Abstract
Objectives
There is debate about the extent to which imprisonment deters reoffending. Further, while there is a large literature on the effects of imprisonment, methodologically sound and rigorous studies are the exception due to problematic sample characteristics and study designs. This paper assesses the effect of imprisonment on reoffending relative to a prison diversion program, Community Control, for over 79,000 felons sentenced to state prison and 65,000 offenders sentenced to Community Control between 1994 and 2002 in Florida.
Methods
The effect of imprisonment on recidivism is examined within one-, two-, and three-year follow-up periods using Logistic Regression, Precision Matching, and Propensity Score Matching.
Results
Findings indicate that imprisonment exerts a criminogenic effect and that this substantive conclusion holds across all three methods.
Conclusions
The main contribution of this study is that various methods yield results that are at least in a similar direction and support overall conclusions of prior literature that imprisonment has a criminogenic effect on reoffending compared to non-incarcerative sanctions. Limitations and directions for future research are noted.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Defiance theory contends that the effect of an imposed punishment is contingent on how an offender perceives the punishment and the extent to which s/he is bonded to the punisher and society (Sherman 1993). However, defiance seems more apt to earlier stages of the punishment process, i.e., arrest, and thus is not described above with respect to the effect of incarceration on subsequent behavior.
Other approaches that do not fit into the approaches used in the current study all share the characteristic that each attempts to construct some sort of counterfactual (Nagin et al. 2009:155–167). Specific examples may be found in Bhati and Piquero (2007), Manski and Nagin (1998), and Loughran et al. (2009).
These authors are most concerned with exact matching on the variable age, because “offending rates are highly age dependent and because the postsanction outcome variable, offending rate or recidivism probability, necessarily must be measured over age.” They conclude that “it is very important in analyses of observational data to compare the postsanction offending rate of an imprisoned individual with that of one or more non-imprisoned individuals who are the same age” (p.138). Several studies match variables to one degree or another (i.e., some subset of variables in one study but not another), but our study does this in a very detailed manner (i.e., exact same age, etc.). Of course, exact matching cannot address the problem of unobserved imbalance among unobserved attributes as it only addresses observed imbalances among observed attributes.
Community Control is an Intensive Supervision Program in which the offender is placed on “house arrest” which involves a home curfew established by the supervising officer. Also, contact standards are more extensive for this population relative to regular felony probation. The offender is restricted to his/her residence, with the exception of being allowed to work, attend treatment, visit the probation office, and limited other occasions that must be approved in advance by the community control officer.
As is true of prior studies in this area, we do not account for the length of time served in our prison sample. This would require a different set of analyses which would incorporate incapacitation effects that is beyond the scope of this paper. We note that based on the prison sample derived from the precision matching described later, the prison group served an average of 22 months incarceration, the median was 18 months, and 84.6% served 36 months or less. This suggests that not accounting for prison exposure time may have a minimal impact on the findings.
The primary offense is identified when multiple current convicted offenses are contained in the sentencing guidelines based on which one results in the highest number of total guidelines points.
Nagin et al. (2009:184) argue that “it is particularly important to have extensive measurements of what the individual has “done” as measured by prior record and characteristics of the conviction offense. Because these factors represent the primary determinants of sentence type and length, it is vital that they be measured on as many dimensions as possible.” Our inclusion of the current offense quantified in nine different crime types and the seriousness of the current offense, based on the sentencing guidelines score and the use of three different indicators of prior record with no truncation of continuous measures, is our attempt at using the data available to us in order to meet their rigorous criteria and account for non-random assignment of sentences.
