The purpose of this study was to develop and
validate a reliable and accurate method for measuring the
effectiveness of offender classification systems to improve the
management of correctional facilities. In the early 1980s, the state
of Hawaii began classifying its prisoners with a newly developed
Federal Bureau of Prisons classification instrument. The goal of this
study was to estimate the validity of this new classification
instrument using Cox's Proportional Hazards model. The results were
then compared to a more traditional statistical procedure to
demonstrate the advantage of the new method in predicting
violence. Two prediction models, one at intake and one after six
months in prison, were used to evaluate the difference between the
traditional and the new methods. The first model, initial
classification (security total), used the sum of four variables and
was the only method that would be used for classification purposes for
the first six months. This security total was taken to be predictive
of violence. After this time, the reclassification prediction model
(custody total) was applied, using the sum of seven different
variables. This custody total variable was used as a major determinant
of reclassification. The two groups of inmates used were (1)
infractors, all inmates who had committed an act of violence while in
the institution from 1977 to September 1980, and (2) noninfractors, a
sample selected from the group of individuals who did not commit major
violations during the time period.
Two inmate samples were drawn from the Hawaii
State Prison case files. The infractors group was composed of 57
inmates who had committed an act of violence while in the institution.
A noninfractors group of 106 inmates was selected from the individuals
who had not committed a major violation during their incarceration.
For the infractors, a classification form was filled out for the
semiannual evaluation period immediately preceding each incident, and
for the evaluation period after the last incident. For the
noninfractors, one form was filled out for the semiannual evaluation
period immediately preceding the sampled time. The variable SECTOT
(the sum of the four initial classification variables) was the major
predictor of violence and was used to fit a proportional hazards model
to predict the initial incident of violence. In addition, a stepwise
analysis was run using the original variables rather than the SECTOT
variable as a further check.
Two samples of prison inmates were used, one group of 57
inmates who had committed infractions and another group of 106 inmates
who had no reported infractions.
All prison inmates in the Hawaii State Prison (now the
Oahu Community Correctional Center).
Inmates at the Hawaii State Prison.
Hawaii State Prison (now called the Oahu Community
administrative records data
Research variables include (a) initial
classification: offense (severity), expected length of incarceration
(sentence), type of prior commitments, and history of violence, and
(b) reclassification: percentage of time served, involvement with
drugs/alcohol, mental/psychological stability, most serious
disciplinary report, frequency of disciplinary reports, responsibility
that the inmate demonstrated, and family/community ties. In addition,
the collection supplies information on race and sex of inmates,
sentence limitation, history of escapes or attempts, previous
infractions, entry, reclassification, and termination dates (month and
year), and custody level.