Oregon’s Transitional Leave Program and Recidivism

Early Release Programs

Compared with research on many other topics in corrections, we know relatively little about the impact of leaving prison early. The so-called “good-time” credits have been used for decades to incentivize positive behavior in prison, but rigorous research on these policies is lacking. A small number of studies have looked at the effect of leave in the context of establishing employment (i.e., work release) or reconnecting with family (i.e., home release). Cheliotis (2008) systematically reviewed the early literature in this field and concluded that these programs lead to small reductions in recidivism. Several methodological limitations qualify this assertion. First, most of these studies failed to account for other factors connected to program participation that might impact outcomes. In other words, individuals granted work or home release might have been at lower risk from the outset. Second, the available studies vary greatly with regard to the length of leave granted and the follow-up period for assessing recidivism. Third, some studies combined transitional leave with additional programming requirements (e.g., vocational training and drug treatment), raising questions about the independent effect of early release. Fourth, a few studies included temporary leave programs, where releasees must return to custody each night or on weekends. Finally, many of the studies reviewed (see Joo et al., 1995; Sims & O’Connell, 1985; K. A. Wright & Rosky, 2011) are comparatively old, reflecting criminal justice practices and community characteristics that may no longer apply.

Several more recent studies have addressed some but not all of these limitations. Weisburd and colleagues (2017) studied a program for low-risk individuals in Israel that combined work release with therapeutic interventions. They used propensity score modeling (PSM), a statistical procedure capable of balancing groups on numerous potentially confounding factors, to match 547 comparison subjects with 547 granted work release. Releasees in the latter group were significantly less likely to be rearrested over a 12-month follow-up. Of course, the results from a study of Israeli imprisonment should be viewed within their context that differs from the U.S.’s., with the most pronounced difference being the substantially shorter sentences in Israel.

Duwe (2015) used PSM to study a work release program for low-risk individuals in Minnesota. Participants were transferred to a local control facility (e.g., county jail, residential facility) and allowed to work or seek vocational training in the community. Additional supervision measures included random drug testing, and some participants were required to participate in counseling. The treatment group consisted of 1,785 persons granted work release and an equal number who served their planned sentence. Work release was associated with small reductions in rearrest, reconviction, and reincarceration. The opposite association was seen with technical violations: individuals granted a work release were significantly more likely to violate their release conditions.

Bales and colleagues (2016) similarly evaluated Florida’s work-release program. Participating AICs were housed in a release facility and allowed to work in the community. The researchers compared 27,463 individuals granted work release with 15,911 individuals who were eligible for the program but did not participate. People granted work release were less likely to be rearrested after 12 months and less likely to be reconvicted. Additional analyses showed that these differences persisted even after including statistical controls to account for preexisting differences between the two groups. No meaningful difference was observed in the 1-year reincarceration rate.

Forced Releases

A related area of research concerns early release from prison to address overcrowding. Sims and O’Connell (1985) studied 1,674 individuals released roughly six months early from prisons in Washington State in response to a court order. Compared with persons having a traditional release, those released early had nearly identical reincarceration rates after 1 year. Joo and colleagues (1995) conducted a similar study in Texas, where a federal court mandated early parole in response to prison overcrowding. Individuals released early were more likely to be reincarcerated after 12 months than a baseline cohort. J. Austin (1986) studied an early-release program in Illinois that sought to relieve overcrowding. The study involved a random sample of 752 individuals granted early leave and 338 individuals released prior to the implementation of the leave program (i.e., comparison group). A slightly smaller percentage of the former group was rearrested over the 12-month follow-up period.

In addition to generating inconsistent findings, all three of these forced-release studies have methodological problems that decrease our confidence in the findings. Specifically, these studies failed to adequately control for preexisting differences between those granted early release and those serving their planned sentence. Under ideal circumstances, the AICs would be randomly assigned to different groups to rule out competing explanations for the observed findings. An alternative strategy involves statistically controlling for possible differences between the treatment (i.e., early release) and comparison groups (i.e., normal release).

