The Effect of Facility Security Classification on Serious Rules Violation Reports in California Prisons: A Regression Discontinuity Design

Abstract

Objectives

Prison facility security classification is intended to recognize differences across inmates with regard to the propensity to commit misconduct and to appropriately house inmates with varying levels of violent and/or antisocial behavior while they are incarcerated. The intent of security classification is to increase safety for staff and other inmates, but little is known about the effect of security classification on prison misconduct.

Methods

Using administrative records of roughly 60,000 inmates in the California Department of Corrections and Rehabilitation (CDCR), this study attempts to identify the relationship between security classification and rules violation reports using a regression discontinuity design.

Results

Results indicate that inmates placed in Level II (medium security) prisons are approximately 11 percentage points more likely to be written up than inmates placed in Level III (close security) prisons, and that the difference is driven almost entirely by a higher likelihood of write ups for the lowest level offenses like bartering and gambling. In contrast to prior work, this study does not detect an effect of Level IV (maximum security) prisons on rules violation reports.

Conclusions

The fuzzy regression discontinuity design allows for a rigorous way to estimate the causal effect of facility security classification on rules violation reports in California prisons, providing an evidence base for policy-makers facing capacity constraints within the prison system while at the same time updating the extant literature on the effects of an important feature of prison structure on inmate outcomes.

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2
Fig. 3
Fig. 4

Notes

  1. 1.

    Administrative overrides occur when classification staff place an inmate in a facility that does not match the security level indicated by the classification instrument. Administrative overrides can result in an inmate being placed either above or below the placement suggested by the classification score, but almost always results in placement in a higher level than the classification instrument suggests.

  2. 2.

    Section 1 of the Online Supplement includes examples of CDCR classification forms, a complete list of the factors that comprise the preliminary classification score, the factors that trigger mandatory minimum placement, as well as detailed explanations on how classification scores are calculated.

  3. 3.

    The cutoff values have changed since the data were collected, for simplicity and clarity the text refers to the cutoff values that were in place when the data was collected.

  4. 4.

    During lockdowns programs are canceled for at least 24 hours. Full lockdowns affect all inmates in the facility, whereas partial lockdowns might include a particular housing unit or all inmates in a given race.

  5. 5.

    Though there is no formal definition in the Department Operations Manual or in Title 15, Sensitive Needs Yards are separate areas within facility security levels where some inmates are segregated from the “mainline” population. More than 16,000 inmates in the sample are housed on a sensitive needs yard. Sensitive Needs Yard is sometimes colloquially referred to as “protective custody.” Inmates can either be assigned or can request to be housed on a sensitive needs yard. Inmates may be housed in a sensitive needs yard for several reasons, including, but not limited to, because they have dropped out of a gang, have been convicted of a sex offense (especially one involving children), because of their sexual orientation, or because they have a high-notoriety case.

  6. 6.

    In the context of California prisons, it is essential to use preliminary classification score and not final placement score as the running variable in the regression discontinuity design. A detailed explanation is included in the online supplement (Section 2).

  7. 7.

    In this case, the treatment is moving to the lower of two adjacent facility levels compared to the “control condition” which is to remain at the higher of two levels.

  8. 8.

    Formal estimates of the first stage relationship between preliminary score and facility security level placement are available on request.

  9. 9.

    As can be seen from the distribution of the preliminary and placement scores (Online Supplement, Section 2). There are so many inmates who qualify for the mandatory minimum at the Level II threshold there are often inmates with scores just above the threshold who are moved down because they are eligible. This is most likely because they are often seeking as many inmates as possible for minimum custody status (Level I). These inmates are generally used for labor at the Level IV prisons (most Level IVs have an adjoining Level I) and because of the turnover of Level I inmates (non-violent, non-sex offenses with relatively short sentences) they are often looking for eligible inmates. It is most likely the case that California prison administrators will assign inmates with Level II points if they meet all the other requirements for being housed outside the secure perimeter— especially if there is need for a specific project at a prison.

  10. 10.

    As a robustness check, I also estimated the results using local polynomial regressions and the results are consistent. Results are available upon request.

