This research was supported by a grant from the Social Science and Humanities Research Council of Canada (SSHRC/CRSH).
The study findings were presented at the annual conference of the Association for the Treatment of Sexual Abusers, in Los Angeles, CA, in 2022.
The authors would like to acknowledge the significant contribution of: Fanny Audet-Paradis, Justine Daigle, Mariane Fay, Marc Gauthier, and Jeffrey Mathesius. The research team would also like to thank Mélissa Gravel, Nicolas Lebel-Carrier, and Brigitte Béland for their help and support in retrieving some of the material for this review.
Please address correspondence to: Patrick Lussier, School of Social Work and Criminology, Université Laval, Pavillon Charles-De Koninck, 1030 Avenue des sciences-humaines, Quebec City, Quebec, Canada, G1V0A6. Email: [email protected]
In the past, the Canadian government followed in the footsteps of its American counterpart by enacting “sex offender laws”. More recently, however, the Canadian criminal justice system has taken a different approach to the issue of sex offender recidivism (SOR), focusing on treatment, rehabilitation, and risk management. This evidence-based approach has been criticized for not doing enough to prevent convicted offenders from sexually reoffending. This criticism has not been addressed empirically, leaving open the question of whether this Canadian policy shift is associated with changes in the rate of sexual recidivism. The present study uses a meta-analytic framework to look at 185 Canadian-based studies involving over 50,000 offenders, making it possible to combine 226 sexual recidivism rates. After controlling for factors such as follow-up length and the independence of samples, weighted pooled recidivism rates have declined since the 1970s by more than 60%. This trend may have gone unnoticed because it is not related to the year of publication but to the period in which the data were collected. The findings have significant implications for criminal justice practices including the importance of using risk assessment tools that are regularly calibrated to reflect the evolution of sexual recidivism rates over time. Although the current study cannot provide firm conclusions about the factors responsible for this gradual drop, several hypotheses are discussed. Knowledge-based criminal justice practices, better training for professionals, and improvements in treatment programs may have had a subtle and cumulative impact on sexual recidivism rates. The importance of examining period effects on SOR using a comparative and international perspective is discussed.
Keywords: corrections, meta-analysis, longitudinal, period effect, policy, recidivism, sex offenders, sexual recidivism
For the times they are a-changin’
Bob Dylan, 1964
For more than 80 years, sex offender recidivism (SOR) has generated fear, media attention, public outrage and frustration with the criminal justice system. The public’s response has been to pressure the government to increase public safety (e.g., La Fond, 2005; Lieb et al., 1998; Lussier et al., 2021; Petrunik, 2002). As part of their attempts to deal with SOR, American and Canadian governments enacted a series of sex offender laws (SOL) aimed at preventing sexual recidivism. While the policies of the American and Canadian governments were similar in the beginning (Sexual Psychopathy Laws; e.g., Chenier, 2003), their paths have since diverged (e.g., Petrunik, 2003). The America policy context is characterized by stringent measures aimed at deterring reoffending (e.g., public sex offender registries, public notifications, and residency restriction laws). A strong research component followed that involved examining the rationale of these policies and their impact on offenders’ lives (e.g., Cohen & Jeglic, 2007; Zimring, 2009). The Canadian path has had at least three major turning points: (1) correctional-based research, collaboration with universities, and an emphasis on evidence-based studies (e.g., Lussier & Gress, 2014; Olver et al., 2013; Wormith & Olver, 2002); (2) recognition of the importance of community-based sentencing options (as opposed to an overreliance on custodial sentences) guided by evidence-based research on community risk management (e.g., Andrews, Bonta & Wormith, 2006; Reid, 2020) while reaffirming the importance of correctional-based treatment intervention for those receiving custodial sentences; and; (3) a well established and strong tradition of research on sexual violence and abuse that has led to, among other things, the development of therapeutic approaches, risk assessment methods and tools, specialized training, and classification models (e.g., Cormier & Simons, 1969; Hanson et al., 1993; Marshall & Barbaree, 1990; Proulx et al., 1997; Quinsey et al., 1981; Seto & Barbaree, 1999). These turning points may explain, at least in part, why Canadian policies to prevent sexual recidivism moved from stringent and populist measures to evidence-based initiatives. What remains unclear is whether emphasizing evidence-based initiatives has been associated with changes in SOR rates.
