You are on page 1of 12

Journal of Criminal Justice 74 (2021) 101815

Contents lists available at ScienceDirect

Journal of Criminal Justice


journal homepage: www.elsevier.com/locate/jcrimjus

The role of employment as a mediator in correctional education’s impact on


recidivism: A quasi-experimental study of multiple programs
Dror Walk a, *, Noam Haviv a, b, Badi Hasisi a, David Weisburd a, c
a
The Hebrew University of Jerusalem, Israel
b
Ashkelon Academic College, Ashkelon, Israel
c
George Mason University, Fairfax, VA 22030, USA

A R T I C L E I N F O A B S T R A C T

Keywords: There is ample evidence that prison-based educational programs reduce the rate of recidivism among formerly
Correctional education incarcerated individuals and increase their employment prospects. However, little empirical research has focused
Employment on examining the mechanism that links correctional education with these two outcomes. Moreover, the versatile
Recidivism
programs investigated to date make it difficult to distinguish between effective and less effective programs. Using
PSM
Mediator
propensity score matching, we compared post-release employment and recidivism among Israeli formerly
incarcerated individuals who participated in four different education programs. Findings demonstrated two
patterns: participants in the 12th grade and 8th grade programs showed better outcomes in both recidivism and
employment than non-participants, whereas participants in the 10th grade and 6th grade programs only showed
better employment outcomes. The mediating effect of employment on recidivism in the 12th grade and 8th grade
programs demonstrated the dramatic role of employment in the desistance process. Our study suggests that
education programs with more direct content or skills linked to employment would increase the overall impact of
education on successful reentry.

1. Introduction and social interaction upon their release (Gordon & Weldon, 2003).
There is ample evidence that prison-based educational programs
The rate of illiteracy among incarcerated individuals is higher than in reduce the rate of recidivism among formerly incarcerated individuals
the general population, and fewer of them have a high-school diploma or and increase their employment prospects (Brazzell et al., 2009; Lahm,
academic degree (Creese, 2015; Vacca, 2004; Wolf Harlow, 2003). It 2009; Vacca, 2004). However, a considerable portion of these studies
therefore comes as no surprise that correctional education programs suffer from a weak methodology that prevents them from attributing the
have been offered by correctional facilities since their inception. For positive effects reported to the program activities described (Wilson
instance, in the United States, 84% of prisons offer some form of et al., 2000). A recent meta-analysis carried out by Bozick et al. (2018)
educational program to their inmates, and 98% of federal correctional tried to clarify the overall picture by using studies that employed quasi-
facilities offer almost every kind of educational program available experimental designs where the treatment and comparison groups were
(Brazzell et al., 2009). In federal facilities, 27% of incarcerated in­ similar on baseline characteristics. They found that the rate of recidi­
dividuals participate in educational programs (Travis et al., 2014), and a vism among participants in correctional education programs was 32%
similar participation rate was found in surveys conducted by the Bureau lower on average than that of comparison groups, and employment was
of Justice Statistics (Turner, 2018). In time, the variety of contents increased due to participation in corrective education programs by 12%
offered by educational programs spanned from basic education and lit­ on average. Overall, the positive effect is true for all forms of education,
eracy through elementary and high-school level programs to vocational including Adult Basic Education courses, GED/high school courses,
training, life skill acquisition, and even higher education programs vocational college courses, and college courses.
(Cecil et al., 2000). They provide incarcerated individuals with disci­ The rationale behind correctional education programs in prisons is
plinary contents and normative values that aim to encourage partici­ twofold. First, participants acquire normative values and ways of
pants to abandon crime, and engage in legitimate forms of occupation thinking to help them cope with the moral dilemmas that had previously

* Corresponding author.
E-mail address: dror.walk@mail.huji.ac.il (D. Walk).

https://doi.org/10.1016/j.jcrimjus.2021.101815
Received 25 November 2020; Received in revised form 12 April 2021; Accepted 12 April 2021
Available online 4 May 2021
0047-2352/© 2021 Elsevier Ltd. All rights reserved.
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

led them to crime (Andrews & Bonta, 2003; MacKenzie, 2008). Second, group. Another is to directly measure motivation and compare partici­
incarcerated individuals who participate in educational programs and pants and non-participants with similar degrees of motivation (e.g.,
vocational training acquire practical tools that may help to improve Steurer et al., 2001).
their chances of finding honorable, profitable employment (Lockwood In a study conducted among 3170 inmates in Maryland, Minnesota,
et al., 2012). Yet very little empirical research has focused on examining and Ohio, US, 43.3% of respondents participated in correctional edu­
the mechanism that links correctional education with post-release cation. Respondents’ outcome measures three years after their release
outcome measures in order to discern between these two rationales were better than non-participants’: 16% fewer arrests, 23% fewer con­
(MacKenzie, 2006). victions, and 31% fewer reincarcerations (Steurer et al., 2001). The
Similarly, very few studies have examined both outcome measures – authors noted that their study structure did not control for self-selection,
recidivism and employment. Perhaps some education programs affect and therefore the incarcerated individuals’ characteristics could also
both outcome measures, while others only affect one. The interplay explain some of the differences found between the two groups. The
between the two outcome measures has yet to become clear as well: do authors tried to overcome this issue by handing out motivation ques­
education programs affect both recidivism and employment directly, or tionnaires to all study participants, showing that no differences in
do they only affect employment, which, in turn, affects recidivism? motivation were found between the correctional education participants
Moreover, no study has examined this interplay by comparing several and control group. A study conducted in Canada found that, of 654
education programs. There may, for instance, be two education pro­ formerly incarcerated individuals who took part in an academic study
grams that affect both outcome measures, but do so in different ways: program, 25% were reincarcerated within 3 years of their release,
while one program affects both recidivism and employment indepen­ compared to some 50% among the general formerly incarcerated indi­
dently, the other affects the former vicariously through the latter. vidual population in Canada (Duguid & Pawson, 1998). The authors
The current study aims to overcome these limitations, while attempted to solve the issue of self-selection by using a recidivism
attempting to encompass the aspects found scattered in the literature. To assessment tool. According to this instrument, the likelihood of rein­
accomplish this purpose, we analyze the impact of four correctional carceration among the program participants’ group was 42%, indicating
education programs offered by the Israeli prison system on recidivism that, while this group was indeed less prone to returning to crime than
and employment. To understand the program-employment-recidivism the general population of formerly incarcerated individuals, correc­
interplay, we will employ mediating effect analysis, thereby tional education had a beneficial effect that further reduced their risk of
improving the understanding of the mechanisms explaining desistance recidivism.
from crime. A better understanding of this interplay may enrich It seems that most evidence points to correctional education’s posi­
knowledge on the mechanisms of desistance, and, in practical terms, tive effect; however, many studies rely on a structure that does not
assist correction services to better design and prioritize their various optimally address prior differences between the program participants’
education programs. group and other formerly incarcerated individuals, failing to adequately
match control groups to participant groups. Moreover, the growing
2. The evidence on correctional education effectiveness number of studies conducted over the last few decades makes it harder
to obtain a clear overview of the literature. It is therefore becoming
There is ample evidence that prison-based educational programs increasingly more acceptable to base an understanding of the big picture
reduce the rate of recidivism and improve employment status among on meta analyses that summarize multiple studies.
formerly incarcerated individuals (Adams et al., 1994; Anderson & Wilson et al. (2000) reviewed 33 studies with 53 comparisons, and
Moore, 1995; Brazzell et al., 2009; Duguid, 1982; Gerber & Fritsch, reported a general 22% decline in recidivism (measured in two-thirds of
1995; Lahm, 2009; Vaaca, 2004; Wilson et al., 2000). This impressive the studies by reincarceration), as well as a certain improvement in
impact is not trivial, since the conditions of incarceration provide a employment. When the data were analyzed by program type, findings
learning environment that is far from ideal. Prisons are filled with every indicated an 18% drop in recidivism due to GED (general educational
kind of individual, whose varying needs and levels of education do not development) and ABE (adult basic education) programs (grouped
render them equally motivated to participate in learning programs. together for reasons of size), a 26% decline due to PSE programs, and a
Moreover, prison classrooms tend to be small and crowded. Incarcerated 22% decrease due to vocational training. It should be noted that these
individuals’ movement, too, is limited by means of security and sur­ meta-analyses contained studies with varying degrees of methodological
veillance, leaving very little time for actual learning in the classroom. rigor, including some where control groups were not matched to
Finally, prison regulations prohibit individuals’ use of the internet, participant groups. Under such circumstances, one cannot definitively
limiting their access to online databases and libraries (MacKenzie, attribute the decline in recidivism to the programs’ effect, for it may be
2008). explained, at least in part, by the different characteristics of the incar­
A major concern when evaluating correctional program using non­ cerated individuals who participated in the programs, e.g., motivation
randomized designs is selection (or self-selection) bias. Participants’ (Cecil et al., 2000). Wilson et al. (2000) put this simply: “These studies
pre-program advantages over non-participants may inflate outcome rule in the plausibility that these programs have a casual impact but fail
measure estimates, thereby overstating the treatment effect. For to rule out plausible alternative explanations”. MacKenzie later updated
example, if those who participate in a program are more motivated to this meta-analysis, adding new studies, and restricting her overview to
make positive changes in their lives than those who do not participate in studies published from 1980 onwards. She ultimately reviewed 22
such programs, they may also be less likely to re-engage in criminal studies, and reached a similar conclusion to that made by the original
activity, and therefore the measured effect cannot be attributed solely to meta-analysis, i.e., that on average, correctional education reduces
the program but also to participants’ merits (Kim & Clark, 2013). Se­ recidivism (MacKenzie, 2006).
lection bias may also shrink the outcome measure estimate whenever Steve Aos and his team from the Washington State Institute for Public
incarcerated individuals choose to participate in a program as means of Policy (WSIPP) published a meta-analysis of various incarcerated in­
retaining privileges instead of improving their skills, and are therefore dividuals’ rehabilitation programs, among them 21 educational ones.
more likely to recidivate due to their other characteristics (Brewster & They ensured that all studies included were either experimental or
Sharp, 2002). There are several ways of overcoming selection bias in quasi-experimental with matched control groups. The latter contained
observational data studies. The most popular one is propensity score respondents who either did not participate in educational programs or
matching (PSM), which will be presented in the method section of this received treatment as usual. The authors adjusted the effect size to the
paper. This technique helps to select individuals for the comparison methodological quality of the study, thus reducing it by certain per­
group who share similar characteristics with members of the treatment centages for studies that were not of the highest level: quasi-

