The performance of several indicators in detecting recall bias

Self-reported exposure data are vulnerable to differential recall, which threatens the internal validity and thus the credibility of the study. When a proxy measure of the exposure of interest is available, it may be possible to check for the presence of recall bias due to misclassification of the exposure information. We present a number of indicators of recall bias for case-control studies, and we describe their behavior by means of computer simulations. The indicators are all based upon a comparison between self-reported exposure data and data on job titles grouped into exposure-specific matrices. The indicators are based upon the assumption that job titles are reported without differential misclassification for cases as well as controls. We also assume that exposure data are reported with the same accuracy by workers in different jobs. Under these assumptions, the proposed indicators will provide valuable-although not conclusive-signs of recall bias. The principles elaborated in this paper can be applied to other situations in which self-reported exposure data are used.