Bias of retrospective reporting to combat missingness in longitudinal research

Abstract

Longitudinal studies are invaluable for understanding psychological processes over time but are often plagued by attrition, potentially compromising study results. Researchers may address this issue by supplementing their samples with new participants. Supplemental sampling refers to the addition of new participants to a study at later waves to compensate for attrition, such that they do not participate at the first time point. In doing so, researchers may ask participants to provide retrospective reports of their past experiences. Although this approach offers an intuitive solution, it is unclear whether or not this approach may introduce bias if the retrospective reporting is not accurate. Our research employs a simulation-based approach to systematically investigate how inaccuracy in retrospective reporting affects longitudinal studies. Six potentially influential factors are manipulated in the simulation study, including sample size, effect size, size of supplemental sample, missing data mechanism, proportion of attrition, and the type and level of retrospective reporting bias. Conditions are compared against a baseline with complete, non-biased data and conditions where no retrospective reporting is included. Results indicated that estimation accuracy decreases significantly with biased retrospective reporting. Alternative approaches to retrospective reporting with supplemental samples are discussed.