Canada’s first SOLs were enacted in the 1940s in response to growing concerns regarding SOR. The Criminal Sexual Psychopath Law (CSPL), enacted in 1948 and heavily influenced by Massachusetts’ Sexual Psychopath Law, gave judges the power to impose indeterminate custodial sentences on individuals convicted of sex crimes, during which they would receive specialized therapy (Greenland, 1972). At the time, it was believed that the relatively stable nature of underlying risk factors (e.g., social anxiety, feelings of inadequacy, masculine insecurities, early traumatic experiences) made SOR almost inevitable (e.g., Pascoe, 1961), suggesting that prison sentences alone would not have a deterrent effect. In Canada, the CSPL was enacted before the development of evidence-based strategies, and so psychiatrists were not well positioned to offer appropriate services. Indeed, psychiatrists recognized the state of treatment strategies at the time and were critical of both the CSPL and the term "sexual psychopath", which was later changed to dangerous sexual offender. Scholars and psychiatrists expressed frustration with the law highlighted a series of ethical, legal, clinical, and practical issues with SOL (e.g., Cormier & Simons, 1969; Greenland, 1972; Marcus, 1966)1. For example, despite the CSPL being predicated on the use of specialized treatment, it took almost 20 years before such programs were gradually introduced in Canadian prisons. In the meantime, “sex offender treatment” consisted mainly of unstructured group psychotherapy sessions in psychiatric units and outpatient clinics and there was little to no evidence to support this approach (e.g., Marcus & Conway, 1969; Turner et al., 1958). Legislation to deal with dangerous offenders would eventually be described as instituting confinement disguised as therapy (see Price, 1970).
While some described Canada’s experience with dangerous offender legislation as a complete failure (Greenland, 1984), others argued that the laws might have been successful if they had not been misdirected (Jakimiec et al., 1986). For example, despite the CSPL being intended for dangerous sexually violent offenders, Cormier and Simons (1969) observed early on, paradoxically, that such persons were not considered good candidates for a sexual psychopath designation given their poor treatment prognosis. That paradox was further discussed by Price (1970) who highlighted a dilemma that still resonates today: policymakers assume that SOR is inevitable, scholars argue that SOR is rare, and practitioners are stuck in between, forced to apply an imperfect law using the means and knowledge available to determine offenders’ dangerousness. At the time of the first Canadian SOL, there was very little research in Canada on sexual offending, let alone prospective longitudinal research on SOR2. One barrier to developing effective policies was that offenders designated as “sexual psychopath” or “dangerous sex offenders” were reluctant to participate in research, possibly due to fear of attracting attention from peers (Marcus & Conway, 1971). Knowledge about sexual offending was based mainly on individuals in mental health settings or under the care of psychiatrists and thus (a) not representative of all individuals convicted of a sex crime and (b) based on psychiatrists’ intuition and clinical judgement rather than empirical research. Moreover, some criteria for admission to a mental health setting (e.g., paraphilia, personality disorder) are also risk factors for SOR, which could lead to higher recidivism rates. It was only during the mid-1970s that correctional-based sex offender treatment programs gradually emerged across Canada. These programs were based on the experiences of psychiatrists and thus not aimed at all individuals convicted of a sexual offense but rather those who demonstrated paraphilic sexual preferences (Freund et al., 1972). The first Canadian therapeutic models advocated for the use of behaviorally-oriented techniques to modify deviant sexual preferences (e.g., Marshall, 1971; 1973). In effect, the first therapies assumed that (a) most persons involved in sexual offenses had deviant sexual preferences and (b) such preferences were responsible for their perpetration of a sexual offense. Reliable empirical evaluations of such assumptions were lacking.
By the end of the 1970s, nearly 40 years after the first SOL, Canada had yet to develop a clear national strategy for providing therapeutic services to persons convicted of sexual offenses. Not only were there no clear guidelines regarding the provision of treatment for sex offenders in prison but clinical assessment practices were also inconsistent (Wormith, 1986). Clinical research on individuals convicted of a sex crime increased and the content and therapeutic approach of treatment programs gradually expanded during the 1980s to reflect the heterogeneity of the offender population and the diversity of risk factors. More specifically, the strictly behaviorally oriented perspective made room for more cognitive and psychoeducational methods (for a review, Marshall & Laws, 2003). Treatment programs no longer focused exclusively on changing offenders’ sexual arousal and interests. This shift was influenced, among other things, by the recognition that sexual offending was multifactorial (e.g., Hall & Hirschman, 1991) and that effective treatment required dealing with a number of cognitive (e.g., cognitive distortions), affective (e.g., empathy toward victims), behavioral (e.g., anger management, social competence), interpersonal (e.g., attachment and intimacy issues) and sociocultural (e.g., sexism) factors (e.g., Marshall, Laws & Barbaree, 1990). These changes were influenced by new theories regarding the origins and development of sexual offending (e.g., Langevin & Lang, 1985; Laws & Barbaree, 1990; Malamuth, 1986; Marshall & Barbaree, 1984; Quinsey, 1984). While there were still few publications on SOR during the 1980s, researchers working to develop treatment programs were laying the foundation for prospective longitudinal studies regarding the impact of treatment on SOR.
