A Rapid Evidence Assessment of the effectiveness of prison education in reducing recidivism and increasing employment

Recent and current policy

All prisons in England and Wales provide education. Over the last 15 years, the focus of policy has been on improving the literacy and numeracy rates of prisoners. This focus can be traced back to the findings of the 2002 report by the Social Exclusion Unit on how to cut re-offending rates (Social Exclusion Unit, 2002). The report outlined nine key factors in re-offending with education and employment being top of the list, alongside physical and mental health, family networks, housing, and drug and alcohol misuse. The report found that many prisoners’ basic skills were poor, with 80 per cent having writing skills, 65 per cent numeracy skills, and 50 per cent reading skills at or below the level of an 11-year-old child (Social Exclusion Unit, 2002). These figures were attributed to a Home Office statistical release, but in recent years scholars have questioned their accuracy (see, for example, Nichols, 2016). In the years since, a variety of reviews and government reports have re-emphasized the role of education in prisons (see, for example, Education and Skills Committee, 2005).

The years 2015–16 saw a renewed focus on prisons as a place of reform. In September 2015, Michael Gove, the then Secretary of State for Justice, announced a review of education in adult prisons whilst his successor, Elizabeth Truss, published a White Paper on Prison Reform in November 2016 (Ministry of Justice, 2016). Gove’s commissioned review was led by Dame Sally Coates and its recommendations were far-reaching and ambitious (Coates, 2016). If followed, they would constitute a fundamental shift in the way education works in prisons in England and Wales. In particular, Coates recommended that prison governors have responsibility for commissioning and delivering education and they be held accountable for outcomes through publication of prison league tables. Coates also pointed to a huge drop in those studying Level 3 (AS/A level) and Open University qualifications and cites the restrictions placed by the Ministry of Justice (MoJ) on such studies as the reason. She recommended a core prison curriculum including English, maths and ICT delivered to Level 3 and above. Coates highlighted the need to tailor education towards employment and the role of local and national employers to help shape vocational education in prisons.

The Government White Paper on Prison Safety and Reform (Ministry of Justice, 2016) indicates that the Ministry of Justice will seek to incorporate these recommendations into an overall change in emphasis in the role of prisons. The White Paper states that prisons are, first and foremost, places that ‘exist to deprive people of their liberty and so punish them for crimes’ (p. 20). The White Paper goes on to outline four further purposes of prison. These are public protection, to maintain safety and order, to reform offenders (as a means of preventing offenders from committing more crimes), and to prepare prisoners for release. Should the White Paper progress through Parliament successfully, it will be against these standards that prison governors and staff are measured.

In the White Paper, education comes under ‘reform’ and the government will seek to use measures of progress in English and maths to track the ‘distance travelled’ by prisoners during their time in custody. Further to this, they will also look at the number of qualifications and employment-focused accredited courses they have undertaken and completed during custody. Unfortunately, this could potentially miss the point of education. Looking at the number of qualifications obtained whilst in prison will encourage providers to deliver low-level, short-term education courses that could end up being a ‘tick-box’ exercise (something the White Paper explicitly states they wish to move away from).

It seems education will be seen as part of a range of interventions available to prisoners. The White Paper briefly mentions the streamlining of interventions under a case management system whereby these can be delivered in a suitable order to work with each other rather than compete to recruit people. It explicitly states which aspects of the Coates Review they will enact. First, governors will be made fully responsible for commissioning education in their prisons (rolled out once current contracts end during the course of the 2017–18 financial year). Second, all prisoners’ learning needs will be assessed upon entry to the prison in order to develop a personalized learning plan (processes similar to this are already in place and this does not constitute a fundamental change to procedure). Third, a core common curriculum will be developed, with English and maths at the heart of it to ensure that prisoners can pick up their educational progress when they move around the estate. Finally, governors will be encouraged to work with local organizations to ensure the educational provision is tailored to the needs of the workforce and for education providers to develop ‘through the gate’ skills such as interviewing techniques.

Having recounted the current policy climate, this review of extant literature provides a timely update to the discussion of what is known about the relationship between engaging in prison education and post-release outcomes.