An example of the lack of information relating to the precision matching process is the study of the effectiveness of Florida’s Community Control program versus prison in terms of recidivism conducted by Smith and Akers (1993). Their Table 1 (p.276) displays differences in summary statistics across the two punishment groups with sample sizes of 228 in the Community Control group and 266 in the prison group. The extent of the description of the matching process is captured in the statements, “in addition to the criminal comparability of two groups, as measured by the sentencing score, matching was also done on offense type in this study to increase even further the equivalency of the two groups. Offenders were also matched on age and previous violations of the two groups. However, offenders could not be matched on race or prior felonies and significant differences remained between the Community Control and prison groups on these two variables” (p.277). There are 228 Community Control cases and 266 in the prison group after matching. These different sample sizes intimate that more than one case in both groups was retained in instances in which the two groups matched on all the matching variables. This is further evidenced by the fact that the percentages across the two groups for categories of race and offense, which were matching variables, differed. If precision matching was conducted, as done in our study in which only one case from the prison and one case from Community Control samples were retained when there was more than one case in both groups that matched on all of the matching variables, the percentages across the groups would have been identical. However, this is not explicitly noted by the authors. In another study, Scarpitti and Stephenson (1968) matched youth who received probation (n = 943) to those who were placed in a non-residential group center (n = 100), a group center in which the youth resided (n = 67), and those placed in a state reformatory (n = 100). The groups differed significantly on race, school status and five characteristics of their families (Table 1). Three variables were used in the matching process; an index of family socioeconomic status, an index of delinquency history, and race. Only 44 boys matched across all four programs; however, descriptive statistics are not presented to indicate the level of equivalency of the two groups across the matching variables. We recognize that the lack of detail may have to do with publication requirements; nevertheless, we think that when matching and equivalency issues are involved more detail is best.
In the analysis to follow using precision matching, it will be evident that the steps detailed above were replicated using different matching variables based on the inclusion of a varying number of covariates used in the matching process. An occurrence that appears on the surface to be counter-intuitive is that the number of cases in which precision matching generates will increase or decrease when more variables are added to the precision matching variable. Here, we provide an example to explain why this occurs. Let us assume we match the prison and Community Control datasets based on a matched variable, which includes five control variables, and the number of cases in the prison group that have a unique value on the matching variable is 10, and the number of cases with the same unique value on the matching variable in the Community Control group is 15. The final dataset of matched cases across the two sanctioning groups will be one after merging the two groups on unique values on the match variable and retaining one of these records. When a sixth variable is added to the match variable, by definition, the same number of cases will have the same value across the two groups based on the five matching variables, and one or more of the prison group cases can have the same matching variable value as one or more of the cases in the Community Control group. After retaining only one case from each unique value on the matching variable, the number of cases in the final dataset of matched prison and Community Control records will now be two. When a seventh or subsequent variables are added to the matching variable, the potential of adding additional cases to the final dataset continues, resulting in the final dataset with more cases than when fewer variables are used in the matching process. In contrast, the occurrence of a decrease in the number of matched cases as the number of matching variables increases occurs in the following circumstances. If there are no cases in the prison or Community Control samples that have the same value on an additional matching variable when increasing the number of matching variables by one, the cases that matched on the next lowest number of matching variables will no longer be included in the final matched dataset.
Although somewhat counterintuitive, the reason why there are incremental increases in the number of cases in which precision matches were obtained from Matching #1 (n = 12,088) to Matching #4 (n = 20,314) is because there are additional matching variables included. The reason for this occurrence is explained in endnote 9. The reason the number of cases decline from the Matching #4 model to the Matching #5 model is a result of the inclusion of the dichotomous variable of whether cases were recommended for a prison sentence under the guidelines (1) or a non-prison sanction (0). In contrast, the variables added in the models for Models #2, #3, and #4 are continuous variables which allow for more possible matches.
Nagin et al. are emphatic about the importance of matching on age: “we regarded it as very important in analyses of observation data to compare the postsanction offending rate of an imprisoned individual with that of one or more non-imprisoned individuals who are the same age” (p.138). This assertion is grounded in the large number of studies that consistently find age as one of the best predictors of recidivism and that the judiciary makes punishment choices that result in different sanctioning populations that are substantively different in the relative age of offenders. In our study, the average age of those sentenced to prison is 34.5 versus 30.4 among those in Community Control, a substantive difference that requires serious attention in the analysis to achieve across group equivalency. To determine if there is empirical support for this assertion, we conducted precision matching analysis to assess if this emphasis on retaining age as a continuous measure was warranted. Specifically, and following Nagin et al.’s five-critical-variable recommendation, five different matching variables were created in which all included sex, race, Hispanic, current offense, and prior felonies, and each contained a different measure of age [a continuous measure in years, age as below or above the median (30 years), age in increments of 10 years (19 or younger, 20 to 29, etc.), and measures based on the percentage distribution into terciles, quartiles, and quintiles]. Comparisons in recidivism probabilities across the prison and Community Control groups created through precision matching using these five different approaches to quantifying age revealed evidence of a criminogenic effect of imprisonment regardless of how age was operationalized. However, the matching based on the continuous age measure produced the only significant effect and was substantively greater than the other four group comparisons using less precise measures of age. These results, which strongly support Nagin et al.’s concern over age, are available upon request.
