Abstract
Loss of subjects from an experimental study can seriously undermine the interpretability of results. Although the degree of damage depends on several factors, the authors argue that the reasons for attrition should receive greater attention in the analysis and interpretation of results than has heretofore been customary. The authors describe a method for the collection and analysis of collateral data, i.e., measurements of the occurrence and frequency of events in subjects' lives that may affect outcome measures and/or lead to attrition. Once collected from both lost and retained subjects, these data were incorporated into a six-step procedure to evaluate the bias resulting from attrition. How collateral data are employed with new techniques for missing data estimation is also described.