Page 1

Correctional Education as a Crime Control Program by

Audrey Bazos and Jessica Hausman UCLA School of Public Policy and Social Research Department of Policy Studies

Prepared for the United States Department of Education Office of Correctional Education John Linton, Director


Executive Summary




Correctional Education










Appendix A: Recidivism Studies


Appendix B: Effect Size Reduction


Appendix C: The Link Between Education and Crime


Appendix D: Cost per Crime Prevented: Method


Appendix E: Savings due to Prevented Re-incarcerations: Method


Appendix F: Potential Critiques


Appendix G: An Alternative Method of Calculating Cost-Effectiveness (not included in findings)




Spending on prison education programs fell even as prison populations and budgets soared during the 1980s and 1990s. In the current fiscal situation many states find themselves in, additional cutbacks in correctional education programs are expected. Do those reductions make sense, either from a crime-control perspective or from a long-term budget perspective? We know expanding prison populations works at reducing crime, but with a very high price tag. Prison capacity expansion has been estimated to prevent 60,000 to 340,000 crimes per year with a respective cost of 200 million to 5.5 billion dollars. Several studies have shown that prison education programs also significantly reduce crime. Once correctional education participants are released, they are about 10 to 20 percent less likely to re-offend than the average released prisoner. This study compares the cost-effectiveness of these two crime control methods educating prisoners and expanding prisons. One million dollars spent on correctional education prevents about 600 crimes, while that same money invested in incarceration prevents 350 crimes. Correctional education is almost twice as cost-effective as a crime control policy. Additionally, correctional education may actually create long-run net cost savings. Inmates who participate in education programs are less likely to return to prison. For each reincarceration prevented by education, states save about $20,000. One million dollars invested in education would prevent 26 re-incarcerations, for net future savings of $600,000.


INTRODUCTION State and federal funding for correctional education programs was significantly reduced throughout the 1990s while the total incarcerated population increased1. Many states, such as Florida, Illinois, Kentucky, Iowa and California, are further slashing correctional education budgets due to the current budget crisis2. These states and others, such as Ohio, Michigan and Kansas, are closing prisons to make the necessary cuts in state spending3. Budgets need to be cut. But states have a responsibility to protect public safety by controlling crime. How can states prevent crime with limited resources and save money in the long run? We will investigate the possibility that prison education programs are the answer.

CORRECTIONAL EDUCATION There are two basic types of correctional education programs – vocational training and literacy development. Vocational training courses focus on the acquisition of skills that are directly transferable to a workplace, such as appliance repair. Literacy development courses are loosely based on the traditional classroom model centered around the improvement of reading and math skills. There are two main reasons why researchers in this field believe in-prison education can reduce future criminal activity. The first involves the impact of increased cognitive skills on changes in behavior and the second contends that participants can learn how to live a crime-free life by participating in education courses.


Harrison and Beck, 2002. King and Mauer, 2002. 3 Justice Policy Institute, “What the States are Doing,” http://www.justicepolicy.org/article.php?id=27 2



Available evidence suggests that education programs in correctional

facilities increase literacy4. Increased educational attainment generally is associated with increased income, even among those with relatively low cognitive skills5,6. And increased income is associated with a decreased incidence of crime7. This can be explained because people choose between committing crimes and pursuing employment in the labor market. The risks associated with committing crimes are larger when having a job pays more, or getting a job is easier. As a result, choosing to commit a crime is a less attractive option to those who could earn more money with a legal job. An increase in an individual’s educational attainment is therefore likely to be associated with increased earnings, which is in turn associated with a decreased level of criminal activity. [See Appendix C] •

Socialization: Prison education programs give inmates the opportunity to learn “prosocial norms” by providing an enclave removed from the “criminal subculture” predominant among inmates8. Interacting with educators can familiarize inmates with the norms that law-abiding citizens observe while also reducing the feeling of “alienation that inmates tend to experience while in prison9.” The resulting improvement in social skills can make it easier for inmates to find and hold a job upon release, which in turn reduces their likelihood of re-offending.


Based on data from Minnesota, California and Texas, 3 states that keep track of the literacy progress of the inmates in correctional education. For more information, access the following websites: http://www.corr.state.mn.us/pdf/2001annualreport.pdf, http://www.tea.state.tx.us/adult/tables2002/ins3.pdf, http://www.cde.ca.gov/adulteducation/datacollect/fedprogdata/fedstudentdata99-00.html 5 Tyler, Murnane and Willett, 2000. 6 According to Tyler (2003), basic cognitive skills are those that most individuals obtain before completing the ninth grade. 7 Lochner and Moretti, 2002. Controlling for many factors, including age, cohort and state of residence. 8 Harer, 1995, p.2.


The vast majority of studies that have studied the impact of in-prison education on future criminal activity have found that participation reduces crime. We focused our analysis on the three studies that took the strongest methodological approach at investigating this relationship10. The first found that six months worth of participation in an education course among federal prisoners was responsible for a 15.7% reduction in re-arrests, even after accounting for a number of other factors known to predict recidivism11. The second found that Wisconsin inmates who complete a high school or adult basic education course are 20% less likely than the average offender to be re-incarcerated, again controlling for characteristics that predict recidivism12. Both of these studies used statistical regression methods to determine the impact of correctional education. Most notably, they both controlled for a number of factors believed to predict recidivism rates, such as age, race, and length of sentence. The study of federal prisoners even controlled for factors such as substance abuse and employment upon release from prison. [See Appendix A] Unfortunately, we were unable to include either the federal or Wisconsin study in our analysis. The complexity of the federal prison system, in that federal prisoners are contracted out to state and private prisons, prevented us from obtaining accurate correctional education budget estimates for the federal prisoners studied. While we were able to obtain the correctional education budget for Wisconsin institutions, the data provided did not differentiate between the budget for vocational courses and the budget for high school and ABE courses. As Piehl only


Harer, 1994, p.37. Our criteria for a strong methodological approach are the existence of a comparison group and some accounting for other characteristics that could predict recidivism. 11 Recidivism is defined as a relapse into criminal behavior. Though no exact measures of this exist, re-arrest and re-incarceration are regarded as the best estimate in correctional data. 12 Piehl, 1995. 10


analyzed the impact of adult basic education and high school programs on recidivism, we were unable to apply the available budget data to her findings. We base our calculations of cost per crime prevented on the findings of a third study – the ‘Three State Recidivism Study’ conducted by the Correctional Education Association. This study compared the re-arrest, re-conviction and re-incarceration rates of correctional education participants to non-participants in Maryland, Minnesota and Ohio three years after their release13. [See Appendix A] The following reductions in recidivism were found for each state: TABLE 1: REDUCTIONS IN RECIDIVISM RATES IN ‘THREE STATE’ STUDY State Maryland: Minnesota: Ohio: [See Appendix B]

% Reduction in re-incarceration

% Reduction in re-conviction

% Reduction in re-arrest

16.2% 33.3% 22.6%

13.5% 29.4% 21.2%

5.263% 22.22% 13.79%

Average % reduction in recidivism 11.6% 28.3% 19.2%

As the findings of the ‘Three State’ study were comparable to the other two studies which utilized sound research methodology, we were confident in using the ‘Three State’ study as the foundation of our analysis. Further, we were able to collect budget and enrollment data for Maryland, Minnesota and Ohio correctional education programs, which made it possible to conduct a cost-effectiveness analysis. In order to be conservative in our findings, we discounted the effect size found in the ‘Three State’ study by 50% before calculating our cost effectiveness and budget savings figures. [See Appendix B]


In the “Three State Recidivism Study,” correctional education included adult basic education, high school degree courses, GED courses, post-secondary academic programs, life skills and pre-release classes and vocational training.


