CoChat
Sign Up
PreprintCited by 17Criminal Justice and Corrections AnalysisCriminal Justice and PenologyCrime Patterns and Interventions

Prison Rehabilitation Programs: Efficiency and Targeting

Open access
Arbour, William, Lacroix, Guy, Marchand, Steeve·2021·SSRN Electronic Journal
DOI 10.2139/ssrn.3761992
Open this paper in CoChat — free account

Abstract

Increasing evidence suggests that incarceration can improve inmates' social reintegration under certain circumstances. Yet, the mechanisms through which incarceration may lead to successful rehabilitation remain largely unknown. This paper finds that participation in social rehabilitation programs can significantly reduce recidivism only when inmates are evaluated by an assessment tool which allows to identify their criminogenic needs. This suggests that targeting criminogenic needs is crucial for successful rehabilitation. We also find that inmates with a high risk score or who exhibit procriminal attitudes benefit little from participation. To control for selection, we instrument participation with program availability at the time and place of incarceration. We also use a coefficient stability approach to test for omitted variable biases. Both approaches yield similar results.

Verified 2026-10-23 · 1 reference not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

Participation in social rehabilitation programs while incarcerated significantly reduces recidivism, but only for inmates whose criminogenic needs were evaluated with the LS/CMI assessment tool. The paper argues that targeting programs to assessed needs is crucial, while high-risk inmates and those with procriminal attitudes benefit little or not at all.

Key findings

  • Among LS/CMI-evaluated inmates, program participation reduces recidivism within three years by about 9 percentage points: OLS with full controls gives −9.4 pp, Oster (2019) unrestricted estimator gives −9.2 pp, and causal random forest ATE gives −8.18 pp.
  • No significant overall effect is found for non-evaluated inmates; by program type, non-evaluated inmates show a significant effect only for self-development programs (−6.3 to −8.7 pp), while evaluated inmates show significant effects for all categories except “other.”
  • Heterogeneity is substantial: inmates with high total LS/CMI scores (≥24) show no clear program effect, medium-risk inmates (scores 8–23) show about −10 pp, a one-standard-deviation increase in procriminal attitude moves the effect toward zero by 3.4–4.8 pp, and ages 30–40 increase effectiveness by about 5.5 pp. Out-of-bag predictions show −16.6 pp below the median versus −4.9 pp above the median.
  • Targeting evidence: higher LS/CMI education/employment scores increase odds of education or job-skills programs; higher alcohol/drug scores increase odds of addiction programs; assault crimes increase odds of violence programs; drug offenses increase odds of addiction programs.

Methods

  • Administrative data from DACOR, Quebec’s Ministry of Public Security, covering three male provincial prisons from 2007 to 2019; main OLS sample has 15,047 observations, 3,898 participants, and 8,767 LS/CMI scores. Recidivism is defined as an incarceration sentence from another conviction within three years after release.
  • Baseline OLS regressions of recidivism on program participation interacted with LS/CMI evaluation, controlling for LS/CMI component scores and other covariates. Robustness to unobserved selection uses Oster (2019)’s coefficient-stability approach.
  • Heterogeneous treatment effects are estimated with causal random forests (Wager and Athey, 2018) and best linear projections of conditional average treatment effects on observables.

Limitations

  • The paper acknowledges potential selection on unobservables and uses Oster (2019) to argue omitted-variable bias is unlikely, but that method relies on assumptions about Rmax and δ.
  • Evident from the text: the design is observational, program participation is voluntary, and there is no randomized assignment; the sample is limited to male provincial prisons in Quebec with short sentences, so findings may not generalize to other populations or jurisdictions.
  • Evident from the text: recidivism is measured as a subsequent incarceration sentence, follow-up windows vary, and some high-risk subgroup estimates have large confidence intervals with no clear effect.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.2139/ssrn.3761992

Cites

56 of 57 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

14 of 57 cited sources checked · 7 supported · 1 flagged · accuracy not yet measured

  • Premisecitation
    § Contributions to the Literature · p. 6

    Heller et al. (2017) provide evidence that behavioral interventions may reduce recidivism among young offenders.

