ABSTRACT
In California, incarceration in the state prison system is in part organized by security level. The higher the security level, the more restrictive the setting. Upon arrival at a reception center, new inmates are scored within a classification system that is used to determine the appropriate level of security. In this paper, we report on the development and testing of a new inmate classification scoring system. Over 20,000 inmates took part in a randomized experiment in which half were assigned to their housing using the existing scoring system and half were assigned to their housing using the new scoring system. There were two key out- comes: (I) potential mismatches between the number of inmates assigned to different security levels and the available beds and (2) reports of inmate misconduct.
We conclude there to be some potential crowding problems, but that the new scoring system is much better than the old scoring system in sorting inmates by the likelihood of misconduct. We also conclude that some predictors popular in the past are no longer effective (e.g., marital status), while some new predictors are extremely powerful (e.g., gang activity), and that one can build in a number of mandatory housing placements (e.g., for sex offenders) and not degrade the overall effec- tiveness of the new classification system. Finally, the new classification system is shown to be more user-friendly than the existing classification system and well received by the staff responsible for implementing it.