K. A. Wright and Rosky (2011) did the latter when they used PSM to control for possible differences between 434 low-risk individuals subject to a forced release in Montana and 434 serving their planned sentence. On average, the former group had 22 months shaved off their sentence. Recidivism was measured in two ways: (a) reincarceration for a new offense and (b) reincarceration for a new offense or technical violation. The early release and comparison groups were nearly identical on the first metric but differed significantly on the second. The authors attributed this finding to a backlash by probation officers overwhelmed by the sudden increase in their caseload.

Length of Stay Research

A third body of research that indirectly addresses the potential impact of early prison releases concerns the relationship between time served in prison (i.e., length of stay [LOS]) and recidivism following discharge. An early review of studies in this area found no convincing evidence for a relationship between LOS and recidivism (Nagin et al., 2009). Most of these studies failed to adequately control for selection bias, raising concerns about the validity of the findings. More recent studies use PSM, regression discontinuity, or natural experiments to compare reoffending rates across samples of prison releasees who served shorter or longer sentences. Most of these studies find that LOS has no consistent relationship with the likelihood of recidivating after accounting for criminal history, crime severity, and relevant demographics (Al Weswasi et al., 2022; Leymon et al., 2022; Loughran et al., 2009; Rhodes et al., 2018; Snodgrass et al., 2011; Wermink et al., 2018). Other studies suggest a more nuanced relationship between LOS and reoffending. For example, Mears and colleagues (2016) found a curvilinear relationship with sentences less than 2 years (i.e., increasing and decreasing recidivism) and no impact of LOS for sentences between 25 and 60 months. Meade and colleagues (2013) also found a curvilinear relationship in their study, with recidivism rising slightly as sentences increased from 1–12 months to 13–24 months and then decreasing incrementally with longer prison stays. Statistical testing revealed that the only significant difference in recidivism was for individuals serving more than 5 years. Finally, Rydberg and Clark (2016) found that longer periods of incarceration were associated with a higher likelihood of parole revocations and a lower likelihood of reincarceration for a new crime.

In summary, the currently available research suggests that an “early” release from prison, whether via a planned release to support reintegration, a forced release due to overcrowding, or a shorter initial sentence, probably has a negligible impact on reoffending. That said, the number of studies available to support this conclusion and the scope of this research remains limited. Policymakers considering early release as a strategy for decarcerating could benefit from additional scientific evidence. Of particular importance is the impact, if any, of small reductions in LOS across a broad range of persons. Similarly, studies are needed to assess the effect of shortened prison stays independent of mandatory participation in concurrent programming (e.g., temporary housing, vocational training, drug treatment, and counseling). Oregon’s STTL program provides an excellent opportunity to address these research needs.

Short-Term Transitional Leave in Oregon

STTL has existed in Oregon since 1989. The program was designed to improve reentry planning and reintegration in the community following an individual’s release from prison. Under the original policy, AICs were responsible for requesting leave and developing a release plan. If approved, they were released up to 30 days early and supervised by the county’s community corrections agency. Violation of release conditions could result in a range of sanctions, including a return to prison for the duration of the original sentence. Whether by design, a lack of interest, or limited awareness of the program, relatively few individuals participated under this version of STTL. Between 2011 and 2013, only 15.6 persons were released early per month. This amounted to only 4.1% of all releases.

With the passage of HB 3194 in 2013, the amount of STTL available increased from 30 days to 90 days. The state legislature also mandated a change in the administration of the policy. The Department of Corrections (DOC) is required to actively identify those eligible for the program and assist them in preparing a transition plan that addresses housing, employment, and re-establishing family connections. ¹ The receiving county’s community corrections agency must approve the release plan and coordinate implementation efforts with the DOC. The number of persons released under the revised STTL policy increased substantially, from 97.3 per month in 2014 (23.8% of all releases) to 113.7 per month in 2017 (28.6%).

