An earlier study conducted for CDC (Berk and de Leeuw, 1998) evaluated the existing inmate classification score system by which inmates were classified and then placed in different levels of security. While the data suggested that overall the procedures were placing inmates roughly consistent with Department's expectations, there was also evidence that improvements in the system could be made.

In this paper, we address how the inmate classification system was revised and then describe a randomized experiment to test the new system against the old one. Over 20,000 inmates took part in the experiment, which included two years of follow-up data. Key outcomes to be examined were the amount and type of misconduct in prison and the implications of the new system for prison crowding; would the current distribution of beds by security levels suffice if inmates were distributed to levels differently? They study is one of the largest randomized trials ever undertaken, and certainly the very largest criminal justice randomized experiment. We will focus not just on the design, but its implementation in a prison setting where placement decisions can have very serious consequences. Finally, we report the central findings and policy implications.