Estimating income equity in social health insurance system

The first results of the U.S. reforms with enhancing insurance coverage demonstrate that owing to universal health insurance in Massachusetts reform of 2006 the spending from the Health Safety Net Fund wend down 40 percent, and consumers enjoy higher health status, improved physical and mental health (Courtemanche and Zapata 2012). Similarly, the access to public insurance by the selected low-income consumers in Oregon experiment of 2008 has led to higher use of primary, preventive and inpatient care, lower out-of-pocket medical expenditures and resulted in better self-reported physical and mental health (Finkelstein et al., 2012). However, the focus on specific subpopulation of consumers would shed light on potential income inequity with respect to the most vulnerable groups.

The purpose of this paper is to capture potential consumer heterogeneity in estimating income equity in health care access and spending. We exploit the novel dataset for adult consumers in Japanese social health insurance system, where the semi-century of universal coverage has led to increased utilization and better health status (Kondo and Shigeoka 2013).

The methodological novelty of the paper is twofold. First, we apply a latent class approach (Deb and Trivedi, 1997), that better encompasses unobservable consumer characteristics than subjective health assessment, to a multi-part model which distinguishes between non-users, the users of inpatient and outpatient care. Secondly, to address retransformation problem in the second part of the models, we use generalized linear models with latent classes (Greene, 2007).

Our sample is the 2009 data for health care expenditure by non-elderly adult consumers (a subsample of the Japan Household Panel Survey by Keio University). The unique feature of this recently launched household survey is the fact that it distinguishes between inpatient and outpatient health care expenditure, as well as between expenditure covered and non-covered by social health insurance.

Since poverty lines vary in each Japanese municipality and municipality information are unavailable, we employ income quintiles so that the lowest quintile approximated the low income group (OECD, 2009). Individual characteristics included age, gender, education, employment, and type of residence. Health status is taken into account with a binary variable for low health condition, Ben-Sira’s (1982) psychological distress index, and body mass index. Binary variables for drinking, smoking, doing sports, and having checkups reflect health-related life styles. We add a dummy for National Health Insurance, owing to additional high-cost medical benefits for the poor in this plan.     

Our analysis indicates adequate goodness-of-fit for the generalized linear model with inverse Gaussian distribution family, while gamma, Weibull, and lognormal distribution families did not fit our data. Consumers separate into two latent classes both in the generalized linear model and OLS model for logged health care expenditure.

Overall, the results of the estimations reveal horizontal equity of health care access and spending in Japanese health insurance system. However, horizontal inequity may be found in health insurance premiums and the stop-loss effect (i.e., the prevalence of catastrophic health care coverage).