Offender Rehabilitation: Examining Changes in Inmate Treatment Characteristics, Program Participation, and Institutional Behavior

Abstract

The efficacy of offender rehabilitation has been a topic of much debate over the past few decades. While much of the corrections literature has focused on program effectiveness, less attention has been placed on the expansion and delivery of services to incarcerated offenders, and whether the advances that have been made with regard to rehabilitation have changed the nature of treatment delivery to inmates. Using data from three time points collected as part of the Bureau of Justice Statistics (BJS) survey of state inmates, this paper examines the criminogenic needs of offenders and how those needs have changed over time, the role inmate needs play in driving participation in institutional programs, and whether inmates with unmet treatment needs commit a disproportionate number of institutional infractions. The results suggest that inmate needs have changed substantially over the past decade, with the most extensive needs concentrated in a small proportion of inmates. Consequently, correctional institutions are not always been able to match offenders to the appropriate services, which may have a direct impact on institutional safety.

Keywords:

• prisons
• treatment programs
• institutional misconduct
• offender needs

Acknowledgments

The author would like to thank especially John R. Hipp for his helpful suggestions on earlier drafts. The author would like to thank also Susan Turner and the anonymous referees for their helpful comments.

Notes

1. The prison population grew from just over 743,000 in 1991 to more than 1.4 million in 2004 (Bureau of Justice Statistics, 2009).

2. Although one survey was administered between 1980 and 1990 (1986), the consequences of key legislation such as the 1986 Anti‐Drug Abuse Act would likely not have been captured in that survey year. Furthermore, changes in the survey itself across survey waves complicated the ability to include earlier survey waves, as the precision of the measures used in this analysis deteriorated.

3. Andrews and Bonta (2003) also delineated the “big four” risk factors, antisocial attitudes, antisocial associates, history of antisocial behavior, and antisocial personality pattern, which are included in the “big eight”.

4. Many of these risk factors (e.g., antisocial attitudes, antisocial associates, history of antisocial behavior, and antisocial personality) are often factors that are addressed within a specific treatment modality, such as cognitive behavioral therapy, multimodal programs or skill‐based programs; these programs typically teach offenders how to resist antisocial behaviors. Hence, these factors are often rolled into the treatment modality of a larger program, such as substance abuse treatment.

5. For example, history of antisocial behavior is not a dynamic criminogenic need and is, therefore, not susceptible to programmatic interventions (MacKenzie, 2006).

6. Some researchers have asserted that mental health is a significant factor in criminality in that it often drives criminogenic needs. While mental health may be an important factor, it is outside the scope of this paper. Furthermore, consistent measures of mental health were not available across the three waves used in this study, and therefore, it could not be included.

7. The prison population of 2.1 million includes federal and state prisoners and inmates in local jails (Bureau of Justice Statistics, 2006).

8. It should be noted that the growth in the prison population has slowed in more recent years. For example, between 2007 and 2008, the prison population increased by less than 1,000 (1,517,867 and 1,518,559 for 2007 and 2008, respectively), while this increase was more than 44,000 between 2005 and 2006 (1,448,344 and 1,492,973 for 2005 and 2006, respectively).

9. Prison education programs received a boost in 1991, when the federal mandatory literacy level was increased from an eighth‐grade level in 1985 to a high school diploma or GED level (MacKenzie, 2006). This federal standard may have had some influence on state facilities, with about 26% of state prison systems mandating education programs (Di Vito, 1991; MacKenzie, 2006). However, this policy change was seemingly contradicted in 1994, when restrictions to post‐secondary educational programs were enacted. Inmates had previously been able to attend prison college programs through the use of Pell grants, which were designed to assist families with low or middle incomes with college; however, after much criticism, the passage of the 1994 Crime Bill excluded inmates from receiving and using Pell grants for college programs (Batiuk, Moke, & Rountree, 1997; MacKenzie, 2006).

10. Between 1991 and 2004, the prison population grew by over 620,000 individuals (Bureau of Justice Statistics, 2009).

11. The sample weighting factors were applied to account for the probability of a respondent being selected for the sample, variable non‐response rates across selection strata, and inmate characteristics.

12. For details on the sample design and weighting procedures for each survey year, please refer the Survey of State and Federal Correctional Facilities Resource Guide, available at: http://www.icpsr.umich.edu/NACJD/sisfcf/.

13. The final weight was not used in the analyses; instead, standard errors were corrected by clustering on facility. Applying both the final weight and clustering by facility over‐corrected the standard errors and produced incorrect results; based on reviewer comments, only the clustering was applied for the final analyses.

14. A number of previous studies have validated the measures used by the LSI‐R. See Andrews and Bonta (1995) for a review of research regarding the predictive validity of the LSI‐R.