Educational and employment backgrounds of prisoners and links with re-offending

Data regarding the educational needs of the prison population are complex and varying. This section provides an overview of recent reports of the educational levels and employment histories of prisoners in England and Wales.

The Surveying Prisoner Crime Reduction (SPCR) was a longitudinal cohort study of 1457 adult prisoners sentenced to between one month and four years in custody in England and Wales in 2005/06 (Minstry of Justice, 2014). The survey found that around a third of prisoners (32%) reported being in paid employment in the four weeks prior to custody and 51 per cent reported being in paid employment during the 12 months prior to custody (Brunton-Smith and Hopkins, 2014). Both of these figures are considerably lower than the UK general employment rate, which was 75 per cent in 2006 for those of working age (Ministry of Justice, 2010). Many prisoners report needing help finding a job on release (48%), with education (41%), and to improve work-related skills (40%), and agreed that having a job would help them stop reoffending (68%) (Hopkins, 2012). A lower likelihood of reconviction has been associated with having been employed in the year before custody (40% compared with 65% not in employment) and having a qualification prior to custody (45% compared with 60% for those who did not have a qualification) whilst a higher rate of reconviction is associated with having been in receipt of benefits in the year before custody (58% compared with 42% for those not in receipt of benefits) (Hopkins, 2012).

The most comprehensive and up-to-date study of the educational levels of prisoners found by the reviewers comes from the Centre for Education (CfE) in the Criminal Justice System at the University of Central London (Creese, 2015). This study reports on the literacy and numeracy levels of prisoners in England and Wales and employed data from the prison education providers as opposed to using MoJ data. In prisons in England and Wales, educational providers are required to conduct initial assessments of the English and maths needs for all prisoners entering the estate (unless they have been tested in the six months prior to prevent repeat testing). The study used data from four educational providers comparing the levels of literacy and numeracy in various prisons to the national levels (as measured by the Skills for Life national survey conducted by BIS in 2012) (Creese, 2015). Their figures show that 85 per cent of the general population have literacy skills at Level 1 or Level 2 compared to only 50 per cent of prisoners. Their findings regarding numeracy, however, are very different. Up to, and including, Level 1, prisoners out-perform national statistics, with 79.4 per cent of prisoners having Entry Level 3 numeracy skills and above, compared to 76.4 per cent of the general population. It should be noted that this gap is much larger at Entry Level 3, where prisoners outperform the national average consistently. However, at Level 2, only 9 per cent of prisoners are recorded as having achieved Level 2 numeracy whilst 21.8 per cent of the general population have done so.

These studies demonstrate that statistics regarding the educational levels of prisoners vary. However, there is general agreement that prisoners are, on average, less well qualified, less skilled and often come from difficult educational backgrounds.

Previous reviews of the evidence

The Ministry of Justice describes evidence on the effectiveness of employment/education programmes in reducing re-offending as ‘mixed/promising’ (Ministry of Justice, 2013). A number of systematic reviews have examined the impact of employment and/or education programmes on re-offending and/or employment. These include: Pearson and Lipton (1999); Bouffard et al. (2000); Wilson et al. (2000); and Visher et al. (2005). Reviews we have identified undertaken in the last ten years are as follows:

MacKenzie (2006) updated a meta-analysis originally conducted by Wilson et al. (2000). Meta-analysis found that the odds of recidivating were 16 per cent lower among academic education programme participants than non-participants. The meta-analysis also showed that vocational programme participation was associated with a reduction in recidivism with the odds of recidivating 24 per cent lower among vocational programme participants than non-participants.

A review of previous meta-analyses highlights the varied quality of research, with few studies being able to assert a confident causal connection between education and recidivism or post-release employment rates. However, these previous meta-analyses do indicate a general positive impact of education in prison. On the whole, research indicates that education reduces recidivism and increases employability. With respect to employment, there is some evidence that there is an impact. However, studies of employment are less common and the evidence less robust. Measuring employment post-incarceration is complex and the lack of research found reflects this. Many of the identified reviews are now dated, with most having been undertaken in or before 2006. Of the two more recent studies, one is not a systematic review (Hall, 2015) and the other only includes studies from the USA (Davis et al., 2013). The current REA adds to the list above by providing a more up-to-date review of literature and provides an opportunity to identify any recent UK studies that might be included in such a review.