2
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

experimental studies; studies that examined demonstration programs; or stable, the vast majority of studies have examined specific types of ed­
those that were self-evaluated by program providers. The effects found ucation programs, thereby limiting their generalizability. Only a small
were far more modest than those of previous meta-analyses, but number of them had looked into several programs offered by the same
remained significant nonetheless: an average reduction of 7% in recid­ correctional system. Such a study has a methodological advantage over
ivism (conviction for new offense) among participants of education meta-analyses describing the effect of studies conducted in different
programs, and a 9% drop among those who underwent vocational prison systems, since the latter consist of differing operating practices,
training (Aos et al., 2006). thereby producing a biased estimate of the effect. The general conclu­
Bozick et al. (2018) analyzed correctional education evaluations sion in recent years that “educational programs work”, or that of the
published between 1980 and 2017. Their analysis was also restricted to past, that “educational programs do not work”, may lead policymakers
studies with both a participant and comparison group. It reviewed 57 to the wrong decisions. For instance, the general positive effect may, in
studies (containing 81 comparisons) whose outcome measure was fact, reflect the remarkable effectiveness of certain programs, and the
recidivism, and 22 studies (containing 26 comparisons) with employ­ complete ineffectiveness of others. Similarly, the absence of a general
ment as their outcome measure. Upon selecting the participant group, positive effect does not rule out some programs’ possible effectiveness.
the authors applied the stricter intent-to-treat approach, regarding Batiuk et al. (2005) demonstrate such a scenario. Their study shows that,
incarcerated individuals as program participants if they signed up for although no overall positive effect on reducing recidivism was found for
the program, whether they completed it, or even attended it, or not. This the four programs it examined, the college program had significantly,
approach is aimed at policymakers looking to examine the cost- and rather remarkably, reduced recidivism by more than 60%; an effect
effectiveness of allocating resources to such programs on an organiza­ overshadowed by the absence of other programs’ effect, namely basic
tional level, and is not restricted to identifying the effectiveness of the education, secondary education, and vocational training.
treatment among graduates of the program alone. The Maryland Sci­ Very few studies have examined both outcome measures – recidivism
entific Methods Scale (SMS) (Farrington et al., 2002) was used to and employment – in the same correction system. Recidivism is regarded
overcome selection bias, and level 1 studies were eliminated. The as the dominant measure when assessing successful rehabilitation, and
findings of this study indicated that the rate of recidivism among par­ most rehabilitative efforts are directed at the minimization thereof. Its
ticipants in correctional education programs was 32% lower on average dramatic impact not only affects the individual’s life, but the safety and
than that of comparison groups. The effect was similar in size on all wellbeing of society as a whole. Nevertheless, this measure greatly limits
quality levels of studies (levels 2 to 5). When effect was analyzed by the perception of rehabilitation. Does the fact that a formerly incarcer­
program type, findings indicated that all programs contributed to a ated individual is no longer involved in criminal activity (or, to be more
significant decline in recidivism, with PSE programs being the most accurate, is no longer caught engaging in such activity) necessarily
effective and reducing recidivism by 49%. As for employment as an indicate that he or she has successfully reentered society, and has been
outcome measure, the authors concluded that, overall, employment was satisfactorily rehabilitated (Lipton et al., 1975)? Alternatively, if a
significantly increased due to participation in corrective education formerly incarcerated individual manages to keep a day job for years
programs, although the effect size was more modest – an average of post-release, but is periodically arrested on suspicion of misdemeanor
12%. The authors noted that the level of confidence associated with the (possibly due to excessive law enforcement among formerly incarcer­
findings on employment as an outcome measure was lower, since the ated individuals), has he necessarily failed to be rehabilitated? It seems
methodological quality of many of the studies reviewed was relatively that a study that examines both employment and recidivism simulta­
low. When effect was analyzed by program type, vocational training and neously is necessary to unveil the mechanism of desistance, since “the
academic programs were significant contributors to post-release mechanism that leads employment to declines in recidivism has become
employment, with the former being more effective in achieving this increasingly questioned” (Bushway & Apel, 2012, p. 38).
outcome. A rare example of a study that combined examining various educa­
In order to limit the “file drawer problem” effect – the potential bias tional programs with different measures is that of Duwe and Clark
of study findings in the absence of studies with non-significant findings – (2014). The authors investigated recidivism and employment among
Bozick et al. (2018) searched an array of sources in literature to obtain more than ten thousand formerly incarcerated individuals who partici­
official program evaluation reports not published in journals, working pated in GED and PSE programs in Minnesota. A variety of recidivism
papers, research briefs, theses, and dissertations. They revealed that, measures were used in this study, as well as a range of employment
even when other potential studies with a small effect size or no effect at measures. The study findings indicated heterogenous effects for the
all are added, the positive effect of correctional education remains sig­ programs: the PSE program reduced three of the four recidivism mea­
nificant both with regard to the decline in recidivism and the increase in sures; the most remarkable of them being a 24% drop in reincarceration.
employment. An analysis of the possibility that studies with larger The participants’ employment rate, however, was no higher than the
numbers of respondents have greater impact on the general effect size comparison group’s, although their total work hours and total wages
was examined using a leave-one-out method, and indicated that the were higher. By contrast, the GED program did not reduce the recidivism
general finding is stable with respect to the positive effect of correctional measures, but did improve the participation group’s employment rate
education on the outcome measures of formerly incarcerated individuals compared with the comparison group. Though this study tested various
(Bozick et al., 2018). programs against multiple outcome measures, it did not examine the
Interestingly, the economic analysis conducted by the WSIPP also relationships between them.
found that the cost-benefit ratio on PSE programs was exceptionally Only a study that combines an examination of both recidivism and
high, in addition to such programs’ superiority, as indicated by the meta employment as outcome measures while attempting to clarify the
analyses mentioned above: for every single USD spent on a PSE program, interplay between them can allow for a true exploration into the
there was a 19.74 USD return, making it the most cost-effective program mechanism of rehabilitation and crime desistance. Judd and Kenny
of the dozens offered by way of rehabilitation in the adult criminal (1981) stated this idea clearly: “When a treatment is judged effective…
justice system examined by this organization (Washington State Insti­ it is usually informative to examine the mediating process that produces
tute for Public Policy, 2017). those effects. Such a process analysis is an attempt to specify the causal
chain responsible for the observed treatment effects. With a process
3. Limitations in studies on education programs’ effectiveness analysis one asks not whether a treatment produced the desired effects,
but rather how it did so” (p. 603). Accordingly, we should ask: Was the
Although the general findings on correctional education programs’ effect of correctional education on recidivism direct, i.e. do participants
effectiveness in reducing recidivism and increasing employment seem desist from crime due to the pro-social values acquired while partaking

