Sample attrition and multivariate analysis in the Retirement History Study

The attrition of respondents in panel studies of older people can create bias in data analysis. Changes in the examination of attrition are suggested. Bivariate and multivariate tests comparing those eligible to respond and those actually continuing in the panel are reported for five waves of data from the Retirement History Study. The results indicate that variable relationships in the data examined are relatively free of the bias that can be created by respondent attrition. Thus, attrition from samples of older respondents probably is less of a factor than suggested by previous studies reporting on frequency distributions.