Drop-out or loss to follow-up of participants in a study because of refusal to continue, failure to locate, or other problems. The concern is that non-random drop-out may make the sample no longer representative of the original population. In studies concerned with evaluating the effects of a particular treatment, attrition can be a major problem. The problem is that participants in the treatment group come to dislike the nature of the treatment, resulting in a disproportionate number of them leaving compared to the control group. In studies with infants, ‘drop-outs’ create a problem that has to be accounted for (viz., do they differ in any relevant ways from those who remain in the study?). Thus, it may be the case that infants who do not complete the study are in some way temperamentally different from those who do, and as a consequence having a bearing on the generalizability of the findings. In general, a resultant statistical problem is how best to account for the missing values (something that is claimed for multilevel modeling).
See Generalization, Longitudinal studies, Multilevel modeling (MLM), Panel studies