Do Parole Technical Violators Pose a Safety Threat? An Analysis of Prison Misconduct

Parole Revocation

In 2010, by year-end, there were 1.6 million prisoners in state and federal institutions (Guerino, Harrison, & Sabol, 2011), and more than 93% of these inmates will eventually be released (Petersilia, 2003). Subsequently, the number of inmates returning to society is now 4 times greater than it was 20 years ago (Harrison & Karberg, 2003; Travis & Visher, 2005). Studies have consistently shown that outcomes for released prisoners are not positive; roughly 30% of released inmates are rearrested within 6 months postrelease (Petersilia, 2003), two thirds of inmates are rearrested within 3 years, half reconvicted (P. A. Langan & Levin, 2002), and as many as 40% of inmates return to prison within 3 years (Pager, 2003).

Unfortunately, little is known about parolees who return to prison for parole revocations. The readmission of parolees to prison has been identified as a significant channel in spurring mass incarceration (Clear, 2007; Mauer, 2005, Petersilia, 2003; Simon, 2000). Parolees return to prison through two mechanisms: either through the commission of a new offense or revocation of parole due to violating a technical condition of parole. The number of prison admissions for parole violations has increased at a much faster rate than the increase in admissions for new offenses. Between 1980 and 2000, state prisons experienced a sevenfold increase in the number of prison admissions for parole violations (Travis & Lawrence, 2002). Travis and Lawrence (2002) note that the number of prison admissions for parole violations in 2000 was roughly equal to the total admissions to state prisons in 1980. However, currently there is a dearth of information about the behavior of individuals after returning to incarceration (Petersilia, 2003; Travis, 2007), as research focusing on technical violators is very limited (Travis, 2007; see Lin, Grattet, & Petersilia, 2010; Steen & Opsal, 2007). Understanding the differences in behavior between these two groups of prisoners is relevant for informing prison operations and parole decisions.

Method

Data and Sample

The data utilized for the present study stem from official prison records provided by a large southern state’s primary corrections agency. The data included information on offender and current offense characteristics, complete disciplinary histories while incarcerated, and information on past convictions. We limited the analysis to male inmates returning to prison for at least the second time (new inmates were excluded as they are a qualitatively different group, so this was important to our study to aid in better aligning treatment and control groups) and who returned to prison between May 1, 2004, and May 31, 2006, and who were still incarcerated on August 15, 2009. ² This approach allowed for us to observe inmate behavior, via official disciplinary records, over a complete 3 years of incarceration beginning with each inmate’s specific entry date; the period for which misconduct is most likely to occur (see Adams, 1992; Flanagan, 1983; Griffin & Hepburn, 2006; Morris & Worrall, 2010). Inmates from the sample were each assigned to 1 of 47 general population prison units in the state. In this particular state, units are not mutually exclusive in terms of security level. Inmates housed in state jails, transfer facilities, private prisons, and substance abuse treatment facilities were excluded from the analysis. Also excluded were inmates sentenced to capital punishment or life without parole. This strategy resulted in a total of 24,331 inmates eligible for analysis. ³ Descriptive statistics are presented in Table 1.

OPEN IN VIEWER

Table 1.

Descriptive Statistics.

	Technical violators (n = 7,655)				Full sample (N = 24,331)
Variable	M	SD	Minimum	Maximum	M	SD	Minimum	Maximum
Treatment
Technical violator	—	—	—	—	0.31	0.46	0	1
Outcome variables
Violent	0.01	0.11	0	1	0.11	0.31	0	1
Property	0.05	0.21	0	1	0.37	0.48	0	1
Drug related	< 0.01	0.06	0	1	0.05	0.21	0	1
Covariates
Education	6.6	3.86	0	13	6.6	3.85	0	13
IQ Score	91.54	13.69	57	148	91.05	13.62	48	148
Age	42.9	9.8	17.71	78.16	38.53	9.95	17.5	83.07
Black	0.48	0.5	0	1	0.44	0.50	0	1
Hispanic	0.21	0.41	0	1	0.26	0.44	0	1
Priors	1.56	0.74	1	5	1.6	0.81	1	5
Sentence length (log)	9.02	1.43	6.59	16.11	8.56	1.63	5.2	16.11
Physical stature	27.98	4.58	16.73	53.92	28.34	4.84	11.3	53.94
Gang member	0.11	0.31	0	1	0.16	0.37	0	1
Drug offense	0.27	0.44	0	1	0.26	0.44	0	1
Property offense	0.38	0.48	0	1	0.23	0.42	0	1
Sexual assault history	0.14	0.35	0	1	0.16	0.37	0	1
Married	0.22	0.41	0	1	0.19	0.39	0	1