Another consideration is that some are ineligible for STTL due to statutory restrictions or DOC administrative rules. Regarding the former, the policy stipulates that an individual must serve at least 6 months in prison before being eligible for STTL. Moreover, certain index convictions are excluded from consideration (e.g., Murder I or II, Manslaughter I, Assault I, Kidnapping I, Rape I, Burglary I, Robbery I, and Arson I). ² With regard to the DOC’s rules, as of 2017, individuals can only participate if they are classified as minimum custody, they had not failed a transitional leave during their current or preceding custody, they had no serious institutional misconducts in the prior 15 months, they committed no new crimes during their current incarceration, they had some form of housing available in their release county, and their sentencing judgment did not disallow an early release.

Current Study

Our first objective was to assess whether there are differences in recidivism between individuals receiving STTL and those completing their original sentence (RQ1). More specifically, we use PSM and marginal means weighting to compare recidivism outcomes for individuals (n = 4,898) granted STTL in Oregon between 2011 and 2017 versus a matched sample (n = 4,898) that did not participate in the program (non-STTL). Based on the extant literature, we hypothesize that being released early will have no appreciable connection to recidivism as measured through re-arrest, re-conviction, and re-incarceration for a new crime. We expect slightly higher rates for the STTL group regarding technical violations. This hypothesis results from prior research suggesting that community corrections officers subject early releasees to additional scrutiny and/or apply a different threshold when issuing violations (e.g., Duwe, 2015; K. A. Wright & Rosky, 2011). Our second research question (RQ2) is exploratory and concerns the relationship between the length of STTL granted and release outcomes. Individuals granted roughly 30 days of early release are compared to those granted approximately 90 days to determine whether there is a relationship between the “dose” of STTL and recidivism.

Method

Propensity Score Modeling

Although a randomized controlled trial (RCT) is considered the “gold standard” in evaluation research, it is not feasible in many criminal justice settings due to ethical, legal, or practical considerations. Accordingly, we rely on the increasingly popular technique of propensity score modeling (PSM) to reduce bias between the primary treatment (STTL) and comparison (no-STTL) groups and between the 20 to 40 and 80 to 100 day STTL subgroups. Studies in criminal justice settings find that PSM reliably replicates the results generated in RCTs (see C. M. Campbell & Labrecque, 2022). The current study used two related but different PSM applications to address the two main research questions.

Research Question 1

As might be expected, we found notable differences between the STTL and non-STTL groups before applying PSM (see Pre-PSM columns in Table 1). For example, the STTL group was disproportionately female, White, and older. Likewise, the STTL group had a lower proportion of individuals who committed a person crime as their most serious index (i.e., current) offense. They were significantly less likely to have evidenced behavioral problems during their incarceration. Finally, on average, the STTL group spent about 7 months less in prison.

OPEN IN VIEWER

Table 1:

