Censored Regression in Response to the Distributional Realities of Crime and Justice Measures

Abstract

Regression modeling is the predominant analytic approach utilized in crime and justice studies. In its standard form, however, this analytic approach makes demands of data that often are ill‐suited to the realities of measuring criminal behavior and justice responses to it. Censored dependent variables that contain a strong “floor” or “ceiling” in their observed values represent one case in point. This paper considers the prevalence of such variables in crime and justice research, highlights the implications of the problem, and discusses some potential alternatives. In doing so, it presents examples from the extant literature to illustrate key principles. The discussion concludes by offering some practical guidelines for application and information on available software tools.