Assessing the non-random sampling effects of subject attrition in longitudinal research

The potential effects of attrition in longitudinal research are addressed and a procedure for assessing its effects is recommended. We recommend that researchers assess the effects of subject attrition on their data by assessing: (1) the presence of non-random sampling using multiple logistic regression, (2) mean differences on the study's variables between those who responded and did not respond to the subsequent data collection, (3) the restriction or enhancement of variances, and (4) changes in relationships among variables due to attrition. We demonstrate the procedure using data collected from a random sample of employed adults in the US regarding job satisfaction, job characteristics, demographics, and mood. In our data, subject attrition led to nonrandom sampling, affected the means and variances of some of the variables, but did not affect the relationships among the variables. The effects of subject attrition may be sample specific, but the procedure recommended for assessing its effects may be used in other data sets and substantive areas.