take account of the within cluster
  correlation

15
Sample weights

• Stratification, over-sampling, non-response and
  attrition can all lead to a sample that is not
  representative of the population
• Sample weights are the inverse of the probability
  that an observation is a sample member
• Sample weights must be applied to get unbiased
  estimates of population means, etc. and correct
  standard errors
• Should also be applied in descriptive
  regressions

16
Stata computation
Set the sample design parameters svyset locality
pwwgt, strata(strata) Estimate the mean and
get the correct SE svy mean vacc, over(quint)
17
Child Immunization Rates by Household Consumption
Quintile, Mozambique 1997
No allowance for sample design

With sample weights
Quintile Mean s.e. Deff
poorest 0.531 0.017 1.694
2 0.629 0.019 2.196
3 0.621 0.019 2.117
4 0.708 0.024 3.416
richest 0.843 0.014 1.488
Total 0.654 0.009 2.138

n 6447
No. strata 1
No. PSUs 6447
Quintile Mean s.e. Deff
poorest 0.545 0.014 1.000
2 0.659 0.014 1.000
3 0.708 0.013 1.000
4 0.805 0.011 1.000
richest 0.892 0.008 1.000
Total 0.728 0.006 1.000

n 6447
No. strata 1
No. PSUs 6447
18
Child Immunization Rates by Household Consumption
Quintile, Mozambique 1997

With stratification and clustering
With stratification
Quintile Mean s.e. Deff
poorest 0.531 0.017 1.630
2 0.629 0.019 2.164
3 0.621 0.019 2.075
4 0.708 0.024 3.366
richest 0.843 0.014 1.456
Total 0.654 0.008 1.942

n 6447
No. strata 21
No. PSUs 6447
Quintile Mean s.e. Deff
poorest 0.531 0.028 4.469
2 0.629 0.033 6.577
3 0.621 0.026 4.014
4 0.708 0.029 5.092
richest 0.843 0.018 2.485
Total 0.654 0.017 8.313

n 6447
No. strata 21
No. PSUs 273