Results

As noted, the first step involved the estimation of t tests between technical violators and new crime inmates among the raw (unmatched) sample. These results are presented under the “Unmatched” heading of Table 2. The results demonstrate that every covariate’s mean value, aside from education, is significantly different between the unmatched treatment and control groups. Because we find such differences, we now have strong evidence suggesting that we need to match nontechnical violators to technical violators who had very similar (near equivalent) probabilities of returning to prison on a technical violation. The estimates from the logistic regression model, using the treatment indicator as the outcome to establish propensity of treatment scores (i.e., predicted probabilities of treatment), were used to match technical violators (one-to-one, without replacement) to a nontechnical violator who had a near-equivalent probability of self-selecting into the treatment (caliper = .05). Ultimately, 6,563 of 7,595 (86.4%) technical violators were successfully matched to 6,920 nontechnical violators. In Table 2 are t test statistics for each covariate after matching showing that all covariate differences prematching were eliminated via the matching process. These results allow us to preliminarily assume that the matched groups are equivalent (i.e., balanced) regarding the propensity for being a technical violator. Furthermore, standardized bias statistics for each covariate were reduced to below 20, thus corroborating the evidence that we were successful in matching inmates.

OPEN IN VIEWER

Table 2.

Balance Statistics Among Technical and New Crime Inmates, Pre/Postmatching t-Tests (Nearest-Neighbor Method).

	Unmatched			Matched			Logit model predicting technical violation propensity score
	Technical violator	New crime	t-value	Technical violator	New crime	t-value	Coefficient	SE
Education score	6.60	6.60	0.09	6.66	6.67	−0.10	0.18 *	0.00
IQ score	91.5	90.8	3.79 *	91.2	91.15	0.26	0.00	0.00
Age	42.9	36.5	48.67 *	40.9	41.48	−1.17	0.07 *	0.00
Black	0.48	0.42	8.41 *	0.45	0.45	0.37	0.11 *	0.04
Hispanic	0.21	0.29	−13.06 *	0.23	0.23	−0.40	−0.08	0.04
Priors	1.56	1.62	−5.45 *	1.58	1.57	0.75	−0.27	0.02
Sentence length (log)	8.97	8.25	37.75 *	8.84	8.67	6.41 *	0.29 *	0.01
BMI	27.9	28.5	−7.80 *	28.2	28.15	0.52	−0.02 *	0.00
Gang member	0.11	0.19	−16.60 *	0.12	0.12	0.24	−0.19	0.05
Drug offender	0.27	0.25	3.55 *	0.29	0.29	0.80	0.69 *	0.04
Property offender	0.38	0.16	7.56 *	0.29	0.30	−1.16	1.43 *	0.04
Sex offender	0.14	0.18	−7.01 *	0.15	0.14	0.55	−0.15 *	0.04
Married	0.22	0.18	7.79 *	0.22	0.21	0.93	−0.18	0.04
Constant							−5.52	0.18

NOTE: BMI = body mass index. Standardized bias statistics are below 20 for all predictors in the matched sample.

^*p < .001.

Presented in Table 3 are our postmatched group t-test statistics across each of the three outcome variables (violent, property, and drug-related inmate misconduct). As shown, the mean probability of each form of misconduct for both matched and unmatched groups is significantly higher for nontechnical violators (i.e., recidivists). Inmates who return to prison for a new offense appear to be significantly more likely to engage in violent, property, and drug-related misconduct compared with inmates who return to prison on a technical violation, all else constant.

OPEN IN VIEWER

Table 3.

Differences in Misconduct Between Technical/New Crime Inmates Pre and Postmatching.