    “Heller et al. (2017) provide evidence that behavioral interventions may reduce recidivism among this population. They exploit three RCTs in Chicago in which at-risk youths and juvenile delinquents participated in programs and find a significant reduction in arrests and reincarcerations.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Heller et al. (2017) source
  • Premisecitation
    § Contributions to the Literature · p. 5

    Bhuller et al. (2020) provide causal evidence that inmates who serve longer sentences in Norway are less prone to reoffend.

    “Bhuller et al. (2020) use a judge leniency design and provide causal evidence that inmates who serve longer sentences in Norway are less prone to reoffend.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Bhuller et al. (2020) source

      The Introduction and Section 5 report judge-stringency IV evidence that incarceration in Norway reduces reoffending. But the comparison is incarceration versus no incarceration, not longer versus shorter sentences among inmates. Section 5.1 reports time served for the incarceration group without establishing that serving a longer sentence itself reduces reoffending. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
  • Premisecitation
    § 1 Introduction · p. 4

    Increasing evidence suggests prison establishments that emphasize rehabilitation can reduce recidivism.

    “There is, however, increasing evidence suggesting prison establishments that emphasize rehabilitation can reduce recidivism (Landersø, 2015; Mastrobuoni and Terlizzese, 2019; Bhuller et al., 2020; Lotti, 2020; Hjalmarsson and Lindquist, 2020).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Landersø, 2015 source
    • cites Mastrobuoni and Terlizzese, 2019

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Bhuller et al., 2020 source
  • Premisecitation
    § Contributions to the Literature · p. 7

    Assessment tools used to predict recidivism are now widespread in most jurisdictions.

    “Assessment tools used to predict recidivism are now widespread in most jurisdictions: bail and sentencing judges (Stevenson and Doleac, 2019; Albright, 2019), parole board members (Berk, 2017), and probation officers (Kopf, 2014) use the assessment scores to determine the level of risk.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Stevenson and Doleac, 2019 source
    • cites Albright, 2019

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Berk, 2017 source
  • Premisecitation
    § 4 Discussion and Policy Implications · p. 32

    Research has documented convincing evidence of substantial rehabilitative effects in Norway, Denmark and Sweden.

    “and research has documented convincing evidence of substantial rehabilitative effects (Landersø, 2015; Bhuller et al., 2020; Hjalmarsson and Lindquist, 2020).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Landersø, 2015 source
    • cites Bhuller et al., 2020 source

      The Introduction and Concluding Remarks report that incarceration in Norway discouraged crime and encouraged employment, largely among people who were not working before incarceration. The paper calls this a proof of concept and says evidence from other settings is needed to assess generalizability; it does not establish rehabilitative effects in Denmark or Sweden. Page numbers were not provided in our copy. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites Hjalmarsson and Lindquist, 2020

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. 33
  • May combine two claims. 37

  • Classified as probably not a substantial claim.

    Premisecitation
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    Age is invariably negatively correlated with criminal activities.

    “This effect could be partly mechanical: an extensive literature shows how age is invariably negatively correlated with criminal activities (see, e.g., Landersø et al. (2017)).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Landersø et al. (2017) source

      The introduction (p. 4) describes an almost universally observed crime-age profile in which crime rates rise until about age 18–20 and decline thereafter. That contradicts the claim that age is invariably negatively correlated with criminal activity. (LLM, conf=0.90)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

    Premisecitation
    § Contributions to the Literature · p. 5

    Others argue that prison spells might act as a school of crime.

    “Others argue that prison spells might act as a school of crime in which networks of criminals share knowledge and influence one another (Bayer et al., 2009; Stevenson, 2017).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Bayer et al., 2009 source
    • cites Stevenson, 2017 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § Contributions to the Literature · p. 5

    Some find that lengthy prison sentences may deplete human capital and alter inmates' capacity to reintegrate into the labor market.

    “Some find that lengthy prison sentences may deplete human capital and alter inmates' capacity to reintegrate into the labor market (Lochner, 2004; Aizer and Doyle Jr, 2015; Mueller-Smith, 2015).”
    Quote in the paperNot a substantial claim2 of 3 sources checked
    • cites Lochner, 2004 source
    • cites Aizer and Doyle Jr, 2015 source

      Section 2.1 proposes that juvenile incarceration can hinder human and social capital accumulation, including social capital that aids job search. The Introduction likewise says incarceration during adolescence may reduce future wages. But the paper does not establish that lengthy sentences cause these effects or measure released inmates’ ability to reintegrate into the labor market; its findings concern juvenile incarceration and high-school completion. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites Mueller-Smith, 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § Contributions to the Literature · p. 6

    Blattman et al. (2017) show that behavioral therapy can reduce violent crimes.