INCARCERATION Obviously, there are ways to reduce crime other than educating prisoners. Increasing incarceration rates and lengthening prison sentences through “tough on crime” legislation prevent crime by incapacitating perpetrators. William Spelman surveyed a large body of research on crime statistics and criminal activity and found that a one-percent increase in prison populations would prevent 60,000 to 100,000 crimes per year14 for a total cost of over $200 million a year. We used these figures as a standard against which to compare the effectiveness of correctional education. RESULTS Cost Effectiveness Comparison What we attempt to do in the analysis that follows is compare the cost per crime prevented by correctional education to the cost per crime prevented through incarceration. In other words, if a state has a million dollars to invest in crime control, which method will prevent more crimes – educating inmates or keeping them imprisoned longer?15 The following information was collected in order to make this comparison; •

The average cost per participating inmate was calculated by dividing the total correctional education budgets of the three states’ programs by the total number of participants in all three states. The average cost per participant is $1,400.

In order to determine how participating in education classes might affect recidivism, we first needed to know the average number of crimes committed by the offender who never


Spellman, 1994, p.225 Note what we are not examining here. We do not consider other approaches to crime control, such as community-based programs for ex-convicts on parole or probation which may also be cost-effective. Unfortunately, little research has been done on the outcomes of these programs. We also do not address the social implications of correctional education programs – what they may do for the families of offenders or the communities to which they 15


receives these services. Based on self-reported criminal activity and crime statistics, William Spelman estimated that increasing prison capacity by 1% would prevent about 80,000 crimes per year. Using this estimate, we concluded that the average offender commits 9 crimes per year.16 •

According to the results of the ‘Three State’ study, correctional education is responsible for a ten percent reduction in recidivism (with the 50% discount applied). When we apply this rate to the nine crimes committed by the average offender, we see hat correctional education prevents about one crime per offender per year.

Therefore, the cost per crime prevented by correctional education is about $1,600.

This can now be compared to the cost of crime prevention through incarceration. The cost to incarcerate one inmate is $25,000 a year.

Imprisoning one offender prevents about nine crimes, so we get a cost per crime prevented through incarceration of $2,800.

Cost per Crime Prevented Comparison Average annual cost of education per inmate: Expected average number of crimes per offender per year: Reduction in recidivism due to education: Crimes prevented upon release per participant:

$1,400 9 10% .9

Cost per crime prevented by correctional education:


Annual cost to incarcerate one inmate:


Average number of crimes prevented by incarcerating one inmate: Cost per crime prevented by incarceration:

9 $2,800

return. Rather, the available research allows us to evaluate prison education programs as a crime control policy against a predominant alternative – expanding prison populations.


Framed in another way, we see that a $1 million investment in incarceration will prevent about 350 crimes, while that same investment in education will prevent more than 600 crimes. Correctional education is almost twice as cost effective as incarceration.17 [See Appendix D] To determine the strength of our cost-effectiveness findings, we conducted a sensitivity analysis. In our initial analysis we discounted the effect size of the ‘Three State’ study findings by 50%. Here we wanted to determine by how much we would need to discount the effect size of the study’s findings in order for the costs of education to break even with those of incarceration. What we found is that we would have needed to discount the effect size by a total of 72% in order for the costs of each crime prevention method to break even. Another way to look at this is that correctional education would only have to be responsible for a 6% reduction in recidivism for its costs to break even with those of incarceration. Research shows that the true effect of correctional education on reductions in recidivism is somewhere between ten and twenty percent. Budget Savings By preventing crimes, in-prison education is also preventing a number of future reincarcerations. Specifically, the ‘Three State’ study found that correctional education was able to reduce re-incarcerations by about 24%. Applying the fifty percent discount we applied to all the results from this study gives us a 12% reduction in re-incarcerations. By dividing the three


This estimate is congruent with other research that estimates the annual number of crimes committed by previously incarcerated felons. See Appendix D. 17 A second method of estimating the cost-effectiveness of correctional education was developed using the findings of RAND’s 1994 ‘Three Strikes and You’re Out’ study. Using substantially lower estimates of crimes committed per offender and looking only at California crime rates this study found that various sentencing alternatives for repeat offenders result in a cost ranging from $14,000 to $20,000 to prevent one serious crime. Using RAND’s estimates for crimes committed per offender, we found that it would cost $7,000 to prevent one serious crime through correctional education. This method indicates that correctional education is two to three times as costeffective as incarceration. [See Appendix G]


states’ total correctional education budgets by the number of prevented re-incarcerations, gives us a cost of $38,500 per prevented re-incarceration. A million dollars invested in correctional education can prevent 26 future re-incarcerations. However, if a state chooses not to invest in correctional education these future reincarcerations will not be avoided. With the average incarceration lasting 2.4 years18 at a cost of $25,000 per year, 26 incarcerations will cost the state almost $1.6 million dollars. Since avoiding these incarcerations through correctional education only costs $1 million, a state can gain a net savings of $600,000 in future costs for every $1 million it invests in correctional education today. Clearly, spending on prison education saves states money in the long run due to the prevented re-incarcerations of its participants. But states will not save this money if they do not make this investment – prisoners will just keep coming back. [See Appendix E] With these findings we again conducted a sensitivity analysis. The findings of the ‘Three State’ study would need to be discounted by a total of 78% in order to bring the future savings of correctional education down to zero. In other words, correctional education would only need to reduce re-incarcerations by 3% in order to eliminate future savings.

RECOMMENDATIONS Don’t cut what works States have a responsibility to ensure public safety, and cannot abandon this duty in times of budgetary crisis. Cutting funding for prison education programs is a bad decision. These programs prevent more crimes than increasing incarceration rates and lengthening sentences – and cost far less. They even save states money on correctional budgets by reducing the number of offenders who return to prison in the future.


Keep detailed, accurate records at correctional facilities Correctional education administrators should record as much information about their programs as possible. Crucial data includes the number of hours per week prisoners and teachers spend in the classroom, how prisoners perform on a literacy assessment before and after participating in the program, how many participants drop-out of classes, and how much each type of program (e.g. adult basic education, vocational education) costs per year. With these data, researchers would be able to determine the amount of classroom time necessary to create gains in literacy and decreases in recidivism. This would inform an effort to produce a maximally cost-effective correctional education system. Invest in further research Several large-scale studies should be conducted to add to the body of research on the effects of correctional education. These studies would need to have the following components: •

Random assignment: The most fundamental improvement that could be made to correctional education research would be random assignment – to randomly place in either a treatment or control groups those inmates who wish to participate in education programs. To make this politically feasible, find states with long waiting lists for correctional education19. Use the gap between supply and demand for these programs to justify the random selection and assignment of program participants.

Participant characteristics: Participants in both groups – treatment and control – should be within three years of release, but have no less than a year left on their sentences. The treatment group will need ample time to actively participate in their classes. Some matching will need to be done between the treatment and control

18 19

Bureau of Justice Statistics, 2001. Lawrence, Mears, Dubin, and Travis, 2002


groups to ensure that there are no significant differences in age, race, criminal history, incarcerating offense and education level. This will require a large sample size. •

Programmatic details: Researchers must clearly state the types of correctional education programs that have been implemented in their study. They should investigate the effects of participation in each type of program separately, such as vocational or adult basic education, so that conclusions can be made about which approach best reduces recidivism at the lowest cost.


Follow-up: Program participants and non-participants should be tracked for three years.

Information about how often recidivists are re-arrested per year (as opposed

to whether or not they were arrested in a given time period) should be collected to get a more accurate estimate of the true recidivism rate. •

Size: The population size (and funding) required for this type of study depends on the assumptions made. If we believe that the true effect of education on recidivism is around 20%, we would need 550 participants and an equally sized control group in order to have enough power to find significant results. If we believe that the true effect of education on recidivism is only 10%, we would need 1,800 participants with an equally sized control group in order to have enough power to find significant results.


