Adam Wagstaff

1
Photos from Hans Kemp
Health insurance for the poor in Vietnam An
impact evaluation of Vietnams health insurance
program
Adam Wagstaff Development Research Group East
Asia HD, The World Bank
2
Introduction

• Policy and program issues
• Lack of health insurance in China and Vietnam
  following de-collectivization of agriculture
• New policy of public finance of free health care
  for the poor by enrolling them in social
  insurance
• Substantive issues
• Health insurance literature focuses on negative
• Paper looks at risk-reduction associated with HI,
  and positive consequences from it
• Methodological issues
• Paper uses propensity score matching (PSM) with
  pre- and post-intervention data to estimate
  impact of health insurance
• Empirical findings policy implications

3
Policy/institutional issues
Policy issues

• In China and Vietnam, cooperative health
  insurance collapsed after de-collectivization of
  agriculture
• In both countries concern over affordability of
  health care, esp. among rural poor
• People encouraged to enroll in Vietnams health
  insurance (VHI) programcompulsory for certain
  groups
• Decision 139 mandates and supports provinces to
  enroll poor in VHI (or make alternative
  arrangements for them)
• What will impact of enrollment among 139
  beneficiaries be on key outcomes?

4
Costly care, high spending
Policy issues
5
Impoverishing too
Policy issues
6
Impoverishing too
Policy issues
Out-of-pocket payments for health care pushed
2.6m Vietnamese into poverty in 1998. Increased
headcount by 23 and poverty gap by 25
7
VHI before decision 139
Policy issues

• Set up in 1993, reformed in 1999
• Compulsory scheme for formal sector workers,
  civil servants, etc.
• Voluntary schemecurrently attracts mostly school
  kids students
• By 1998, ?15 enrolled 60 compulsorily
• Coverage against inpatient costs, fees incurred
  in outpatient care less generous coverage for
  voluntary members

8
How decision 139 will change coverage
Policy issues
9
Health insurance issues
Substantive issues

• Much of the health insurance literature
  emphasizes the negative
• Moral hazard
• Adverse selection
• Recent work emphasizes
• Risk-reduction benefits of insurance, and
  positive consequences of this
• Lower precautionary savings
• Better health outcomes
• Difficulty of measuring true moral hazard

10
Evaluation with non-experimental data
Methodological issues
Participation in program Participation in program Participation in program
Outcome D1 Yes D0 No
Y1 outcome with treatment ? ?

Y0 outcome without treatment ? ?
Difference effect of treatment on treated
Difference bias
11
Propensity score matching as approach to reducing
bias
Methodological issues
Component of bias Strategy to reduce bias
Participants and non-participants differ in relevant respectsi.e. have different Xs Compute probability of participation as function of Xs, P(X). Match participants and non-participants on P(X). Compute mean difference in outcomes between matches (single difference or SD)
For some participants, there are no comparable non-participants Confine comparisons to region of common support of P(X)
Outcome differences not attributable to treatment might remain even after conditioning on Xs and confining attention to common supportproblem of selection bias In cross-section, nothing can be done. With pre- and post-intervention data, compute difference between mean change among participants and mean change among non-participants (double difference or DD). This allows for time-invariant selection on unobservables effect
12
Data variables
Empirical results

• Data from Vietnam Living Standards Survey
• High proportion of HHs interviewed in 1993 were
  re-interviewed in 1998
• Outcomes variables
• Contact probability
• Volume of services used (1998 data only, so can
  do only single difference PSM)
• Out-of-pocket payments
• Non-medical HH spending
• Child health, measured through anthropometrics
  (underweight, etc.)

13
Probit model for participation
Empirical results

• VHI enrollment depends on
• Whether in school ()
• Employed
• Communist party, government, army, social
  organization, state-owned company ()
• Private company (-)
• Income ()
• Education ()
• Urban ()
• Commune fixed effects

14
Descriptives of probability, before after
matching
Empirical results
Predicted probability of coverage Predicted probability of coverage Predicted probability of coverage Predicted probability of coverage
cases Mean Std. Dev. Min Max
Before matching Uninsured 14537 0.12736 0.13665 0.00011 0.98705
Before matching Insured 3015 0.38192 0.25335 0.00477 0.99989
After nearest neighbor matching Uninsured 3015 0.38189 0.25326 0.00477 0.98705
After nearest neighbor matching Insured 3015 0.38192 0.25335 0.00477 0.99989
After caliper matching with 0.001 caliper Uninsured 2775 0.34330 0.22151 0.00477 0.98705
After caliper matching with 0.001 caliper Insured 2775 0.34330 0.22150 0.00477 0.98768
15
Histograms of probabilities, before and after
matching
Empirical results
Uninsured Insured
16
PSM results 1 (DD SD)
Empirical results
Sample Estimator Outcome Effect T-stat
Sample DD Out-of-pocket payments 4.582 0.19
Inpatients SD Inpatient costs -738.18 -1.69
Inpatients SD Out-of-pocket payments -1102.73 -2.42
Sample SD Inpatient costs 10.09 1.04
Sample SD Non-hospital costs 15.50 0.89
Sample DD Contact probability 0.040 2.26
Sample DD Weight-for-age kids lt 10 0.203 1.98
Sample DD Weight-for-height kids lt10 0.215 1.90
Sample DD Non-health consumption 387.53 5.37
DDdouble difference SDsingle difference
17
PSM results 2 (SD)
Empirical results
Sample Sample Poorest quintile Poorest quintile
Effect T-stat Effect T-stat
Total visits 0.017 0.22 0.095 0.57
Hospital visits 0.051 3.85 0.030 1.72
Inpatient nights 0.973 3.55 0.216 0.78
CHS visits 0.025 1.85 0.069 1.21
Polyclinic visits 0.000 0.00 0.009 0.28
Private visits -0.001 -0.03 -0.052 -1.67
Traditional healers -0.010 -0.75 -0.004 -0.13
Pharmacy visits -0.056 -0.94 0.043 0.30
18
Conclusions

• PSM useful for program evaluationuse panel data
  and diffs-in-diffs estimator if possible
• VHI increases contact probability, volume of use

• No impact on out-of-pocket payments
• Effect on non-medical consumptionreflects risk
  reduction?
• For hospital care, smallest impact of VHI among
  the poor
• Extrapolation to 139 difficultpoorest quintile
  estimates most relevant but NB no copayments