For public safety reasons, the public and policymakers routinely consider recidivism rates to be an important, if not paramount, measure of the corrections system's performance. This report demonstrates how certain factors influence recidivism rates for inmates released from Department facilities. This information provides proper context for interpreting results of evaluations that may use recidivism rates to assess the Department's performance.
Knowing how much these factors collectively influence recidivism suggests how much room there might be for policy or program interventions to reduce recidivism. This information also allows the Department to know whether increases or decreases in recidivism are due to changes in characteristics of the inmate population released as opposed to changes in policies, programs, or resource allocations. With this knowledge, recidivism rates can serve as correct, useful baselines for measuring whether and how well policies or programs expected to reduce recidivism actually work.
The combined and relative effects of these factors were determined using a statistical procedure (proportional hazard regression) that analyzes all factors' effects on recidivism simultaneously. This allows us to know several things about the factors:
For information on the statistical procedures used to evaluate these prediction models and estimate the effects of factors on recidivism, see Statistical Analysis.
One way to show how much these factors, as a group, affect recidivism is to test how well they improve the ability to predict inmate recidivism. Table 3 below shows the predictive power of the model for each gender and recidivism measure at three follow-up periods. For example, consider the male reoffense rate at 36 months follow-up. Using only the base rate of male reoffending one would correctly guess whether a male reoffends in 50.7% of cases—not much better than a coin toss. However, using a statistical model with factors that influence recidivism, one correctly identifies whether a male reoffends in 64.6% of cases: an improvement of 13.9 percentage points. The predictive power of these combined factors is determined not by the percentage of cases the model correctly classifies, but rather by how much higher that percentage is than chance guessing. These factors predict recidivism from 6.3 to 15.1 points better than chance does, depending on the cohort, recidivism measure, and follow-up period of each model.
|Table 3. Summary of Predictive Power of Combined Factors|
|Follow-up Period Analyzed||Prediction Component||Reoffense||Reimprisonment|
|18 MONTHS||Correct by Chance||55.8%||62.5%||76.0%||86.8%|
|Correct by Model||68.2%||73.4%||86.0%||93.1%|
|36 MONTHS||Correct by Chance||50.7%||50.2%||53.9%||66.2%|
|Correct by Model||64.6%||61.9%||68.0%||79.0%|
|60 MONTHS||Correct by Chance||74.1%||66.6%||60.3%||50.4%|
|Correct by Model||85.1%||79.2%||75.4%||64.9%|
Knowing the combined effect of factors on recidivism does not tell us which factors influence rates more than others. The combined effects of these factors can be broken out into how much variation in recidivism each individual factor accounts for. Table 4 below ranks the factors from most (1st) to least (up to 18th) influence for each gender cohort and recidivism measure over the entire 60-month follow-up period. These rankings indicate the relative importance of each factor. The rankings allow one to know which factors affect recidivism more or less than others for the release cohorts as a group. The rankings do not necessarily mean that a particular factor will be similarly important for subgroups of each cohort.
|Table 4. Rank of Individual Factor Effects|
|Age at Release (years)||2 **||2 **||2 **||2 **|
|Race (Black)||7 **||11 ||5 **||6 |
|Ethnicity (Hispanic)||14 ||12 ||17 ns||13 ns|
|High||15 ||14 ns||15 ||17 ns|
|Low||10 **||15 ns||7 **||15 ns|
|Post-Release Supervision||4 **||7 **||14 **||18 ns|
|Time in Prison (months)||6 **||3 **||8 **||8 |
|Disciplinary Reports (#)||5 **||4 **||4 **||7 |
|Education Level (grade)||3 **||10 ||3 **||5 |
|Prior Recidivism (#)||1 **||1 **||1 **||1 **|
|Most Serious Career Crime:|
|Homicide||13 **||9 ||13 **||12 ns|
|Sex / Lewdness||9 **||13 ns||12 *||14 ns|
|Robbery||18 ns||17 ns||10 **||9 |
|Other Violent||17 ns||16 ns||18 ns||11 ns|
|Burglary||12 **||6 **||6 **||3 **|
|Property Crimes (#)||8 **||8 **||9 **||10 |
|Drug Crimes (#)||11 **||5 **||11 **||4 **|
|Weapons Crimes (#)||16 ns||18 ns||16 ||16 ns|
|Males:||prior recidivism, age, education level, disciplinary reports, property crimes, drug crimes, homicide (worst crime), high custody, other violent (worst crime).|
|Females:||prior recidivism, age, drug crimes, ethnicity, sex/lewdness (worst crime), low custody.|
|Males:||supervision after release, sex/lewdness (worst crime), ethnicity, time in prison.|
|Females:||supervision after release, time in prison, disciplinary reports, homicide (worst crime), high custody, property crimes.|
|Males:||robbery (worst crime), burglary (worst crime), low custody, race.|
|Females:||robbery (worst crime), education level, race, other violent (worst crime), burglary (worst crime), weapons crimes.|
The relative order of influence each factor has on recidivism is different from the direction and actual size of that influence. Some influential factors raise recidivism rates, while others lower rates; and some factors raise or lower rates more so than others. The general relative effect of a factor can be described in one of two ways, depending on the nature of the factor variable. Table 5 summarizes the general effect of each factor.
For indicator variables (e.g., black) measured as a dichotomy (black or non-black), the hazard ratio from each proportional hazard regression model are interpreted as in the following examples. Holding other factors constant, black males are 29% more likely to reoffend and 31% more likely to be reimprisoned than non-black males. Black females are 13.5% less likely to reoffend and 21.7% less likely to be reimprisoned than non-black females, holding other factors constant.
For continuous variables (e.g., age), parameter estimates from each proportional hazard regression model are interpreted as in the following examples. Holding other factors constant, for males each year older at release lowers the reoffense likelihood 3.2% and lowers the reimprisonment likelihood 3.3%. For females, each year older at release lowers the reoffense likelihood 2.9% and lowers the reimprisonment likelihood 3.4%, holding other factors constant.
For recidivism rate curve charts illustrating each factor's individual effect, click on the particular cell in the table below. These rates reflect the partial effect of each factor on recidivism after controlling for other factors' influences. The charts demonstrate how much each factor influences recidivism rates. Each curve chart shows the relationship between a factor or factors and recidivism rates over the full follow-up period (60 months). A separate chart is provided for effects on reoffense and reimprisonment for males and females. In some cases for convenience, than effects of more than one factor are displayed on a single chart. Factor data for continuous variables (e.g., age) are grouped into categories only to facilitate displaying the basic relationship each has to recidivism rates. For an explanation of how to read these charts, see Recidivism Rate Curves. For more information on the methodology and statistical procedure used to measure these factors' relative influence on recidivism, see Statistical Analysis.
These interpretations of the general effect of factors are relative. The percentages in Table 5 do not refer to actual percentage point reductions in recidivism rates. The rates in the associated charts are hypothetical, assuming that all other factor effects are held constant. To view actual recidivism rates for inmates in factor categories, see Tables of Actual Rates by Factor - Unadjusted.