REFERENCES Aos, S., Phipps, P., Barnoski, R., Lieb, R. 2001. “The Comparative Costs and Benefits of Programs to Reduce Crime.” Version 4.0. Washington State Institute for Public Policy. Beder, H. 1999. “The Outcomes and Impacts of Adult Literacy Education in the United States.” National Center for the Study of Adult Learning and Literacy, Harvard Graduate School of Education. Retrieved January 14, 2003, from www.gse.harvard.edu/~ncsall/research/report6.pdf. Blumstein, A., J. Cohen, and D. Nagin (eds.). 1978. “Deterrence and Incapacitation: Estimating the Effects of Criminal Sanctions on Crime Rates.” Washington, D.C.: National Academy of Sciences. Card, D. 1994. “Earnings, Schooling, and Ability Revisited.” Princeton University, Industrial Relations Section, Working paper #331. Retrieved January 14, 2003, from http://www.irs.princeton.edu/pubs/pdfs/331.pdf. Gendreau, P., Goggin, C., and Little, T. 1996. “Predicting Adult Offender Recidivism: What Works!” (User Report No. 1996-07). Ottawa: Department of the Solicitor General of Canada. Greene, J. and V. Schiraldi. 2002. “New Prison Policies for Times of Fiscal Crisis.” Washington, D.C.: Justice Policy Institute. Retrieved March 18, 2003, from http://www.justicepolicy.org/downloads/CuttingCorrectly.pdf. Greenwood, P.W., C.P. Rydell, A.F. Abrahamse, J.P. Caulkins, J. Chiesa, K.E. Model, and S.P. Klein. 1994. “Three Strikes and You’re Out: Estimated Benefits and Costs of California’s New Mandatory-Sentencing Law.” Santa Monica, CA: RAND. Grogger, J. 1998. “Market Wages and Youth Crime.” Journal of Labor Economics v16, n4: 754-791. Haigler, K., C. Harlow, P. O’Connor, and A. Campbell. 1994. “Literacy Behind Prison Walls.” National Center for Educational Statistics. Retrieved March 8, 2003, from http://nces.d.gov/publs94/94102.pdf. Harer, M.D. 1994. “Recidivism Among Federal Prisoners Released in 1987.” Federal Bureau of Prisons, Office of Research and Evaluation. Retrieved January 14, 2003, from http://www.bop.gov/orepg/oreprrecid87.pdf. Harer, M.D. 1995. “Prison Education Program Participation and Recidivism: A Test of the Normalization Hypothesis.” Federal Bureau of Prisons, Office of Research and Evaluation. Retrieved January 14, 2003, from http://www.bop.gov/orepg/orepredprg.pdf.


Harrison, P.M. and A.J. Beck. Bureau of Justice Statistics Bulletin: Prisoners in 2001. Retrieved December 3, 2002, from http://www.ojp.usdoj.gov/bjs/pub/pdf/p01.pdf. Kessler, D. and S. Levitt. 1998. “Using Sentence Enhancements to Distinguish Between Deterrence and Incapacitation.” National Bureau of Economic Research (NBER) Working Paper 6484. Retrieved March 10, 2003, from http://www.nber.org/papers/w6484. King, R.S. and M. Mauer. 2002. “State Sentencing and Corrections Policy in an Era of Fiscal Constraint.” Washington, D.C.: The Sentencing Project. Retrieved March 18, 2003, from http://www.sentencingproject.org/policy/pub9091.pdf. Lawrence, S., D.P. Mears, G. Dubin, and J. Travis. 2002. “The Practice and Promise of Prison Programming.” Washington, D.C.: Urban Institute. Retrieved December 5, 2002, from http://www.urban.org/UploadedPDF/410493_PrisonProgramming.pdf. Levitt, S. 1995. “Why do Increased Arrest Rates Appear to Reduce Crime: Deterrence, Incapacitation, or Measurement Error?” National Bureau of Economic Research (NBER) Working Paper 5268. Retrieved March 10, 2003, from http://www.nber.org/papers/w5268. Lochner, L. and E. Moretti. 2002. “The Effect of Education on Crime: Evidence from Prison Inmates, Arrests, and Self-Reports.” National Bureau of Economic Research (NBER) Working Paper 8606. Retrieved January 14, 2003, from http://www.econ.ucla.edu/moretti/lm46.pdf. Minnesota Department of Corrections. 2002. “2001 Annual Performance Report.” Retrieved December 5, 2002, from http://www.corr.state.mn.us/pdf/2001annualReport.PDF. National Center for Education Statistics. (n.d.) “Prose Literacy and Sample Items.” Retrieved March 4, 2003, from http://nces.ed.gov/naal/defining/measprose.asp. National Institute for Literacy. (n.d.) “Correctional Education Facts.” Retrieved December 5, 2003, from http://novel.nifl.gov/nifl/facts/correctional.html. Pastore, A.L. and K. Maguire (eds.). 2001. Sourcebook of Criminal Justice Statistics [Online]. Retrieved January 15, 2003, from http://www.albany.edu/sourcebook. Piehl, A.M. 1995. Learning While Doing Time. Kennedy School Working Paper #R94-25. Electronic copy received from author. Spelman, W. 1994. Criminal Incapacitation. New York, N.Y.: Plenum Press. Steurer, S., L. Smith, and A. Tracy. 2001. “Three State Recidivism Study.” Correctional Education Association. Retrieved December 5, 2002, from http://www.ceanational.org/documents/3StateFinal.pdf.


Texas Education Agency. “Participant Progress, Separation and Attendance by Educational Functioning Level.” 2002. Retrieved January 13, 2003, from http://www.tea.state.tx.us/adult/tables2002/ins3.pdf Tyler, J., R. Murnane, and J. Willett. 2000. “Cognitive Skills Matter in the Labor Market, Even for School Dropouts.” National Center for the Study of Adult Learning and Literacy, Harvard Graduate School of Education. Retrieved January 14, 2003, from www.gse.harvard.edu/~ncsall/research/report15.pdf. Tyler, J. 2003. “Basic Skills and the Earnings of Dropouts.” Brown University Department of Economics, Working Paper 2002-09. Electronic copy received from author. Previous version available from http://www.econ.brown.edu/2002/wp2002-09.pdf.


APPENDIX A: RECIDIVISM STUDIES Nearly all the studies on this topic have found that in-prison education program participation significantly reduces recidivism. However, the majority of these studies have serious methodological flaws in that they do not address the issue of selection bias. It could be that the significant results on recidivism were not due to the education program, but that program participants were somehow different from non-participants. Prisoners are not randomly assigned to education programs. Rather, participation to some extent and successful completion to a large extent are voluntary. This calls into doubt the reliability of the findings in many of these studies. The inmates who choose to enroll and actively participate in these programs are precisely the same inmates who may be more inclined to stay out of prison upon release. The reduction in recidivism would therefore be due to a positive selection into the education programs by prisoners that are less likely to recidivate to begin with. The three studies used in our analysis – Prison Education Program Participation and Recidivism: A Test of the Normalization Hypothesis conducted by Miles D. Harer, Ph.D. of the Federal Bureau of Prisons, Learning While Doing Time conducted by Anne Morrison Piehl of the Kennedy School of Government and Three State Recidivism Study conducted by the Correctional Education Association – all address the selection bias issue.


1. Prison Education Program Participation and Recidivism: A Test of the Normalization Hypothesis, by Miles D. Harer of the Federal Bureau of Prisons (1995) Dr. Harer’s hypothesis is that prison education programs are “normalizing” in that they reduce “prisonization” by nurturing pro-social norms. Prisonization refers to the alienation of the prisoner from the prison staff and management, and from the larger society due to the five pains of imprisonment: isolation form the larger community, lack of material possessions, blocked access to heterosexual relationships, reduced personal autonomy, and reduced personal security (Sykes 1958). In addition, many inmates are dedicated to a “criminal subculture” which in combination with the deprivations of imprisonment leads to a “normative orientation hostile to prison management and supporting a continuation of criminal behavior after release from prison (Irwin and Cressey, 1962, p.1). Harer argues that normalizing programs, such as prison education “facilitate humane treatment of inmates, open lines of communication between staff and inmates, and provide opportunities for diversion from the pains of imprisonment in ways that legitimate and reinforce law-abiding norms” (p.2). The data used in this analysis are from a 35-percent random sample (N=1,205) of all inmates who were released, between January 1 and June 30, 1987, directly from Federal prison or through halfway houses to the community in the United States and who had received prison sentences greater than 3 months. The analysis is conducted on a sub-sample (N=619) of this group and only contains persons having a prison stay of more than one year. Recidivism is defined as a re-arrest or parole revocation within three years after release. Multivariate regression is used to control for all variables thought to predict program participation and recidivism. The intention of the multivariate approach is to “identify and measure all the individual and environmental factors thought to influence both program


participation and recidivism, and control for these measures in a regression model when assessing program effects.” With all of the following factors controlled for, the multivariate analysis found that educational program participation significantly reduces the hazard of recidivating. •