3
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

in the program? Or, alternatively, is the effect of correctional education facilities, as well as treatment programs for versatile offense types, such
on crime desistance mediated by improved employment in the sense that as drug abuse, intimate violence, and sex offenses. Moreover, studies
program participants improve their chances of finding paid work, sub­ indicate that both American and Israeli correctional officers support
sequently minimizing their exposure to other individuals engaging in education and treatment programs, both as a rehabilitative tool and an
crime, and are less tempted to break the law? Batiuk et al. (1997) sought organizing principle for prison management (Gideon et al., 2015;
the answers to these questions in the heat of the Pell grant crisis that, Phelps, 2011). However, there are differences in the organizational
during the 1990s, denied incarcerated individuals the possibility of cultures of the two systems. The American model of running a correc­
participating in prison-based post-secondary education programs. The tional facility is based on a disciplinary approach that combines clear cut
rates of recidivism and employment among nearly 100 inmates in an rules coupled with both spatial and psychological distance between staff
Ohio State facility who completed an academic prison-based program, and incarcerated individuals. This practice may stem from a national
and received an associate’s college degree, were compared with incar­ social deviance approach, but also has roots in a tough reality in which
cerated individuals from the same prison who had a secondary school gangs are attempting to gain control of prison management (Skarbek,
level education. Using a series of logistical regression analyses, the au­ 2014). The Israeli model, by contrast, is based on mutual respect be­
thors concluded that the minimizing effect of the college program on tween staff and incarcerated individuals, where disobedience is
recidivism was mediated by employment, rather than being direct. In discouraged by loss of privileges (Dervan, 2011). This practice may have
other words, the educational program improved participants’ chances of its roots in Jewish tradition, which is very dominant in prison-based
finding paid work following their release, which subsequently led to correctional activities, whereby correction is a profound principle, and
reduced recidivism. incarcerated persons are perceived as individuals worthy of human
A study examining the interplay of different outcome measures dignity even before they undergo correction (Gideon et al., 2015).
among various education programs is needed in this area of incarcerated As for the reentry process in Israel, several months prior to the
individuals’ rehabilitation. As we illustrate below, it can unveil the completion of two thirds of their sentence, every incarcerated in­
possible heterogenous effects of these programs, help to better under­ dividual’s case is brought before the parole committee (Assy & Menashe,
stand the processes and mechanisms behind rehabilitation, and provides 2014). Following the latter’s discussion, approximately 25% of appli­
a more complete version of the formerly incarcerated individuals’ story. cants are granted an early release with a detailed rehabilitation plan
comprised of full-time employment and group and individual weekly
4. Method therapy sessions. At the time the study was conducted, the three-years
recidivism rate in Israel was 34.1% (Walk & Berman, 2015).
4.1. Israel Prison Service
4.2. Databases
The Israel Prison Service (IPS) governs the imprisonment of all pre-
and post-trial incarcerated individuals in Israel. As data for the present The information contained in this paper was obtained from the IPS
study was being collected, some 12,000 Israeli inhabitants1 were operative database. This system includes information on incarcerated
incarcerated in IPS correctional facilities, 70% of whom were post-trial, individuals’ socio-demographics, criminal history, and the kind of
and 30% detained. The IPS views itself as a security organization with a offense that led to their incarceration. It also contains information on
social agenda, therefore, during their incarceration, incarcerated in­ incarcerated individuals’ conduct while incarcerated, including visits,
dividuals may partake in therapeutic, occupational, and educational furloughs, disciplinary hearings, and participation in employment and
activities. The forms of therapy offered include drug and alcohol reha­ educational programs (Hasisi et al., 2016; Haviv et al., 2019; Haviv &
bilitation programs, and designated programs for individuals who were Hasisi, 2019; Weisburd et al., 2017). The formerly incarcerated in­
incarcerated for committing a sex offense, domestic violence offense, or dividuals’ employment and income measures were recovered from the
white-collar offense. With regard to employment, there are IPS-operated national tax authority’s database. The combination of IPS and tax au­
factories, as well as private ones operating within prison walls. Voca­ thority data was retrieved from the Israeli Central Bureau of Statistics. A
tional training programs in various areas are offered too. study based on Israeli data has a qualitative advantage as Israeli data­
As for education, there is a wide range of education programs on bases are central, containing all official information on all Israeli citi­
offer that are compatible with individuals’ various levels of education, zens. It can therefore avoid the inherent weakness of US-based studies
as well as religious study programs. Many of the incarcerated individuals that rely on state-specific databases, and do not cover the criminal or
had dropped out of school and did not complete their high-school edu­ occupational activity of formerly incarcerated individuals in other US
cation. There is also an Arab minority in Israel comprising some 20% of states (Duwe & Clark, 2014).
the population, the representation of which in the incarcerated popu­ The study was conducted on data on all post-trial male incarcerated
lation is approximately 45%. These individuals often lack the Hebrew individuals who are Israeli inhabitants released from an IPS prison be­
language skills crucial for their professional and occupational integra­ tween 2004 and 2012 – a period of nine years. After eliminating all
tion into society. Thus, prison-based educational programs provide a individuals who did not report how many years of education they had
golden opportunity to complete one’s education, and acquire Hebrew upon incarceration, or were sentenced to less than three months (and
language skills, thereby improving one’s chances of finding a respect­ subsequently were not eligible to participate in correctional education
able, income-generating job post-release. programs), the file used in the present study consisted of 37,231 incar­
Below we describe the extent to which prison education programs in cerated individuals, of whom 8444 (23%) had participated in correc­
Israel are similar to those offered in the United States. At the same time, tional education programs between 2004 and 2012.
it is important at the outset to define similarities and differences be­
tween the American2 and Israeli correctional systems. Both systems offer 4.3. Education programs
relatively developed educational and vocational programs within their
The Israel Prison Service views education as a basis for integration
into employment and treatment activities in prison. The process of
1
An additional 5000 individuals were incarcerated for involvement in terror identifying prospective participants in correctional education therefore
activities, the majority of which are inhabitants of the Palestinian Authority. begins upon arrest. Detainees complete a test to ascertain their level of
2
It appears to be somewhat superficial to present a unified correctional education, the results of which are used to determine which educational
American model since the differences between regional and state variations are program they should attend. Since close to half of incarcerated in­
increasingly widening (Phelps, 2012). dividuals in Israel are members of the Arab minority, and do no list

4
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

Hebrew as their mother tongue,3 many lack education on various levels. Individuals’ basic socio-demographic characteristics: age and marital
The IPS offers the following five formal education programs based on the status (married vs. non-married), and two measures specific to the Is­
curricula and examinations formulated by the Israeli Ministry of Edu­ raeli context – whether the individual was Jewish (i.e., a member of the
cation’s Division of Adult Education: (1) a class for illiterate individuals majority ethnic group comprised of 80% of the general population and
acquiring basic reading skills; (2) a class for 6th grade level education 55% of the incarcerated population), and whether he was an immigrant
teaching basic Hebrew and mathematics; (3) a class for 8th grade level (vs. native). (2) A series of variables measuring the inmates’ criminal
education, where participants must pass the Israeli Ministry of Educa­ backgrounds and the characteristics of their current offense: number of
tion’s tests for 8th grade (equivalent to ABE); (4) a class for 10th grade prior incarcerations, length of current incarceration sentence, year of
level education teaching mathematics, reading comprehension, com­ release, and type of present offense individual was imprisoned for –
puter skills, and geography, where participants must pass several exams violent, sexual, drug or property. (3) two indicators specific to the IPS:
provided by the Israeli Ministry of Education; (5) a class for 12th grade whether or not the offense of which individuals had been convicted fit
level education teaching reading comprehension, creative writing, civil the court definition for violent offense, and whether or not they had
studies, computer skills, and geography, where participants must pass been evaluated by the IPS as having committed domestic violence.
internal and external examinations formulated by the Israeli Ministry of Overall, 13 variables were used to determine the probability of incar­
Education (equivalent to GED as well as some exit exams). Incarcerated cerated individuals’ participation in each education program.
individuals can also take post-secondary and academic studies, howev­ To match comparison members to participants, a single match
er, the number of participants in such programs is limited, and therefore approach was used with a caliper of 0.01, and without replacement.
these programs were not evaluated in our study. Table 1 presents the PSM details of the various education programs
examined in this study.
4.4. Analytical plan The models’ assumption whereby both treatment and comparison
groups shared the same probability of participating in the education
To examine the possible interplay of different outcome measures programs is clearly supported by the almost-identical means of pro­
among various education programs, we employed a two-stage analysis. pensity scores found for these groups. Moreover, following matching, no
During the first, we explored the main effects of the education programs significant differences remained between participant and comparison
on post-release employment and recidivism. In order to minimize se­ groups in the various variables. The bias measure, developed by Rose­
lection bias, propensity score matching was used. During the second nbaum and Rubin (1985), has also dramatically decreased following
stage, we employed a mediating analysis in order to understand the matching: none of the predictors exceeded 20, the value proposed by the
possible role of post-release employment in reducing recidivism. In the developers to signify an evident unbalance (see also Duwe & Clark,
following sections we will elaborate on these two procedures. 2014). As shown, the PSM procedure produced groups that are balanced
on known included characteristics.
4.4.1. Propensity score matching Indeed, unlike random assignment to treatment and comparison
Propensity score matching (PSM) was used to match a comparison groups, adjustment for the propensity score does little to balance un­
group of individuals who did not attend correctional education to the observed covariates such as relevant unreported background charac­
participants group (Austin, 2011; Jordan, 2012). Propensity scores for teristics and motivation to desist and find reputable employment;
each education program were calculated for both participants and however, this matching method was vastly superior to other matching
comparison group members. Both groups were considered eligible for methods (Joffe & Rosenbaum, 1999). In turn, because PSM capitalizes
the specific program since they reported having completed the preced­ on prediction of the propensity and not on controls for specific variables,
ing education level upon admittance to prison (e.g. 10 or 11 years of unobserved covariates strongly related to included variables will in­
schooling for the 12th grade program). For the participants group in the crease the validity of the propensity score outcome (Weisburd et al., in
beginner class, designed for illiterate individuals, no comparison group press). To assess how robust the study results are to potential unob­
could be established due to the absence of information on low education served confounding, we used one of the most common sensitivity ana­
levels. This program was therefore eliminated from the present study, lyses, the Rosenbaum method.
which ultimately contained four groups – 12th grade, 10th grade, 8th
grade, and 6th grade. Our approach to comparison groups is the intent- 4.4.2. Mediating effect
to-treatment estimate, which is the accepted approach for simulating The effectiveness of correctional education programs was examined
experimental design effect. However, in practice, this approach could using two outcome measures: recidivism and employment. Recidivism
inadvertently mask the treatment-on-the-treated estimate in correc­ was measured using two parameters: reincarceration and rearrest within
tional programs due to high drop-out rates. Cho and Tyler (2010) had three years of release from prison. Employment was measured by
pointed this issue out when they found an employment effect for ABE number of months of employment, as well as a dichotomous measure
programs only when their comparison group was comprised of high (worked/did not work) – both within the first tax year following release
school dropout inmates who participated in the said programs but from prison.
involuntarily dropped out within 21 days (because of reallocation or In an effort to understand the mechanism of desistance from crime,
release). Fortunately, this interesting point may be overlooked in the the interaction between recidivism and employment was examined. We
present study since the dropout rates in the programs examined were explored whether the effect of correctional education was direct on both
below 10%. recidivism and employment, or whether employment mediated the ef­
A considerable number of variables was used in the present study’s fect of correctional education on recidivism. Baron and Kenny (1986, p.
propensity score models, some of which have been found to strongly 1177) set the basic practice for identifying a mediating effect using three
predict recidivism outcomes in previous correctional treatment evalu­ regression models: (1) regressing the mediator (employment in our
ation models (e.g., Kim & Clark, 2013; Langan & Levin, 2002), while study) on the independent variable (education program in our study);
others seemed particularly important in the Israeli context (Weisburd (2) regressing the dependent variable (recidivism in our study) on the
et al., 2017). The variables may be divided into three main groups: (1) independent variable; and (3) regressing the dependent variable on both
the independent variable and on the mediator. If the effects of the first
two regressions are significant, and in the third, the effect of the program
3
Most occupational and business activity in Israel is conducted in Hebrew, will decrease while the effect of employment will increase, then
and therefore the education required in order to integrate into society consists employment would be defined as a mediating factor. But, whereas in the
first and foremost of knowledge in Hebrew. context of linear regression models, the comparison of estimated

