Longitudinal data are essential to research in criminology and criminal justice. Despite attrition's implications for validity, understanding its sources is underexplored empirically. We examine the correlates of retention using covariates organized into domains of prediction, prevention, and projection. Data from the LoneStar Project, a three-wave longitudinal reentry study of 802 males recently released from prisoners in Texas, were analyzed to examine the correlates of proximal, distal, and any study retention. The best correlates of study retention are prevention techniques used by researchers to reduce attrition. In contrast, only a few covariates traditionally associated with attrition and no covariates used for attrition projection were related to retention. What researchers do matters more for retention than the characteristics of individuals they are trying to retain. The findings underscore how researchers can improve study retention in longitudinal research while also correcting for non-random attrition in current longitudinal data sources.
Keywords: high-risk; longitudinal research; panel studies; reentry; retention.