    “Blattman et al. (2017) study the effect of behavioral therapy and show that it can reduce violent crimes.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Blattman et al. (2017) source

      The Abstract (p. 1) says cognitive behavioral therapy initially reduced crime and violence, supporting the broad point that the intervention can help. But the Results (p. 21) say effects on disputes and fights were not statistically distinguishable from zero; the paper does not establish a reduction in violent crimes as a distinct outcome. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation
    § 3.4 Heterogeneity, or Who to Target? · p. 24

    Athey and Imbens (2016), Wager and Athey (2018), and Athey et al. (2019) developed random forests methods designed to capture honest heterogeneity in treatment effects, circumventing overfitting issues inherent to traditional methods and thus avoiding mistaking statistical noise for heterogeneity.

    “In particular, Athey and Imbens (2016), Wager and Athey (2018), and Athey et al. (2019) developed random forests methods designed to capture honest heterogeneity in treatment effects, circumventing overfitting issues inherent to traditional methods and thus avoiding mistaking statistical noise for heterogeneity.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Athey and Imbens (2016) source

      The Abstract (pp. 1–2) and §2 (p. 4) describe a method for detecting treatment-effect heterogeneity that separates tree construction from effect estimation to reduce bias from adaptive fitting. But this paper develops honest regression-tree methods, not random-forest methods; the random-forest characterization is not supported for Athey and Imbens (2016). (LLM, conf=0.89)

      Experimental — accuracy not yet measured
    • cites Wager and Athey (2018) source
    • cites Athey et al. (2019)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § Contributions to the Literature · p. 6

    Balafoutas et al. (2020) show that a simple intervention asking inmates to reflect on incarceration increased social aptitudes.

    “Balafoutas et al. (2020) randomly asked inmates to reflect on their incarceration and show that this simple intervention increased the inmates' social aptitudes.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Balafoutas et al. (2020) source

      Section 3.3 (Priming) says inmates were randomly assigned to write either about their time in prison and its effect on their social behavior or about a neutral topic. Section 4.3 (Effects of the priming intervention) reports increased trust, reciprocity, and cooperation toward people outside prison, but not toward other inmates. The paper measures choices in economic games, not a general increase in inmates’ social aptitudes. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    Semenova and Chernozhukov (2020) provide a method for estimation and inference of the best linear projections of CATEs on observable characteristics.

    “Semenova and Chernozhukov (2020) provide a method for estimation and inference of the best linear projections of CATEs on observable characteristics.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Semenova and Chernozhukov (2020) source

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given
    § Contributions to the Literature · p. 6

    Different types of programs (educational, vocational, and therapy) can be beneficial if well targeted.

    “also suggests that different types of programs (educational, vocational, and therapy) can be beneficial if well targeted.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 2.2 Social Rehabilitation Programs · p. 9

    The program data comprise participants in more than 150 unique programs.

    “The program data comprise participants in more than 150 unique programs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given
    § 2.1 Prisons in Canada · p. 8

    Individuals typically spend only two-thirds of their sentence incarcerated.

    “individuals typically spend only two-third of their sentence incarcerated”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.3 Effects by Program Types · p. 21

    An increase in the LS/CMI alcohol and drug score significantly increases the odds of participating in an addiction-related program, although the effect is only slightly statistically significant.

    “Second, an increase in the score measuring alcohol and drug problems significantly increases the odds of participating in an addiction-related program, although this effect is only slightly statistically significant.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingreasoning
    § 2.5 Descriptive Statistics · p. 11

    Less than 10% of individuals in the sample are granted parole.

    “Less than 10% of individuals in our sample are granted parole.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.2 Robustness to Unobserved Selection · p. 19

    Treatment effects estimated from the controlled regression barely move, although adding the controls substantially increases the regressions' R2.

