Incorporating Co-offending in Sentencing Models: An Analysis of Fines Imposed on Antitrust Offenders

Analyses of sentencing (and other criminal justice processes such as the decision to prosecute, plea bargaining, and contact with the police) often use the isolated individual as the unit of analysis. However, the criminal justice system often processes either offenses or court cases rather than persons. If court cases always involved one individual, this would have little impact. However, offenses involving co-offending—two or more persons acting together—comprise a substantial proportion of criminal activity (Reiss, 1980, 1986). Depending on the prevalence of co-offending, it may be very likely that two or more individuals involved in the same case will be selected as members of the same sample of criminal justice or criminological data. Unless it can be shown that both the individual-level variables of co-offenders and their error terms are mutually independent, analyses based on methods such as ordinary least-squares multiple regression would violate the underlying assumptions of such models. However, alternatives to linear models assuming either type of independence are available. Among the most useful of these are mixed models, specifically those assuming compound symmetry. This is illustrated with an analysis of fines imposed on criminally convicted antitrust offenders. These models may yield results which are substantially different than those from models which ignore co-offending. In a model of fines imposed on antitrust offenders, models which ignore co-offending generally overstate both estimates and statistical significance of offense-level variables and understate those of offender-level variables.