Differences Between STTL and Non-STTL Cases Pre and Post-Statistical Matching

Measures	Pre-PSM			Post-PSM
	Non-STTL	STTL	Bias	Non-STTL	STTL	Bias
	(n = 18,574)	(n = 5,347)		(n = 4,898)	(n = 4,898)
Demographics
Sex
Male	88.3	82.3	21.6	83.3	83.0	0.8
Race/Ethnicity
White	73.5	82.5	3.4	82.0	81.6	1.1
Black	9.4	8.5	24.3	8.2	8.7	1.8
Hispanic/Latinx	12.6	5.6	17.1	6.2	6.0	0.6
Avg. age (SD)	36.2 (11.2)	37.6 (11.0)	10.4	37.4 (11.1)	37.5 (11.1)	0.2
Region
Central Eastern	11.6	9.6	6.6	10.2	10.0	0.9
Metro	61.8	60.3	3.1	60.5	60.6	0.1
Northwest Coastal	13.3	14.4	3.3	13.3	13.9	1.7
Southwest	13.3	15.7	6.8	15.9	15.6	1.0
Criminal History
2+ prior prison releases	20.8	18.3	6.3	18.3	18.4	0.2
Avg. age at first arrest (SD)	23.4 (8.5)	24.0 (9.0)	5.9	23.9 (8.9)	23.9 (8.8)	0.4
Avg. # felony convictions (SD)	5.5 (5.3)	5.8 (5.4)	3.8	5.7 (5.2)	5.8 (5.5)	1.4
Most Serious Index Offense
Person crime	25.2	20.7	10.6	22.8	21.6	3.1
Property crime	30.8	34.7	8.2	34.9	34.7	0.3
Severity on sentencing grid > 7	16.8	12.7	11.7	14.1	13.1	3.1
Institutional conduct
1+ major nonviolent infractions	44.1	21.0	50.9	28.0	22.2	13.3
1+ days spent in disciplinary seg.	40.3	18.7	48.7	24.8	20.0	11.5
Avg. years in prison (SD)	2.1 (2.4)	1.5 (1.4)	27.4	1.8 (1.7)	1.5 (1.5)	15.9
Criminogenic needs (LS/CMI)
Family
Low	23.7	29.2	12.6	27.5	28.7	2.6
Education
Very low	12.1	22.6	28.2	19.4	21.3	4.9
Recreation
Moderate	15.6	20.6	12.9	19.5	19.9	1.0
Drug rehab.
Low	30.4	37.4	14.8	36.6	36.5	0.1
Associates
Moderate	17.9	22.1	10.5	21.3	21.7	1.0
Attitudes
Low	15.1	19.5	11.6	18.8	19.5	1.7
Moderate	16.6	21.5	12.6	21.3	21.4	0.2
Antisocial patterns
Low	19.3	25.9	15.7	24.7	25.1	0.9
Model Balance Summary
Covariates with sig. diff. by group	69.2%			10.3%
Mean standardized bias	5.3			1.7
Covariates with bias ≥ 20	3.4%			0.0%
Covariates with bias ≥ 10	13.8%			0.5%
Area Under the Curve (AUC)	.85			.54

Note. This table highlights the largest differences between the treatment and comparison groups. The list is not exhaustive as there were over 117 measures consisting of 589 categories that were compared pre- and post-PSM application. A full table can be provided upon request. Reported as a percent with means and standard deviations italicized. STTL = Short-Term Transitional Leave; PSM = propensity-score matching; LS/CMI = Level of Service Case Management Inventory.

To address these differences between the STTL and non-STTL groups, we employed a one-to-one greedy match with a caliper ⁵ within each of the five temporal release cohorts. These temporal cohorts attempt to capture the changes and usage of Oregon’s STTL program over time. The propensity score was conditioned using a binary logistic regression, including 117 measures and 588 subcategories based on the individual’s demographics, criminal history, index offense, institutional behavior, and LS/CMI need domains.

The rightmost columns in Table 1 document the outcomes of this matching procedure. First, it is important to note the difference in sample size between pre and post-PSM samples. In one-to-one matching, cases on either side that do not have a match are dropped from the analysis. This resulted in a reduction of 449 cases (8.4%) from the STTL group and 13,676 cases (73.6%) from the non-STTL group, leaving us with 4,898 cases in each group. ⁶

Table 1 also provides the balance metrics used to assess the overall performance of the match. ⁷ When assessing the performance of a PSM technique to remove the selection bias between groups, it is important to ensure that multiple balance metrics are checked while recognizing that no single metric provides an omnibus performance test. We follow metrics offered and used in prior literature (P. C. Austin, 2009; C. M. Campbell & Labrecque, 2022; Hamilton et al., 2016) which include 1) significance testing, 2) standardized percent bias, and 3) the Area Under the Curve statistic (AUC). ⁸ Considering the totality of the evidence across all the metrics examined, it suggests a successful match overall. Each of the three metrics listed is discussed below in more detail.

Although statistical significance testing (a) is generally a poor way of assessing a model’s performance as it depends on sample size, it is a common practice in PSM spanning multiple disciplines. With the general aim of PSM to simulate an RCT, the focus for this metric is to reduce the percentage of covariates statistically significant to the level observed in an RCT, which is 5% or less. Although our PSM technique did not achieve this, it was remarkably close, especially considering the biased (pre-PSM) sample possessed significant differences across 69% of the 117 covariates.

