Loss to folow-up in a sample of Americans 70 years of age and older: The LSOA 1984-1990.

Loss to follow-up is a problem in longitudinal samples, and the literature on response rates in panels of older persons suggests that they may be more vulnerable to nonrandom attrition and its consequent biases. The event history approach used in this study to determine the correlates of nonresponse addresses important shortcomings of previous analyses by incorporating time-varying covariates. Nonresponse is not random; persons of older ages, lower education, who live alone, rent (not own), have more functioning impairments, or have another sample person in the household are more likely to become nonrespondents. However, correction accounting for the effect of these correlates of nonresponse, as well as unobserved characteristics potentially affecting nonresponse, suggests that the association between these characteristics and the probability of nonresponse is not large enough to introduce bias. While these results are not portable to other analyses or panels, they do indicate that in this case, significant nonrandom nonresponse does not bias all related analytic results.