  11. 11.

    Local linear regression estimation of the regression discontinuity is appropriate in this context despite the discrete nature of the running variable. Lee and Lemieux (2010) note that estimating the conditional expectation of the outcome at the cutoff requires extrapolation to some extent, even in the case of a continuous running variable. As a consequence, “the fact that we must do so in the case of a discrete assignment variable does not introduce particular complications from an econometric point of view, provided the discrete variable is not too coarsely distributed” (Lee and Lemieux 2010, 336).

  12. 12.

    In order to estimate the regression discontinuity design using local linear regressions, I need to choose the kernel, the weighting function used to estimate the density of a random variable, and the bandwidth, the range of the running variable to be included in the analysis. As noted in the extant literature, the choice of the kernel has little impact on the regression discontinuity estimate. Though, Fan and Irene (1996) show that the triangle kernel is optimal for estimating the local linear regression at the cutoff. Following their suggestion, I use the triangle kernel throughout the study.

  13. 13.

    Because the California prison system has multiple thresholds along the range of the running variable, the maximum bandwidth for the estimation is constrained by the distance between the threshold values. Selection of bandwidth for the II/III cutoff is constrained by the score distance between the Level I/II cutoff at 19 and the Level II/III cutoff at 28. So, bandwidth cannot exceed 8 without extending to the margin of the next threshold. Because preliminary scores have no upper bound, the maximum potential bandwidth of 24 for the Level III/IV cutoff is constrained by the lower bound, the Level III cutoff at 28 points.

  14. 14.

    In the context of this analysis, the statistical significance of the results does not depend on the choice of bandwidth, when the results are significant they are significant at all values of the bandwidth.

  15. 15.

    Though there is no optimal bandwidth in the case of the regression discontinuity using a discrete running variable, in some ways the discrete nature of the assignment variable simplifies the problem of bandwidth choice because estimates can be computed at all possible values of the running variable (Lee and Lemieux 2010).

  16. 16.

    McCrary (2008) is perhaps the more well known test for continuity in the running variable, but I do not use it here because it is best suited for continuous running variables. In cases where the running variable is discrete, the McCrary Test can over- or under-reject the smoothness condition (Frandsen 2016).

  17. 17.

    A detailed explanation of the density test is included in the online supplement (Section 3).

  18. 18.

    Graphical representation of the continuity in the baseline covariates is shown in the online supplement (Section 4).

  19. 19.

    “Serious rules violation report” is the technical term for the misconduct reports in this administrative data set. Although the reports range in severity by class, all the reports are considered serious by the California Department of Corrections and Rehabilitation.

  20. 20.

    Though I do not present the results here, I also tested the effect of security level assignment on the number of rules violation reports acquired over the course of the review period. However, given that there is very little variation in the number of rules violation reports (more than 99% of the sample has 3 or fewer rules violation reports) the estimate of the effect of facility level placement on the number of violations may not be meaningful from a policy standpoint. Given that only 8% of the distribution has two or more rules violation reports during the review period, the policy relevant question would appear to be to estimate the effect of facility level placement on the likelihood of a rules violation report. Placement in Level II (medium security) relative to Level III (close security) does have a significant positive effect on the number of rules violation reports, but the effect is very small. Formal results are available on request.

  21. 21.

    This is the finest level of detail available in the data. A detailed listing of examples of A-F violations can be found in California Code of Regulations, Title 15 \(\S\)3323.

  22. 22.

    It is worth noting, that the most severe A1/A2 violations are a rare outcome and despite the large sample sizes at small values of the bandwidth, I would be under-powered to detect anything short of a very large effect size for this outcome.

  23. 23.

    All models were estimated with up to a third order polynomial with the exception of bandwidth equal to 2 which could only accommodate a quadratic term. Formal results are available upon request.

  24. 24.

    Formal results are available in the online supplement (Section 5).

  25. 25.

