Evaluating health consumers' preferences stability through joint estimation of revealed and stated health insurance preferences data

Abstract

This paper presents a variant of the mixed logit model, in the form of a panel-like error components mixed logit that relies on a multinomial logit formulation of the weighted logit formula, as opposed to the usual conditional logit representation; then uses the model to evaluate the consistency of consumers' preferences for health insurance by jointly modelling stated health insurance preferences with revealed health insurance choices of respondents from the 2007 Medical Expenditure Panel Survey (MEPS). Estimation is implemented within the Bayesian paradigm using Markov Chain Monte Carlo (MCMC) methods, and the results suggest that 2007 MEPS respondents do present stable preferences for health insurance. In fact, respondents who initially express health insurance as not worth its cost are found to be 23.28% less likely to be privately insured and 81.53% less likely to be publicly insured. On the other hand, those initially expressing health insurance as worth its cost are found to be 21.72% more likely to be privately insured and 81.68% more likely to be publicly insured.