As more variables are added to the precision matching, the less criminogenic the samples become. This may indicate that precision matching may reduce the matched samples to a subpopulation of offenders who are higher risk Community Control offenders and lower risk prison offenders.
Noteworthy is the fact that the number of cases which met the matching criteria using 15 variables (Match #2) increased incrementally from 6,044 in each sanctioning group to 10,157 when 17 of the 18 variables were utilized. This is counter to the concerns that some have raised when using precision matching to derive equivalent groups. Selltiz et al. (1959:105) state that the “more precise the matching, and the greater the number of factors on which matching is to take place, the greater the number of cases for which no match is available.” We speculate that this is due to the very large number of cases (144,416) which were available to derive the matching groups. Still, the inclusion of the final variable, whether prison was the recommended sentence, did result in a decrease in the number of cases from 10,157 in each group to 9,356. More generally, an anonymous reviewer observed that, as more matching variables are added in Table 6, the average seriousness of current and past criminal history decreases. The policy implications may be that the matched data only are conclusive for offenders with lower seriousness levels (current and past criminal history) and it is difficult to achieve matches (balance) for offenders in this population who get a prison sentence and have much more serious levels of criminality.
A reviewer asked where, on a continuum of criminal risk based on the total sentencing guidelines points offenders who matched and did not match, fell. They raised the concern that precision matching may result in a subpopulation of offenders comprised of higher risk Community Control offenders and lower risk prison offenders. The Appendix Table presents the number and percentage of prison and community control cases within 17 ranges of the guidelines points and the number and percentage of community control cases that matched to a prison case based on the precision matching model #5 and those that did not match. The comparison was not possible to conduct based on the PSM analysis because the offender identification number and sentencing date were not retained in the resulting dataset and, therefore, it was not possible to merge the matched cases with the total sample using these two variables. The Appendix Table shows that, while a higher percentage of the community control cases are in the lower range of the total sentencing guidelines point distribution than prison cases, there are reasonable numbers and percentages of cases at the higher end of the distribution and that matched cases were retained at all levels of the total guidelines points. Additionally, while a higher percentage of the community control cases matched to prison case in the lower ranges of the guidelines points, which is expected given the higher number of cases in these groups, there is a similar percentage that matched as the point ranges increased to the maximum category of 100+. This indicates that the final matched sample is not overly dominated by higher risk Community Control cases and lower risk prison cases.
Balance was not achieved using the entire sample of 144,416 cases. Several methods of transforming and truncating the distributions of the covariates were conducted which did not solve the imbalance problem. To investigate whether this was a sample size issue, we selected a random sample of 500 cases from each of the two groups and were able to attain balance. However, PSM models were generated using the entire sample and conclusions derived from these results are substantively the same as what is reported here from the random sample.
References
Allison, P. D. (1991). Logistic Regression Using the SAS System: A Practical Guide. Cary: SAS Institute.
Anderson, D. B., Schumacker, R. E., & Anderson, S. L. (1991). Release characteristics and parole success. Journal of Offender Rehabilitation, 17, 133–145.
Apel, R. J., & Sweeten, G. (2010). Propensity score matching in criminology and criminal justice. In A. R. Piquero & D. Weisburd (Eds.), Handbook of Quantitative Criminology (pp. 543–562). New York: Springer.
Bales, W. D., & Mears, D. P. (2008). Inmate social ties and the transition to society: Does visitation reduce recidivism? Journal of Research in Crime & Delinquency, 20, 1–35.
Bales, W. D., Bedard, L. E., Quinn, S. T., Ensley, D. T., & Holley, G. P. (2005). Recidivism of public and private state prison inmates in Florida. Criminology & Public Policy, 4, 101–127.
Baumer, E. (1997). Levels and predictors of recidivism: The Malta experience. Criminology, 35, 601–628.
Beck, A. J., & Shipley, B. E. (1987). Recidivism of Young Parolees. Bureau of Justice Statistics – Special Report. Washington, DC: U.S. Department of Justice.