	Unmatched			Matched
	Technical violator	New crime	t-value	Technical violator	New crime	t-value
Violent misconduct	0.0145	0.1533	−33.64 *	0.0132	0.1162	−24.52 *
Property misconduct	0.0454	0.5150	−78.89 *	0.0523	0.4334	−56.85 *
Drug-related misconduct	0.0036	0.0651	−21.47 *	0.0041	0.0457	−16.33 *

*

p < .001.

To determine whether propensity scores may be biased due to the influence of an unmeasured covariate/s (i.e., hidden bias), we conducted the sensitively assessment protocol recommended by Rosenbaum (2002), and demonstrated by Guo and Fraser (2010; see also Becker & Caliendo, 2007). Put simply, the approach provides estimates (γ) of the magnitude for which an omitted covariate would need to render the PSM results nonsignificant. In short, the test estimates how powerful an omitted covariate must be to potentially invalidate the results of the analysis. As noted by Becker and Caliendo (2007), gamma values near 1.0 suggest that a PSM analysis is highly sensitive to hidden bias. The further from 1.0 that gamma is, the less sensitive the model is to unmeasured influences on the probability of treatment. The sensitivity analysis was carried out for each outcome. Significant gamma estimates were 8.5, 15.5, and 9.0 for violence, property, and drug-related outcomes, respectively. Contextually, for example, this means that an unmeasured covariate would have to increase the odds of being a technical violator by 8.5 times to render the analysis for the violence outcome nonsignificant. Overall, the results of the sensitivity analysis suggest that propensity scores for each outcome were highly robust to hidden bias.

The final step in our analysis was the IPTW regression analysis. Because the outcomes were binary and the data nested, we estimated multilevel logistic regression (aka HGLM [hierarchical generalized linear models]) models where the sole predictor was the weighted treatment effect. The results (omitted to save space) overwhelmingly suggest that returning to prison via a technical violation markedly reduces an inmate’s probability of engaging in any form of misconduct. However, differences relative to the type of misconduct were found. Such differences are clearly visualized in Figure 1, which presents the predicted probabilities stemming from the multilevel IPTW model results. As shown, differences in expected participation in property-related types of misconduct are the greatest, followed by violence and then drug-related misconduct. Nontechnical violators are substantially more likely to engage in any type of misconduct, as defined here, compared with technical violators.

OPEN IN VIEWER

Figure 1.

Predicted probabilities of misconduct between matched treatment and control groups.

In sum, the results presented here were based on a thorough and multitiered counterfactual analysis and were demonstrated to be highly robust to hidden bias. Moreover, we were able to match almost 7,000 pairs of inmates from a complete inmate population on the treatment variable who returned to prison over a specific period of time, thereby reducing the influence of possible cohort effects. It is clear that returning inmates who are technical violators, as opposed to new crime violators, are at a much lower risk of misconduct. In the following section, we provide a discussion of how these findings may apply to theory, policy, and future research on prison offender management.

Discussion

To date, there has been very limited empirical research addressing behavioral differences between groups of individuals returned to prison postrelease. Furthermore, within the realm of literature addressing prison misconduct, no other known study has examined the differences based on mechanism of return to prison. In furthering the understanding of these behavioral differences, the current study benefited from the use of statistical techniques that controlled for bias in the effect of returning to prison for a technicality versus a new crime. Based on the official disciplinary records from male inmates serving a sentence in the prison system of a large southern state after readmission, the findings presented here suggest that technical violators are significantly less likely to engage in any form of prison misconduct. These differences varied by type, with the largest differences in property-related misconduct, followed by violence, then drug-related misconduct. Again, due to the use of the PSM, these results are net of any selection bias and were reinforced by the results from the IPTW regression analysis.

The differential variation observed in the predicted probabilities for misconduct type between technical violators and new offenders is a finding that suggests the need for future research in this area. Further examination of the reasons for these differences was not possible based on data limitations and capacity of this study. As it is unknown what the exact parole violation was, it is impossible to assess any relationship between type of violation and type of misconduct.