    “Treatment effects estimated from the controlled regression (equation (6)) barely move, although adding the controls substantially increases the regressions' R 2 .”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 3.3 Effects by Program Types · p. 21

    We reject the null hypothesis that all programs for non-evaluated inmates have the same effect, with a p-value less than 0.05.

    “Furthermore, we reject the null hypothesis that all programs for non-evaluated inmates have the same effect with a p -value of less than 0 . 05 .”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 2.5 Descriptive Statistics · p. 12

    Participants tend to be slightly younger than non-participants.

    “We also find that participants tend to be slightly younger than non-participants”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 2.5 Descriptive Statistics · p. 11

    For the lowest total LS/CMI scores, the rate of recidivism is close to zero; the recidivism rate increases steadily with the total score and comes close to one for the highest total scores.

    “For the lowest total scores, the rate of recidivism is close to zero. The recidivism rate increases steadily with the total score and comes close to one for the highest total scores.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.1 Effect of Participating in Programs · p. 16

    Controlling for each LS/CMI component separately increases the R2 to 0.142 while leaving the estimate of the effect of programs unchanged (-9.6 percentage points for evaluated participants).

    “Column (d) controls for each component of the LS/CMI separately instead of using the total score, which increases the R 2 to 0 . 142 while still leaving our estimate of the effect of programs unchanged (-9.6 percentage points for evaluated participants).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingreasoning
    § 2.5 Descriptive Statistics · p. 11

    The vast majority of inmates in the sample are released near their planned date.

    “This is not surprising in our context, because the vast majority of inmates are released near their planned date.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.3 Effects by Program Types · p. 21

    For inmates not evaluated, only programs related to self-development have a significant effect on recidivism, with an effect between -6.3 and -8.7 percentage points.

    “For inmates not evaluated, we only find a significant effect for programs related to self-development (between -6 . 3 and -8 . 7 percentage points).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 2.5 Descriptive Statistics · p. 11

    The highest peak of the LS/CMI total score distribution is around 29, where close to 70% of inmates recidivate within three years.

    “The highest peak is around 29, where close to 70% of inmates recidivate within three years.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.2 Robustness to Unobserved Selection · p. 19

    Setting δ=1, the value of Rmax that would be needed to estimate a treatment effect of 0 is substantially above those considered appropriate by Oster (2019), and in some cases a negative treatment effect is still found for Rmax=1.

    “We obtain values again substantially above those considered appropriate by Oster (2019) and, in some cases, still find a negative treatment effect for R max = 1 .”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Oster (2019) source
    • The authors state this as an assumption: The estimated index of observable variables predicts the true direction of the relationship between this index and participation.
    • The authors state this as an assumption: The estimates in the randomized experiments analyzed by Oster (2019) are truly causal so that they should survive the above correction.
    • The authors state this as an assumption: Control variables already included in the baseline regression are not part of the confounding set (the set of observables proportionally related to unobservables).

  • May combine two claims.

    Findingown data
    § 3.2 Robustness to Unobserved Selection · p. 19

    Regardless of the follow-up window, there are large and significant effects of participation on recidivism and low values of R2.

    “Regardless of the follow-up window, we find large and significant effects of participation on recidivism and low values of R 2 .”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 2.5 Descriptive Statistics · p. 12

    The share of individuals belonging to an indigenous ethnic group is higher among participants than non-participants.

    “and the share of individuals belonging to an indigenous ethnic group is higher among participants.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.1 Effect of Participating in Programs · p. 16

    Including the full vector of controls increases the R2 to 0.19 but leaves the estimate of the effect of programs essentially unchanged (-9.4 percentage points).

    “Finally, column (e) includes our full vector of controls. This further increases our R 2 to 0 . 19 but leaves our estimate of the effect of programs essentially unchanged (-9.4 percentage points).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 2.5 Descriptive Statistics · p. 11

    For the LS/CMI components with the strongest relationships, the lowest score predicts a recidivism rate of around 20%, while the highest score predicts one of around 80%.

    “For these components, the lowest score predicts a recidivism rate of around 20%, while the highest score predicts one of around 80%.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.4 Heterogeneity, or Who to Target? · p. 24

    The gap in recidivism between participants and non-participants widens as the risk score increases up to a score of 12, narrows as the score increases further, and closes entirely at a score of 32.

