Nancy Mathiowetz & Roger Tourangeau

Joint Program in Survey Methodology

Thursday, May 31, 2001

Abstract

Measures of disability are included on a number of national surveys, such as the National Health Interview Survey and the American Community Survey. In addition, the Census 2000 Long Form included several disability items. Data from a variety of sources suggest that these and similar batteries of disability questions are prone to a number of serious measurement problems: Self-respondents give different answers to disability questions than proxy respondents do; the context in which the disability questions are embedded can make a noticeable difference in the answers; small changes in wording or format across surveys can lead to big differences in the estimated rates of disability; and a substantial proportion of those classified as disabled in one interview are classified as not disabled in re-interviews conducted a few months later. This talk reports the findings from an experimental study designed to investigate the effects of alternative features of disability questions and response status (self v. proxy) on both the overall rate of disability and the consistency of responses to disability questions over time.