Dynamics of health insurance enrollment in Vietnam, 2004–2006

Abstract

Vietnam is undertaking health financing reform in an attempt to achieve universal health insurance coverage by 2014. Changes in health insurance policies doubled coverage between 2004 and 2006, yet about one-fifth of the insured in 2004 dropped out of the health insurance system by 2006. Using longitudinal data from VHLSS 2004 and 2006, this paper models static and dynamic health insurance choices and highlights the importance of income and education in determining the movement in or out of a particular scheme. The models show significant adverse selection where individuals with poor health are more likely to be insured.

Keywords:

• health insurance
• adverse selection
• Vietnam

Acknowledgements

The authors thank the anonymous referee, Trevor Breusch, Alison Booth, Deborah Cobb-Clark and Amy Liu and seminar participants at the Australian National University and 9th International Workshop on Pension, Insurance and Saving at University Paris Dauphine for useful comments. The usual disclaimers apply.

Notes

1. The health insurance fund switched from a surplus of 1989 billion VND in June 2006, which was accumulated over the previous ten years, to a deficit of 1200 billion VND at the end of 2006. In 2007, the health insurance fund also experienced a deficit of 1650 billion VND. In 2008, deficit was at 1700 billion VND. In 2009, the deficit was estimated at 2000 billion VND.

2. The 2008 Health Insurance Law increases the contribution rate to 6% of salary, in which employees pay 2% of their salaries and employers 4%. In addition, health insurance is compulsory for students from 2010.

3. The respective figures are calculated using Table 2 where 36% = 0.5/1.4 and 14% = 0.2/1.4.

4. We use total household per capita expenditure (including health expenditure) to proxy for household income since it captures the household permanent income more precisely. We experiment with excluding health expenditure from total expenditure and found that the results are almost the same. The results for this robustness check will be available upon request.

5. The VHLSS06 provides a comprehensive description of an individual's overall functional health on the basis of vision, aural, remembering or concentrating, ambulation (ability to get around), dexterity (use of hands and fingers) and communication attributes. For each attribute, four possible responses are recorded: not difficult, a little difficult, very difficult and impossible. We classify an individual as one with any limitation in functional ability if having a little difficulty or more in any of above attributes.

6. Information on smoking, chronic disease or disability is only available in the VHLSS06. We make use of our individual panel to assume that individuals who report having ever smoked or having any chronic disease or being disabled in 2006 also did so in 2004.

7. Risk-rating of health insurance premiums means insurers can differentiate premiums according to assessed true risk. Due to this premium risk-rating practice, the most common finding in empirical studies in health insurance in developed countries is that healthy people are more likely to be covered by private health insurance (see Doiron, Jones, and Savage (2008) for a review).

8. Information on whether a commune is classified as a beneficiary of the 135-Program is available in the commune information section. In both surveys, the commune questionnaire is asked for all communes in rural areas and some communes in urban areas. Although commune information is not available for all communes, the fact that all communes covered by the 135-Program are in rural or remote areas allows us to use commune questionnaire to identify Program-135 communes. We do not include a variable indicating whether the household is identified by the commune as poor in the regressions since this variable is highly correlated with our household expenditure quintile.

9. Estimates are broadly similar for 2004 so we do not report them for brevity.

10. For all MNL models, we test for a crucial assumption of the MNL model known as Independence of Irrelevant Alternatives (IIA) assumption. The MNL model assumes that the odds of any pair of outcomes are determined without reference to the other outcomes that might be available. We have computed a Hausman-type test (Hausman and McFadden 1984) of IIA and this shows that the assumption of IIA assumption is not violated in almost all cases. We also experiment with an alternative multinomial probit model which does not require the IIA assumption. Results from the multinomial probit model are quantitatively similar to those from the MNL model. We also implement a Wald test for combining outcome categories. The Wald test statistics show that we can confidently (with chi-square statistics >100 in all cases) reject the null hypothesis that outcome categories can be collapsed. Results of these tests will be available upon request.

11. Our approach is similar to Wolfe and Goddeeris (1991) or Cutler, Lincoln, and Zeckhauser (2010) where we all study the impact of past characteristics on the current insurance decision. This paper, however, makes a significant improvement in this approach by looking at the impact of past characteristics on the dynamics of health insurance status.

12. Estimates for initial variables are largely the same for 'baseline' and extended models, indicating that 'change' variables are exogenous in extended models. Estimates for 'baseline' models are not reported for brevity but will be available upon request.

13. We do not separate this state further by defining the destination scheme because in some cases, the number of individuals moving into any specific scheme is so small that the MNL loses its precision.

14. According to the join circular number 10/2008/TTLT-BYT-BTC dated 24/09/2008, households with per capita income above the poverty line but not over 130% of the poverty line are defined as near poor. These households are subsidized at least 50% of health insurance premium.