• • • • • • • • • • • • • • • • • • • • •

Salient Factor Score (SFS): “The SFS is an 11-score statistical prediction device developed by the U.S. Parole Commission and Federal Bureau of Prisons, and is used along with measures of offense severity and prison behavior to set inmate parole dates. The SFS ranges from 0, poor risk of post-release success, to 10, or very good risk of postrelease success. The following characteristics of inmates are incorporated into the SFS: number of prior convictions, number of prior jail or prison commitments, age at current offense, length of commitment- free period before commencing the current offense, whether under criminal justice supervision (e.g., parole, probation) at commencement of the current offense, and heroin dependence. Each of these items is scored numerically and their sum equals the inmate’s SFS” (Pg. 18 – Fed prisoners released in ’87). SFS is a powerful indicator of post-release success (Harer, 1994). Number of prior convictions. Number of prior convictions and SFS are used separately in two different models because there is a very high correlation between the measures. Race Gender Education Program Participation – coded 1 if completed .5 or more courses per each 6 months of prison term, 0 if otherwise. Property offense as incarcerating offense Heroin user five or more times in the 2 years prior to admission to prison Missing information on heroin use Alcohol abuser evidenced by prior arrests for alcohol related crimes. Missing information on alcohol use. Number of school years completed at time of admission to prison for the instant offense. Worked or was student full time for at least 6 months during 2 years prior to admission. Under criminal justice supervision at time of incarcerating offense. Misconduct charges Received social furlough Number of days served in prison Employed at release from prison. Controlled for in order to isolate normalization effects since education programs probably also enhance employment prospects, which in turn has been shown to decrease recidivism (Grogger). Released through a halfway house Age at time of release Resided with spouse at time of release Honorable and dishonorable discharge from military service Urban socio-economically deprived community background measured by household income, population size and percent of the population that was black by zip code.


The researcher found that males, persons for whom there is missing information on heroin dependency, and older inmates are all less likely to participate and persons with higher educational attainment at admission, persons released through a halfway house, and persons who were in the military and discharged other than for honorable or dishonorable reasons were all more likely to participate. For 98.5% of persons taking .5 courses or more, there is a nearly matching propensity score among persons taking less than .5 courses during their prison term. In other words, the participants in these programs look quite similar to non-participants. Also important to note is that the distribution of SFS’s among those who successfully completed any education courses during their prison term is the same as the SFS distribution for the entire sample. To estimate reduced recidivism due to education course participation, Harer used a logistic regression model using all the variables controlled for in the multivariate approach to predict the log odds of recidivating in the 3-year follow up period. Then, with the coefficients from the regression, and by setting the control variables at their sample means, he estimated recidivism rates for the sample under the condition that no inmates took at least .5 education courses during each 6 months of the prison term and under a second condition that all inmates took courses at that rate or higher. Under the first condition (no participation), the estimated recidivism rate is 45.73 percent and under the second condition (total participation), the recidivism rate is 38.54 percent, reflecting a difference of 7.19 percentage points. This translates into a 15.7% reduction in recidivism due to participation in education courses.


2. Learning While Doing Time, by Anne Morrison Piehl of the Kennedy School of Government, (1995)

Through a data set on Wisconsin inmates, Anne Morrison Piehl demonstrated that completion of adult basic education and high school education programs is significantly associated with lower recidivism – defined as re-incarceration in a Wisconsin prison within 26 months. Instrumental variables using prisoner attributes are used to correct for possible positive selection bias. However, the results give no indication of significant selection bias. The primary data set used for this analysis is the monthly data file from the Wisconsin Department of Corrections which includes all 32,000 terms served in the Wisconsin prison system from January 1, 1980 until June 1990. This data contains information on the start and end date of each term of imprisonment, type of admission and release, conviction offenses, sentence length, time served, educational background (including test scores and attainment). A sample of 1,473 released men (212 of who were program completers) from this population were used in the study. All men were followed for 26 months. The researcher compared the 212 inmates who completed an education program to three groups; the full sample of 1,473, those eligible for education (N=662) and those with no high school degree (N=676). Eligibility is primarily determined by lack of a high school degree, although 135 inmates who had completed high school were deemed eligible because of their low test score upon arrival to prison. Piehl first showed that completing an education program is associated with a longer duration between prison terms controlling for age, race, criminality (number of prior prison terms), sentence in months, whether or not the inmate was serving an old sentence (parole


revoked), prison security level (medium vs. maximum), high school graduation status and tested grade level. (Note that age, criminality, and type of sentence served are crucial components of the SFS calculation described in the previous study and are shown to be an extremely reliable predictor of recidivism.) Results were statistically significant at the 95% confidence level for all three comparison groups. When she examined the probability of recidivism for the same three comparison groups, similar results were found. For each comparison group, education completers are 20% points less likely to recidivate when controlling for the same variables listed previously. All three findings were significant at the 95% confidence interval. While a number of instrumental variable analyses did “not provide overwhelming evidence against selection bias, given the small sample size they also do not provide evidence of substantial selection bias” (pg. 19). In Wisconsin, prisoners are mandated to either participate in an education or work program. There are strong incentives for an inmate to participate in an education program if he has been deemed to need it. Prison administrators strongly encourage inmates to go to school and there are minimum standards for entry to many vocational programs; unqualified inmates must first complete an educational course. While recommendations for inmates’ educational needs are not perfectly implemented, due to lack of program space and inmate unwillingness to participate, the practice of mandating either school or work somewhat attenuates self-selection concerns. Unfortunately, we were unable to include either the federal or Wisconsin study in our analysis. The complexity of the federal prison system, in that federal prisoners are contracted out to state and private prisons, prevented us from obtaining accurate correctional education budget estimates for the federal prisoners studied in Harer’s analysis. While we were able to


obtain the correctional education budget for Wisconsin institutions, the data provided only the total amount spent on prison education generally. We were therefore unable to determine the budget for each type of program. As Piehl only analyzed the impact of adult basic education and high school programs on recidivism, we were unable to apply the available budget data to her findings.


3. Three State Recidivism Study, by Correctional Education Association, 2001 This study collected data on 3,170 inmates who were released from Maryland, Minnesota and Ohio prisons in late 1997 and early 1998 and followed for three years after release. Each state was asked to generate a list of prisoners to be released within a set period of time until lists of 1,200 names from each state were collected. TABLE 2: RECIDIVISM STUDY COHORT Maryland Minnesota Ohio Total for all states

Participants 275 574 524 1,373

Non-participants 610 477 710 1,797

Totals 885 1,051 1,234 3,170

Data was collected through a number of sources. Overall, 500 variables were collected on each of the study participants. •

The Inmate Pre-Release Survey was a self-report instrument “designed to gather

information on inmate demographics, family information, prior employment data, adult and juvenile criminal histories, educational experiences both prior and during incarceration, participation in programs other than education, motivational factors, release plans (including post-release residence, employment and criminal justice information)” (p.27). •

The Educational/Institutional Data Collection Form was developed to collect data directly

from the institution. The information gathered included current incarcerating offense, sentence length of current incarceration, basic demographic information, number of felony arrests and convictions, major institutional infractions, programming (e.g., substance abuse, counseling) and employment participation, and prerelease information. Data on type of educational or vocational programming, level of participation during enrollment, number of diplomas and or certificates received, and whether enrollment was mandatory or voluntary was unavailable.


Criminal History Data was collected for three years after release measuring 1) re-arrest,

re-conviction, and re-incarceration rates; 2) time to re-arrest, re-conviction, and re-incarceration; and 3) re-arrest, re-conviction, and re-incarceration offenses. •

Employment Data for releasees was obtained through the state departments of labor.