15. The 10 subscales consist of the following: criminal history, education/employment, financial, family/marital, accommodation, leisure/recreation, companions, alcohol/drug problem, emotional/personal, and attitudes/orientation. Although the LSI‐R is used to determine the overall risk and needs of an individual, the various subscales provide a measurement of criminogenic need for each respective domain. There are a number of domains not included in this analysis because they do not necessarily map onto specific programs. For example, there may not be a specific program addressing leisure time or companions; however, these issues might be addressed in a substance abuse program as a means of identifying situations in which substance abuse occurs. Furthermore, specific programs addressing leisure time, family/marital issues, accommodation, financial, etc., were not captured in the BJS survey and are therefore a limitation of the data itself.

16. Since each subscale provides a measurement of criminogenic need for each respective domain, the fact that the other domains are not captured should not compromise the efficacy of using the LSI‐R to define treatment needs for education/employment and substance abuse.

17. As one reviewer pointed out, these questions do not ask respondents about illegal employment. There may be some respondents, for example, who were not seeking employment because they were satisfied with their illegal income. The BJS data do not capture these issues, and therefore it is unclear the number of people who might be misclassified. However, research does show that individuals with limited job skills, unemployment, and earning potential tend to commit a disproportionate number of crimes (Freeman, 1996; Short, 1997). Therefore, it is likely that individuals earning an illegal income (via criminal activity) have educational and employment needs that could be improved through programming.

18. When tested as a latent construct, criminal history was highly correlated (<.80) with the latent construct for drug and alcohol needs. Consequently, the criminal history indicators were tested independently.

19. Length of incarceration was calculated by subtracting the admission date from the interview date to obtain the total number of days an inmate had been incarcerated. The difference was then logged to account for the skewed distribution.

20. Some recent research has looked at the influence of staff characteristics on institutional misconduct (Camp et al., 2003; McCorkle et al., 1995), but much of the literature does not consistently include staff characteristics in their analyses (see Camp et al., 2003; Jiang, 2005; Steiner, 2008; Steiner & Wooldredge, 2008; Wooldredge & Steiner, 2009). Variables regarding staff composition are not available in the Survey of State Inmates and are therefore not included in the model here; however, this should not pose an omitted variable bias problem. Research has shown that while the inclusion of staff characteristics may be statistically significant, their inclusion does not change the significance or magnitude of effect of the variables when staff characteristics are not included (Camp et al., 2003).

21. Conventional commitment measures also include employment status and high school education, measures already included in the model (see Harer & Langan, 2001; Wooldredge, Griffin, & Pratt, 2001).

22. The security level of the institution is usually included in studies examining institutional misconduct (see Cao, Zhao, & Van Dine, 1997; Jiang and Fisher‐Giorlando, 2002; McCorkle et al., 1995). A limitation of the BJS data is that security level measures are only available in the 1991 and 1997 surveys. To test the potential bias of not including these variables, we included a measure of security level in our final models and tested it against the results reported here for 1991 and 1997. Security level was operationalized as a series of dummy variables: maximum and medium security level, with minimum security serving as the reference category. The results showed that security level was generally significant throughout our models; however, its inclusion did not dramatically alter the magnitude of the parameter estimates for the other variables in the model. There were slight changes in the size of some of the parameter estimates, but these changes were nominal. None of the significant parameter estimates reported here changed sign or significance. Two parameter estimates in 1991 (the effect of total number of current offenses on drug and alcohol treatment and probation on prison employment) and one parameter estimate in 1997 (the effect of male on vocational program) not significant in the results reported here became significant when security level was included in the model. This suggests that, while security level is an important and significant predictor of both treatment participation and institutional misconduct, the omission of these variables in the results reported here does not bias the findings. This may be due to the fact that individual characteristics associated with security level, such as the number or prior incarcerations, severity of offense, and commitment offenses have already been taken into account. A recent study by Camp and Gaes (2005) suggests that when inmates with the same risk of committing institutional misconduct were assigned to either a low security prison or a higher security prison, they had an equal probabality of committing misconduct. This infers that the security level of the prison may not have an effect on an inmate’s propensity to commit an infraction, other factors being equal.

23. Raw population counts were not readily available in the data; however, a base weight for males and females was calculated and included in the dataset for each participant. This weight indicates the number of inmates each person represents. Using this logic, each inmate was multiplied by their respective weight. This value was then summed up by facility in order to achieve an approximation of the population for each respective institution.

24. All fit indices and χ ² values are the averaged values across imputations. The results were not sensitive to the imputations.

25. The fit statistics validate the relationship between the factor loadings and the underlying latent constructs, with a CFI and TLI of .952 and .946, respectively, and an RMSEA of .059. The obtained χ ² value was 6102.036, df = 120.

26. As indicated above, the mean value of both latent constructs was set to zero in 1991 in order to assess whether mean change across years is statistically significant.