Research questions

The primary research question was: What impact does education in prison have on rates of recidivism and employment?

To investigate this further we addressed the following sub-questions:

What research is currently in existence around the link between education in prison and recidivism/employment?

Characteristics of the studies relevant to the objectives of the review

To address the research questions identified for this REA, we searched for reports of studies which evaluated the effect on re-offending and post-release employment of educational and vocational training interventions delivered in a custodial setting to adult prisoners. ¹ We based our definition of ‘education and vocational training delivered in a custodial setting’ on the definition developed by Davis et al. (2013). We define an educational or vocational programme as one that includes a basic skills (e.g. literacy, numeracy, language or ICT), an academic or a vocational curriculum taught by an instructor and designed to lead to the attainment of a certificate, licence, degree or other academic or vocational qualification widely recognized in the country in which the programme is delivered. Educational programmes could be delivered either face-to-face or at a distance (i.e. through an online learning environment) or by a mix of both. Studies were excluded if they:

were not published in the English language;

Process for identifying studies for inclusion

From the initial bibliographic database keyword searches 4304 studies were identified. Their titles and abstracts were reviewed by one of two reviewers to ascertain whether they were an evaluation of a relevant intervention (as defined above). Each reviewer checked approximately one-quarter of the other reviewer’s work to monitor inter-rater reliability. At the end of this process 184 studies were categorized as ‘meeting’ or ‘partially meeting’ the intervention criteria. Two different reviewers then reviewed these titles and abstracts, bringing the list down to 29 papers that clearly met the inclusion criteria and 59 that partially met the criteria. These 59 papers were then retrieved in full and reviewed more closely, and five were included in the ‘met’ category. Five papers could not be retrieved and were therefore excluded from this study. From these searches, there were now 35 studies included as meeting the inclusion criteria. Grey literature searches revealed a further seven studies, taking the total to 42 studies. The review of key systematic reviews showed a further 38 relevant studies. Six of these were excluded during a review of titles and abstracts and 15 could not be retrieved. (Most studies not retrieved were ‘grey’ literature studies that could not be located due to the references provided being websites that are no longer available.) A further six did not match our criteria. In total, a further 11 studies were added from reviewing other systematic reviews, the majority of which were grey literature. This meant the total number of papers identified from abstracts was 53.

Following a more detailed review of each of these papers, a further ten papers were excluded. Of those excluded, two were first authored by Duguid and excluded because, in the sample of 28 studies, three studies were authored or co-authored by Duguid (Duguid, 1997; Duguid and Pawson, 1998; Duguid et al., 1998) and all related to the same set of findings. As such, only one of these studies has been included in the meta-analysis. A further six studies were excluded because the evaluation framework they employed provided insufficient methodological rigour. We only included studies that employed an evaluation design that could be characterized as Level 2 or above on the Scientific Methods Scale (Sherman et al., 1998) as modified for reconviction studies by Harper and Chitty (2005) (see Figure 1). This is a slightly lower benchmark than many of the systematic reviews discussed above. This was to allow for the possibility of more UK studies meeting the inclusion criteria. Two papers were excluded because, from the data presented, the research design was not clear.

OPEN IN VIEWER

Figure 1.

Scientific Methods Scale adapted for reconviction studies (Table 1.2 in Harper and Chitty, 2005; see: https://www.medcalc.org/calc/odds_ratio.php; accessed March 2016).

All of these papers were added to the data capture template for a more in-depth review. At this stage 15 were excluded as closer inspection revealed they did not meet the criteria; six were themselves a review of previous literature and have been included in the review of previous meta-analyses above; and four could not be located. As such, the total number of studies included in this review is 28.