5
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

Table 1
Matching process for various education programsa
Education No. of participants (mean No. of non-participants No. of matched participants No. of matched non- No. of predictors found different
program propensity score) (mean propensity score) (mean propensity score) participants (mean propensity before and (following) matching
score)

12th grade 526 (0.13) 9302 (0.06) 349 (0.12) 349 (0.12) 4 (0) / 13
10th grade 435 (0.14) 6965 (0.07) 244 (0.13) 244 (0.13) 5 (0) / 13
8th grade 2580 (0.69) 1729 (0.59) 677 (0.62) 677 (0.63) 4 (0) / 13
6th grade 2537 (0.84) 714 (0.72) 270 (0.74) 270 (0.75) 3 (0) / 13
a
The appendix lists the detailed matching process for the four education programs. It shows that, for all programs, the matching procedure achieved a balance
between the treatment and comparison groups in all variables.

coefficients is straightforward, comparing the effects of nested nonlinear the observed confounders have been dealt with using matching
probability models, such as logit, is not. In these latter models, co­ methods. Nevertheless, other methods yield similar results (Liu et al.,
efficients may differ not only because of confounding, but due to the 2013). This approach enables one to determine how strongly an un­
rescaling of the model too. Several solutions for this problem have been measured variable must influence the selection process to undermine
proposed, and we employed the KHB method (in Stata), which, in fact, the treatment effect (Becker & Caliendo, 2007). The Gamma statistic
extends the decomposability properties of linear models to nonlinear denotes the odds of differential assignment due to unobserved variables,
probability models (Kohler et al., 2011). and the further its value is from 0, the more the model is insensitive, or
robust against bias. If, for example, the significant level of the bounds
5. Results falls between the Gamma values of 1.6 and 1.7, then bias may only arise
in the presence of an unobserved variable that increases the odds of
5.1. Observed outcome measures for the four education programs participating in the program by 60% to 70%. Where both treatment and
response variables are binary, the Mantel-Haenszel test is used to
Table 2 presents the outcome measures for the four education compare the odds ratios (Aakvik, 2001; Mantel & Haenszel, 1959).
programs. Table 3 presents the bounds for the outcome measures using a 0.10
In the 12th grade program participants showed significantly better marginal significance threshold across the various education programs.
outcomes than non-participants in both recidivism and employment. Overall, our results remain significant, with gamma values between
The odds ratios for reincarceration and rearrest among non-participants 1.5 and 1.8 (in the various outcome measures) in the 12th grade pro­
are more than double those of participants, and the same is true for the gram; between 1.1 and 1.2 in the 10th grade program; between 1.2 and
odds ratio for employment during the first year following release among 1.8 in the 8th grade program, and between 1.2 and 1.4 in the 6th grade
participants. Average months of employment was significantly higher program. Despite the lack of clearly defined thresholds for a model to
among participants compared to non-participants, with a medium effect meet a strong ignorability assumption, the results of the sensitivity test
size. In the 10th grade program participants showed significantly better conducted in this study are in line with other studies in the correction
outcomes than non-participants in their measures of employment but field (Chen et al., 2020; Haviv & Hasisi, 2019; Weisburd et al., 2017).
not of recidivism. The odds ratio for employment during the first year However, the results also suggest that our models, like those of other
following release was about 1.5 times higher for participants compared quasi-experimental studies, should be interpreted with some caution,
to non-participants, and the average months of employment was since they cannot account for unobserved covariates not included in the
significantly higher for them too, with a small effect size. In the 8th models.
grade program participants showed significantly better outcomes than To sum up: the four educational correction programs we analyzed
non-participants in both recidivism and employment. The odds ratio for demonstrated two patterns of effect – participants in the 12th grade and
reincarceration and rearrest were about 30% higher for non-participants 8th grade programs showed significantly better outcomes than non-
compared to participants, and the odds ratio for employment during the participants in both recidivism and employment, whereas participants
first year following release among the latter was more than double that in the 10th grade and 6th grade programs only showed significantly
of the former. The average months of employment was also significantly better employment outcomes. The next step was to determine whether,
higher for participants compared to non-participants, with a medium for the effective programs, the program and employment function
effect size. In the 6th grade program participants showed significantly independently to decrease recidivism, or whether perhaps employment
better outcomes than non-participants in measures of employment but mediated the program’s effect.
not of recidivism. The odds ratio for employment during the first year
following release was almost double among participants compared to 5.3. The mediating effect of employment
non-participants, and the average months of employment was signifi­
cantly higher among them too, with a small effect size. To determine the mediating effect of employment on recidivism, we
employed the KHB method for binary response models. Since both
5.2. Sensitivity analysis recidivism measures (reincarceration and rearrest) and both employ­
ment measures (any employment during the first year, and number of
Since one cannot estimate the magnitude of selection bias with employment months during the first year following release) yielded very
nonexperimental data, PSM, like other observational data analysis similar effects (see Table 2), we employed reincarceration as the
methods, is constrained by the degree to which variables that differen­ dependent variable and any employment during the first year following
tiate selection factors can be identified. In general, a large number of release as a mediator. Program participation served as the independent
relevant covariates may support the assumption of equivalence between variable.
the groups (Shadish, 2013). However, if there are unobserved variables Table 4 presents the parameters for the KHB method results for all
that simultaneously affect assignment into treatment and the outcome four education programs.
measures, a hidden bias may arise to which matching estimators are not The mediating effects were found to be statistically significant for
robust. To estimate the degree to which any significant results hinge on three of the four education programs. However, since mediation is
the ignorability assumption (i.e., sensitivity), Rosenbaum (2002) pro­ meaningless unless the total effect is significant, only the mediating
posed the bounding approach, which is the most frequently used when effects for the 12th and 8th grade programs were established.

6
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

Table 3
Mantel and Haenszel bounds for outcomes measures of the four education

2.55** (0.22)
(OR) / t-test
programs

(Cohen’s d)

0.40 (0.89)

0.02 (10.3)
Chi-square

12.95***
12th grade 10th grade 8th grade 6th grade

(1.89)
program program program program

3 yrs. reincarceration 1.6–1.7 n.s. 1.2–1.3 n.s.