Only information on quarterly wages reported by the employer and state industry codes (SIC’s) of the employer were available. Unfortunately, multivariate regressions were not run on this wealth of information. Rather, averages on these variables are provided separately for participants and non-participants. On the whole, the two populations look quite similar. Where they statistically significantly differ at the .01 level are illustrated in the following table: TABLE 3: DIFFERENCES BETWEEN PARTICIPANTS AND NON-PARTICIPANTS FOUND BY PRE-RELEASE SURVEY Age Race White African American Hispanic Held legal job for one year or more during lifetime Family history of incarceration in prison or jail Crime which serving current incarceration Violent Property* Drug/alcohol Age at first arrest Served time in a juvenile facility Number of felony arrests prior to current arrest* Number of times prior to this sentence been in jail Number of times placed on probation Number of times placed on parole/release* Number times been in prison Completed high school GED Test of Adult Basic Education Scores (TABE)* Place to live upon release * NOT statistically significant.

Participants 30.8

Non-Participants 32.6

40.1% 52.9% 3.3% 55.0% 57.4%

36.4% 57.7% 1.7% 64.4% 51.7%

50.2% 26.5% 17.5% 18.56 44.5% 5.05 3.57 2.64 1.67 2.35 10.2 13.7 8.0 86.9

37.9% 29.8% 24.5% 20.09 34.1% 4.69 3.73 2.74 1.74 2.62 20.3 17.3 7.8 82.7


No differences were found between participants and non-participants on motivational factors. These motivational factors included: • • • • • • • • •

Motivation to prepare for a job or vocational training Motivation to get a job, a better job, or higher pay Motivation to improve job performance Motivation to contribute better to my family or community Motivation to help children with homework Motivation to become less dependent on others Motivation to make others feel better about self Motivation to look good to prison or parole officials to get out Motivation to get a better situation in prison Statistical regression was not used in this study to control for possible selection bias.

However, examining the data collected gives us an insight as to whether one of the groups is at greater risk of recidivating than the other. Of highest concern here would be that the nonparticipant group is at a higher risk of recidivating independent of any effects correctional education might have on the participants. The CEA found that younger age, inconsistent job history, volatile family history, younger age at first arrest, higher levels of having served time in a juvenile facility, and lower high school/GED achievement put the participants at greater risk of recidivism. Higher rates of minority status, higher number of property crimes as incarcerating offense, higher numbers of prior incarcerations (only slightly) and lower rates of having a place to live upon release put the non-participants at greater risk of recidivism20. While these findings do not give us conclusive results as to which group is more likely to have higher recidivism rates, it does show us that educational participants have just as many high risk factors as nonparticipants. Unfortunately the degree to which each of these risk factors is associated with recidivism is unknown. 20

A report issued by the Department of the Solicitor General of Canada ranked predictors of adult offender recidivism. The strongest predictors, ranked in order of predicting power are: “criminogenic” need variables (i.e., attitudes, values, and behaviors that support a criminal lifestyle), criminal history – both as an adult and juvenile, social achievement (employment, education), age/gender/race, and family factors.


The following tables show the recidivism results for participants and non-participants in the ‘Three State’ study. TABLE 4: CORRECTIONAL EDUCATION OUTCOME DATA N



Maryland Re-arrest Re-conviction Re-incarceration

840 840 840

54% 32% 31%

57% 37% 37%

Minnesota Re-arrest Re-conviction Re-incarceration

1025 1025 1025

42% 24% 14%

54% 34% 21%

Ohio Re-arrest Re-conviction Re-incarceration

1234 1234 1234

50% 26% 24%

58% 33% 31%


Re-arrest Re-conviction Re-incarceration




3099 3099 3099

48% 27% 21%

57% 35% 31%


APPENDIX B: EFFECT SIZE REDUCTION Thus far, none of the studies examining the effect of correctional education on recidivism have been experimental. Because there has not been random assignment of inmates into control and treatment groups, we cannot know that the reductions in recidivism witnessed are truly due to the education programs. While a number of attempts have been made to account for this, there may yet still be some unmeasured traits among participants that are correlated with lower recidivism rates. It is for this reason that less confidence can be placed in the cause-and-effect conclusions made about correctional education’s ability to reduce recidivism. In order to account for this, we have followed the lead of the ‘Washington State Institute of Public Policy’ (WSIPP) by discounting the effect size of the ‘Three State’ study’s results by 50%. WSIPP developed a system to rate the methodology and resulting reliability and validity of studies examining the effects of programs intended to reduce crime (‘The Comparative Costs and Benefits of Programs to Reduce Crime,’ WSIPP, 2001). Based on their criteria21, we estimate that the ‘Three State’ study would score a ‘3’ on a scale of 1 to 5, wherein: “A ‘3’ indicates an evaluation where the program and comparison groups were matched for pre-existing differences in key variables. There must be evidence presented in the evaluation that indicates few, if any, significant differences in these variables. Alternatively, if an evaluation employs statistical techniques (e.g. logistic regression) to control for pre-existing differences, and if the analysis is successfully completed, then a study with some differences in matched preexisting variables can qualify as a level 3 study” (WSIPP, pg. 40).


Although the ‘Three State’ study was published subsequent to the WSIPP analysis, a conversation with the lead author of the study – Steve Aos, led me to conclude that it would also score a ‘3’ for methodological design.


Discounting the recidivism rates gives us the following results: TABLE 6: DISCOUNTED RECIDIVISM RATES (50%) State Maryland: Minnesota: Ohio:

% Reduction in re-incarceration

% Reduction in re-conviction

% Reduction in re-arrest

16.2% 33.3% 22.6%

13.5% 29.4% 21.2%

5.263% 22.22% 13.79%

Average % reduction in recidivism 11.6% 28.3% 19.2%

50% discount of reduction in recidivism 5.82% 14.2% 9.60%

Note: The true discount due to poor methodological design may be more or less than 50%.


APPENDIX C: THE LINK BETWEEN EDUCATION AND CRIME According to research, increased educational attainment is associated with a decreased incidence of crime22. This can be explained because people choose between committing crimes and pursuing employment in the labor market. The risks associated with criminal activity bear a greater cost when the alternative to crime, having a job, pays more. As a result, choosing to commit a crime is a less attractive option to those who could earn a greater amount in the labor market. The association between education and crime can also be derived from research that indicates that increased cognitive skills23 are associated with increased income, and that increased income is associated with decreased crime. Several prominent studies establish that there is a positive relationship between cognitive skills and income24. Key findings include: •

The payoff to a minor difference in measured cognitive skills is between $1,000 and $1,400 in annual income among those with relatively low cognitive skills. This return to skills is even greater among those who failed the GED. Individuals who scored slightly higher but still failed the test earned significantly more — $2,000 for men and $3,000 for women — than those who scored lower and failed the test25.

Basic cognitive skills are rewarded with higher income in the labor market26.


Lochner and Moretti, 2002. Controlling for many factors, including age, cohort and state of residence. The term “cognitive skills” refers to an individual’s mental abilities due to a combination of innate ability and educational attainment. It is generally supposed that cognitive skills can be increased through instruction. Assessments that quantify cognitive skills include the SAT, GED, Air Force Qualifying Test (AFQT), or the National Adult Literacy Survey (NALS). 24 Card, 1994; Tyler, Murnane and Willett, 2000; Tyler, 2003. 25 Tyler, Murnane and Willett, 2000. 26 According to Tyler (2003), basic cognitive skills are those that most individuals obtain before completing the ninth grade. 23



The effects of schooling on income are large, and may be underestimated by up to 30%27.

