Do health interview surveys yield reliable data on chronic illness among older respondents?

Previous research evaluating quality of health interview survey data has generally relied upon comparisons of household interview data with medical records or other external sources of information. However, 'gold standards' are not always satisfactory or available. This paper illustrates an alternative approach to the evaluation of data quality - examination of the reliability of reports of chronic conditions in longitudinal surveys. The data come from national samples of older Americans (First National Health and Nutrition Examination Survey Epidemiologic Followup Study, 1971-1975, 1982-1984, 1986) and older Taiwanese (Survey of Health and Living Status of the Elderly, 1989, 1993, 1996). The results show that, among respondents who reported a chronic condition at a given interview, the likelihood that the condition was acknowledged at the subsequent interview was higher for hypertension and diabetes than for arthritis and stroke. Low levels of consistency for stroke appear to result partly from the poor wording of questions. In Taiwan, younger, more educated persons and those experiencing severe conditions were somewhat more likely to acknowledge the condition at follow-up compared with their respective counterparts. Women and persons of high cognitive status in the United States and respondents in both countries who used a proxy to report the occurrence of a stroke were also more likely to acknowledge conditions at follow-up.