The findings presented here not only have implications for prison management but also for parole revocation decisions. Empirical studies on recidivism that have included institutional misconduct have reported a significant positive association both for rates of reoffending and early failure (DeLisi, 2003; Huebner & Berg, 2011; Matsueda et al., 1992; Trulson, DeLisi, et al., 2011; Trulson, Haerle, et al., 2011). Situated within the context of a positive correlation between prison misconduct and future criminal offending, these results have implications for policy makers and prison and parole administrators in enacting policies and making decisions on best practices for reducing prison populations.

Although it was beyond the scope of this article to provide a test of the mechanisms that may be at work in the relationship between prison misconduct and future criminal offending, the available theoretical perspectives provide a framework for espousing a positive relationship. The link between misconduct and propensity for reoffending is based on the logical continuation of a criminal career framework and life-course theory in the explanation of criminal trajectories and desistance (DeLisi, Trulson, Marquart, Drury, & Kosloski, 2011; Sorensen & Davis, 2011; Trulson, DeLisi, Caudill, Belshaw, & Marquart, 2010; see also Bushway, Piquero, Broidy, Cauffman, & Mazerolle, 2001; Laub & Sampson, 2003; Piquero, Farrington, & Blumstein, 2003; Sampson & Laub, 1997, 2003). Empirical evidence supports the positive relationship between prior criminal activity or incident reports and future prison misconduct (Camp et al., 2003; Cunningham & Sorensen, 2007; Gendreau et al., 1997; Walters, 2007) and suggests that prison misconduct is associated with recidivism (Camp & Gaes, 2005; DeLisi, 2003; Heil, Harrison, English, & Ahlmeyer, 2009; Huebner & Berg, 2011; N. P. Langan, Camp, William, & Saylor, 2004; Trulson, DeLisi, et al., 2011; Trulson, Haerle, et al., 2011). This begins to underscore the relevant impact that prison misconduct has on predicting future recidivism. Indeed, to the extent that prison has a criminogenic effect on behavior, it is possible to speculate that returning parole violators to prison may enhance their likelihood of engaging in prison misconduct and therefore increase the risk of future offending. ⁶

As the issues surrounding prisoner reentry and prison population reduction continue to gain momentum, it may be necessary to view prisoner misconduct outside of the framework of adaptation to incarceration and place it in the context of entire criminal careers. Understanding patterns of misconduct within and between types of inmates promotes identification of offenders who may pose less of a risk to public safety. Incarceration, while commonly viewed from the standpoint of incapacitation, and therefore as a break in criminal offending, is often anything but.

One of the main issues with this line of logic is the claim that not all forms of prison misconduct are considered to be “real” crimes. If many of the behaviors eligible for disciplinary infractions are committed outside the prison environment, they would not be subject to criminal justice intervention. Restrictions on personal property, not complying with grooming standards, being out of place, as well as rules about contact with visitors or communication with the outside world are enforced, and violations can result in disciplinary action while incarcerated. However, there are a number of forms of misconduct that would be considered criminal, including various forms of assault, drug possession and trade, and murder. These distinctions have led to the questioning of the theoretical relevance of prison misconduct on future behavior and whether it is based on the same criminal propensity for general offending, as it is possible that criminal behavior and institutional misconduct may have differing causal origins.

The available empirical evidence, however, has not supported this claim. Even though all misconduct is not criminal, studies have found evidence to suggest that all forms of misconduct can be indicative of a propensity toward criminal behavior (Camp et al., 2003; DeLisi et al., 2011; Sorensen & Davis, 2011). Camp and Gaes (2005) concluded that prison misconduct and criminal behavior are positively correlated, based on findings that criminal behavior and prison misconduct are both predicted by prior criminality. Furthermore, they reasoned that this conclusion is logical when working within a life-course framework. In addition, findings presented by French and Gendreau (2006), in their meta-analysis of the impact of correctional treatment programs on reducing misconduct, signified that inmate misconduct and criminal behavior share common predictors. Results showed that the most effective prison programs in reducing institutional misconducts were also able to produce a reduction in recidivism rates, leading to the assertion that “[this result] reinforces the view that prison misconduct behavior is a reasonable proxy for antisocial behavior in the community” (French & Gendreau, 2006, p. 210).