    As can be seen in Fig. 4 there is a change in slope on either side of the cutoff. Changes in slope can suggest an interaction effect, or a nonlinearity in the relationship between the running variable and the outcome. Concerns about nonlinearity can be assuaged by modeling the regression discontinuity using local polynomial regressions. The substantive conclusions at the Level III/IV cutoff do not change based on the local polynomial regressions. Results are available on request.

  26. 26.

    Unfortunately, the data cannot accommodate a specific breakdown of the violations in Division E and F.

  27. 27.

    Under-reporting by correctional officers in Level III is consistent with the data and seems to be most plausible if the observed difference is driven by officer and not inmate behavior, given that the only other explanation that would be consistent with the data pattern is that officers in Level II are over-reporting inmate misconduct. Under reporting in Level III is not only consistent with the data patterns, but also keeps the measurement vs. inmate behavior explanation in the realm of officer discretion.

  28. 28.

    As previously demonstrated, there is no evidence of manipulation across the cutoff at the score threshold between the levels, and while inmates had some control over their classification score once they have been re-assessed (in the sense that they have some control over their behavior), they still have imprecise control over the running variable, which is what is required for a valid regression discontinuity design (Lee and Lemieux 2010).

  29. 29.

    For example, inmates who have a lower preliminary score but are placed in the higher facility security level because of a binding mandatory minimum. In the language of Angrist et al. (1996), those with a binding mandatory minimum would be the “always takers,” because they will always be housed in the higher security level regardless of their preliminary score. The “never takers”, those with preliminary scores who are not housed in some other housing unit, the hospital or a treatment unit, have been excluded from the sample. Finally, the “defiers” are those with a preliminary score which would suggest a higher level and they are actually housed in a lower level. “Defiers” are uncommon at the Level II/III and Level III/IV threshold. They are most common at the Level I/II threshold because so many inmates are held at the Level II because of binding mandatory minimums (see the online supplement).

  30. 30.

    This resulted in substantial numbers of inmates moving down in facility security level, which likely had major budget implications. Unfortunately, there was no information available to calculate the security level specific costs that would allow for interpretation of the results in a cost-benefit framework.

References

  1. Angrist JD, Imbens GW, Rubin DB (1996) Identification of causal effects using instrumental variables. J Am Stat Assoc 91(434):444–455

    Article  Google Scholar 

  2. Bayer P, Hjalmarsson R, Pozen D (2009) Building criminal capital behind bars. Q J Econ 124(1):105–147

    Article  Google Scholar 

  3. Berecochea JE, Gibbs JB (1991) Inmate classification a correctional program that works? Eval Rev 15(3):333–363

    Article  Google Scholar 

  4. Berk RA, de Leeuw J (1999) An evaluation of california’s inmate classification system using a generalized regression discontinuity design. J Am Stat Assoc 94(448):1045–1052

    Article  Google Scholar 

  5. Berk RA, Ladd H, Graziano H, Baek J-H (2003) A randomized experiment testing inmate classification systems. Criminol Public Policy 2(2):215–242

    Article  Google Scholar 

  6. Blevins KR, Listwan SJ, Cullen FT, Jonson CL (2010) A general strain theory of prison violence and misconduct: an integrated model of inmate behavior. J Contemp Crim Justice 26(2):148–166

    Article  Google Scholar 

  7. Brown v. Plata, 131 S.Ct. 1910, 563 U.S. 493, 1796 Ed. 2d 969 (2011)

  8. Brown EG (2017) State of california full budget summary. http://www.ebudget.ca.gov/FullBudgetSummary.pdf. Accessed 13 Sept 2017

  9. Camp SD, Gaes GG (2005) Criminogenic effects of the prison environment on inmate behavior: some experimental evidence. Crime Delinq 51(3):425–442

    Article  Google Scholar 

  10. Camp SD, Gaes GG, Langan NP, Saylor WG (2003) The influence of prisons on inmate misconduct: a multilevel investigation. Justice Q 20(3):501–533

    Article  Google Scholar 

  11. Chen KM, Shapiro JM (2007) Do harsher prison conditions reduce recidivism? A discontinuity-based approach. Am Law Econ Rev 9(1):1–29

