A systematic review of criminal recidivism rates worldwide: 3-year update

Background: Comparing recidivism rates between countries may provide useful information about the relative effectiveness of different criminal justice policies. A previous 2015 review identified criminal recidivism data for 18 countries and found little consistency in outcome definitions and time periods. We aimed to update recidivism rates in prisoners internationally. Methods: We conducted a systematic review of criminal recidivism rates in prisoners and followed PRISMA guidelines. Using five bibliographic indexes, we carried out non-country-specific and targeted searches for 50 countries with the largest total prison populations. We included reports and studies of released prisoners that reported re-arrest, reconviction and reincarceration rates. Meta-analysis was not possible due to multiple sources of heterogeneity. Results: We identified criminal recidivism information for 23 countries. Of the 50 countries with the largest prison populations, 10 reported recidivism rates for prisoners. The most commonly reported outcome was the 2-year reconviction rate. We were able to examine reconviction between different time periods for 11 countries and found that most reported small changes in official recidivism rates. Overall, for 2-year follow-up period, reported re-arrest rates were between 26% and 60%, reconviction rates ranged from 20% to 63%, and reimprisonment rates varied from 14 to 45%. Conclusions: Although some countries have made efforts to improve reporting, recidivism rates are not comparable between countries. Criminal justice agencies should consider using reporting guidelines described here to update their data.


Invited Reviewers
Recidivism rates (or rates of repeat offending) are often used as a measure of effectiveness of prison systems and post-release offender management programmes (Ministry of Justice, 2017). The comparison of recidivism rates between countries and regions may provide useful information about relative effectiveness of different sentencing and rehabilitation policies. However, the operational definitions of recidivism may vary significantly between countries. In a previous systematic review, recidivism rates among prisoners worldwide, published before December 2014, were examined (Fazel & Wolf, 2015) and differences in outcome definitions, reporting practices and their comparability between countries were outlined. In addition, a proposed reporting guideline to facilitate international comparisons of recidivism statistics was published.
Here, we provide an update on recidivism rates in prisoners worldwide.

Methods
This review builds up on the methods of the previously published study by Fazel & Wolf (2015). We expanded the search to other databases and modified the search strategy. We searched SAGE, MEDLINE, EMBASE, PsycINFO, PsycARTICLES for the last 10 years (from 01.01.2008 until 23.07.2019) with no language restrictions. The keywords included the names of the 50 countries with largest prison populations in absolute terms (World Prison Brief, 2018) and a list of commonly reported outcomes ( Figure 1). Google Scholar and Google Web were used for subsequent targeted searches. In addition, we scanned reference lists of included documents. In case of multiple reports identified for the same country, we extracted the most recent data. Studies for geographical regions within the country were included if the national information were unavailable or dated.
We included cohorts where reconviction, re-arrest, and re-imprisonment rates in released prisoners were reported. We excluded studies of recidivism in individuals receiving non-custodial sentences or in heterogeneous samples of offenders without data for a subgroup of released prisoners. If no new data had been identified for a particular country, we reported the rates from the original review (Fazel & Wolf, 2015). Due to heterogeneity in outcome definition and time periods, meta-analysis was not conducted.
DY and SS conducted the search and independently extracted the data on country, sample selection, definitions of outcomes and rates. Uncertainties were checked with SF. The publications in languages other than English were translated with the assistance of native speakers, who were either employees or students at Oxford University.

Results
We identified 28 publications that reported recidivism rates in released prisoners from 25 countries (Table 1 and Table 2 (Yeoman, 2015). In addition, we identified several publications that reported cross-sectional data on recidivism (i.e. how many current prisoners had previous convictions; from Brunei, Finland, Ghana, India, Russia and Thailand) but these did not provide information on time at risk and were excluded.
All included reports were conducted on general populations except for the studies from Italy (n = 479) and Latvia (n = 442). For Italian and Latvian samples, we estimated 95% confidence intervals, assuming normal distribution (provided in parentheses).
For all reported outcomes, a two-year follow-up period was the most commonly used. As shown in Table 2, the two-year re-arrest rates ranged from 24% (Singapore) to 60% (USA), two-year reconviction rates ranged from 20% (Norway) to 63% (Denmark), and two-year reimprisonment rates ranged from 14% (Oregon, USA) to 45% (Australia) (see Table 3 for two-year reconviction rates from included countries).