    “The gap widens as the risk score increases up to a score of 12 and narrows as the score increases further, closing entirely at a score of 32 (at this point, inmates are deemed highly risky).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 2.5 Descriptive Statistics · p. 12

    There are statistically significant differences in average scores for other LS/CMI components between participants and non-participants, but the differences are small compared with the standard deviations of scores.

    “However, we find statistically significant differences in average scores for other components. The differences are, however, small compared with the standard deviations of scores.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 2.5 Descriptive Statistics · p. 11

    Few inmates are at either extreme of possible LS/CMI total scores.

    “Few inmates are at either extreme of possible scores.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    For the youngest age group (29 and below), point estimates suggest a similar effect of programs to that of the 30 to 40 age group, although the effect relative to the oldest group is not statistically significant.

    “As for the youngest age group (29 and below), our point estimates suggest a similar effect of programs than that of the 30 to 40 age group, although the effect relative to the oldest group is not statistically significant.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.4 Heterogeneity, or Who to Target? · p. 29

    The effect of programs is statistically significant for individuals with a procriminal attitude score between 0 and 2, but not for those who score higher.

    “The effect of programs is significant for individuals who score between 0 and 2 23 but not for those who score higher.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    The results support the finding that medium-risk individuals benefit the most from programs: treatment effects are -10 percentage points for offenders with risk scores ranging from 8 to 23.

    “The results support the finding that medium-risk individuals benefit the most from programs: we estimate treatment effects of -10 percentage points for offenders with risk scores ranging from 8 to 23.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 3.4 Heterogeneity, or Who to Target? · p. 29

    Within the group of offenders with a total LS/CMI score of 24 and above, the estimated treatment effects are close to zero and are not statistically significant.

    “However, within this group, the estimated treatment effects are close to zero and are not statistically significant”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    Beyond an LS/CMI score of 24, participating in programs does not clearly affect recidivism, but the confidence intervals are large.

    “Beyond an LS/CMI score of 24, which corresponds to high and very high risk, participating in programs does not clearly affect recidivism, but it is worth noting that the confidence intervals are large.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § Contributions to the Literature · p. 5

    Recent papers have shown that incarceration in rehabilitation-friendly prisons can be beneficial.

    “Recent papers have shown that incarceration in rehabilitation-friendly prisons can be beneficial.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.3 Risk Evaluation · p. 9

    The LS/CMI is a widely implemented proprietary assessment tool in North America.

    “The LS/CMI is a widely implemented proprietary assessment tool in North America (Andrews et al., 2000)”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Andrews et al., 2000

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisereasoning
    § 1 Introduction · p. 4

    Selection issues may occur if (self-)selected participants differ systematically from non-participants.

    “Selection issues may occur if (self-)selected participants differ systematically from non-participants.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § Contributions to the Literature · p. 6

    Lotti (2020) finds that a reform in England and Wales shifting from strict warehousing to rehabilitating young offenders decreased recidivism substantially.

    “Lotti (2020) exploits a major reform in England and Whales which marked a shift from strict warehousing to rehabilitating of young offenders. The reform is found to have decreased recidivism substantially.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Lotti (2020) source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § Contributions to the Literature · p. 6

    Mastrobuoni and Terlizzese (2019) find that open prisons yield lower recidivism.

    “Mastrobuoni and Terlizzese (2019) compare (harsh) closed to (rehabilitation-oriented) open prison regimes on recidivism. The endogeneity of assignment is accounted for by using variations in nearby prisons' overcrowding, and they find that open prisons yield lower recidivism.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Mastrobuoni and Terlizzese (2019)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisenone given
    § 1 Introduction · p. 4

    It is still unclear which specific factors are truly beneficial.

    “While this nascent literature provides convincing evidence that favorable prison conditions may reduce recidivism, it is still unclear which specific factors are truly beneficial.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § Contributions to the Literature · p. 6

    Hjalmarsson and Lindquist (2020) show that time spent in prison allows inmates with mental health issues to engage in therapy and programs, resulting in long-lasting health benefits and reduced recidivism.