Descriptive findings (28 papers)

The type of education evaluated by the studies reviewed varied considerably. This section provides an overview of the papers included in this review. The purpose is to describe the type of education the studies relate to, the methods of analysis they employed, the jurisdictions to which they relate, and the ways in which they measured the outcomes of recidivism and employment. The included studies pertained particularly to adult education. However, many did not specify the age of the prisoners in their study and instead stated ‘prisoners’ or ‘adults’ and supplied a mean age. Three studies included only male offenders, one study included only females, ten did not specify the gender of the participants and the remaining papers included both male and female prisoners.

As nearly all of the studies included are from North America, many considered the impact of Adult Basic Education (ABE), General Education Diploma (GED) attainment, post-secondary education and/or vocational education. In the UK the ABE is equivalent to basic skills (literacy and numeracy) whilst GED attainment is sufficient to gain a college placement and is (broadly) equivalent to degree-entry qualifications (such as A-Levels). Post-secondary education refers to degree-level qualifications. Of the 28 papers reviewed, 16 focused on a single type of education whilst nine covered a range of educational levels. Three studies were unclear about the nature of the education they were evaluating. Of the 16 that focused on a single type of education, 10 focused on post-secondary, or degree-level, qualifications and three on GED preparation or achievement. Only one study focused on basic skills education and two on vocational education. The nine studies that covered a range of qualifications included a combination of ABE, GED, post-secondary and vocational qualifications. In total, four included basic skills, eight GED, three post-secondary and six vocational education. In general, the studies included in this review were poor in providing information regarding the nature and content of the education delivered. None of the studies discussed the number of people on the course or how the course was delivered. Cronin (2011) and the three studies by Duguid and colleagues (Duguid, 1997; Duguid and Pawson, 1998; Duguid et al., 1998) specified that college or university faculty members as opposed to teachers in the prison delivered the educational courses and the Batiuk (1997) study specified on-the-job training.

In general, there were two types of quantitative investigation into the impact of prison education on post-release outcomes. Here, we have categorized studies as involving a ‘general’ investigation or a ‘specific’ investigation. ‘General’ investigations (15/28) took the form of looking at a cohort of released prisoners from a particular place (usually a state in the USA), categorizing them as educational ‘participants’ and educational ‘non-participants’ and then comparing the recidivism/employment rates. Second, there were ‘specific’ investigations (11/28) that focused on a particular programme, or set of programmes, run in one or more prisons. The final two studies took a different form, with one comparing outcomes of two education departments to state-wide recidivism rates and the other being a report of an earlier paper. Attempts were made to source the original paper; however, we were unable to locate it and, as such, chose to include the summary report.

The designs of the studies also varied. The majority (14 studies out of 28) had a non-random control vs. treatment design. Many of these (eight studies out of 28) looked at the ‘general’ impact of prison education. For example, the Batiuk et al. (2005) report looked at all prisoners released in the state of Ohio, USA, between 1989 and 1992. The researchers then compared the recidivism rates of those who had engaged in prison education compared to those who had not. Other studies (four studies out of 28) used a ‘matched’ control group where the intervention group was compared to a control group that was matched on a range of criteria (such as age or sentence length). Three studies compared the recidivism rates of prison education participants to their expected or predicted recidivism rates. Two studies used survival analysis as part of their method, which involved tracking a cohort of prison education participants and measuring how many ‘survive’ without returning to prison within a given time period. None of the studies employed randomization as part of their study design. Due to the complexities inherent to the study of offender learning in prison, there is limited opportunity to conduct randomized control trials (RCT) in a prison setting (Pettus-Davis et al., 2016). RCT levels of study of offender learning in prison are infrequent as a result.