3 yrs. rearrest 1.5–1.6 n.s. 1.2–1.3 n.s.
participants

3.15 (4.58)
Any employment during 1.8–1.9 1.1–1.2 1.8–1.9 1.4–1.5
the first year following

34.6%

48.3%

41.0%
Non-

release
6th grade program

No. of employment 1.6–1.7 1.2–1.3 1.7–1.8 1.2–1.3


months1 during the first
Participants

4.20 (4.79)
year following release
1
For the sake of simplicity, we split the number of employment months into
31.9%

49.0%

56.8%
two categories according to the median value of each education program.
7.35*** (0.40)

Table 4
(OR) / t-test
(Cohen’s d)
Chi-square

KHB method results determining mediating effect of employment for the four
10.47***

10.33***

43.85***
(0.67)

(0.69)

(2.11)

education programs
Total effect Direct effect Mediating effect

12th grade program 0.65 (0.19)** 0.48 (0.19)** 0.17 (0.05)**


participants

2.82 (4.19)

10th grade program 0.33 (0.21) 0.24 (0.21) 0.09 (0.05)


8th grade program 0.46 (0.12)*** 0.31 (0.12)* 0.15 (0.03)***
36.6%

52.7%

42.7%
Non-

6th grade program 0.17 (0.19) − 0.03 (0.20) 0.20 (0.06)**


8th grade program

*p < .05 ** p < .01 *** p < .001.


Participants

4.67 (4.84)

6. Discussion
27.8%

43.4%

61.1%

The current study aimed to overcome the shortcomings found in


literature on correctional education effectiveness. With very few studies
4.31* (1.47)

2.29* (0.21)

investigating several educational programs across various outcome


(OR) / t-test
(Cohen’s d)

0.76 (0.81)

0.62 (1.19)
Chi-square

measures relevant to reentry, the clear positive effect found may inad­
vertently be masking heterogenous effects, and inhibiting a look into the
mechanism of desistance. Indeed, the present study findings indicate
two patterns for different education programs: while the 12th and 8th
Recidivism and employment outcomes for participants and non-participants in four education programs

participants

2.93 (4.38)

grade programs demonstrated impressive effects on reducing recidivism


and increasing employment, the 10th and 6th grade programs had
33.3%

50.6%

41.7%
Non-
10th grade program

modest effects in increasing employment, and were found not to be


effective in reducing recidivism. The reduced recidivism effects of the
12th and 8th grade programs were found to be mediated by employ­
Participants

3.90 (4.71)

ment. Does this complicated picture enrich our understanding of the


28.9%

54.9%

51.3%

mechanism explaining formerly incarcerated individuals’ desistance


from crime?
5.27*** (0.45)
(OR) / t-test
(Cohen’s d)
Chi-square

6.1. The overall effect of education programs on employment and


13.00***

17.63***

26.86***

recidivism
(0.45)

(0.45)

(2.29)

The overall effect of the education programs examined on employ­


ment comes as no surprise, much like the effect some of them had on
participants

3.58 (4.64)

recidivism as well, since many previous studies indicated this positive


32.7%

53.5%

46.8%

impact (Bozick et al., 2018). However, it is impressive to realize that


Non-
12th grade program

these programs, which are neither therapeutic (e.g., drug treatment) nor
practical (e.g., employment programs) in their nature, demonstrate
Participants

5.79 (5.08)

positive effects. The robustness of these programs is a well-rooted


18.1%

34.2%

66.8%

insight in IPS practices, whereby every incarcerated individual with


* p < .05 ** p < .01 *** p < .001.

educational gaps must complete an education program before enrolling


for other programs. The superior effect found for two of the programs
Any employment during the
first year following release
No. of employment months

could serve the education branch to insist that individuals study until
following release (SD)

they complete the more potent programs, and only then move on to
during the first year
3 yrs. Reincarceration

psycho-social or employment programs.


Employment is an obvious starting point in the reentry process
3 yrs. Rearrest

because it is the major “routine activity” of most individuals released


from prison (Bucklen & Zajac, 2009; LeBel et al., 2008). The present
Table 2

study findings reveal that all four education programs examined had
considerably improved their participants’ employment status during

7
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

their first year following release from prison. This is no small achieve­ 6.3. The mediating effect of employment
ment as the said programs were theory-based, and not employment-
based, and in light of the latter’s effect, as demonstrated in literature, This complex understanding of the impact of certified prison edu­
being less than encouraging (Visher et al., 2005). cation programs serves to explain not only their positive effect due to
Bushway and Apel (2012) have suggested that incarcerated or both their symbolic (the certificate) and instrumental (employment)
formerly incarcerated individuals’ participation in employment-based values, but their interaction too, i.e., the mediating effect employment
programs serves as a low-risk indicator for employers, thereby sowing has on the link between the program and lower recidivism rates. Indeed,
the seeds for future integration into society. Every employer tries to the program itself has a significant impact on reducing recidivism,
identify success indicators in prospective employees, and for formerly however, when employment (achieved in part due to program partici­
incarcerated individuals, a history of success in rehabilitation programs pation) is taken into consideration as well, weight shifts from the edu­
strongly suggests improved odds of successful work placement. While cation program to employment. In other words, when the program’s
the education programs examined by the present study are not directly instrumental value (employment) is introduced, the magnitude of its
aimed at improving employment, they may serve employers as in­ independent symbolic value’s (the diploma) contribution to reduced
dicators of their participants’ perseverance, as well as their ability to recidivism lessens. Thus, for example, in the 8th grade program, the rate
complete given tasks, and accept authority – thereby improving their of recidivism among non-participants is 1.35 higher than that of par­
chances of getting hired. ticipants (36.6% vs. 27.8% respectively), however, when only the un­
Nevertheless, only two out of the four education programs examined employed are taken into account (thereby reducing the effect of
in the present study have been shown as effective in both improving employment on recidivism), the difference between the two groups
employment and lowering recidivism. The program data used here only drops almost entirely, reaching a mere 1.05 (44.4% vs. 42.1%
attests to a medium negative effect size between employment and respectively).
recidivism (r between − 0.18 to − 0.34). Indeed, studies have shown that This complex connection therefore appears to have some practical
correlations between employment and recidivism are not simple, implications on the development and offering of prison education pro­
particularly since many formerly incarcerated individuals are only grams. By nature, such programs are not designed to improve partici­
employed part-time in low-paying jobs (Geller et al., 2006), making pants’ employment status, as they were originally intended for 14-year-
integration into society as proud, independent members more difficult. olds (in the 8th grade program example, equivalent to ABE) and 18-year-
It may therefore be suggested that employment is of substantial olds (in the 12th grade program example, equivalent to GED). This is
importance beyond its effect on reducing recidivism. It is an expression certainly true for the prison program consisting of basic subjects such as
of released individuals’ return to normative life, and thus any significant Hebrew, English, and mathematics, as opposed to subjects with practical
effect achieved due to program participation may prove to be of value such as technology and computers. The insight emerging by the
considerable value. In that sense, incarcerated individuals are among present study, whereby the greater impact on reduced recidivism can be
“people from all walks of life [who] pursue education for much the same traced to employment, requires program providers to increase the
reason: because it provides opportunity” (Brazzell et al., 2009, p. 41). instrumental value of their offering, by focusing, for example, on the
necessary skills in the labor market today. Even prison programs must be
adapted to this dynamic world in which information and communica­
6.2. The certification effect of the 8th and 12th grade programs tion technology is transforming the nature of how work is conducted, as
well as the meaning of social relationships, subsequently requiring the
The various levels of education programs appear to be of a linear successfully employed to have different skills (Binkley et al., 2012).
nature, since they merely reflect students’ baseline level. For example, These changes are not only limited to jobs with a high educational
individuals who completed 10 or 11 years of schooling will be offered a profile, but certainly apply to the lower-level jobs for which formerly
12th grade program, whereas individuals who completed 8 or 9 years of incarcerated individuals are often hired. It is therefore important to
schooling will be offered a 10th grade program. How, then, can the incorporate language and information skills into these education pro­
impressive potency of the 8th and 12th grade programs be explained? It grams, such as writing emails, using word processor and electronic
seems that these programs’ added value lies in the formal certification spreadsheet software, as well as encourage participants to share infor­
from the Ministry of Education they grant, serving not only a bureau­ mation with colleagues and enhance their creativity, as these are
cratic entity, but probably having a significant psychological effect on necessary skills in employment today (Van Laar et al., 2017).
program graduates as well, with some practical subsequences. Prison
education officers who were asked to explain the study findings 6.4. Between the 12th and 8th grade programs
confirmed the notion of a “certification effect”. Graduates from the 8th
and 12th grade programs were very proud and excited to receive their Both 12th and 8th grade programs had a significant effect on
formal diplomas, that were often the first they had ever obtained. Thus, reducing recidivism and increasing employment prospects. However,
a bureaucratic gesture serves as a signal of their competence, and of their while the magnitude of the mediator is quite similar in both programs
being successful individuals in normative society. No less importantly, (reducing the log odds of recidivism by 0.17 and 0.15 for 12th and 8th
the diploma enables owners to apply to many jobs offers for which they grade programs respectively), there is a tremendous difference in the
would otherwise not have been eligible. direct effect of the program on recidivism (reducing the log odds of
The impact of certified education programs may be viewed through recidivism by 0.48 and 0.31 for 12th and 8th grade programs respec­
the prism of Ryan and Deci’s (2000) self-determination theory (SDT). tively – more than 1.5 times). It seems that, compared with the 8th grade
They maintain that behaviors driven by external motivation cover a program, the 12th grade program is much more potent in reducing
spectrum of motivation levels that are distinguishable by the in­ recidivism on its own, independently of the extra effect of employment.
dividual’s level of autonomy. It seems that individuals’ participation in Thus, when only unemployed participants are taken into account
certified education programs is a manifestation of external motivation’s (thereby excluding employment’s contribution to reduced recidivism),
level of autonomy. However, the diploma has symbolic value – it con­ the rate of recidivism among non-participants in the 12th grade program
firms graduates’ competence, and rehabilitates their self-worth – is 1.24 times that of participants (37.5% vs. 30.2% respectively),
thereby amounting to a more autonomous and self-determined form of whereas the difference between the two groups in the 8th grade program
extrinsic motivation. At the same time, the certification also improves is small, just 1.05 times (44.4% vs. 42.1% respectively). How, then, can
graduates’ chances of integrating into the labor market, thereby meeting the difference in significant effect between the two programs be
the beneficial aspect of external regulation. explained?

