Misclassification bias in estimates of bereavement effects

Prospective studies that examine marital status differences in health and mortality frequently fail to update information on marital status in statistical models. The authors illustrate how the resulting misclassification of marital status can produce substantial bias in estimates of bereavement effects associated with widowhood. They use as their main source of data the Longitudinal Study of Aging (LSOA), 1984-1990, a national survey of persons aged 70 years and older. The estimates are based primarily on 3,192 respondents who were married and cohabiting with their spouses at the time of the baseline survey and who could be matched to their spouses' records. Comparisons of widowhood status derived from matched spouse records with reported marital status recorded in LSOA interviews demonstrate that reliance on interview information leads to substantial misclassification of marital status. Results from survival models indicate that estimates of marital status effects are sensitive to whether and how marital status is updated after baseline interviews. For example, updating marital status in hazard models from interview information alone indicates a protective effect of widowhood, whereas classifying widowhood on the basis of spouses' death records reveals a significant bereavement effect (relative mortality risks between 1.3 and 1.4).