Does Missing Data in Studies of Hard-to-Reach Populations Bias Results? Not Necessarily

Abstract

Missing data are always an issue in community-based longitudinal studies, calling into question the representativeness of samples and bias in conclusions, the research has generated. This may be due to the difficulty of implementing random sampling procedures in these studies and/or the inherent difficulty in sampling hard-to-reach segments of the population being studied. In fact, the ability to accurately study hard-to-reach populations in light of potential bias created by missing data remains an open question. In this study, missing data are defined as both failure to interview potential research participants identified in the sampling frame and failure to retain enrolled research participants longitudinally. Using the sample from the Mobile Youth Survey, a multiple-cohort, longitudinal study of adolescents living in highly impoverished neighborhoods in Mobile, Alabama, we examined sample representativeness and dropout to determine whether missing data led to a nonrepresentative, and therefore, biased sample. Results indicate that even though random procedures are not strictly used to draw the sample, (a) the sample appears to be largely representative of the population that was studied, and (b) attrition is largely uncorrelated with characteristics of those who dropped out. This suggests that it is possible to study with validity hard-to reach populations in community settings.