Recently, some have called for research following the life-course perspective that focuses mainly on criminal behavior occurring in the community to be integrated with research on antisocial behavior in prison under a life-course importation model of inmate behavior (DeLisi et al., 2011; Sorensen & Davis, 2011). DeLisi et al. (2011) argued that the same criminogenic processes and traits that are at work in influencing delinquency remain when those same individuals are incarcerated. Results from negative binomial regression models of 2,520 institutionalized juvenile males, indicating that youths with lengthier and more serious delinquent careers were more involved in misconduct, supported their importation model hypothesis. Also finding support for a behavioral continuity thesis, Sorensen and Davis (2011) found some degree of significant relationship between crime of conviction and one’s propensity to commit violent misconduct in prison in their study of inmates from the Texas Department of Criminal Justice.

While it is understood that no guarantees can be made regarding future human behavior, our results, we suggest, identify a group of inmates who are less likely to pose a serious risk to safety. Failure of some individuals who have committed technical violations to remain crime-free would be expected; however, as a whole, this group of offenders is still far less likely to commit misconduct than those who were readmitted for a new offense. It is important to keep in mind that an overwhelming majority of these offenders will be released from prison at some point; so basing release or readmission decisions on likelihood of short-term criminal propensity is a logical choice. The public and policy makers, when forced to reduce the prison population, need to weigh the risks of those offenders who may be committing technical violations versus those who have committed crimes that may justify their imprisonment. Using up prison resources and space on parole violators does not appear to be the most prudent choice.

As with other studies, ours is not without other potential limitations. The data used in the analysis were from a single, yet populous, southern state. Without replication from different state inmate populations, caution should be used in generalizing the findings presented here. Furthermore, only males were included in the sample and so conclusions regarding the behavior of female inmates cannot be drawn. The findings were also based on an analysis of official disciplinary infractions. The use of official data comes with its own set of limitations, and this study is not exempt. It is possible that these official reports are biased and do not reflect all misconduct that occurs in prison. However, there is reason to believe that official misconduct records are reflective of actual misconduct. Daggett and Camp (2009), in their analysis of official misconduct records and inmate survey data, concluded that official records are representative of inmate reporting and perceptions of prison safety. Furthermore, our data did not capture misconduct reported during prison sentences occurring prior to the offense of record (i.e., the start of our observation period). Thus, we must assume that the probability of misconduct resets upon entry to prison for a subsequent term, net of the effects of our control variables. It is important to note that although a substantial assumption, the attributes we do control for are strongly correlated to misconduct, past, present, and future, which does help to limit this concern, and our strong sensitivity analysis go a long way to support the robustness of our findings. ⁷ Nonetheless, this component will indeed vary across inmates and should be the topic of future research (i.e., does the probability of misconduct reset upon reentry to prison, in and of itself).

The use of any incarcerated population is also a function of official intervention and processing. Evidence has shown that judges continue to exercise discretion in their decisions on who to return to prison for parole violations (Lin et al., 2010).Therefore, the possibility exists that the included sample is not representative of all technical violators, only those who get caught, are reported, have their parole revoked and are returned to prison. To the extent that judges and parole officers’ decisions not to report or send parole violators to prison is based on their opinion that those parolees do not pose as significant of a risk, we feel our results are enhanced as strong significant results were found even among those who were deemed a greater threat and subsequently readmitted.

Conclusion

The use of prisons in the United States has a valid purpose and place, and it is likely that their use and significance will not diminish. The large numbers of individuals who continue to be cycled through the prison system underscore the need to better understand their behavior in and out of prison, and the relationship in between. The operation of the criminal justice system, in general, and the correctional system, specifically, is a costly endeavor. Therefore, it behooves policy makers and correctional administrator to reserve the use of prison for those who most warrant the use of this sanction. As prison administration and management continues to be an increasingly costly endeavor, risk prediction becomes increasingly important in making in/out decisions. Our findings show that technical violators exhibit far less problematic behavior when reincarcerated, suggesting that the reduction or elimination of returning technical violators to prison may be an effective way to reduce the prison population without increasing the risk posed to the community. This is not to say that technical violations should be ignored or not face some form of penalty. The implication would be that greater use of alternative punishments that keep offenders in the community should be used.