    Article  Google Scholar 

  12. Cohen LE, Felson M (1979) Social change and crime rate trends: a routine activity approach. Am Sociol Rev 44(4):588–608

    Article  Google Scholar 

  13. Eck JE (1995) A general model of the geography of illicit retail marketplaces. In: Eck J, Weisburd D (eds) Crime and place. Criminal Justice Press, Monsey, pp 67–93

    Google Scholar 

  14. Fan J, Irene G (1996) Local polynomial modelling and its applications. Chapman and Hall, London

    Google Scholar 

  15. Farabee D, Grattet R, McCleary R, Raphael S, Turner S (2011) Expert panel study of the inmate classification score system. Technical report, State of California Department of Corrections and Rehabilitation

  16. Frandsen B (2016) Party bias in union representation elections: testing for manipulation in the regression discontinuity design when the running variable is discrete, vol 38. Advances in econometrics. Emerald Publishing, Bingley

    Google Scholar 

  17. Gendreau P, Goggin CE, Law MA (1997) Predicting prison misconducts. Crim Justice Behav 24(4):414–431

    Article  Google Scholar 

  18. Hewitt JD, Poole ED, Regoli RM (1984) Self-reported and observed rule-breaking in prison: a look at disciplinary response. Justice Q 1(3):437–447

    Article  Google Scholar 

  19. Huebner BM (2003) Administrative determinants of inmate violence: a multilevel analysis. J Crim Justice 31(2):107–117

    Article  Google Scholar 

  20. Imbens GW, Lemieux T (2008) Regression discontinuity designs: a guide to practice. J Econom 142(2):615–635

    Article  Google Scholar 

  21. Kreager DA, Young JT, Haynie DL, Bouchard SDR, Martin Zajac G (2017) Where “old heads” prevail: inmate hierarchy in a men’s prison unit. Am Sociol Rev 82(4):685–718

    Article  Google Scholar 

  22. Lee DS, Lemieux T (2010) Regression discontinuity designs in economics. J Econ Lit 48(2):281–355

    Article  Google Scholar 

  23. Lerman AE (2009) The people prisons make: effects of incarceration on criminal psychology. In: Raphael S, Stoll M (eds) Do prisons make us safer?: The benefits and costs of the prison boom. Russell Sage, New York

    Google Scholar 

  24. Lerman AE (2013) The modern prison paradox. Cambridge University Press, New York

    Google Scholar 

  25. MacDonald JM (1999) Violence and drug use in juvenile institutions. J Crim Justice 27(1):33–44

    Article  Google Scholar 

  26. McCorkle RC, Miethe TD, Drass KA (1995) The roots of prison violence: a test of the deprivation, importation, and not-so-total institution models. Crime Delinq 41(3):317–331

    Article  Google Scholar 

  27. McCrary J (2008) Manipulation of the running variable in the regression discontinuity design: a density test. J Econom 142(2):698–714

    Article  Google Scholar 

  28. Morris RG, Carriaga ML, Diamond B, Piquero NL, Piquero AR (2012) Does prison strain lead to prison misbehavior? An application of general strain theory to inmate misconduct. J Crim Justice 40(3):194–201

    Article  Google Scholar 

  29. Nguyen H, Loughran TA, Paternoster R, Fagan J, Piquero AR (2016) Institutional placement and illegal earnings: examining the crime school hypothesis. J Quant Criminol 33(2):207–235

    Article  Google Scholar 

  30. Osgood DW, Wilson JK, O’Malley PM, Bachman JG, Johnston LD (1996) Routine activities and individual deviant behavior. Am Sociol Rev 61(4):635–655

    Article  Google Scholar 

  31. Page J (2011) The toughest beat: politics, punishment, and the prison officers union in California. Oxford University Press, Oxford

    Google Scholar 

  32. Petersilia J (2006) Understanding california corrections. Technical report, California Policy Research Center, University of California

  33. Porporino FJ, Doherty PD, Sawatsky T (1987) Characteristics of homicide victims and victimization in prisons: a Canadian historical perspective. Int J Offender Ther Comp Criminol 31(2):125–136