    “Finally, Hjalmarsson and Lindquist (2020) study the effects of Swedish reforms that lengthened sentences. They show that time spent in prison allows inmates with mental health issues to engage in therapy and to participate in various programs, resulting in long-lasting health benefits and reduced recidivism.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Hjalmarsson and Lindquist (2020)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 1 Introduction · p. 4

    Wilson et al. (2000) and Davis et al. (2013) conclude that evidence suggests participation decreases recidivism and increases employment, but research designs are too poor to provide reliable estimates.

    “In their meta-analysis of correctional rehabilitation programs for adult offenders, Wilson et al. (2000) and Davis et al. (2013) both conclude that, while evidence suggests that participation decreases recidivism and increases employment, the research designs are deemed too poor to provide reliable estimates of their impact.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Wilson et al. (2000) source
    • cites Davis et al. (2013) source

  • May combine two claims.

    Premisereasoning
    § 1 Introduction · p. 4

    Observed and unobserved differences between the two groups may partly explain differences in post-release outcomes, thus biasing the programs' causal effects.

    “Observed and unobserved differences between the two groups may partly explain differences in post-release outcomes, thus biasing the programs' causal effects.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § Contributions to the Literature · p. 6

    These studies provide indirect evidence that rehabilitation programs are perhaps responsible for the observed link between incarceration and post-release outcomes.

    “These studies provide indirect evidence that rehabilitation programs are perhaps responsible for the observed link between incarceration and post-release outcomes.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § Contributions to the Literature · p. 5

    Papers focusing on the United States and using randomly assigned judges find no effects of incarceration on economic outcomes or recidivism.

    “However, papers focusing on the United States and using randomly assigned judges find no effects of incarceration on economic outcomes (Kling, 2006) or on recidivism (Green and Winik, 2010).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Kling, 2006 source
    • cites Green and Winik, 2010 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § Contributions to the Literature · p. 6

    Tobón (2020) shows that exogenous assignment to newer Colombian prisons substantially reduces recidivism.

    “Tobón (2020) shows that exogenous assignment of inmates from older to newer Colombian prisons-which are less crowded and offer better conditions, services, and rehabilitation programs-substantially reduce recidivism.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Tobón (2020) source

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given
    § Contributions to the Literature · p. 5

    Incarceration may decrease crime through incapacitation and deterrence.

    “Conversely, incarceration may decrease crime through incapacitation and deterrence in addition to providing a unique opportunity for intervention.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § Contributions to the Literature · p. 7

    Mastrobuoni and Terlizzese (2019) argue that targeting might play an important role, as they do not find that high-risk offenders benefit from moving from closed prisons to open prisons like their lower-risk counterparts.

    “Mastrobuoni and Terlizzese (2019) argue that targeting might play an important role, as they do not find that high-risk offenders benefit from moving from closed prisons to open prisons like their lower-risk counterparts.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Mastrobuoni and Terlizzese (2019)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given
    § 2.5 Descriptive Statistics · p. 12

    The higher share of indigenous individuals among participants is explained by some programs being specifically targeted for this population in one of the three prisons.

    “The higher share of individuals belonging to an indigenous ethnic group among participants is explained by some programs being specifically targeted for this population in one of our three prisons.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § Contributions to the Literature · p. 6

    Macdonald (2020) shows that eliminating early release opportunities decreases recidivism, likely through reduced rehabilitation efforts in educational programs.

    “Macdonald (2020) exploits a reform in Arizona that eliminated judges' discretion in the decision to release inmates early. He shows that eliminating early release opportunities decreases recidivism, likely through reduced rehabilitation efforts in educational programs.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Macdonald (2020)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § Contributions to the Literature · p. 6

    Kuziemko (2013) presents evidence that inmates respond to canceled parole eligibility by decreasing rehabilitation effort, including reducing program participation, thus increasing recidivism likelihood.

    “Kuziemko (2013) exploits a 1998 policy reform which canceled parole eligibility for convicts in the state of Georgia. She presents evidence that inmates respond by decreasing their rehabilitation effort, including reducing their program participation while incarcerated, thus increasing recidivism likelihood.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Kuziemko (2013) source

  • May combine two claims.

    Premisereasoning
    § Contributions to the Literature · p. 6

    Overall, convincing evidence that programs can benefit inmates exists but remains scarce.