These findings indicate that educational attainment and cognitive skills are rewarded with higher income in the labor market, even for those individuals on the lower end of the distribution. Therefore, criminal behavior is more costly to those with higher cognitive skills because they have more to gain in the labor market than they can expect to gain from crime. This may result in reduced criminal behavior. In fact, research has corroborated this relationship28. This association between education and crime offers an opportunity for policy makers to dedicate resources to improving cognitive skills as a crime control policy. Much of this research on the payoff to education considers traditional K-12 education instruction. However, we can also apply the fundamental reasoning of the preceding analysis to the population of adult learners. Investment in children and youth is certainly critical, but there may be payoffs to targeting the adult population, as well. Adult education Before arguing for an investment in adult education as an effort to reduce crime, it is important to answer a key question: can adults improve their cognitive skills by attending academic courses? Several large evaluations have shown that adults benefit from education services in different ways. In some programs, adult learners have demonstrated employment and earnings gains29. Other research shows that participants are likely to self-report gains in basic cognitive


Card, 1994. Grogger, 1998. 29 Beder, 1999. 28


skills30. However, these studies do not use pre- and post-testing to establish that real improvements in literacy have occurred. Literacy The National Adult Literacy Survey (NALS) defines literacy as “using printed and written information to function in society, to achieve one's goals, and to develop one's knowledge and potential31.” Five levels of literacy are commonly referred to in the field. The NALS describes them as follows: • • •

“Level 1: Individual can read a little but not well enough to fill out an application, read a food label, or read a simple story to a child. Level 2: Individual usually can perform more complex tasks such as comparing, contrasting, or integrating pieces of information, but usually not higher-level reading and problem-solving skills. Levels 3 through 5: Individuals usually can perform the same types of more complex tasks on increasingly lengthy and dense texts and documents.32”

Separate evidence directly from state outcome reports does suggest that adult education programs improve literacy. For example, California and Ohio released data that indicates that participants in adult education programs gained at least one level of literacy33. California’s statistics show that between 45% and 65% of adult learners gained a literacy level after participating in an adult basic education program34. Ohio’s reporting illustrates that over half of adult learners completed a literacy level. Approximately 35% of them completed more than one level35.


Ibid. The National Assessment of Adult Literacy defines literacy as http://nces.ed.gov/naal/defining/defining.asp 32 http://www.nifl.gov/nifl/faqs.html#measure 33 There are several different definitions of literacy used in the adult education field. The NALS scale uses a 1-5 scale that rates the ability to find and analyze information embedded in text (prose), the level of mathematical skill (quantitative) and the ability to understand and complete everyday forms (document ). Many states have their own measures of literacy. Some are similar to the NALS scale but several more levels, such as the system in California. Others, such as Minnesota, follow the scale that compares skills to the standards related to grades K-12. 34 http://www.cde.ca.gov/adulteducation/datacollect/fedprogdata/fedstudentdata99-00.html 35 http://www.ode.state.oh.us/ctae/adult/able/annual/EnrollmentCompletionProgressSummary2001.asp 31


This data illustrate two things. First, it does seem that adult education programs are effective at improving cognitive skills. Second, states have a long way to go to improve their reporting of adult education outcomes. The U.S. Department of Education is leading efforts to standardize reporting, but state governments use many scales other than the NALS. This makes it difficult to compare the literacy gains attributed to programs in different states. The literacy of inmates The NALS found that the literacy of the prison population differs substantially from the “household” – or, free, population (see Table 1 below). Approximately 67.5% of the incarcerated population functions at the two lowest levels of prose literacy, compared to 47% of the household population. The average member of the household population scores a 3 on the prose portion of the NALS assessment, whereas the average prisoner scores a 2. The difference between these two levels is observed in analytic skills that are necessary to process information in the often indirect ways it is presented in the real world. TABLE 7: PERCENTAGE OF POPULATION AT EACH LITERACY LEVEL36 NALS Prose Literacy Level 1 2 3 4 5

Incarcerated population 30.5 37 26 6 0.5

Household population 20 27 32 17 3

Clearly, the prison population has a great need for adult basic education – in fact, their need is greater than that of the household population. However, skeptics argue that inmates are not capable of increasing their cognitive skills. Evidence suggests the contrary.




Literacy outcomes Available evidence suggests that education programs in correctional facilities have a positive impact on cognitive skills. In Minnesota, 28% of enrollees in prison education programs increased a grade level in one year37. In California, approximately 37% of adult learners in the correctional population gained a literacy level in one year38. In Texas, participants in prison education programs demonstrated gains as well; an average of 36% of inmates completed one literacy level in one year. And an average of 21% of inmates completed two or more levels in the 2001-2002 fiscal year39.


http://www.corr.state.mn.us/pdf/2001annualreport.pdf http://www.cde.ca.gov/adulteducation/datacollect/fedprogdata/fedstudentdata99-00.html 39 http://www.tea.state.tx.us/adult/tables2002/ins3.pdf 38


APPENDIX D: COST PER CRIME PREVENTED: METHOD Our cost-effectiveness analysis compares the reduction in crime40 associated with correctional education to that of prison expansion. Data collected gave us the following information on correctional education program enrollments and budgets for each state in the ‘Three State Recidivism’ study: TABLE 8: 1997 STATE CORRECTIONAL EDUCATION ENROLLMENTS AND BUDGETS State Maryland Minnesota Ohio

Cumulative Enrollment41 14,523 2,293 26,885

Annual budget42 $11,857,298 $7,832,029 $40,231,253

This data was collected from the following sources: Maryland Enrollment and budget: Mark Mechlinski, Field Director, MD Correctional Education. Minnesota Enrollment: Estimate - 2,781 inmates participated in ABE, vocational education and academic higher education by the end of third quarter, FY2001 (‘Three State,’ pg. 59), which accounted for 43.9% of that year’s total prison population. 43.9% of 1997’s prison population is 2,293. We are assuming there were no significant changes in the proportion of inmates enrolled in correctional education between 1997 and 2001. Budget: Jamie Friesen Nordstrom, Accounting Manager, MN Department of Corrections. Ohio Enrollment: Ohio Central School System: www.drc.state.oh.us/web/educatio.htm. Budget: Estimate based on FY 1999 actual budget of $37,491,900 (Ohio State Government Book, 4th Edition, pg. 599) less $750,000 for ‘Youthful Offenders program,’ less 3% for annual budget increase (Richard Ebin, Project Manager, OH Department of Rehabilitation and Correction.) 40

Throughout the majority of this study when “crime” or “serious crime” is referred to, we are speaking of the seven FBI index crimes. These crime categories are murder, rape, robbery, assault, burglary, theft, motor vehicle theft and arson. However, when we are making comparisons to the RAND data we differentiate between ‘serious crime’ and ‘total crime.’ Only murder, rape, robbery, assault, 60% burglaries and arson are defined as ‘serious crimes.’ This is because California law does not consider theft, motor vehicle theft and 40% of burglaries serious crimes. All of these crimes combined are defined as ‘total crimes.’ 41 We chose to use total cumulative correctional program participants per year rather than average participants per day (i.e., number of program “slots”) in our calculations because no information was reported on the ‘length of participation’ of education ‘participants’ in the ‘Three State Recidivism’ study. Participants, therefore, could be inmates who attended classes for a week and then dropped out. We felt that ‘total cumulative participants’ would be a better approximation to the actual types of participants in the study than would ‘average participators per day.’ 38 All dollar amounts are converted to 2003 dollars to provide a standard rate to compare cost data from different years.


Comparing a 1% increase in prison bed capacity to correctional education William Spelman estimated that a 1% increase in current prison bed capacity would reduce crime by .12 percent to .20 percent, which translates into 60,000 to 100,000 serious crimes prevented per year43. This one-percent increases prison populations by 9,000 inmates, at a cost of $25,000 each – for a total of $225,000,000 a year. If incarcerating 9,000 additional offenders result in a reduction of 60,000 to 100,000 crimes, then the number of crimes prevented per offender ranges from 6.67 to 11.11 per year. 60,000 / 9,000 = 6.67 crimes per year 100,000 / 9,000 = 11.11 crimes per year We will use the mid-point of this range, 8.89 crimes per year in our analysis. If 9,000 offenders are committing an average of 8.89 crimes per year, then the cost per crime due to an increase in the prison population is $2,812. $225,000,000 / (8.89 * 9,000) = $2,812 This estimate is in line with other research conducted on the average number of serious crimes committed per offender. Based on survey research previously conducted on inmates, Spelman estimated that six offenses per year are committed by anyone that has ever been arrested for a serious crime. For those who have been arrested two or more times, a cohort more representative of the prison population as a whole, 15 to 20 offenses are committed per year. Offenses and crimes are not interchangeable, however. Consider the case of two men working together to rob a bank. In this case, “each one has committed an offense, but there is only one crime”44. Based on Spelman’s figures, we estimate that there is an average of 1.56 offenders per crime committed. Reanalysis of existing research allowed Spelman to estimate that 1.73

43 44

Spellman, 1994. RAND, 1994, p. 17.


offenders are involved in the average personal crime and 1.39 offenders are involved in the typical adult property crime. An average of these two numbers gives us 1.56 offenders per crime. If the number of offenses per offender ranges from 6 to 20, then applying this 1.56 offenders per crime estimate tells us that offenders commit somewhere between 3.85 and 12.8 crime per year. 6 / 1.56 = 3.85 20 / 1.56 = 12.8 In order to compare the cost-effectiveness of correctional education to that of prison expansion, we will use Spelman’s 8.89 crimes per offender per year estimate in our calculations. Table 9 shows this rate applied to the findings of the ‘Three State Recidivism Study’ to assess correctional education’s comparative cost-effectiveness.