8
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

One could suggest that the 8th grade program contains very basic (2019) examined two Jewish religious programs offered in Israeli
academic content and therefore does not share the same independent prisons, and reported that inmates who participated in both consecu­
potential to reduce recidivism. However, how can such a basic program tively displayed reduced recidivism. It seems, therefore, that something
even lead to employment, lacking any content designed to achieve such about the 12th grade program gives it added value in reducing recidi­
an effect? Perhaps one could suggest that this program contains all the vism; however, what that something is remains unclear. A more pro­
surrounding components of a normative life, and in fact trains partici­ found analysis of its content, as well as more comprehensive information
pants for normative settings. Despite its basic content, participants in provided by teachers and participants during in-depth interviews, may
this program experience all stages encountered by individuals in reveal the sources of this program’s powerfulness.
normative, achievement-oriented society: coping with authority (the
teacher); experiencing frustration due to lack of understanding; and 6.5. Study limitations
making an effort prior to an exam. Such training teaches individuals
upon their release to speak the cultural language of the labor market, The present study is based on PSM analysis, and is therefore subject
thereby increasing their chances of integrating into it. to the limitations associated with this technique. Propensity score
The impact of the 12th grade program, by contrast, is much more analysis assumes that all variables related to treatment assignment are
powerful in reducing recidivism regardless of employment. What is it included in the vector of observed covariates, or are accounted for
about this program that leads to this effect? Seemingly, the program through strong correlation with observed covariates (Rosenbaum &
consists of basic high-school subjects that cannot directly impact desis­ Rubin, 1985). For example, we were unable to assess the level of par­
tence from crime in themselves. However, having a 12-year education ticipants’ motivation before they began any of the programs, and we did
diploma does open up many options for climbing the social ladder and not have access to additional in-prison information such as furloughs
gaining social status. Program graduates can take part in advanced and disciplinary hearings, or participation in other programs. To assess
vocational training, and even attend college or university. Perhaps the the possible bias due to unobserved covariates, Rosenbaum’s bounding
very availability of so many options, whether realized or not, improves approach was employed. Despite the sensitivity test’s reasonable results,
self-worth, and serves as an advantage in many other areas as well, such these study findings should be interpreted with some caution.
as partner choice and even improved health (Zimmerman et al., 2015). Moreover, the employment outcome measure used in this study –
It seems that, beyond the practical benefit derived from the program, employment in the year following release – is not necessarily the best
it is attended by individuals that had already acquired basic skills, such economic index for reflecting successful reentry. Studies have shown
as verbal communication and abstract conceptualization, thereby that the wages earned by formerly incarcerated individuals, more than
enabling a richer, and more profound discussion of formal subjects in employment itself, predicts reduced recidivism (Grogger, 1998; Visher
class. As part of this program, current affairs are presented, and et al., 2008), for insufficient wages may make it more difficult for
discourse on ethics and values ensues in a manner allowing participants formerly incarcerated individuals to stay on the straight and narrow
to associate content with their life circumstances, and the decisions that (Visher et al., 2011).
led them to crime and incarceration. Moreover, since the 12th grade The present study found that the average number of employment
program is the last of the formal education programs, graduates that months among study participants during the first year following their
have yet to be released from prison may go on to join enrichment pro­ release was 3–6 months. This finding implies low and insufficient in­
grams in which subjects such as criminology and psychology are taught come, making desistance more challenging. After all, it makes sense that
on a popular level, thereby enabling inmates some introspection, as well formerly incarcerated individuals, even those who have a 12-year edu­
as an understanding about what went wrong on their way to becoming cation certificate, will be more likely to find short-term, low-paying
engaged in crime. In that sense, this program follows the twofold work. More detailed and sensitive earning measures, such as hourly
rationale behind correctional education mentioned earlier in this paper: wages and total wages earned (e.g., Duwe & Clark, 2014), as well as
acquiring normative values to help participants cope with moral di­ subjective measures like work commitment (Apel & Horney, 2017), may
lemmas they may face outside prison (Andrews & Bonta, 2003; MacK­ clarify the role played by the household economy and work experience
enzie, 2008) alongside practical tools that may help to improve their in formerly incarcerated individuals’ rehabilitation.
chances of finding honorable, profitable jobs (Lockwood et al., 2012). Finally, this study was conducted using Israeli data, and the rele­
Understanding the added value of the 12th grade program, possibly vance of its insights to other prison systems and national cultures re­
due to the in-depth content it contains, should encourage program de­ quires further examination. We believe that the Israeli system described
velopers to enhance those parts of the programs during which a moral in the method section shares a similar approach and practices with many
and ethical discussion relevant to participants’ lives is held. But can Western prison systems, rendering the outcomes of this study highly
improved moral conduct be expected following ethical studies, partic­ relevant to them as well. At the same time, the collaborative rather than
ularly when participants are convicted of a crime? A 16-lesson formal disciplinary approach to corrections in Israel may have enhanced the
cognitive-moral training program was offered in a North Carolina impacts of education programs.
correction facility, introducing topics such as “How can one make good
ethical decisions? Right versus wrong decisions, and moral courage” 7. Conclusion
(based on the program by Leming & Bom, 2000). The effectiveness of
this program was examined using the change in individuals’ self- One of the concluding sentences of the round table on correctional
reported moral factors, among which were empathy, social re­ education held by John Jay College and Urban Institute members is that
sponsibility, confidence, and reflectiveness. The reported changes “education’s power to transform lives in both tangible and intangible
among program participants compared to the control group were non- ways makes it one of the most valuable and effective tools we may have
significant on most metrics, and the authors admit that it may have for helping people rebuild their lives after incarceration” (Brazzell et al.,
been naïve of them to expect these individuals to undergo a fundamental 2009, p. 42).
shift in ethical thinking following a short intervention (Kropp, 2006). Our findings reinforce earlier studies showing that education pro­
More encouraging findings emerged from an effectiveness analysis of grams can be effective in reducing recidivism and increasing employ­
religious rehabilitation programs offered by the Israel Prison Service. ment. Such twofold effects appeared in two of the four programs
These programs involve some analysis of Jewish religious texts that are evaluated in the present study. Moreover, in them, employment was
linked to incarcerated individuals’ life circumstances, and whose found to mediate the impact of the program on recidivism, with the
running motifs include: choosing between right and wrong; social strongest impact being for individuals who gained employment
accountability, and the importance of moral integrity. Haviv et al. following release. Prison systems to date provide educational

9
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

programming as a general intangible benefit, irrespective of its impacts Acknowledgements


on employment. Our study suggests that a more tangible link to
employment would increase the impact of these programs on successful This study was supported by a grant no. 352/2010 from the Israel
reentry. Prison Service to the Institute of Criminology, The Faculty of Law at the
Hebrew University of Jerusalem.

Appendix
Table A
12th grade program – characteristics of treatment and comparison groups before and after PSM.

Treatment group Comparison group Bias before Treatment group Comparison group Bias after

N = 526 N = 9302 N = 349 N = 349

Demographic Characteristics
Married 33.6% 33.4% 0.6 34.7% 38.4% − 7.9
Nationality (Jewish) 59.7% 59.1% 1.3 59.9% 53.0% 14.0
New immigrant 21.9% 18.8% 7.7 20.3% 20.3% 0.0
Criminal Background
No. of prior incarcerations 0.16 0.42*** 32.0- 0.16 0.12 5.3
Present Incarceration
Age at start of incarceration 33.9 33.7 2.3 33.8 33.7 1.5
Length of sentence (months) 29.7 15.1*** 68.1 30.8 27.8 14.2
Violent offense 43.3% 44.3% − 1.9 41.5% 37.8% 7.5
Drug offense 24.1% 23.4% 1.7 23.8% 25.2% − 3.4
Property offense 35.6% 34.4% 2.4 38.1% 36.4% 3.6
Sex offense 6.3% 5.0% 5.3 6.6% 6.6% 0.0
Year of release 2009.2 2008.0*** 50.1 2009.0 2009.1 − 6.9
Crime Profile
Violence 10.8% 20.8*** − 27.6 12.9% 12.6% 0.8
Domestic Violence 0.19% 0.24% − 1.2 2.9% 5.7% − 6.1
***
p < .001.

