Reporting Heterogeneity Effects in Modelling Self Reports of Health

Abstract

Self-assessed measures of health using Likert-type scales are widely used to assess the health and well-being of populations, and are a feature of household surveys throughout the world. However, the self-reported and subjective nature of these measures means that different people will inherently respond in different ways - a concept known as reporting heterogeneity. In this paper we consider two types of reporting heterogeneity. The first is differential item functioning, which results when individuals systematically differ in their interpretation and use of response categories. The second is middle-inflation bias, which arises when respondents adopt a ‘box-ticking’ strategy - for example, because they are unsure of how to answer survey questions, or because they do not take the surveys they are completing seriously. This type of reporting heterogeneity typically materializes in the form of an artificial build-up of responses in middling response categories. We consider approaches for adjusting for each type of reporting heterogeneity, both in isolation and in combination. The results suggest that self-assessed measures of health are susceptible to both types of reporting heterogeneity, and that failure to account for these nuances may lead to erroneous inference concerning the analysis of self-reported health.