Total crimes per year

Reduction due to education*1

Crimes prevented by education

Correctional education budget

Cost per crime prevented



























Participants Maryland 14,523 Minnesota



*1 The reduction due to education is based on recidivism rates of participants compared to non-participants, with a 50% discount. Again, based on Spelman’s research, cost per serious crime prevented due to a 1% increase in prison population costs $2,812 a year. Average cost per serious crime prevented: Correctional education: $1,564 Incarceration: $2,812 Serious crimes prevented per $1 million invested: Correctional education: 639 Incarceration: 356 Based on these findings, correctional education is 1.8 times more cost-effective at preventing crime than incarceration.


The preceding calculations applied a discount rate of 50% to the findings of the ‘Three State’ study. However, this discount is rather arbitrary. The true discount required to account for flaws in the methodological design may be more than this rate or nothing at all. To account for this unknown, we re-calculated the cost per crime prevented using a range of discount rates on the effect size. TABLE 10: COST PER CRIME PREVENTED BY CORRECTIONAL EDUCATION VARYING THE DISCOUNT ON THE RECIDIVISM RATE Reduction in recidivism

Crimes prevented by education

Cost per crime prevented

19.7% 9.86% 5.48%

76,535 38,306 21,308

$783 $1,564 $2,812

Three States No discount 50% discount 72% discount

Cost per crime prevented by incarceration:


What this table demonstrates is that the effect size discount rate would have to be more than 72% in order for the crimes reduced through incarceration to be more cost-effective than the crimes reduced through prison education. This means that the true effect of correctional education on crime reduction would need to be 5.5% in order for it to “break even” with incarceration. This is far below the ten to twenty percent researchers believe to be the true effect size of education. Another assumption we make in our findings is that the mean number of crimes committed per offender per year is 8.89. While Spelman was able to estimate a range of 6.67 to 11.11 crimes per offender per year, other research he reviewed found that the average may be more like 3.85. Separate research based on inmate surveys conducted by RAND found that the average might actually be around 2.6 serious crimes per offender per year. To account for the possibility that our


8.89 crimes per offender may be an overestimate of the true rate of crime, we tested our results against a number of different rates. These results are listed in Table 11. TABLE 11: COST PER CRIME PREVENTED BY CORRECTIONAL EDUCATION USING RANGE OF CRIME RATES Total crimes per year*1

Reduction in recidivism

Crimes prevented*2

Cost per crime prevented*3

388,502 291,486 168,249 113,623

9.9% 9.9% 9.9% 9.9%

38,306 28,740 16,589 11,203

$1,564 $2,085 $3,612 $5,349

Three States 8.89 crimes/yr 6.67 crimes/yr 3.85 crimes/yr 2.6 crimes/yr Prison Expansion 8.89 crimes/yr 6.67 crimes/yr 3.85 crimes/yr 2.6 crimes/yr

Total crimes per year*4 80,010 60,030 34,650 23,400

Cost per crime prevented $2,812 $3,748 $6,494 $9,615

*1 Number of crimes participants would have committed had they not received education. *2 Total crimes multiplied by 9.9% reduction in recidivism *3 Total state correctional education budget divided by number of crimes prevented. *4 Number of crimes offenders would have committed had they not been incarcerated.

Notice that when we change the assumptions made about the number of crimes committed per participant per year, we are also changing our assumptions about the crime rates of offenders incarcerated by prison expansion. At every crime rate, correctional education will always be twice as cost-effective as prison expansion.


APPENDIX E: SAVINGS DUE TO PREVENTED RE-INCARCERATIONS: METHOD The ‘Three State’ study tracked the re-incarcerations of a sample of offenders; some had participated in correctional education, and some had not. The study reported that 29.7% of nonparticipants were re-incarcerated when tracked three years after release. We applied this percentage to the total number of participants in all three states’ correctional education programs, 43,701. Had these prisoners never participated in any education programming, 12,965 of them would have been re-incarcerated (43,701 * .297). The study found that correctional education was responsible for a 24% reduction in reincarcerations in these three states. When we apply the 50% discount, this becomes an effect size of 12%. By applying this rate to the 12,965 re-incarcerated prisoners, we see that correctional education is responsible for preventing a total of 1,557 re-incarcerations (12,965 * .12). We then divided the total budget for correctional education in the three states ($59,920,581) by the number of re-incarcerations prevented by correctional education (1,557) to find the cost per prevented re-incarceration: $38,481. So, a $1 million investment in correctional education will prevent 26 future re-incarcerations ($1,000,000 / $38,481). We then had to compare this to the cost of an average incarceration to determine whether any future savings were associated with investing in correctional education. According to the Bureau of Justice Statistics, the average length of stay is 2.4 years, with a cost of 25,000 per year45. Therefore, the average incarceration costs roughly $60,000. Incarcerating these 26 offenders would cost the state about $1,560,000. But preventing these incarcerations through correctional education only costs the state $1 million. A state can save almost six hundred thousand dollars in future costs by investing one million in correctional education today.


Pastore and Maguire, 2001.


In an effort to test the sensitivity of our findings, we varied the 50% discount rate we originally applied to the recidivism rate of program participants. The break-even point – where the cost of preventing one incarceration through correctional education was the same as the cost of the average incarceration – was reached when we discounted the effect size by 78%. Another way to look at this is that correctional education’s true effect on reduced re-incarceration would have to be around 3% for a state to realize no future savings – a figure that is much lower than what was found by the Three State study.



In order to make the conclusion that correctional education programs are more cost-effective

than increasing prison beds, we are implicitly making the assumption that the types of crimes committed by education participants are equally or more expensive than the crimes committed by the general offender. For example, if education participants typically commit relatively inexpensive crimes compared to the general prison population, would not be able to conclude that correctional education is more cost-effective. Prison expansion may be preventing fewer crimes, but as a whole, the monetary benefits of preventing those crimes may outweigh those of correctional education. Fortunately, this does not appear to be the case. In both the ‘Three State Recidivism’ study and ‘Learning While Doing Time,’ the correctional education participants are, on average, more serious offenders. By examining ‘crime of current incarceration’ and ‘re-arrest offenses’ among the inmates in the ‘Three State’ study, we see that education participants are more likely to be violent felons. TABLE 12: ‘THREE STATE’ CRIME OF CURRENT INCARCERATION Crime Violent Property Drug/Alcohol

Participants 50% 27% 18%

Non-participants 38% 30% 25%


Participants 30% 23% 21%

Non-participants 24% 23% 22%


Further, we can see that violent crimes are substantially more expensive than property crimes: TABLE 14: AVERAGE COST TO THE VICTIM PER CRIME Crime Violent Robbery Assault Property Burglary Larceny Auto Theft

Cost per victim $14,357 $13,712 $1,564 $203 $3,565

Source. Criminal Incapacitation, Spelman, 1994, pg. 223 All costs expressed in 1989 dollars.

