Skip to main content
Open Access Publications from the University of California

UC Santa Cruz

UC Santa Cruz Electronic Theses and Dissertations bannerUC Santa Cruz



In 1999, Washington State decided to allocate supervision based on offenders' risk characteristics. These risk characteristics were measured by the actuarial instrument ``Level of Service Inventory - Revised.'' In analyzing the data, I discovered the unusual administration of the instrument by the local bureaucracy that resulted in many offenders being bumped to a higher supervision level. Using a regression discontinuity design, I uncover the mechanics of the bumping up process. After cleansing the instrument of the manipulation, I find that the manipulated instrument predicts time to recidivism better than the cleansed instrument. Moreover, using a regression discontinuity design I evaluate whether offenders of similar risk characteristics recidivate less if they receive a more intense supervision level. I find that offenders who received more supervision did not recidivate less. I also examined whether the authorities targeted certain groups of offenders more than others in the process of administering the instrument. Specifically, I tried to find out whether their behavior exhibits signs of what is known in the literature as ``preference or prejudice-based discrimination'' as opposed to ``statistical discrimination.'' I did not find conclusive evidence that certain groups were bumped up more than others. Moreover, even if we accept that certain groups are bumped up more, this can be justified by their higher recidivism rates, which amounts to ``statistical discrimination.'' Therefore, I found no evidence of unequal treatment in the administration of the risk assessment instrument. Finally, I found that I can significantly improve the predictive validity of the LSI-R instrument by having the items weighted. The results were found to be robust in a validation sample as well.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View