One of the most important metrics in gauging the performance of a propensity score method is the standardized percent bias (b) which shows the degree to which the two groups differed on a given measure and/or category (Campbell et al., 2009; Rosenbaum & Rubin, 1985). Generally, the higher the percent bias, the greater difference exists between the groups on an observed measure. If the percentage bias is more than 20%, the groups are expected to harbor too much bias to be compared directly (i.e., without statistical adjustment). The closer to zero the percent bias is, the more similar the groups are. If PSM is successful, it should reduce the percent bias for every measure and category examined to below 20% at the very least and ideally below 10% (for more on the formulas for these metrics, see P. C. Austin, 2009, 2011; C. M. Campbell & Labrecque, 2022). We examined the standardized bias in multiple ways: The overall mean standardized bias, the percent of covariates with bias greater than or equal to 20, and the percentage of covariates with bias greater than or equal to 10. Across the standardized percent bias metrics, particularly the mean, are below 10%, suggesting a remarkable similarity between the groups following the application of PSM.

Finally, the (c) AUC metric is one used to capture the sensitivity and specificity of the scale used to predict a binary outcome. Ideally, before PSM is applied, the propensity score should strongly predict the treatment (i.e., greater than .71, see Rice & Harris, 2005). After PSM is applied, the propensity score should not be able to distinguish the treatment from comparison groups (i.e., an AUC value near .50). We achieved this as shown in Table 1, with the propensity score yielding an AUC of .845 before PSM and .536 after.

Research Question 2

We employed a PSM approach known as marginal means weighting through stratification (MMW-S; Hong, 2010, 2012) to assess the association between the “dosage” or length of STTL received and recidivism. This approach has advantages over other PSM approaches. It is designed to assess ordinal treatment categories. It devises a statistical weight to all cases, removing most, if not all, differences across the doses of how early adults in custody were released from their initial planned release date. As noted earlier, we compared outcomes for two STTL groups, 20 to 40 days and 80 to 100 days, with outcomes for the non-STTL group.

When assessing the performance of the PSM technique for ordinal treatments, there are not as many balance metric tools available as there are for binary treatments. Table 2 provides a breakdown of the available balancing metrics used to determine how well our MMW-S approach reduced the bias and balance of the dosage categories. The metrics included a one-way analysis of variance to examine the propensity score distribution across the dosage categories. Ideally, post-PSM, the F-statistic should be close to or less than 1.0 and not significant, and the percent of covariates significantly predicting the dosage category in an unweighted versus weighted ordinal logistic regression should be near or less than 5%. The AUC should be near .50. The totality of the metrics suggests a successful balance in this analysis, which allows us to draw cautious conclusions about the unbiased effects of dosage categories.

OPEN IN VIEWER

Table 2:

Balance Metrics of Pre- and Post-PSM

STTL Dosage Category	Pre-Weight			Post-Weight
	N (22,277)	Mean propensity Score	SD	N (22,277)	Mean Propensity Score	SD
Non-STTL	17,594	.810	.126	17,594	.788	.133
20 to 40 days	1,615	.741	.134	1,615	.786	.135
80 to 100 days	3,068	.678	.121	3,068	.785	.135
F-statistic		1,552.77			.49
p-value for F		<.001			.611
% of covariates significant		50.0%			8.2%
AUC		.74			.52

Note. PSM = propensity-score matching; STTL = Short-Term Transitional Leave; AUC = Area Under the Curve.

Results

To What Degree Does STTL Influence the Odds of Recidivism?

