Missing Data Problems in Criminological Research: Two Case Studies

This paper considers the problem of missing data in two circumstances commonly confronted by criminologists. In the first circumstance, there is missing data due to subject attrition—some cases drop out of a study. In this context, analysts are frequently interested in examining the association between an independent variable measured at time t(x_t ) and an outcome variable that is measured at time t + 1(y_t + 1); the problem is that the outcome variable is only observed for those cases which do not drop out of the study. In the second circumstance there is missing data on an independent variable of interest for typical reasons (i.e., the respondent did not wish to answer a question or could not be located). In this case, researchers are interested in estimating the association between the independent variable with missing data and an outcome variable that is fully observed. Criminologists often handle these two missing data problems by conducting analyses on the subsample of observations with complete data. In this paper, we explore this problem with two case studies and we then illustrate the use of methods that directly address the uncertainty produced by missing data.