Selective Incapacitation and Public Policy: Evaluating California's Imprisonment Crisis

Chapter 7. Modeling the California Criminal Justice System, Part I: Reproducing and Evaluating the Past

103 Chapter 7 Modeling the California Criminal Justice System, Part I Reproducing and Evaluating the Past This chapter reports the results of the simulation analysis that reproduces the compositional dynamics of the California criminal justice system during the period 1979-1998. The purpose of this chapter is twofold. The first is to demonstrate the validity of the simulation model in terms of its ability to accurately reproduce historical circumstance.The second goal of this chapter is to retrospectively evaluate the California criminal justice system with respect to its success at selectively incapacitating dangerous offenders. Chapter 8 discusses the results of the projection analyses, which are dependent on the plausibility of this baseline model. As chapter 6 explained, dynamic systems simulation modeling is based on the integration of system flows over time.This is accomplished by calculating the values of variables in a complex system of differential and difference equations that are initialized with values based on actual data, and then iterated over time to reproduce an historical sequence of events. System states are represented by differential equations, such as STATE (t) ⫽ STATE (t⫺1) ⫹ INFLOWS ⫺ OUTFLOWS* dt Simply put, the composition of a population state at time (t) is equal to the population at time (t⫺1),1 plus new arrivals, and minus exits. The inflows and outflows are described by rate equations, which represent the proportion of the “sending” state that moves into the “receiving” state at each time step. For example , the equation that models the prison population at time (t) takes the following form: PRISON (t) ⫽ PRISON (t⫺1) ⫹ [(STREET (t) * prison commitment rate from STREET) ⫹ (JAIL(t) * prison commitment rate from jail) ⫹ (PAROLE (t) * return rate)] ⫺ [ (PRISON(t) * parole rate) ⫹ (PRISON(t) * unconditional release rate)] * dt In this case, the inflows consist of new commitments to prison resulting from criminal convictions and recommitments resulting from parole violations; the outflows consist of parole releases and unconditional releases. The rate equations are in turn modified by another quantity, which may accelerate or decelerate the rate over time.2 This elaborated form of the equation is PRISON (t) ⫽ PRISON (t⫺1) ⫹ [ {(STREET (t) * prison commitment rate from STREET) * street conviction rate modifier} ⫹ {(JAIL(t) * prison commitment rate from jail) * jail conviction rate modifier} ⫹ {(PAROLE (t) * return rate) * return rate modifier}] ⫺ [ {(PRISON(t) * parole rate) * parole rate modifier} ⫹ {(PRISON(t) * unconditional release rate) * release rate modifier}] * dt Following these principles, a simulation model is estimated by constructing an equation system that codifies the structural relationships depicted in fig104 Evaluating the Past, Choosing the Future FIGURE 7.1 Structural Model of the California Criminal Justice System ure 7.1 for each of the four hundred fifty population groups, and then compiles the results of the subgroups to represent the entire system.3 The modeling of the subpopulations also allows for the estimation of the historical time-shapes of various system components for a variety of populations of interest. In the present analysis, the most pertinent of these is the dangerousness classification scheme; however, this modeling strategy permits the examination of the system histories of a variety of other sorts of populations, such as black men aged 18–24, Hispanic women, and violent offenders—in fact, system histories can be generated for any population characterized by any of the attributes tracked in the simulation model. THE BASELINE MODEL The following section demonstrates the validity of the simulation model produced by this strategy. Figures 7.2 through 7.6 depict the graphic time series data for each of the system populations modeled, compared with the simulated data for each of those populations.4 One can see that the simulated series mirror the characteristic shapes of the actual data quite closely. In order to provide a target measure for goodness-of-fit, a maximum average Modeling the California Criminal Justice System, Part I 105 FIGURE 7.2 Arrested Population, 1979–1998 Note: Mean difference between series is 0.2% of target value (standard deviation ⫽ 6.5%). FIGURE 7.3 Jail Population, 1979–1998 Note: Mean difference between series is 4% of target value (standard deviation ⫽ 5.7%). FIGURE 7.4 Probation Population, 1979–1998 Note: Mean difference between series is 3% of target value (standard deviation ⫽ 10.7%). 106 FIGURE 7.5 Prison Population, 1979–1998 Note: Mean difference between series is 3% of target value (standard deviation ⫽ 10.0%). FIGURE 7.6 Parole Population, 1979–1998 Note: Mean difference between series...