    “Overall, convincing evidence that programs can benefit inmates exists but remains scarce.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § Contributions to the Literature · p. 7

    The extent to which risk assessments can be used to provide targeted rehabilitation services is less understood.

    “However, the extent to which risk assessments can be used to provide targeted rehabilitation services is less understood (Long et al., 2019).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Long et al., 2019 source

  • May combine two claims.

    Interpretationown data
    § 3.1 Effect of Participating in Programs · p. 16

    Adding the total LS/CMI score only marginally affects the estimate of the effect of programs on recidivism for evaluated participants (-9.3 percentage points), suggesting the estimates are not driven by selection.

    “Remarkably, this only marginally affects our estimate of the effect of programs on recidivism for evaluated participants (-9.3 percentage points), suggesting our estimates of the effect of programs are not driven by selection”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 3.2 Robustness to Unobserved Selection · p. 19

    The controls account for a sizable portion of the variance in recidivism, yet such essential covariates prove to be negligibly correlated with the treatment effect.

    “This indicates that the controls account for a sizable portion of the variance in recidivism, yet such essential covariates prove to be negligibly correlated with the treatment effect.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 3.3 Effects by Program Types · p. 24

    Evaluated inmates are more likely to have their needs recognized and to be assigned to an appropriate program.

    “This further suggests that evaluated inmates are more likely to have their needs recognized and to be assigned to an appropriate program.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § 4 Discussion and Policy Implications · p. 31

    Such evaluations can be useful to target inmates who seem to benefit the most from programs, like those with medium overall risk measures and those with low measures of procriminal attitude.

    “We also show that such evaluations can be useful to target inmates who seem to benefit the most from programs, like those with medium overall risk measures and those with low measures of procriminal attitude.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § 3.4 Heterogeneity, or Who to Target? · p. 24

    The method allows one to compute a treatment effect for each individual in the sample based on an array of characteristics.

    “The method allows one to compute a treatment effect for each individual in the sample based on an array of characteristics.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4 Discussion and Policy Implications · p. 32

    In the United States, the average expenditure per inmate is low--around $33,274, on average, per year in 2015.

    “In the United States, the average expenditure per inmate is low--around $33,274, on average, per year in 2015 (Mai and Subramanian, 2017)”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Mai and Subramanian, 2017

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    Since older inmates are less likely to reoffend in the first place, younger individuals can be expected to gain more from participating in programs.

    “Then, since older inmates are less likely to reoffend in the first place, one can expect younger individuals to gain more from participating in programs.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4 Discussion and Policy Implications · p. 32

    In the provincial prisons of Quebec, the average cost of incarceration is around 56,234 US dollars.

    “In our setting, the cost of incarceration is middle-of-the-road: the average cost of incarceration in the provincial prisons of Quebec, where offenders are incarcerated for mostly misdemeanors and low-level crimes, is around 56,234 US dollars.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Segel-Brown, 2018

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationown data
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    Programs have heterogeneous effects as a function of observable characteristics, and thus targeting might be beneficial.

    “Our tests presented above suggest that programs have heterogeneous effects as a function of observable characteristics and thus targeting might be beneficial.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § 4 Discussion and Policy Implications · p. 32

    In Norway, Denmark and Sweden, average expenditure per inmate in 2015 were around 138,116, 79,096, and 144,139 US dollars, respectively.

    “In Norway, Denmark and Sweden, average expenditure per inmate in 2015 were around 138,116, 79,096, and 144,139 US dollars, respectively, 24”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Aebi et al. (2016)

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 3.4 Heterogeneity, or Who to Target? · p. 28

    Procriminal attitudes are deemed supportive of crime and are averse to treatment.

    “first, inmates with risks and needs associated with a procriminal attitude have lower treatment effects because these attitudes are deemed supportive of crime and are averse to treatment (Andrews et al., 2000).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Andrews et al., 2000

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

63 of 63

  • The authors state this as an assumption.

    The estimated index of observable variables predicts the true direction of the relationship between this index and participation.

    “we only need to assume that our estimated index of observable variables predicts the true direction of the relationship between this index and participation.”

  • The authors state this as an assumption.

    The estimates in the randomized experiments analyzed by Oster (2019) are truly causal so that they should survive the above correction.

    “assuming that the estimates in those papers are truly causal so that they should survive the above correction”

  • The authors state this as an assumption.

