NCJ Number
217855
Journal
Journal of Experimental Criminology Volume: 2 Issue: 1 Dated: Spring 2006 Pages: 23-44
Date Published
2006
Length
22 pages
Annotation
This paper argues that the instrumental-variables (IV) methods used by economists to solve omitted-variables bias problems in observational studies also solve the major statistical problems (treatment dilution and treatment migration) that arise in imperfect criminological experiments.
Abstract
An analysis of the Minneapolis domestic violence experiment (MDVE) shows that IV estimation generated outcome effects that were approximately one-third larger than with the traditional analytical method. Treatment migration and treatment dilution are features of the MDVE one of the most influential randomized trials in criminological research. It focused on resolving the debate as to whether mandatory arrests of suspected perpetrators of domestic violence would be more effective in deterring such assaults compared to other police responses. This was done by analyzing outcomes for various types of police responses to domestic violence calls. An analysis that contrasts outcomes according to the treatment delivered is likely to be misleading, because all features of a given situation are not taken into account when correlating outcomes with specified and measured variables. A commonly used approach for the analysis of randomized clinical trials with imperfect compliance is to compare subjects according to original random assignment, ignoring compliance entirely. This is known as an "intention-to-treat" (ITT) analysis; however an ITT effect provides a poor predictor of the average causal effect of similar interventions in the future in the event that future compliance rates differ substantially. This article explains how the IV framework is the simplest and most robust solution to the treatment-dilution and treatment-migration problem. 5 tables and 42 references