Self-assessed health (SAH) measures are widely used in models of health and health inequalities. Such models assume that SAH is a reliable measure of health status. We utilise a unique feature of a national longitudinal survey to examine the consistency of responses to a standard SAH question that is asked twice to the same individual in close temporal proximity in up to three waves (2001, 2009, and 2013). In particular, we analyse whether the consistency of responses varies with personal characteristics. The main analysis sample includes 18,834 individual-year observations. We find that 57% of respondents provide inconsistent reports at least once. Characteristics that are associated with significantly higher inconsistencies are age, education, cognitive ability, and time between responses. The results suggest that there are systematic differences in the ability of individuals' to self-evaluate and summarise their own health. Consequently, failure to account for such error may lead to large estimation biases in models of health outcomes, particularly with respect to the relationship between education, cognitive ability, and health.
Keywords: Australia; Cognition; Health reporting; Mixed-effects multinomial logit model; Panel data; Self-assessed health; Socioeconomic status.
Copyright © 2017 Elsevier Ltd. All rights reserved.