Beck, A. J., & Shipley, B. E. (1997). Recidivism of Prisoners Released in 1983. Bureau of Justice Statistics – Special Report. Washington, DC: U.S. Department of Justice.
Becker, S., & Ichino, A. (2002). Estimation of average treatment effects based on propensity score. The Stata Journal, 2, 358–377.
Berube, D., & Green, D. P. (2007). The effects of sentencing on recidivism: Results from a natural experiment. Unpublished manuscript. New Haven: Yale University.
Bhati, A., & Piquero, A. R. (2008). Estimating the impact of incarceration on subsequent offending trajectories: Deterrent, criminogenic, or null effect? The Journal of Criminal Law and Criminology, 98, 207–254.
Blumstein, A., Cohen, J., & Nagin, D. S. (1978). Deterrence and Incapacitation: Estimating the Effects of Criminal Sanctions on Crime Rates. Washington, DC: National Academy Press.
Bontrager, S., Bales, W. D., & Chiricos, T. (2005). Race, ethnicity, threat and the labeling of convicted felons. Criminology, 43, 589–622.
Chiricos, T., Barrick, K., Bales, W. D., & Bontrager, S. (2007). The labeling of convicted felons and its consequences for recidivism. Criminology, 45, 547–582.
Cochran, W. G. (1965). The planning of observational studies of human populations (with discussion). Journal of the Royal Statistical Society, Series A, 128(2), 134–155.
DeMaris, A. (1992). Logit Modeling: Practical Applications. Thousand Oaks: Sage.
Drago, F., Galbiati, R., & Vertova, P. (2009). The deterrent effects of prison: Evidence from a natural experiment. Journal of Political Economy, 117, 257–278.
Florida Department of Corrections and Office of the State Courts Administrator. (2010). Florida Criminal Punishment Code: Scoresheet Preparation Manual.
Green, D., & Winik, D. (2008). The effects of incarceration and probation on recidivism among drug offenders: An experimental approach. Unpublished manuscript. New Haven, CT: Yale University Press.
Heckman, J. J., & Hotz, V. J. (1989). Choosing Among Alternative Nonexperimental Methods For Estimating The Impact Of Social Programs: The Case Of Manpower Training. University of Chicago, Economics Research Center.
Helland, E., & Tabarrok, A. (2007). Does three strikes deter? A nonparametric estimation. Journal of Human Resources, 42, 309–330.
Kohl, R., Hoover, H. M., McDonald, S. M., & Solomon, A. L. (2008). Massachusetts Recidivism Study: A Closer Look at Releases and Returns to Prison. Washington D.C.: Urban Institute-Justice Policy Center.
Kraus, J. (1974). A comparison of corrective effects of probation and detention on male juvenile offenders. British Journal of Criminology, 14, 49–61.
Kraus, J. (1978). Remand in custody as a deterrent in juvenile jurisdiction. British Journal of Criminology, 18, 285–289.
Kraus, J. (1981). The effects of committal to a special school for truants. International Journal of Offender Therapy and Comparative Criminology, 25, 130–138.
Kubrin, C. E., & Stewart, E. A. (2006). Predicting who reoffends: The neglected role of neighborhood context in recidivism studies. Criminology, 44, 165–197.
Langan, P. A., & Levin, D. J. (2002). Recidivism of Prisoners Released in 1994. Washington, D.C.: U.S. Bureau of Justice Statistics.
Langan, P. A., Schmitt, E. L., & Durose, M. R. (2003). Recidivism of Sex Offenders Released from Prison in 1994. Bureau of Justice Statistics. Washington, DC: U.S. Department of Justice.
Loughran, T. A., Mulvey, E. P., Schubert, C. A., Fagan, J., Piquero, A. R., & Losoya, S. H. (2009). Estimating a dose–response relationship between length of stay and future recidivism in serious juvenile offenders. Criminology, 47, 699–740.
Maltz, M. D. (1984). Recidivism. New York: Academic Press.
Manski, C. F., & Nagin, D. S. (1998). Bounding disagreements about treatment effects: A case study of sentencing and recidivism. Sociological Methodology, 28, 99–137.
Mears, D. P., & Bales, W. D. (2009). Supermax incarceration and recidivism. Criminology, 47, 801–836.
Menard, S. (1995). Applied Logistic Regression. Thousand Oaks: Sage.