The cost-effectiveness of correctional education previously calculated may actually be underestimated to the extent that education participants tend to commit more serious (or more expensive) crimes than the average offender. ‘Learning While Doing Time’ corroborated the finding that correctional education participants are more serious offenders. It does not provide data on crime of incarceration nor recidivating crime, however it does provide data for length of sentence and prior penal experience. Even though those who complete an education course had fewer prior penal experiences (28%) than the sample on average (49%), they had longer than average sentences, 64.71 months, compared to 58.07 months for the entire sample. Because prior convictions can affect the length of subsequent sentences, we would predict that the full sample, on average, should have longer sentences due to their prior records. However, we see the opposite – that those who complete an education course have the longer sentences - indicating that the crimes for which they were incarcerated are, on average, more serious crimes than those committed by the average prisoner. More serious crimes are also more expensive crimes.



One could also make the argument that the population of education participants is simply

less inclined to commit crime than is the average offender upon release. Because the ‘Three State’ study did not control for criminality indicators, we cannot know that the participants were not less likely to recidivate from the beginning, independent of the effects of education. What we can see, however, is how the two groups match up on criminality indicators, or indicators thought to predict the future criminal patterns of offenders. One would think that some of the best predictors of criminality in the future would be criminal history factors. The ‘Three State’ study allows us to compare the differences between participants and non-participants on these factors. TABLE 15: CRIMINALITY INDICATORS Indicator

Participants N=1,342

Non-participants N=1,757

Total Pop. N=3,099

No. of felony arrests No. of times in jail No. of times in prison

Mean 5.05 3.57 2.35

Mean 4.69 3.73 2.62

Mean 4.8459 3.66 2.5





Number of prior felony arrests indicates that the participants have a higher degree of criminality, while number of prior incarcerations in jail and prison would indicate the opposite. Across all three categories the differences are minute; although overall it appears as if the nonparticipants have a higher degree of criminality. However, this difference diminishes when we take into account the age differences between the participants and non-participants. That the nonparticipants are almost 2 years older on average than the participants means that they have had that much more time to commit crimes and get caught. In two years, the participants may well have the same means on these criminality indicators as the non-participants.


These indicators are simply that – predictors of future criminal behavior. Such indicators only go so far in predicting actual future criminal behavior. It may be that there are some unknown characteristics of the participant group that make them less likely to recidivate. It is precisely for this reason that the effect sizes for the reductions in recidivism found in the ‘Three State’ study were reduced by 50%, which may be overestimating or underestimating the true effect of the differences between participants and non-participants.


APPENDIX G: ALTERNATE METHOD OF CALCULATING COST-EFFECTIVENESS RAND calculated the cost per crime that would be prevented due to implementation of various ‘three strikes’ laws in California46. In this appendix, we apply the crime rates they predicted per offender to our cohort of correctional education participants to compare the costeffectiveness of these two crime control strategies. Based on surveys of California prison inmates in 1993, RAND found that the rate at which different offenders commit crime varies greatly47. RAND divided incarcerated offenders “in half on the basis of the rate at which they have committed crimes”48. The more frequent offenders are called high-rate offenders and the other half are called low-rate offenders. However, in the final model used by the researchers, this proportion of low- and high-rate offenders was modified. They found that the incarcerated population consisted of 40% low-rate offenders and 60% high-rate offenders “when the model was used to estimate initial conditions (as of the end of 1993) by running it to its steady state using constant incidence scenario”49. This was done because the researchers decided that their “model’s results must match what is known about the numbers of active felons, offenses, arrests, and people incarcerated, as well as the rate at which people released from prison are rearrested”50. Their estimated proportion of high- and low-rate offenders had to be modified in order to correspond with what is actually known about these offenders’ offense rates. It is important that the distribution of high- and low-rate offenders we use in our analysis coincides with that used by the RAND researchers. RAND determined the number of crimes prevented by different mandatory sentencing laws by estimating the number of additional crimes these laws would prevent over and above what the current law is already preventing. In order to 46

RAND, 1994. Ibid., p. 12. 48 Ibid., p.12. 49 Ibid., p.66. 47


determine the ‘costs per crime prevented’ for each of the new sentencing laws, it was necessary to determine the number of crimes already being prevented. In order to compare the number of crimes prevented by correctional education to the number of crimes prevented by mandatory sentencing laws it was necessary that we use the same base crime rate used by the RAND researchers. Therefore, 60% of our incarcerated education participants will be classified as high-rate offenders and 40% will be classified as low-rate offenders. TABLE 16: CRIMES PER OFFENDER PER YEAR Type of Offender Low-rate: High-rate:

Type of Felony Violent and Serious Other .24 .37 4.13 6.47

Total Index .60 10.60

RAND classified California crimes according to the FBI index felonies. Violent crimes include murder, rape, robbery and assault. Serious, but not violent crimes include burglary, theft, motor vehicle theft and arson. Serious crimes include all violent and serious crimes. Other crimes include drug and other. In determining the cost-effectiveness of correctional education for the ‘Three State’ study participants we: •

Classified 60% of the participants as high-rate and 40% as low-rate offenders.

Determined the number of serious crimes predicted to be committed by this sample in the first year after release based on RAND’s number of serious crimes committed per offender per year.


Ibid., p.53.


Estimated the number of serious crimes prevented by multiplying the total number of predicted serious crimes by the effect size (reduction in recidivism) of each state’s correctional education program.

Calculated the cost per serious crime prevented by dividing the education budget by the total number of crimes prevented.

Results for each of the states can be found in table 17.



Maryland Low-rate participants High-rate participants Total participants Minnesota Low-rate participants High-rate participants Total participants Ohio Low-rate participants High-rate participants Total participants


Serious crimes per participant per year*2

Total serious crimes per year

Reduction in recidivism*3

Serious crimes prevented due to education











































Correctional education budget

Cost per serious crime prevented









*1 RAND found that the incarcerated population consisted of 40% low-rate offenders and 60% high-rate offenders. *2 Crimes per high-rate and low-rate offenders based on RAND 1993 survey of California prisoners. *3 ‘Reduction due to education’ based on recidivism rates of participants compared to non-participants.


These results can be compared to RAND’s estimates of cost per crime prevented for five different mandatory sentencing laws for repeat offenders: TABLE 18: COST PER CRIME PREVENTED BY VARIOUS MANDATORY SENTENCING LAWS Law Jones Three-Strikes Jones Second-Strike Only Jones Violent Only Rainey Three-Strikes Guaranteed Full Term

Crimes Prevented

Cost per Crime Prevented (in ’03 dollars)

338,000 287,000 220,000 267,000 342,000

$20,238 $17,630 $14,651 $18,499 $16,016

Brief descriptions of these alternative laws follow. California adopted the Jones 3-strikes law in 1994. These alternative-sentencing laws may be comparable to laws adopted by other states in the 1990’s. TABLE 19: DESCRIPTIONS OF MANDATORY SENTENCING ALTERNATIVES. Alternative

2nd Strike Sentence

3rd Strike Sentence

Jones 3-strikes Jones 2nd strike only Jones violent only

Double the nominal sentence Double the nominal sentence Double the nominal sentence if violent

25 yr to life

Rainey 3-strikes

Extra 10 yr for 2nd violent felony incarceration Extra 5 yr

25 yr to life

Guaranteed full term

Double the nominal sentence 25 yr to life if violent

Extra 10 yr for 3rd serious conviction; 20 yr to life for 3rd violent felony incarceration

Is Prison Required? Yes, if sentence is enhanced Yes, if sentence is enhanced Yes, for violent 2nd or 3rd strike Yes, for serious on 2nd or 3rd strike Yes, if serious

“Good Time” Allowed Limited to 20 percent after 1st strike Limited to 20 percent after 1st strike Limited to 20 percent for violent 2nd or 3rd strike None if violent, or 3rd strike None if serious


Based on the results from the ‘Three State Recidivism’ study we can see that correctional education is 2 to 3 times as cost-effective as increasing sentences for repeat offenders. Average cost per serious crime prevented: Correctional education: $6,959 Increased sentences for repeat offenders: $14,651 to $20,238 Serious crimes prevented per $1 million invested: Correctional education: 144 Increased sentences for repeat offenders: 49 to 68


Profile for UCLA Luskin School of Public Affairs

Correctional education as a crime control program  

Correctional education as a crime control program