Reporting bias and heterogeneity in self-assessed health. Evidence from the British Household Panel Survey

Abstract

This paper explores reporting bias and heterogeneity in the measure of self-assessed health (SAH) used in the British Household Panel Survey (BHPS). The ninth wave of the BHPS includes the SF-36 general health questionnaire, which incorporates a different wording to the self-assessed health variable used at other waves. Considerable attention has been devoted to the reliability of SAH and the scope for contamination by measurement error; the change in wording at wave 9 provides a form of natural experiment that allows us to assess the sensitivity of panel data analyses to a change in the measurement instrument. In particular, we investigate reporting bias due explicitly to the change in the question. We show how progressively more general specifications of reporting bias can be implemented using panel data ordered probit and generalised ordered probit models. Our results suggest that the distribution of SAH does shift at the ninth wave but there is little evidence that this varies with socio-economic characteristics at an individual level.