NCJ Number
85295
Date Published
1982
Length
34 pages
Annotation
Forty-four States responded to Florida's survey on forecasting methods for corrections agencies, expressing concern over the reliability of linear and multiple regression approaches and discussing several simulation modeling techniques. Florida describes its Simulated Losses/Admissions Model (SLAM).
Abstract
Nearly half of the respondents were using simple linear regression of historical population points, 16 percent used multiple regression, and 14 percent had used simulation modeling techniques. The survey did not discover any particular pattern in the degree of sophistication in projecting methods, but did find widespread displeasure with simple linear regression when the data base was the previous inmate population. A major shortcoming of either linear or multiple regression is that neither explains cause-and-effect relationships among variables that influence population growth, such as population at risk and unemployment. Most corrections forecasters consider linear regression unreliable unless policy variables affecting growth or decline are nondynamic. Multiple regression techniques are somewhat better. A review of simulation modeling used in a few States indicates that flow models are more reliable in generating forecasts that bear some resemblance to actual growth patterns, particularly when the State criminal justice system is involved in policy changes. An overview of different modeling techniques implemented by the Florida Department of Corrections in the 1970's covers SIMMODG, SPACE, JUSSIM (Justice System Interactive Model), and a model developed by ABT, Inc. It also discusses their shortcomings relative to Florida's forecasting efforts. Specifications for a simulation model, based on experiences with these methodologies, are outlined. Other lessons learned in developing SLAM are detailed. Tables, formulas, and graphs are provided.