Table B
10th grade program – characteristics of treatment and comparison groups before and after PSM.

Treatment group Comparison group Bias before Treatment group Comparison group Bias after

N = 435 N = 6965 N = 244 N = 244

Demographic Characteristics
Married 35.4% 32.8% 5.5 34.4% 29.5% 10.4
Nationality (Jewish) 61.6% 59.6% 3.1 62.3% 61.1% 2.5
New immigrant 20.9% 19.5% 3.5 22.5% 24.6% − 5.1
Criminal Background
No. of prior incarcerations 0.28 0.75*** − 42.5 0.29 0.36 − 6.7
Present Incarceration
Age at start of incarceration 33.1 33.6 − 4.9 33.1 33.0 − 0.8
Length of sentence (months) 26.8 15.7*** 50.9 26.2 26.7 − 2.5
Violent offense 41.4% 43.8% − 4.9 38.1% 38.5% − 0.8
Drug offense 28.7% 23.7%* 11.4 31.1% 30.3% 11.4
Property offense 32.0% 36.1% − 8.8 29.5% 28.3% 2.6
Sex offense 5.5% 4.0% 7.2 4.9% 3.3% 7.7
Year of release 2009.2 2007.9*** 59.9 2009.0 2009.0 − 5.5
Crime Profile
Violence 13.1% 20.8*** − 20.7 13.9% 15.2% − 3.3
Domestic Violence 0.0% 0.5% − 9.9 0.0% 0.0% 0.0
* p < .05 *** p < .001.

Table C
8th grade program – characteristics of treatment and comparison groups before and after PSM.

Treatment group Comparison group Bias before Treatment group Comparison group Bias after

N = 2580 N = 1729 N = 677 N = 677

Demographic Characteristics
Married 33.1% 34.2% − 2.4 38.8% 35.2% 7.8
Nationality (Jewish) 58.8% 59.1% 0.6- 58.6% 58.8% 0.3-
New immigrant 19.0% 19.8% − 2.1 20.8% 20.7% 0.4
Criminal Background
No. of prior incarcerations 0.33 0.84*** − 42.1 0.53 0.45 6.2
Present Incarceration
Age at start of incarceration 33.8 33.9 − 0.5 34.2 33.8 3.3
Length of sentence (months) 25.0 15.5*** 46.6 18.1 17.2 4.6
(continued on next page)

10
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

Table C (continued )
Treatment group Comparison group Bias before Treatment group Comparison group Bias after

N = 2580 N = 1729 N = 677 N = 677

Violent offense 46.1% 44.4% 3.4 41.7 43.9% − 4.5


Drug offense 24.5% 24.2% 0.7 24.7% 24.5% 0.3
Property offense 35.1% 34.1% 2.1 31.3% 35.4% − 8.7
Sex offense 4.9% *3.6% 6.3 3.4% 4.0% − 2.9
Year of release 2008.3 2007.9 19.7 2007.4 2007.6 − 10.5
Crime Profile
Violence 17.8% 20.6* − 7.0 23.2% 21.7% 3.8
Domestic Violence 0.2% 0.2% − 1.7 0.1% 0.3% − 3.4
* p < .05 *** p < .001.

Table D
6th grade program – characteristics of treatment and comparison group before and after PSM

Treatment group Comparison group Bias before Treatment group Comparison group Bias after

N = 2537 N = 714 N = 270 N = 270

Demographic Characteristics
Married 31.1% 32.8% − 3.6 40.4% 37.8% 5.6
Nationality (Jewish) 60.0% 58.5% 2.9 54.1% 57.0% − 6.0
New immigrant 18.5% 19.2% − 1.8 17.4% 19.3% − 4.7
Criminal Background
No. of prior incarcerations 0.29 0.88*** − 47.7 0.65 0.63 1.5
Present Incarceration
Age at start of incarceration 34.0 33.5 3.7 34.5 33.4 9.8
Length of sentence (months) 23.6 16.2*** 38.7 18.9 19.5 − 2.9
Violent offense 44.6% 47.2% − 5.2 45.2% 45.9% − 1.5
Drug offense 23.5% 21.4% 4.9 18.9% 20.7% − 4.4
Property offense 36.8% 33.3% 7.2 25.6% 27.4% − 3.9
Sex offense 5.3% 4.1% 6.0 7.8% 5.6% 10.5
Year of release 2009.0 2007.9*** 48.2 2007.2 2007.4 − 8.5
Crime Profile
Violence 20.9% 21.6% − 1.6 23.3% 25.2% − 4.5
Domestic Violence 0.2% 0.3% − 1.7 0.7% 0.4% 7.6
***
p < .001.

References Brazzell, D., Crayton, A., Mukamal, D. A., Solomon, A. L., & Lindahl, N. (2009). From the
classroom to the community: Exploring the role of education during incarceration and
reentry. Urban Institute.
Aakvik, A. (2001). Bounding a matching estimator: The case of a Norwegian training
Brewster, D. R., & Sharp, S. F. (2002). Educational programs and recidivism in
program. Oxford Bulletin of Economics and Statistics, 63(1), 115–143.
Oklahoma: Another look. The Prison Journal, 82(3), 314–334.
Adams, K., Bennett, K. J., Flanagan, T. J., Marquart, J. W., Cuvelier, S. J., Fritsch, E., &
Bucklen, K., & Zajac, G. (2009). But some of them don’t come back (to prison!): Resource
Burton, V. S. (1994). A large-scale multidimensional test of the effect of prison
deprivation and thinking errors as determinants of parole success and failure. The
education programs on offenders’ behavior. The Prison Journal, 74(4), 433–449.
Prison Journal, 89, 239–264.
Anderson, S. V., & Moore, E. A. (1995). Evaluation of the impact of correctional education
Bushway, S. D., & Apel, R. (2012). A signaling perspective on employment-based reentry
programs on recidivism. Bureau of Planning and Evaluation: Ohio Department of
programming. Criminology & Public Policy, 11, 21–50.
Rehabilitation and Correction.
Cecil, D. K., Drapkin, D. A., MacKenzie, D. L., & Hickman, L. J. (2000). The effectiveness
Andrews, D., & Bonta, J. (2003). The psychology of criminal conduct (4th ed.). Anderson
of adult basic education and life-skills programs in reducing recidivism: A review
Publishing.
and assessment of the research. Journal of Correctional Education, 55, 207–226.
Aos, S., Miller, M., & Drake, E. (2006). Evidence-based public policy options to reduce future
Chen, M. Y., Liu, Y., & Zumbo, B. D. (2020). A propensity score method for investigating
prison construction, criminal justice costs, and crime rates. Washington State Institute
differential item functioning in performance assessment. Educational and
for Public Policy.
Psychological Measurement, 80, 476–498. https://doi.org/10.1177/001316441
Apel, R. J., & Horney, J. (2017). How and why does work matter? Employment
9878861.
conditions, routine activities, and crime among adult male offenders. Criminology,
Cho, R. M., & Tyler, J. H. (2010). Does prison-based adult basic education improve
55, 307–343.
postrelease outcomes for male prisoners in Florida? Crime & Delinquency, 59,
Assy, R., & Menashe, D. (2014). The catch-22 in Israel’s parole law. Criminal Justice and
975–1005.
Behavior, 41(12), 1422–1436.
Creese, B. (2015). An assessment of the English and maths skills levels of prisoners in England.
Austin, P. C. (2011). An introduction to propensity-score methods for reducing the effects
UCL Institute of Education (Centre for Education in the Criminal Justice System).
of confounding in observational studies. Multivariate Behavioral Research, 46,
Dervan, L. E. (2011). American prison culture in an international context: An
399–424. https://doi.org/10.1080/00273171.2011.568786.
examination of prisons in America, the Netherlands and Israel. Stanford Law & Policy
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in
Review, 22, 413–428.
social psychological research: Conceptual, strategic, and statistical considerations.
Duguid, S. (1982). Rehabilitation through education: A Canadian model. Journal of
Journal of Personality and Social Psychology, 51, 1173–1182.
Offender Counseling Services Rehabilitation, 6(1–2), 53–67.
Batiuk, M. E., Lahm, K. F., McKeever, M., Wilcox, N., & Wilcox, P. (2005). Disentangling
Duguid, S., & Pawson, R. (1998). Education, change, and transformation: The prison
the effects of correctional education: Are current policies misguided? An event
experience. Evaluation Review, 22(4), 470–495.
history analysis. Criminal Justice, 5(1), 55–74.
Duwe, G., & Clark, V. (2014). The effects of prison-based educational programming on
Batiuk, M. E., Moke, P., & Rountree, P. W. (1997). Crime and rehabilitation: Correctional
recidivism and employment. The Prison Journal, 94(4), 454–478.
education as an agent of change – A research note. Justice Quarterly, 14(1), 167–180.
Farrington, D. P., Gottfredson, D. C., Sherman, L. W., & Welsh, B. C. (2002). The
Becker, S. O., & Caliendo, M. (2007). Sensitivity analysis for average treatment effects.
Maryland Scientific Methods Scale. In L. W. Sherman, D. P. Farrington, B. C. Welsh,
The Stata Journal, 7, 71–83.
& D. L. MacKenzie (Eds.), Evidence-based crime prevention (pp. 13–21). Routledge.
Binkley, M., Erstad, O., Herman, J., Raizen, S., Ripley, M., Miller-Ricci, M., & Rumble, M.
Geller, A., Garfinkel, I., & Western, B. (2006). The effects of incarceration on employment
(2012). Defining twenty-first century skills. In P. Griffin, B. McGaw, & E. Care (Eds.),
and wages: An analysis of the fragile families survey. Center for Research on Child
Assessment and teaching of 21st century skills (pp. 17–66). Springer.
Wellbeing. Working Paper 2006-01-FF.
Bozick, R., Steele, J., Davis, L., & Turner, S. (2018). Does providing inmates with
education improve postrelease outcomes? A meta-analysis of correctional education
programs in the United States. Journal of Experimental Criminology, 14(3), 389–428.