Amendments from Version 2
Figure 2 was amended. In the previous version, the 1-year reconviction rate in Estonia had been plotted incorrectly. We are grateful for our readers who have alerted us of this. The figure now represents the correct data. No other changes were made.
Any further responses from the reviewers can be found at the end of the article We additionally compared reconviction rates examined in the previous review (Fazel & Wolf, 2015) with updated information (Table 4). Such comparisons were possible for 11 countries (Denmark, France, Germany, Iceland, Singapore, Republic of Ireland, Sweden, Singapore, UK: England and Wales, UK: Northern Ireland, UK: Scotland).

Discussion
In this systematic review, we have presented worldwide prisoner recidivism rates and found that only 10 out of 50 countries with the largest prison populations reported recidivism statistics for cohorts of released prisoners. This finding suggests the lack of systematic and open approach towards recidivism research in many countries, despite its importance for public safety and health. In addition, Although some jurisdictions have made efforts to increase comparability of recidivism statistics (e.g., Northern Ireland implemented the same reconviction criteria as England and Wales), overall recidivism rates remain difficult to compare between countries because of significant variations in outcome definitions and reporting practices. In particular, when reporting reconviction rates, certain jurisdictions with lower rates (e.g Norway and North Carolina) operationalise recidivism as both an offence and conviction that have to occur during a specified follow-up period. This definition of recidivism is thus contingent on the length of court proceedings, and reconviction rates are typically lower when compared to jurisdictions that allow additional time after the follow-up period for court proceedings (and convictions) to be finalised (see Figure 2). For two countries that were included in the original 2015 review, no new published data was identified (Finland and Norway).

Reconviction Reimprisonment
The new crime has to happen during a follow-up period and extra 2 years are allowed for finalisation of the sentence.

Re-arrest
Being a suspect of a crime.

Reconviction
New criminal conviction during a follow-up period.
1, 2, 5 years No precise data provided that would allow for estimation of the cohorts' sizes.

Reconviction
The offence and conviction both have to happen during a follow-up period 2 years

France
Ministère de la Justice, 2013

Reconviction
The offence and conviction both have to happen during a follow-up to be counted as recidivism.
1, 2, 3, 4, 5, 6 years Follow-up period starts next calendar year from the year of initial conviction. Follow-up may overlap with time in prison.

Hans-Jörg & Jörg-Martin, 2014 Reconviction
The offence and conviction both have to happen during a follow-up to be counted as recidivism.

Reconviction
New criminal conviction during a follow-up period 2 years Includes prisoners in Vernd ("halfway house", type of parole)

Ireland, Republic of Central Statistics
Office, 2016

Reconviction
To be counted as a recidivism event, an offence has to occur within a follow-up period and a conviction has to happen within two years after the offence.

Reimprisonment
Receiving a new prison sentence during a follow-up period.

Reconviction (or initiation of proceedings)
A new criminal charge that did not results in acquittal or other technical dismissal during a follow-up period.

Reconviction (or initiation of proceedings)
A new criminal charge that did not results in acquittal or other technical dismissal during a follow-up period 1, 2, 3 years

years
Includes Drug Rehabilitation Centre inmates. No precise data provided that would allow for estimation of the cohort's size.

Reimprisonment
Receiving a new prison sentence during a follow-up period.

Spain - Catalonia
Area of Research and Social and Criminological Formation, 2015

Reimprisonment
Receiving a new prison sentence during a follow-up period.

years
Sweden Swedish National Council for Crime Prevention, 2018

Reconviction
The new crime has to happen during the follow-up period and extra 3 years are allowed for finalisation of the sentence Data for longer follow-up periods became available and the rates were recalculated.

Reimprisonment
Return to a prison in the same state during a follow-up period.

Re-arrest Reconviction Reimprisonment
To be accounted for, a respective event (new arrest, conviction or reimprisonment) should happen during a follow-up period on the state territory.

USA - Oregon
State of Oregon Criminal Justice Commission, 2018

Re-arrest Reconviction Reimprisonment
To be a accounted for, a respective event (new arrest, conviction or reimprisonment) should happen during a follow-up period on the state territory.
1, 2, 3 years Includes released prisoners on parole and post-release supervision.
* Table 2. Reconviction, re-arrest and reimprisonment rates in released prisoners by country and follow-up period length.   Changes in the outcome definition. 1-and 2-year reconviction rates were used as outcomes in the older report.
In the newer report, 'proven reconviction' is used, which is 1-year reconviction rate with an extra 6-month period to allow for the imposition of a court conviction. The management of individuals' data and the agencies responsible for it have also changed (outlined in the reports' methodology sections).

