NCJ Number
191522
Journal
Sociological Methodology Volume: 28 Issue: 1 Dated: January 1, 1998 Pages: 99-137
Date Published
1998
Length
39 pages
Annotation
This study obtained bounds on treatment effects under weak nonparametric assumptions and explored how the bounds vary with the assumptions imposed; and in so doing, clarifies the source of common disagreements about the magnitudes and signs of treatment effects through a case study of sentencing and recidivism.
Abstract
The empirical analysis used the event-history data on juvenile offenders collected by Utah (National Juvenile Court Data Archive 1992). This report presents several sets of findings and shows how conclusions about treatment effects varied depending on the assumptions made. This report begins by presenting estimates under the assumption of random treatment selections and bounds obtained without making any assumptions about the process that determined treatment selection and outcome. The report then presents bounds obtained under two alternative models of judicial decision making. The "outcome optimization" model assumes that judges make sentencing decisions that minimize the chance of recidivism. The "skimming" model assumes that judges classify offenders as "higher risk" or "lower risk," sentencing the former to residential treatment and the latter to nonresidential treatment. Each model expresses an easily understood hypothesis about judicial decision making, and each has a potentially wide range of social science application. Neither model imposes the poorly motivated functional-form and distribution assumptions that mar parametric latent variable models and instrumental variables methods. Finally, the report brings to bear further prior information in the form of exclusion restriction, as it posits that specified subpopulations of offenders have the same response to treatment but face different treatment selection rules. 5 tables and 58 references