Determinants of accessibility and affordability of health care in post-socialist Tajikistan: Evidence and policy options

Abstract

There is increasing evidence of rising levels of inequality in health care utilisation in the post-socialist countries of Central Asia and the Caucasus. Against this backdrop, we investigate the determinants of accessibility and affordability of health care utilisation in Tajikistan. A modified version of the Andersen Behavioural Model is used to conceptualise the determinants of health care utilisation in Tajikistan. Poisson and Ordered Logit regression models are performed to estimate the determinants of health care utilisation. Empirical results demonstrate that poverty, chronic illness and disability are the most important determinants of health care utilisation and affordability in Tajikistan. Other significant determinants include gender, the level of education of the household head, and the availability of medical personnel at a given population point. These findings suggest an urgent need for health care reform in order to ensure equality in accessibility and affordability for the entire population.

Keywords:

• health care equity
• poverty
• inequality
• Central Asia
• Tajikistan

Notes

1. We use the terms ‘community’ and ‘population point’ interchangeably in this paper. Technically, all these terms refer to the same 208 primary population points of the survey.

2. It is common for transitional and developing countries to use consumption as a proxy for income in the estimation of population welfare, for instance, regarding poverty and inequality (Braithwaite et al. 2000). One reason for this is that income is usually underreported in household surveys. Another reason is that consumption better reflects actual well-being, especially in agricultural countries like Tajikistan, where the consumption of goods that were self-produced in a household is very important for the welfare of the population.

3. Since Poisson regression sometimes does not fit because of overdispersion, we used the likelihood ratio test to test overdispersion (Long and Freese 2006). According to our estimate, the test results strongly suggest the absence of overdispersion in the Poisson regression model (G ²=0.00, Prob > = G ²=0.00). We also estimated negative binomial regression model (nbreg) with the same set of variables to check for overdispersion. The result of the nbreg is identical to Poisson in terms of regression coefficient and t-score. This confirms the absence of overdispersion in the Poisson model (Long and Freese 2006).