27% (2011) 26% (2015)
No exact information about sample size available. Overall, for the countries with updated data available, any changes in recidivism rates over time were small where there were no obvious revisions to reporting practices. This contrasts with reductions in self-reported crime in some surveys in high-income countries such as England and Wales (Office for National Statistics, 2018). Changes in rates were observed in those

Figure 2. One-and two-year reconviction rates in released prisoners in different jurisdictions by usage of additional time after the follow-up period for finalization of court proceedings. Note: *the Netherlands used the initiation of court proceedings that did not result in acquittal or technical dismissal during the follow-up as an outcome.
countries that changed the operationalisation of the outcome or the ways they collected and reported data. One exception to this is the Republic of Ireland, where the reconviction rate decreased by 6% in 3 years in the absence of any obvious changes in reporting practices. During this period, the number of people in the released prisoners' cohort nearly doubled from 5,489 in 2008 (Central Statistics Office, 2013) to 9,339 in 2010 (Central Statistics Office, 2016).
We conclude that international comparisons between countries remain problematic, and the use of a checklist (Appendix 1; Fazel et al., 2019a) may facilitate more consistent and transparent reporting of recidivism rates.

Data availability
Appendix 1, containing the recidivism reporting checklist, is available from OSF.  , 2016 1 , p.3). Furthermore, the production and use of data, such as recidivism rates that is often held by government departments is a major issue of democratic governance (Parsons, 2002 3 , p.145). This review updates findings from a previous one (Fazel & Wolf, 2015 4 ), and expands the searches conducted from 20 to 50 countries with the largest prison populations.
Whilst recidivism rates do have the potential to be used as a standardised measure of prison performance and could be used to compare prisons nationally (and internationally), any choice of outcome should reflect the purpose of prisons, and this is likely to vary internationally. There are, however, several drawbacks to using recidivism rates in this context; Recidivism rates underestimate the true amount (and cost) of crime in society, as a significant (but unknown) amount of crime is unreported/unsolved.

○
Recidivism rates do not tell us anything about whether the new offence committed was more or less serious than the previous one. Therefore, they are a fairly crude measure of effectiveness. Notwithstanding any concerns as to the appropriateness of these comparisons, one of the main findings of this updated review is that researchers are still some way off being able to perform these comparisons; only 10 of the 50 countries reported recidivism rates for prisoners, and due to the heterogeneity in the type of figures produced it was not possible for the authors to produce a meta-analysis. A recidivism reporting checklist is proposed as a means of standardising how countries produce this statistical information and its adoption should be recommended. Is the work clearly and accurately presented and does it cite the current literature? Yes

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
Competing Interests: No competing interests were disclosed.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
Author Response 07 Oct 2019 Denis Yukhnenko, University of Oxford, Oxford, UK

Reviewer 3
1) Recidivism rates underestimate the true amount (and cost) of crime in society, as a significant (but unknown) amount of crime is unreported/unsolved.

Our response:
We agree with this point, although it is not directly to the paper. We have explained the purpose of the paper in the introduction, which does not aim to estimate the amount and cost of all crime in society.
2) Recidivism rates do not tell us anything about whether the new offence committed was more or less serious than the previous one. Therefore, they are a fairly crude measure of effectiveness.
Our response: We agree. Some jurisdictions included in our review reported separate recidivism rates for different types of offences that provide some indication of recidivism severity. We have also included the recommendation to provide more crime categories in our reporting checklist.
3) Recidivism rates only capture instances of failure and do not take into account successes.
Our response: This is a valid point. Success measures could be especially helpful when reporting the outcomes of rehabilitation programmes, but beyond the scope of this paper. Our response: We agree, but again, this is outside the scope of this paper. 5) Recidivism rates often do not tell us if the person was returned to custody, and are therefore limited in terms of calculating the cost of crime to society.' Our response: A helpful point. Using different recidivism outcomes (reconviction, reincarceration, re-arrest) and different sources of information (official crime and healthcare statistics, surveys) can address this limitation. We have revised our reporting checklist so that it includes information on reincarceration and re-arrest (in addition to reconviction), if possible.
Institut national de psychiatrie légale Philippe-Pinel, Montréal, QC, Canada would be more appropriate.  (2013)). More details should be provided regarding the nature of the change in reporting practices.