    Control variables already included in the baseline regression are not part of the confounding set (the set of observables proportionally related to unobservables).

    “Control variables already included in this baseline regression are assumed not to be part of the confounding set (i.e., the set of observables proportionally related to unobservables).”

  • The paper relies on this without support.

    The LS/CMI evaluation and scores are designed in part to predict recidivism and various criminogenic problems.

    “The LS/CMI evaluation and scores are key components in our analysis, as they are designed in part to predict recidivism and various criminogenic problems.”

  • The paper relies on this without support.

    The decision to recidivate will potentially be affected by numerous events happening after inmates leave prison.

    “This is especially relevant in our context, as the decision to recidivate will potentially be affected by numerous events happening after inmates leave prison.”
5 of 5

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Oster-estimator row gives the wrong R-max setting

    Minor

    The Table 5 estimator row says it uses R max = 1.5, whereas the text specifies R max = 1.5 times the controlled R2; Table 5 also lists those resulting R-max values separately.

  • Above-median effect is called marginally significant

    Minor

    The reported estimate of −0.0487 with standard error 0.0323 has a test statistic of about 1.51, below the conventional 10% two-sided significance threshold. Table 9 also gives it no significance star, contrary to the text's description.

  • Program-type coefficient groups are described in reverse order

    Minor

    The text assigns β1 to evaluated inmates and β2 to non-evaluated inmates. Table 6 instead labels the β1 equality test under "Program if not Evaluated" and the β2 test under "Program if Evaluated."

3 of 3

Cites

Flagged first, then by citations

Research works printed but not linked: 9

  1. On the top graph, the estimation contains year and prison fixed effects. The total score is divided into groups with at least 200 observations. A similar pattern is observed when considering more or less categories. On the top and bottom graphs, Best Linear Projection (and 95% CI) 6: Participation Rates and Best Linear Projections of the Treatment Effects on the Score Notes
  2. A Albright, 2019, If you give a judge a risk score, Evidence from kentucky bail decisions. Th John M. Olin Center for Law, Economics, and Business Fellows, 85
  3. D A Andrews, 2000, Level of service/case management inventory: LS/CMI. Multi-Health Systems
  4. Aebi, 2015, for Norway, Denmark, and Sweden are 125,560, 71,905, and 131,035, respectively. We calculated the values in USD$ using an exchange rate of 1.1, the approximate exchange rate at the end of
  5. 2015, We used an exchange rate of 1.39, which was the approximate exchange rate at the end of, This estimate comes from Segel-Brown
  6. C Mai, 2010, The price of prisons: Examining state spending trends
  7. M Mueller-Smith, 2015, The criminal and labor market impacts of incarceration
  8. B Segel-Brown, 2018, Update on costs of incarceration, Parliamentary Budget Officer
  9. B Vose, 2008, The empirical status of the level of service inventory, Federal Probation, 72

Cited by

Sorted by citations

Related

Authors
Arbour, W., Lacroix, G., & Marchand, S.
Affiliations
  • University of Toronto (CA)
  • Université Laval (CA)
  • The University of Melbourne (AU)
Corresponding author
Lacroix, Guy
ORCID
Venue
SSRN Electronic Journal
Publisher
Elsevier BV
Published
2021
ISSN
1556-5068 (ISSN-L)
DOI
10.2139/ssrn.3761992
OpenAlex
W3124954008
Semantic Scholar
619d6e305d9219923c24df95ca000cad9d5f73b1
Cited by
17
References
43
Citations by year
2026: 4 · 2025: 6 · 2024: 3 · 2023: 2 · 2022: 2
Type
posted content
Language
en
Open access
Locations
  • SSRN Electronic Journal (acceptedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • Minerva Access (University of Melbourne) (submittedVersion)
Funders
Ministry of Public Security of the People's Republic of China
Topics
  • Social Sciences › Sociology and Political Science › Criminal Justice and Corrections Analysis
  • Social Sciences › Law › Criminal Justice and Penology
  • Social Sciences › Sociology and Political Science › Crime Patterns and Interventions
Keywords
prison rehabilitation programs, recidivism, social reintegration, criminogenic needs, program targeting, program participation, selection bias, coefficient stability

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-23.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-23.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review