11
D. Walk et al. Journal of Criminal Justice 74 (2021) 101815

Gerber, J., & Fritsch, E. J. (1995). Adult academic and vocational correctional education Mantel, N., & Haenszel, W. (1959). Statistical aspects of the analysis of data from
programs: A review of recent research. Journal of Offender Rehabilitation, 22(1–2), retrospective studies of disease. Journal of the National Cancer Institute, 22(4),
119–142. 71–748.
Gideon, L., Walk, D., & Carmel, T. (2015). Israeli maximum-security prisons (pp. 238- Phelps, M. (2011). Rehabilitation in the punitive era: The gap between rhetoric and
258). In S. C. Richards (Ed.), The Marion experiment: Long-term solitary confinement & reality in U.S. prison programs. Law and Society Review, 45(1), 33–68.
the supermax movement. Southern Illinois University Press. Phelps, M. (2012). The place of punishment: Variation in the provision of inmate services
Gordon, H. R., & Weldon, B. (2003). The impact of career and technical education staff across the punitive turn. Journal of Criminal Justice, 40, 248–357.
programs on adult offenders: Learning behind bars. Journal of Correctional Education, Rosenbaum, P. R. (2002). Observational studies (2nd ed.). Springer.
54, 200–209. Rosenbaum, P. R., & Rubin, D. B. (1985). Constructing a control group using multivariate
Grogger, J. (1998). Market wages and youth crime. Journal of Labor Economics, 16, matched sampling methods that incorporate the propensity score. The American
756–791. Statistician, 39(1), 33–38.
Hasisi, B., Shoham, E., Weisburd, D., Haviv, N., & Zelig, A. (2016). The "care package," Ryan, R. L., & Deci, E. L. (2000). Self-determination theory and the facilitation of
prison domestic violence programs and recidivism: A quasi-experimental study. intrinsic motivation, social development, and well-being. American Psychologist, 55,
Journal of Experimental Criminology, 12(4), 563–586. 68–78.
Haviv, N., & Hasisi, B. (2019). Prison addiction program and the role of integrative Shadish, W. R. (2013). Propensity score analysis: Promise, reality and irrational
treatment and program completion on recidivism. International Journal of Offender exuberance. Journal of Experimental Criminology, 9(2), 129–144.
Therapy and Comparative Criminology, 63(15–16), 2741–2770. Skarbek, D. (2014). The social order of the underworld: How prison gangs govern the
Haviv, N., Weisburd, D., Hasisi, B., Shoham, E., & Wolfowicz, M. (2019). Do religious American penal system. Oxford University Press.
programs in prison work? A quasi-experimental evaluation in the Israeli prison Steurer, S. J., Smith, L., & Tracy, A. (2001). Three state recidivism study. Correctional
service. Journal of Experimental Criminology. Advance online publication https://doi. Education Association.
org/10.1007/s11292-019-09375-0. Travis, J., Western, B., & Redburn, S. (2014). The growth of incarceration in the United
Joffe, M. M., & Rosenbaum, P. R. (1999). Invited commentary: Propensity scores. States: Exploring causes and consequences. The National Academies Press.
American Journal of Epidemiology, 150, 327–333. Turner, S. F. (2018). Multiple faces of reentry. In J. Wooldridge, & P. Smith (Eds.), Oxford
Jordan, K. L. (2012). Juvenile transfer and recidivism: A propensity score matching handbook on prisons and punishment. Oxford University Press.
approach. Journal of Crime and Justice, 35(1), 53–67. Vacca, J. S. (2004). Educated prisoners are less likely to return to prison. Journal of
Judd, C. M., & Kenny, D. A. (1981). Process analysis: Estimating mediation in treatment Correctional Education, 55, 297–305.
evaluations. Evaluation Review, 5, 602–619. Van Laar, E., Van Deursen, A. J. A. M., Van Dijk, J. A. G. M., & De Haan, J. (2017). The
Kim, R. H., & Clark, D. (2013). The effect of prison-based college education programs on relation between 21st-century skills and digital skills: A systematic literature review.
recidivism: Propensity score matching approach. Journal of Criminal Justice, 41(3), Computers in Human Behavior, 72, 577–588.
196–204. Visher, C. A., Debus, S. A., & Yahner, J. (2008). Employment after prison: A longitudinal
Kohler, U., Karlson, K. B., & Holm, A. (2011). Comparing coefficients of nested nonlinear study of releasees in three states. Urban Institute.
probability models. The Stata Journal, 11, 420–438. Visher, C. A., Debus-Sherrill, S. A., & Yahner, J. (2011). Employment after prison: A
Kropp, E. H. (2006). The effects of a cognitive-moral development program on inmates in a longitudinal study of former prisoners. Justice Quarterly, 28, 698–718.
correctional educational environment. Unpublished dissertation, University of Virginia, Visher, C. A., Winterfield, L., & Coggeshall, M. B. (2005). Ex-offender employment
Charlottesville. programs and recidivism: A meta-analysis. Journal of Experimental Criminology, 1,
Lahm, K. F. (2009). Educational participation and inmate misconduct. Journal of Offender 295–315.
Rehabilitation, 48(1), 37–52. Walk, D., & Berman, E. (2015). The recidivism of Israeli prisoners. Research & Strategic
Langan, P. A., & Levin, D. J. (2002). Recidivism of prisoners released in 1994. United States Planning Department, Planning Administration, Israel Prison Service. Publication
Department of Justice. No. RR-03-2015.
LeBel, T., Burnett, R., Maruna, S., & Bushway, S. D. (2008). The "chicken and egg" of Washington State Institute for Public Policy. (2017). Benefit-cost results for adult
subjective and social factors in desistance from crime. European Journal of criminal justice. http://www.wsipp.wa.gov/BenefitCost?topicId=2.
Criminology, 5, 131–159. Weisburd, D., Hasisi, B., Shoham, E., Aviv, G., & Haviv, N. (2017). Reinforcing the
Leming, J., & Bom, P. (2000). Preliminary evaluation of the effect of ethical reasoning impacts of work release on prisoner recidivism: The importance of integrative
program on the socio-moral development of offenders (preliminary evaluation). The interventions. Journal of Experimental Criminology, 13, 241–264.
Institute for Global Ethics. Weisburd, D., Wilson, D., Wooditch, A., & Britt, C. (in press). Advanced statistics in
Lipton, D., Martinson, R., & Wilks, J. (1975). The effectiveness of correctional treatment: A criminology and criminal justice. Springer.
survey of treatment evaluation studies. Praeger. Wilson, D. B., Gallagher, C. A., & MacKenzie, D. L. (2000). A meta-analysis of
Liu, W., Kuramoto, S., & Stuart, E. (2013). An introduction to sensitivity analysis for corrections-based education, vocation, and work programs for adult offenders.
unobserved confounding in nonexperimental prevention research. Prevention Science, Journal of Research in Crime and Delinquency, 37(4), 347–368.
14, 570–580. Wolf Harlow, C. (2003). Education and correctional populations. Bureau of Justice
Lockwood, S., Nally, J. M., Ho, T., & Knutson, K. (2012). The effect of correctional Statistics Special Report No. NCJ 195670. U.S. Department of Justice.
education on postrelease employment and recidivism: A 5-year follow-up study in Zimmerman, E. B., Woolf, S. H., & Haley, A. (2015). Understanding the relationship
the state of Indiana. Crime & Delinquency, 58(3), 380–396. between education and health: A review of the evidence and an examination of
MacKenzie, D. L. (2006). What works in corrections: Reducing the criminal activities of community perspectives. In R. M. Kaplan, M. L. Spittel, & D. H. David (Eds.),
offenders and delinquents. Cambridge University Press. Population health: behavioral and social science insights (pp. 347–384). Agency for
MacKenzie, D. L. (2008, March 31). Structure and components of successful educational Healthcare Research and Quality (Publication No. 15-0002).
programs. Reentry Roundtable on Education, John Jay College of Criminal Justice,
New York.

12

You might also like