3.
It would be helpful to clarify if each study examines a population or a sample (for example, in the Notes section of Table 1).

4.
If a study examines a sample rather than a population, it would be helpful to provide a confidence interval, if available.

5.
Discussion: Given that it is not the objective of the current manuscript, it may be premature to extrapolate on the reasons for a change in rates in the Republic of Ireland.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
Introduction: 1) It would be helpful to expand on the rationale for the review. In the introduction, the authors argue that "recently released prisoners often constitute a high-risk group that commit the majority of violent crimes" and then emphasize the public health burden of violent crime. However, much of the literature shows that recidivism events among recently released prisoners commonly involve justice administration offences (e.g., failure to comply with conditions of release). This may weaken the 'public health burden' argument and should be the subject of discussion in the manuscript. Methods: 1) A justification for the selection of the bibliographic database (MEDLINE) should be provided, given that MEDLINE is generally used for biomedical research.
Our response: This is a very useful comment. Although we followed the choice of database in our previous review, after consideration of this comment, we decided to add several other databases, including SAGE, since many criminological journals are indexed there, and redo the search. We also revised our search strategy to account for this [ Figure 1, 2) The abstract states that "three bibliographic indexes" were used, but this is not mentioned nor expanded upon in the text.
Our response: Thank you for highlighting this. We have added details on the search strategy and revised the list of indexes [see above]. The abstract now states: Old text is underlined): 'Using five bibliographic indexes, we carried out non-country-specific and targeted searches for 50 countries with the largest total prison populations.' 3) "If no new data had been identified for a particular country, we reported the rates from the original review": in what percentage of cases did this occur? How many new or updated estimates were included?
Our response: This was the case for only two countries. We have clarified this in the results section.
Rows 67-69: 'One additional publication (Graunbøl et al., 2010) with data on Finland and Norway was included from the previous review (Fazel & Wolf, 2015), as no new data were identified for these countries.' 4) It is unclear from the author list who "PS" is.
Our response: Added. 5) As per the PRISMA guidelines, it would be helpful to describe the method of data extraction (e.g., independently, in duplicate).

Our response:
The data was instructed independently by two researchers. We added the clarification in the methods section.
Rows 60-61 (old text is underlined): 'DY and SS conducted the search and independently extracted the data on country, sample selection, definitions of outcomes and rates.' 6) According to the reference list, several reports were available in foreign languages only. How were they translated?
Our response: They were translated by the native speakers who were either students or employees of Oxford University. We clarified this in the methods section.
Rows 61-63: 'The publications in languages other than English were translated with the assistance of native speakers, who were either employees or students at Oxford University.' Results: 1) We agree with Reviewer 1 that results are difficult to follow along and that an effort should be made to match up the text with the figures and tables.
Our response: Thank you for highlighting this. We have gone through the text and tables carefully to ensure consistency and increase clarity.
2) The rationale for Table 3 is unclear, given that it repeats information that is also provided in Table 2. Perhaps editing Table 2 or synthesizing the 2-year reconviction rates in the text would be more appropriate.
Our response: Thank you for the suggestion. We wanted to summarise 2-year reconviction rates as it is the most commonly reported outcome, and a separate table is more informative to the reader visually. Also, it follows the approach from the original review, which makes the findings more easily comparable.
3) Table 4 often mentions that there was a change in reporting practices, which often results in considerable changes in rates (e.g., Denmark: 29% (2005), 63% (2013)). More details should be provided regarding the nature of the change in reporting practices.
Our response: We have added brief descriptions of changes in reporting practices to Table  4.
4) It would be helpful to clarify if each study examines a population or a sample (for example, in the Notes section of Table 1). If a study examines a sample rather than a population, it would be helpful to provide a confidence interval, if available.
Our response: Thank you for pointing this out. Only two studies of the included studies examine samples (from Latvia and Italy). We added this information to the Table 1 and provided the information about 95% confidence intervals. 1) Given that it is not the objective of the current manuscript, it may be premature to extrapolate on the reasons for a change in rates in the Republic of Ireland.
Our response: We agree that this is not the main aim of the review. We have shortened this part of the discussion. is in the list of authors.
Our response: Thank you for pointing this out. Corrected. 6) Small item: In Table 1: Austria (Statistik Austria, 2018): "Reconviction: The conviction should happen during a follow-up period. "The use of "should" is confusing.
Our response: A helpful comment. We have rephrased this.