Linear Panel Analysis: Models of Quantitative Change.

immaterial. In repeat victimization there is a tendency to repeat the same ype of c im (or nearly he same), but otherwise the transition matrix is homogeneous. Two papers discuss the benefit of using logistic regression as opposed to linear regression for predicting an individual's probability of success on parole from his characteristics. Carefully constructed models perform somewhat better than simple Burgess scales of predictor variables (counting yes, no or 0, 1, 2 on nine items), but surprisingly, the improvement is too modest to imply that one should use the theoretically correct methods. Two good papers on the deterrence effects of punishment seem dated after the publication of a comprehensive review by a National Academy of Sciences panel (Blumstein, Cohen and Nagin 1978). The family of JUSSIM models is described by Blumstein and Koch. JUSSIM I is a deterministic, steady-state, nonqueuing simulation of flows of individuals (arrestees, defendants, etc.) through stages of the criminal justice system. JUSSIM II expands on JUSSIM I by incorporating the feedback flow of recidivists through subsequent arrests. The paper focuses on JUSSIM III, a version that incorporates a crime generation process and a victim generation process. An unusually clear and useful description is given that shows how to construct the model's input data from available criminal justice system statistics. The final paper describes a fascinating redesign of the Kansas City Preventive Patrol Experiment, by Fienberg, Larntz, and Reiss. This controversial experiment, carried out in 1972-73 by a Police Foundation team headed by Kelling, is remarkable both for the wealth of detail and information it produced and for the vigorous criticisms it provoked. The experiment claimed to show that routine police preventive patrol has no effect on crime rates or citizens' sense of fear or opinion of police services. Although subsequent research has attempted to resolve the residual questions by means different from those suggested here by Fienberg et al., the paper is interesting because it shows how difficult it is to carry out a social experiment with a design adequate to identify reasonable-sized causal effects. The approach suggested here is a randomized controlled field trial with a balanced crossover design.