Table 3 presents the key results of the double robust assessing RQ1, binary logistic regressions predicting recidivism events using the matched groups generated with PSM. ⁹ First, we should note that outcomes for the STTL and non-STTL groups were nearly identical across three of the four recidivism metrics, with rearrest at 50.6% vs. 49.0 (χ² [1, n = 9,796] = 2.61, p = .11), reconviction at 40.7% vs. 40.0% (χ² [1, n = 9,796] = .49, p = .48), and reincarceration at 20.8% vs. 20.1% (χ² [1, n = 9,796] = .86, p = .35). By contrast, the difference in post-release supervision violations was statistically significant at 27.5% vs. 22.7% (χ² [1, n = 9,796] = 29.76, p < .001), a relative difference that is 21.1% higher (27.5%/22.7%) for the STTL group.

OPEN IN VIEWER

Table 3:

Double Robust Regressions Predicting Recidivism Events Within Three Years of Release

Measures	Violation		Rearrest		Reconviction		Reincarceration
	OR	p	OR	p	OR	p	OR	p
STTL (Non-STTL = ref)	1.38	<.001	1.05	.29	1.01	.78	1.06	.27
Release Cohort (C1 = ref)
C2. Dec 2013 to Oct 2014	1.35	.01	1.00	.97	1.04	.71	1.31	.03
C3. Nov 2014 to Oct 2015	1.28	.04	0.90	.27	0.92	.38	1.18	.19
C4. Nov 2015 to Oct 2016	1.24	.07	0.96	.67	0.82	.05	1.05	.71
C5. Nov 2016 to Dec 2017	1.14	.25	0.80	.02	0.76	.01	0.92	.49
Demographics
Sex: Male (Female = ref)	1.56	<.001	1.27	<.001	1.40	<.001	1.62	<.001
Race: (White = ref)
Black	1.24	.03	1.25	.01	1.10	.28	1.00	1.00
Hisp./Latinx	1.02	.89	0.88	.22	0.98	.84	1.15	.26
Other	1.77	<.001	1.81	<.001	1.79	<.001	1.90	<.001
Avg age at release	.90	<.001	0.91	<.001	0.92	<.001	0.89	<.001
Index offense/incarceration
Statutory crime count	1.08	.02	1.00	.96	1.00	.98	1.11	<.001
Robbery crime count	1.25	.03	1.10	.35	1.12	.25	1.04	.69
2 or more releases (1 = ref)	3.13	<.001	1.96	<.001	2.00	<.001	3.10	<.001
Major Nonviolent DR count	1.06	.01	1.09	<.001	1.07	<.001	1.07	.01
Minor DR count	1.20	.04	1.19	.04	1.25	.01	1.13	.18
Distinct visitors	0.95	<.001	0.97	.01	0.98	.03	0.97	.01
Prior prison admissions	0.76	<.001	1.03	.68	0.96	.56	0.89	.19
Criminal history
Diverse crimes in past 5yrs	1.06	.02	1.03	.17	1.03	.28	1.02	.44
Assault/disorderly in past 5yrs	1.10	.01	1.15	<.001	1.10	.01	1.05	.20
Escape/Resist in past 5yrs	1.10	.08	1.09	.10	1.14	.01	1.06	.26
Theft/property in past 5yrs	1.09	.01	1.07	.02	1.12	<.001	1.07	.04
Sex offense non-compliance	1.49	<.001	1.49	<.001	1.52	<.001	1.38	<.001
LS/CMI
History(Very low = ref)
High	1.95	<.001	2.33	<.001	1.99	<.001	2.67	<.001
Very high	2.22	<.001	2.69	<.001	2.27	<.001	3.48	<.001
Family(Low = ref)
High	1.12	.15	1.06	.38	1.09	.24	1.24	.01
Very high	1.11	.31	1.17	.10	1.13	.20	1.19	.11
Education (Very low = ref)
High	1.06	.52	1.05	.52	1.08	.31	1.08	.43
Very high	1.37	.01	1.21	.10	1.28	.03	1.22	.14
Associates(Very low = ref)
High	1.18	.12	1.17	.07	1.12	.21	1.13	.28
Very high	1.24	.03	1.25	.01	1.20	.03	1.23	.06
Drug(Low = ref)
High	1.25	<.001	1.09	.18	1.10	.14	1.24	<.001
Very high	1.21	.02	1.16	.05	1.17	.03	1.14	.13
Antisocial attitudes (Low = ref)
High	1.33	<.001	1.07	.39	1.06	.44	1.05	.60
Very high	1.37	<.001	1.15	.15	1.17	.10	1.12	.31
Antisocial patterns (Very low = ref)
High	1.05	.64	0.95	.53	0.92	.35	0.89	.27
Very high	1.05	.78	1.10	.55	0.97	.86	0.80	.19
Constant	0.24	<.001	1.65	.19	1.50	.31	0.14	<.001
N	9,796		9,796		9,796		9,796
Pseudo R2	.18		.20		.18		.18
χ2	1,939.65		2,680.36		2,362.58		1,753.22
p	<.001		<.001		<.001		<.001
df	102		102		102		102