Nearly all of the studies included in this review were from the United States of America (23 studies out of 28). One was from Canada, one from Australia and one from the UK. The single UK study to meet the inclusion criteria was a report commissioned by the Prison Education Trust from the Justice Data Lab (Ministry of Justice, 2015). As such, the findings from this REA are most relevant to the US context. There are several potential explanations for this US bias. In the UK, the focus of many studies has been on the ‘deficits’ of offenders, thereby providing evidence for the need for education among prisoners rather than evaluations of the efficacy of prison education. Two events in the US might have encouraged a greater focus on evaluation. First, in a review of rehabilitation programmes, Martinson (1974) infamously concluded that ‘nothing works’ in prison reform. After this point, a host of academics engaged in research in an attempt to demonstrate that this was not true. The other event mentioned regularly throughout the papers reviewed here is the removal of PELL grants for US prisoners in 1994. These were grants for poorer students for college level education in the United States. In 1994, the Violent Crime and Control and Law Enforcement Act revoked PELL Grant funding for any incarcerated individual (SpearIt, 2016). Several of the papers in this study pertain to ‘post-secondary’ (what in the UK we would call higher education) programmes in US prisons. Arguably, these papers were a reaction to the revocation of PELL grants, with universities attempting to demonstrate the impact of this type of programming as a means of securing funding for prison-college courses.

The data collection and follow-up periods of each of the studies varied. Several of the 28 studies under discussion here were based on studies that commenced in the 1970s and 1980s, although they were published in the timeframe. Out of the 28 studies reviewed, the participants of 12 studies were released from prison, or engaged in education whilst in prison, in or before 1995. Furthermore, there was variation in follow-up periods employed in the studies. Studies looked at recidivism rates after only one year (including the Justice Data Lab report from the UK – Ministry of Justice, 2015) whilst another followed participants for 13.5 years. The most common follow-up period (12/28) was three years. However, four were unclear what the follow-up period was whilst some studies varied the follow-up periods for the participants. For example, some studies stated that they followed participants for ‘up to’ five years, or a ‘maximum’ of 60 months. This made comparison of findings across studies difficult.

Finally, operationalization of the outcomes of interest also varied. Across studies, recidivism was measured in various ways; six studies used re-arrest, 16 measured re-incarceration, and four used reconviction (of a new crime). Furthermore, five were unclear on their measure of recidivism whilst four used other measures of recidivism including parole violation. Seven of the studies measured employment as an outcome. However, the variety of methods of measuring employment across such a small number of studies demonstrates the complexity of using employment as an outcome. Some studies (3/28) looked at wages earned (measured differently in each study; hourly, monthly and quarterly). One considered employment as being a minimum of 30 hours per week whilst another looked for verification from a parole officer to confirm employment. Two examined a range of factors including type of work, hourly wages and total hours worked.

The information presented demonstrates the varied nature of prison education evaluation. Despite several other meta-analyses and reviews of literature coming before this one, such variation raised questions regarding the suitability of meta-analytic techniques in this field. We also reflected on whether further criteria should be applied (such as the time of the participants’ prison education experience). However, we chose to include all 18 studies (those of the 28 with sufficient information for inclusion) regardless of the follow-up period or time of the participant’s actual prison education experience. There were two key reasons for this decision. First, having a ‘cut off’ date for engaging in education (as opposed to publication date) would go against the normative processes of meta-analysis and complicate the inclusion/exclusion criteria. Second, the original developers of the odds ratio method for meta-analysis employed the technique to retrospectively analyse the link between a behaviour and a disease (e.g. lung cancer and smoking). In Mantel and Haenszel’s (1959) original paper the example retrospectively analyses the likelihood of contracting lung cancer whether a person has smoked, or not smoked, and, if the odds are sufficiently different, then the statistics indicate a relationship. In their process, the relationship holds regardless of how long a person smoked for, how many cigarettes they smoked a day, how long ago they had stopped and, importantly, if they developed lung cancer in the future. Some of the people may be in their 40s while others are in their 70s. Still, the relationship holds. Applying this logic to the work here, odds ratios are applicable because they provide evidence of a relationship regardless of how long prisoners were in education, how much education they had, how long ago they came out of prison and irrespective of if they will recidivate in the future. As such, inclusion of papers with such varying measures was deemed appropriate for this method.

Meta-analysis

As noted above, measures of recidivism and unemployment and follow-up periods varied between studies. However, in line with previous reviews (Davis et al., 2013; Hurry et al., 2006) we combined the different measures in the 18 studies suitable for meta-analysis into single measures of recidivism and employment. To understand the relationship between prison education and recidivism, each study effect size was calculated using the odds ratio. Calculations were cross-validated using an online odds ratio calculator (Harper and Chitty, 2005) and the Metafor package in R (Viechtbauer, 2010), which was also used to calculate the pooled effect and produce image files of the forest plots.

