Use the links below to skip to the specific appendix you wish to view:
Technical Appendix: Data and Methodology
Our analysis of the effect of Corrections' CBT rehabilitation programs on recidivism rates for inmates released in fiscal year 2015–16 drew on data from Corrections and DOJ. During fiscal year 2014–15, Corrections offered all four CBT programs at 11 men's prisons. We compared recidivism rates and other data for inmates that received CBT classes to the data for those that did not. Our analysis examined inmates who completed CBT programs but excluded inmates who completed programs for long‑term offenders, defined as inmates with an indeterminate sentence with the possibility of parole. We used a regression analysis to control for many observable characteristics between the test and control groups in an effort to isolate the effect on the recidivism rates of inmates completing at least half of their assigned CBT classes. Due to limitations in Corrections' data, we were able to examine only the one‑ and two‑year recidivism rates. In developing our methodology we consulted with the director of the Center for Evidence‑Based Corrections at UC Irvine, and a statistician reviewed our statistical model and its results.
Our analysis used data from both SOMS and DOJ. Corrections implemented its SOMS inmate database statewide in October 2014. It was designed to consolidate existing databases and replaced multiple manual paper processes to standardize adult inmate data and inmate population management practices. SOMS includes personal information on individual inmates, such as their date of birth, education level, and current location within the prison system. SOMS also includes information on the programming inmates received, such as when an inmate is enrolled in or completed a rehabilitation class. We also obtained conviction data from DOJ to determine whether an inmate recidivated within two years of his release from prison.
We obtained SOMS data to calculate recidivism rates for inmates released in fiscal year 2015–16. Corrections began using SOMS to track whether inmates completed CBT classes in October 2014. Due to this data limitation, we were able to select only inmates that completed CBT classes from October 2014 through their release in fiscal year 2015–16 to determine whether the inmates recidivated within two years.12 Thus, inmates released in fiscal year 2015–16 are the earliest release group our analysis could examine and still calculate the one‑ and two‑year recidivism rates.
Our analysis required inmates to have completed CBT classes for at least half of their needs as the minimum before being considered to have had their rehabilitative needs met. Corrections assessed inmate needs in four CBT categories (anger management, substance abuse, criminal thinking, and family relationships). Inmates who take classes that address at least half of their assessed needs should have reduced their likelihood to recidivate because Corrections' contracts require vendors to use only evidence‑based programs, meaning that the classes have been shown to be effective. We used half, rather than all needs met, due to the limited number of prisons that offered CBT classes during the time period we examined. If we had focused only on inmates who had all of their needs met, it would have severely limited the sample size for our analysis.
To create the control and test groups for our analysis, we split the cohort from the 11 prisons that were released in fiscal year 2015–16 into inmates who had at least half of their CBT needs met and inmates who had not been assigned to CBT classes. Corrections' policies prioritize inmates' assignment to any CBT classes based on a combination of their COMPAS and CSRA scores. Our analysis examined inmates who were rated as having a moderate to high need in at least one of the CBT categories. Given that our test and control groups represented subsets of the prison population, we took three steps to adjust for the potential differences in characteristics between the two groups. First, we grouped inmates within the test group by the total number of their CBT needs and their level of need for each of the four CBT categories. Second, we matched a selection of inmates from the control group with each member of the test group based on their CBT needs combination. We did this to ensure that we compared inmates who had the same CBT needs and who differed in whether or not they completed at least half of the classes they were assigned. Finally, we controlled for important observable differences between the control group and the test group, including age, self‑identified race, education (as measured by the inmate's most recent TABE reading score and whether or not the inmate had a verified high school diploma or equivalency), crime risk (as measured by the inmate's CSRA score), and prison assignment with variables in our regression analyses, as discussed below.
We calculated the overall recidivism rates between the control and test groups with a regression analysis of the correlation between inmates completing at least half of their assigned CBT classes (treatment) and recidivism rates using probit regressions. Our regressions examined the entire sample of the control and test groups to determine the relationship between completing CBT classes and inmates' one‑ and two‑year recidivism rates while controlling for prison location, age, self‑identified race, education, and crime risk. We examined the correlation between completing CBT classes and the one‑ and two‑year recidivism rates for each subset of our control variables. This means that we conducted individual regressions for prison, self‑identified race, education level, age, and crime risk. Table A shows the results of these regressions, including the percentage change in recidivism for inmates completing CBT classes and the probability (p-value) that the results could be caused by chance. The likelihood of obtaining a result that appears to be significant, but actually occurs by chance, increases with the total number of regressions performed. Nonetheless, we are confident in our analysis that did not find an overall relationship between inmates completing CBT rehabilitation programs and their recidivism rates.
|Factor||Category||P‑value*||Percentage decrease or increase in One‑year recidivism||P‑value||Percentage decrease or increase in Two‑year recidivism|
|Prison||Avenal State Prison||0.806||0.671|
|California City Correctional Facility||0.288||0.821|
|California Institution for Men||0.894||0.656|
|California Men's Colony||0.636||0.136|
|Correctional Training Facility||0.411||0.789|
|Chuckawalla Valley State Prison||0.936||0.162|
|High Desert State Prison||0.471||0.677|
|Ironwood State Prison||0.750||0.162|
|California State Prison, Los Angeles County||0.086†||‑15%||0.063†||‑18%|
|California Substance Abuse Treatment Facility||0.016‡||‑10%||0.000§||‑18%|
|Valley State Prison||0.319||0.732|
|High Risk ‑ Drug||0.086†||11%||0.199|
|High Risk ‑ Property||0.106||0.794|
|High Risk ‑ Violence||0.002§||‑13%||0.001§||‑16%|
|High school diploma/GED||0.278||0.083†||‑6%|
|No High school diploma/GED||0.452||0.592|
|Age||18 to 29 years old||0.255||0.269|
|30 to 39 years old||0.090†||‑5%||0.324|
|40 to 49 years old||0.458||0.604|
|50+ years old||0.707||0.958|
Source: Regression output based on SOMS and DOJ data.