    Article  Google Scholar 

  34. Schwarzenegger A (2006) Prison overcrowding state of emergency proclamation. Technical report, State of California

  35. Shermer LO, Bierie DM, Stock A (2012) Endogeneity in prison risk classification. Int J Offender Ther Comp Criminol 57(10):1248–1274

    Article  Google Scholar 

  36. State of California Legislative Analysts Office (2017) California department of corrections and rehabilitation 2017–2018 budget. http://www.lao.ca.gov/Publications/Report/3595. Accessed 13 Sept 2017

  37. Steiner B, Butler HD, Ellison JM (2014) Causes and correlates of prison inmate misconduct: a systematic review of the evidence. J Crim Justice 42:462–470

    Article  Google Scholar 

  38. Steiner B, Wooldredge J (2008) Inmate versus environmental effects on prison rule violations. Crim Justice Behav 35(4):438–456

    Article  Google Scholar 

  39. Steiner B, Wooldredge J (2009) The relevance of inmate race/ethnicity versus population composition for understanding prison rule violations. Punishm Soc 11(4):459–489

    Article  Google Scholar 

  40. Sykes GM (1958) The society of captives: a study of a maximum security prison. Princeton University Press, Princeton

    Google Scholar 

  41. Thistlethwaite DL, Campbell DT (1960) Regression-discontinuity analysis: an alternative to the ex post facto experiment. J Educ Psychol 51(6):309

    Article  Google Scholar 

  42. Title 15, California Code of Regulations (2012) Crime prevention and corrections: rules and regulations of adult institutions, programs, and parole. http://www.laprogressive.com/wp-content/uploads/2012/09/Title15-2012.pdf. Accessed 22 Jan 2019

  43. Title 15, California Code of Regulations (2016) Crime prevention and corrections: rules and regulations of adult institutions, programs, and parole. https://www.cdcr.ca.gov/Regulations/Adult_Operations/docs/Title15-2016.pdf. Accessed 22 Jan 2019

  44. Trochim WM (1984) Research design for program evaluation: the regression-discontinuity approach, vol 6. Sage Publications, Inc., Beverly Hills

    Google Scholar 

  45. Wooldredge J, Steiner B (2015) A macro-level perspective on prison inmate deviance. Punishm Soc 17(2):230–257

    Article  Google Scholar 

  46. Worrall JL, Morris RG (2011) Inmate custody levels and prison rule violations. Prison J 91(2):131–157

    Article  Google Scholar 

  47. Wortley R (2002) Situational prison control: crime prevention in correctional institutions. Cambridge University Press, Cambridge

    Google Scholar 

Download references

Author information

Affiliations

Authors

Corresponding author

Correspondence to Sarah Tahamont.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

I would like to thank the California Department of Corrections and Rehabilitation and the Expert Panel on the Inmate Classification Score System in California (David Farabee, Ryken Grattet, Richard McCleary, Steven Raphael, and Susan Turner). I would also like to thank Badi Hasisi (the editor), Steve Raphael, Rucker Johnson, Frank Zimring, Brigham Frandsen, Shawn Bushway, Ray Paternoster, Wayne Osgood, Tom Loughran, Wade Jacobsen, Jean McGloin, three anonymous reviewers, and seminar participants at the Bloch School of Management at the University of Missouri, Kansas City; the Hindelang Criminal Justice Research Center at the University at Albany, SUNY; the University of Maryland, and the University of Chicago Crime Lab, New York for helpful comments. All errors are my own. Direct correspondence to Sarah Tahamont, Department of Criminology and Criminal Justice, University of Maryland, 2220 LeFrak Hall, College Park, MD 20742 (email: tahamont@umd.edu)

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (pdf 836 KB)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Tahamont, S. The Effect of Facility Security Classification on Serious Rules Violation Reports in California Prisons: A Regression Discontinuity Design. J Quant Criminol 35, 767–796 (2019). https://doi.org/10.1007/s10940-019-09405-0

Download citation

Keywords

  • Corrections
  • Prison misconduct
  • Administrative data
  • Regression discontinuity