Figure 2 presents a forest plot of the odds ratio and 95 per cent confidence interval for each study. The black square point represents the effect size estimate for each study using the odds ratio. The ‘whiskers’ (thin black lines) represent the 95 per cent confidence intervals which tells us the range of values between which we can be 95 per cent confident that the value actually lies. Therefore, the smaller the ‘whiskers’ the more confident we can be that the findings are a true reflection of impact. The plot illustrates that all studies favour education, meaning delivering education in prison settings has a positive impact on recidivism when compared to the outcomes of a comparison or control group. All studies examined demonstrated this effect; however, two studies (Batiuk et al., 2005; Burke and Vivian, 2001) have confidence intervals that cross the line of nil effect. This means that two studies show that education participation could have a negative effect on recidivism. The pooled odds ratio effect (diamond) is 0.64 (confidence interval between 0.62 and 0.67). This is an overall positive effect of education on recidivism. Importantly, the confidence interval for this is very narrow and does not cross the line of nil effect, so we can be confident that the true impact of education on recidivism is close to that which has been found here.

OPEN IN VIEWER

Figure 2.

Forest plot of the odds ratio for each of the 18 recidivism studies.

We also tested whether there was a statistical difference between odds ratios based on different educational levels. Several studies indicated that the higher the level of education the greater the impact on recidivism. For example, Anderson’s (1995) study indicates that completion of higher-level qualifications such as college-level degrees results in lower recidivism rates than achieving qualifications at lower levels. However, they failed to supply n-values for the control groups to allow for inclusion of this study in this analysis. Therefore, statistical analysis regarding educational levels is inconclusive with further investigation required.

From the 18 studies identified for the recidivism meta-analysis, there are five studies which have examined the effect of prison education interventions on employment outcomes. The forest plot (Figure 3) illustrates the odds ratio for each study. As the desired outcome is an increase in employment (as opposed to a decrease in recidivism), the box and whiskers that sit to the right of the line show a positive effect of education. The overall pooled effect of prison education on employment outcomes is positive, with a pooled odds ratio of 1.24 (confidence intervals between 1.16 and 1.31). However, the overall finding masks some complexity. The first three studies depicted in Figure 3 (Cronin, 2011; Duwe and Clarke, 2014; Hull et al., 2000) show a positive effect of education on employment outcomes. The final two studies (Nally et al., 2011; Steurer et al., 2001) show a negative effect. However, Nally et al. (2011) report that they found that the longer-term effect of education on employment was positive. This illustrates the complexity of the link between education and employment.

OPEN IN VIEWER

Figure 3.

Forest plot of the odds ratio for each of the five employment studies.

Interpreting odds ratios

Odds ratios are calculated using the probability of recidivism/employment for prison education participants compared against non-participants. Finding a pooled ratio of 0.64 means that participating in prison education reduces the likelihood of recidivism by approximately one-third. Note this is distinct from saying that the results show prison education reduces recidivism by a third. Instead, the odds ratio says that a person who has engaged in prison education is two-thirds as likely (0.64 = 64%∼2/3) as someone who has not engaged in prison education to recidivate. Interpreting the findings from the employment studies should be done with more caution, as two out of the three studies sit to the left of the forest plot above (Figure 3). However, the overall pooled odds ratio stands at 1.24. Again, we can interpret this as meaning that those who have engaged in prison education are 24 per cent more likely to find employment than those who have not. This does not mean that the rate of employment is increased by 24 per cent, but that the likelihood of finding employment is increased. In other words, if rates of employment are very low, the impact of education may be very small. From the 18 studies identified for the recidivism meta-analysis it is not possible to compare gender differences. One study was male only (Duguid and Pawson, 1998), one female only (Torre and Fine, 2005), nine were mixed and seven not specified. In cases where the cohort was mixed, in general, results were not presented by demographics.