* The p‑values test the null hypothesis that inmates' assignment to at least half of their needed CBT classes has no effect. The lower the p-value, the more likely you are to reject the null hypothesis. A higher p‑value indicates that you cannot reject the hypothesis. In other words, a low p‑value is likely to be meaningful. Therefore, we only present recidivism increases or decreases for p‑values of 0.10 or less.
† p < 0.1
‡ p < 0.05
§ p < 0.01
Scope and Methodology
The Joint Legislative Audit Committee (Audit Committee) directed the California State Auditor to perform an audit to examine the effectiveness of in‑prison rehabilitation programs at Corrections. Table B outlines the Audit Committee's objectives and our methods for addressing them.
Assessment of Data Reliability
The U.S. Government Accountability Office, whose standards we are statutorily required to follow, requires us to assess the sufficiency and appropriateness of computer‑processed information that we use to support our findings, conclusions, and recommendations. In performing this audit, we obtained Corrections' inmate information to identify inmates' demographics, rehabilitative needs, locations, and program participation. Additionally, we obtained DOJ's statewide conviction information to identify inmates who recidivated. To evaluate these data, we performed electronic testing of the data, reviewed existing information about the data and systems, and interviewed agency officials knowledgeable about the data. However, during our review we identified data limitations. Specifically, program participation information was not centrally tracked until the current system was implemented in October 2014. Additionally, we were unable to identify the needs of some inmates because Corrections does not perform assessments on all inmates. Therefore, we found these data to be of undetermined reliability for the audit purposes. Although this determination may affect the precision of the numbers we present, there is sufficient evidence in total to support our findings, conclusions, and recommendations.
|1||Review and evaluate the laws, rules, and regulations significant to the audit objectives.||Identified and reviewed the laws, rules, and regulations for the Inspector General, C‑ROB, Corrections, and CalPIA that were applicable to Corrections' in‑prison rehabilitation programs.|
|2||Identify the roles and responsibilities of Corrections and any other parties related to the oversight of state‑funded rehabilitation programs and assess the adequacy of their oversight, including the extent to which they ensure appropriate and consistent implementation across institutions.||
|3||Determine whether Corrections has conducted an assessment to determine the level of resources required to meet the rehabilitative needs of inmates in all of Corrections' facilities.||Obtained and evaluated assessments conducted by Corrections from 2012 through 2018 to determine if Corrections is meeting its resource allocation goals and whether those goals are adequate to meet the rehabilitative needs of its inmates.|
|4||Review and evaluate the effectiveness of the rehabilitation programs by determining the following:|
|a. Corrections' justification for the program's use and whether the program is based on evidence and research.||
|b. Corrections' method for evaluating the cost‑effectiveness of the program and whether it has considered investing in an independent oversight entity to perform this function.||
|5||Determine whether Corrections maintains adequate data on its rehabilitation programs—such as enrollment, attendance, and outcomes, by facility—to allow stakeholders to compare the effectiveness of rehabilitation resources across facilities.||
|6||Determine whether Corrections effectively assesses inmates' risks and rehabilitation needs and reviews and evaluates how it ensures that the tools used to identify the needs and risks are valid. Assess the adequacy of Corrections' policies, procedures, and practices for selecting and prioritizing inmates for participation in rehabilitation programs.||
|7||Identify and assess Corrections' performance measures—such as whether inmates' rehabilitative needs were met prior to their release and whether inmates are progressing in their programs—to evaluate the effectiveness of rehabilitative services.||
|8||Determine whether Corrections maintains a waiting list for rehabilitative programs, the number of inmates on the waiting list, how long they have been on the waiting list, their risks, and their needs.||
|9||Review and assess any other issues that are related to the audit.||Interviewed Corrections staff and obtained a complete list of all volunteer programs and determined whether Corrections evaluated these programs for their effectiveness.|
Source: Analysis of Audit Committee's audit request number 2018‑113, planning documents, and analysis of information and documentation identified in the table column titled Method.
The Number of Inmates on Waiting Lists and the Average Time Spent on the List
We reviewed Corrections' waiting list data, and we present the number of inmates on waiting lists as of July 2018 in Table C. This table shows that there were between 18,000 and 31,000 inmates on waiting lists as of July 2018, and that those inmates were on waiting lists for an average of between 211 and 351 days. However, these data are presented for informational purposes only because, as we discuss, we identified significant flaws with Corrections' waiting lists.
|Waiting list that addresses the need for:*||Number of inmates on waiting list||Average number of days on waiting list|
|Not target population||10,820||338|
|Not target population||23,008||351|
|Not target population||19,983||268|
|Not target population||21,524||261|
|Substance abuse disorder treatment|
|Not target population||18,172||289|
|Not target population||13,313||295|
Source: Analysis of SOMS data as of July 24, 2018.
* We excluded programs designated for inmates serving lengthy periods of incarceration, such as life with the possibility of parole.
† Corrections defines its target population as inmates with a moderate to high risk of recidivating and a moderate to high need for the program.
12 Individual CBT classes generally last three to four months and no longer than six months. Inmates would have had the opportunity to take multiple classes before their release during fiscal year 2015–16. Go back to text