Estimation and Testing of Self-Reporting Heterogeneity in Self-Assessed Health Questions

Abstract

We propose a new methodology for statistically testing for reporting heterogeneity in self-assessed health (SAH) questions. We contribute to the literature by quantifying the effect of such heterogeneity in the probability distribution of SAH responses and providing a correction to that distribution that allows us to obtain an unbiased distribution of SAH responses as a function of true health. Our econometric approach extends typical ordered probit formulations for describing SAH responses by incorporating a consistent estimate of individuals' true health. True health is estimated as a latent variable obtained from information on an array of individuals' objective health binary indicators. This additional equation relating unobserved true health and observable health characteristics allows us to disentangle the effect of true health from self-reporting heterogeneity on SAH responses. We propose two different maximum likelihood procedures to estimate each effect consistently and show their performance in finite samples. The application of this methodology to the Understanding Society data set uncovers strong evidence of self-reporting heterogeneity across many individual characteristics. Failing to control for such heterogeneity masks the good health of individuals and exaggerates the occurrence of bad health states, while over(under)-states true health gradient of co-variates.