Note. This is a partial listing of the covariates used. OR = odds ratio; STTL = Short-Term Transitional Leave; LS/CMI = Level of Service Case Management Inventory; DR = disciplinary reports.

The regressions suggest that early prison release is unlikely to impact recidivism as measured through rearrest, reconviction, and reincarceration. ¹⁰ Conversely, in the model predicting technical violations, STTL was a factor. After controlling for other measures, we find that an otherwise average case that experienced STTL still has 38% greater odds of violating the conditions of their supervision than those in the non-STTL group.

Sensitivity analyses suggest that these findings are relatively robust for technical violations and rearrest but less for reconviction and reincarceration. Specifically, an unobserved measure would need to improve the prediction of treatment assignment by 20% for violations (Γ=1.20) and 15% for rearrest (Γ=1.15) to change the result. In contrast, an unobserved measure would need to improve reconviction by just 5% (Γ=1.05) and 4% for reincarceration (Γ=1.04). This suggests that the reconviction and reincarceration models should be interpreted with a greater degree of caution.

Finally, given that we found differences in recidivism across some temporal release cohorts (see Table 3), we explored the possibility of interactions across cohorts and recidivism outcomes. No interactions were detected except for technical violations. Figure 1 provides the probability of violations for the average person on STTL and non-STTL in each temporal cohort with 95% confidence intervals. While technical violations were higher in the STTL group across all five temporal cohorts, the differences in violations over the first three time periods were not significant. Only for Cohorts 4 and 5 did the STTL group have a significantly higher probability of a technical violation.

OPEN IN VIEWER

Figure 1:

Predicted Probability of a Technical Violation Event by STTL Status and Cohort

Does the Length of Transitional Leave Granted Affect Recidivism?

Figure 2 provides the marginal probabilities for each recidivism outcome generated from the binary logistic regressions weighted by the marginal means procedure. Each bar represents the average probability of recidivism at each STTL dosage (i.e., 0 vs. 20-40 vs. 80-100). Reference lines are provided at the lower and upper confidence interval of the “No STTL” category as the baseline of each outcome. STTL “dosages” are deemed to be significantly different from this baseline if the confidence intervals no longer overlap with the reference lines.

OPEN IN VIEWER

Figure 2:

Probability of Each Outcome by STTL-Dosage Group

The only significant effect of our dosage variable was seen with technical violations. The violation graph shows that the non-STTL group (e.g., the baseline) has a 25.5% average chance of committing a technical violation while on supervision. Although the probability of a technical violation increases to 27.9% when someone is released 20 to 40 days early, it is not significantly different from the baseline. It is not until people are released up to and beyond 80 to 100 days early that there is a significant increase (33.4%). Considering the differential rates noted earlier, it is important to interpret the four measures of recidivism together, as it provides a more complete picture of post-prison supervision for someone released on STTL. The fact that the probability of violations increases, unlike that of rearrest, reconviction, and reincarceration, suggests that people are violating the terms of their supervision and are not necessarily committing new crimes.