Miles, T. J., & Ludwig, J. (2007). The silence of the lambdas: Deterring incapacitation research. Journal of Quantitative Criminology, 23, 287–301.
Muiluvuori, M.-L. (2001). Recidivism among people sentenced to community service in Finland. Journal of Scandinavian Studies in Criminology and Crime Prevention, 2, 72–82.
Nagin, D. S., Cullen, F. T., & Jonson, C. L. (2009). Imprisonment and reoffending. In M. Tonry (Ed.), Crime and Justice: A Review of Research (Vol. 38, pp. 115–200). Chicago: University of Chicago Press.
Petersilia, J. (2003). When Prisoners Come Home: Parole and Prisoner Reentry. New York: Oxford University Press.
Petersilia, J., Turner, S., & Peterson, J. (1986). Prison versus probation in California: Implications for crime and offender recidivism. Santa Monica: RAND Corporation.
Piquero, A. R., & Blumstein, A. (2007). Does incapacitation reduce crime? Journal of Quantitative Criminology, 23, 267–286.
Piquero, A. R., Farrington, D. P., & Blumstein, A. (2003). The criminal career paradigm: Background and recent developments. In M. Tonry (Ed.), Crime and Justice: A Review of Research (Vol. 30, pp. 359–506). Chicago: University of Chicago Press.
Pratt, T. C. (2009). Addicted to Incarceration: Corrections Policy and the Politics of Misinformation in the United States. Thousand Oaks: Sage.
Putnins, A. L. (2005). Assessing recidivism risk among young offenders. Australian and New Zealand Journal of Criminology, 38, 324–339.
Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70, 41–55.
Rubin, D. B. (2008). For objective causal inference, design trumps analysis. The Annals of Applied Statistics, 2, 808–840.
Savoainen, J. (1968). Criminal recidivism among felony-level ATI program participants in New York City. New York City Criminal Justice Agency.
Scarpitti, F. R., & Stephenson, R. M. (1968). A study of probation effectiveness. Journal of Criminal Law, Criminology and Police Science, 59, 361–361.
Selltiz, C., Jahoda, M., Deutsch, M., & Cook, S. W. (1959). Research Methods in Social Relations. Holt: Rinehard and Winston.
Sherman, L. W. (1993). Defiance, deterrence and irrelevance: A theory of the criminal sanction. Journal of Research in Crime and Delinquency, 30, 445–473.
Smith, L. G., & Akers, R. L. (1993). A comparison of recidivism of Florida’s community control and prison: A five-year survival analysis. Journal of Research in Crime and Delinquency, 30, 267–292.
Stanz, R., & Tewksbury, R. (2000). Predictors of success and recidivism in a home incarceration program. Prison Journal, 80, 326–344.
Villettaz, P., Killias, M., & Zoder, I. (2006). The Effects of Custodial vs. Non-custodial sentences on Re-offending: A Systematic Review of the State of Knowledge. Philadelphia: Campbell Collaboration Crime and Justice Group.
Visher, C. A., Lattimore, P. K., & Linster, R. L. (1991). Predicting the recidivism of serious youthful offenders using survival models. Criminology, 29, 329–366.
Weisburd, D., Waring, E., & Chayet, E. (1995). Specific deterrence in a sample of offenders convicted of white-collar crimes. Criminology, 33, 587–607.
Wimer, C., Sampson, R. J., & Laub, J. H. (2008). Estimating time-varying causes and outcomes, with application to incarceration and crime. In P. Cohen (Ed.), Applied Data Analytic Techniques for Turning Points Research. London: Routledge.
Wolfgang, M. E., Figlio, R. M., & Sellin, T. (1972). Delinquency in a Birth Cohort. Chicago: University of Chicago Press.
Word, D. L., & Perkins, C. R. (1996). Building a Spanish surname list for the 1990s – a new approach to an old problem. U.S. Census Bureau Technical Working Paper No. 13. Washington, DC: U.S. Department of Commerce.
Zimring, F. E. (1991). The Scale of Imprisonment. Chicago: University of Chicago Press.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Rights and permissions
About this article
Cite this article
Bales, W.D., Piquero, A.R. Assessing the impact of imprisonment on recidivism. J Exp Criminol 8, 71–101 (2012). https://doi.org/10.1007/s11292-011-9139-3
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11292-011-9139-3