Discussion

Oregon statutes explicitly define public safety through three of the four measures in our study—rearrest, reconviction, and reincarceration. While not an official state measure of public safety, we also included a measure of technical violations. As the findings note, the STTL program in Oregon does not appreciably increase recidivism for rearrest, reconviction, and reincarceration and thus does not appreciably reduce reoffending. While STTL does not have a higher rate of failure compared to similar non-STTL cases, half (51.2% of STTL and 51.7% of non-STTL) were rearrested, and nearly half of both groups (41.4% of STTL and 43.2% of non-STTL cases) were reconvicted within 3 years of release, indicating high recidivism rates among all adults leaving prison, regardless of whether they participated in STTL. Thus, the results indicate that STTL does not reduce public safety as measured through Oregon’s official measures of recidivism; it does not improve it either, as rearrest and reincarceration remain high. Of course, the concept of public safety is not completely captured by this study’s four outcome measures; they represent vital criminal justice measures. Further research might want to consider how other measures of justice outcome, such as quality-of-life measures or victim satisfaction, and other conceptualizations of public safety might be affected by STTL.

Conclusion and Policy Recommendations

The analysis indicates that releasing adults from incarceration sooner than their prescribed sentence has generally been successful in Oregon. Despite the state’s expansion of early release over time, effectively reducing the prison population, STTL appears to have a low overall impact on recidivism, neither increasing nor decreasing it, and no appreciable impact on rearrest, reconviction, or reincarceration. Even recognizing that technical violations do rise with longer STTL durations, the results suggest that STTL’s continued use is warranted to reduce the prison population. The results suggest several considerations and policy recommendations.

Although we observed a higher rate of technical violations for STTL participants, the differences exist only during their first year post-release. This further reinforces the notion that “supervision effects” among STTL participants may exist, and the scrutiny is more likely to occur close to release than further out. Further investigation is warranted to assess whether the higher rates of violations among STTL participants reflect actual differential behavior or an unintended result of increased scrutiny. Policymakers may want to look more closely at the reasons for violations and if appropriate sanctions are being applied. A partnership with community corrections and parole officials to lessen potentially increased scrutiny of STTL participants may, to the extent that it exists, help to improve reintegration and success.

In addition, policymakers could look to couple STTL with targeted rehabilitative and reintegration efforts. Rehabilitative and reintegration efforts are likely to be the most effective when coupled with effective tools for assessing needs (e.g., LS/CMI) and available programs that address those specific needs. Thus, one way to improve STTL is by increasing incentives for those eligible, like the accumulation of time off for good behavior or “good time.” For example, the more STTL-eligible persons participate in rehabilitation programs (when available) or vocational training, the earlier one’s release. Another way to improve outcomes is to strengthen the continuity of care between services provided in the institutional settings and those offered in the community. Assuming that releasees are failing supervision because their criminogenic needs are going unmet. The state can emphasize that JRI funds be spent on programs that target such needs rather than viewing intensive supervision as an “evidence-based practice.”

Similarly, if criminogenic needs are unmet, recidivism may remain high due to crucial differences among some specific demographic groups. For instance, differences between men and women may be due to poor access to health care and child care. Poor access to either can reduce the probability of obtaining and sustaining gainful employment. One could make the same argument related to criminogenic needs issues concerning recidivism differences among crime types between STTL and non-STTL participants. Research suggests that parolees benefit from supportive, community-based treatment options (e.g., drug treatment, housing vouchers, and job placement) that are culturally specific (Gideon & Sung, 2011). The state may want to add additional programming or age-appropriate post-prison support to take those at the highest probability of recidivism.

The combination of more early releases granted and for more extended periods has resulted in significant savings in correctional costs. To date, 8,468 Oregonians have been released on STTL with an average of 75.07 days of leave. With it costing Oregon roughly $116 a day per inmate, STTL potentially saves the state as much as $74 million. While the actual savings were lower because some fixed costs were not be reduced, the savings still substantial. As noted, a 120-day STTL release could save the state as much as $13,920 per release. The analysis suggests that an even greater expansion of STTL would increase savings even more while likely not leading to appreciable increases in crime and recidivism beyond increases in technical violations. If effective reentry programming is further integrated into STTL, the cost savings could be coupled with reduced recidivism.

Supplemental Material