Following a series of rare but extreme and violent cases of sexual recidivism against children (e.g., Lieb et al., 1998; Petrunik, 2002), the issue of SOR returned to Canada’s policy agenda in the 1990s. Although scientific evidence indicated that SOR rates were much lower than the public believed (e.g., Furby et al., 1989), there were numerous demands for additional community protections against convicted offenders. The Canadian government responded with a more cautious version of the Sex Offender Registration and Notification (SORN) laws in the United States by implementing a series of measures culminating with the implementation of a national non-public sex offender registry (e.g., Lussier et al., 2021; Murphy et al., 2009; Petrunik, 2003)3. Against the backdrop of these new SOLs, Canadian corrections reaffirmed its commitment to treatment and rehabilitation, this time with an emphasis on evidence-based practices (e.g., Gendreau et al., 1996). A group of Canadian scholars developed the risk-need-responsivity (RNR) model to guide practices and service delivery (e.g., Andrews & Bonta, 1990). The RNR emphasizes risk, risk assessment, and risk management through the identification of risk factors for recidivism and the modification of dynamic risk factors (e.g., Andrews et al., 2011; Zamble & Quinsey, 1997) using a cognitive-behavioral treatment approach (e.g. Porporino, et al., 1991). National guidelines were proposed in the mid 1990s to help develop Canadian sex offender treatment programs (e.g., Williams, 1996). These programs were predicated on the relapse prevention model (e.g., Laws, 1989; Pithers, 1990; Proulx et al., 1996), which maintained that a research component was a critical part of any treatment program (e.g., Abracen & Looman, 2004). This integrated framework made it possible for criminal justice professionals (e.g., psychotherapists, parole/probation officers) to focus on an offender’s current situation, with the goal of preventing a sexual re-offense by intervening on the dynamic risk factors predictive of sexual recidivism (e.g., Hanson & Harris, 2000).
Measurement of SOR became pivotal in a field of research trying to establish its credibility. The seminal study by Furby et al. (1989) suggested that the best way to evaluate the efficacy of sex offender treatment programs was to look at rates of sexual recidivism. Concerns about whether sex offender treatment programs had a significant impact on these rates led researchers to advocate focusing on the risk of sexual recidivism and management of this risk (e.g., Quinsey et al., 1995; Quinsey et al., 1998). Examining sexual recidivism became such an integral component of research on sex offending that some scholars moved away from evaluating treatment programs and towards establishing which risk factors were most important for recidivism, spurring the development and validation of various risk assessment tools (e.g., Boer et al., 1997; Hanson & Thornton, 1999; Quinsey et al., 1998). What started as a way to evaluate treatment programs became a field of study and reinforced SOR as a risk-management issue. This shift was somewhat aligned with RNR principles espoused by in Canadian corrections and involved identifying the base rate of sexual recidivism within samples as well as identifying which individual-level static and dynamic risk factors predicted sexual recidivism (e.g., Hanson & Harris, 2001; Proulx et al., 1997).
By going down the path of actuarial risk assessment, the effectiveness of Canada’s response to sexual offending became contingent on establishing the true base rate of sexual recidivism. An actuarial tool is validated based on a specific base rate of sexual recidivism within a specific sample. An assumption is made that if the sample represents the population from which they were drawn, then findings can be generalized to individual cases to make, for example, decisions about sentencing. Even if this generalizability assumption is met, another assumption is that the base rate of sexual recidivism is stable over time, such that information from the same risk assessment tool can be used to infer the same information at different periods in time. If the base rate of recidivism decreases over time, so to does the predicated probability of recidivism for a particular score on a particular actuarial risk assessment tool. If adjustments for this discrepancy are not made, then risk prediction becomes less accurate (see Abbott, 2017). In effect, a ‘moderate’ score on a risk assessment tool at time t will be less informative of the likelihood of sexual recidivism at time t+X if the base rate of sexual recidivism changed during this time. If base rates change, the utility of actuarial risk assessment tools change. Yet systematic examinations of changes in base rates are poorly developed and focus primarily on factors such as study length and the definition of recidivism (e.g., Furby et al., 1989). Researchers have reported that base rates for juvenile offenders have declined, suggesting the presence of period/cohort effects (for a review, see Caldwell, 2016; REDACTED), but similar research for adult offenders is lacking. Some research suggests that period/cohort effects could also be occurring with adult offenders, which would require significant adjustments to the base rates used in actuarial risk assessments (e.g., Lee & Hanson, 2021). Indeed, research attempting to validate the Static-99R, an actuarial risk assessment tool, reported an overestimation of the risk of recidivism (Helmus et al., 2021). While these findings could simply be the result of cohort effects (e.g., predictors of recidivism could be different across cohorts), the drop in estimated risk by an actuarial instrument could be partly explained by a decline in sexual recidivism rates over time. Although risk assessment is a daily practice in Canadian corrections (e.g., Fréchette & Lussier, 2021), there is little understanding of whether this risk has changed. This is especially concerning given ongoing legal concerns about actuarial risk assessment practices (e.g., Ewert v. Canada).
Past research emphasized the importance of individual-level risk factors (e.g., personality traits, sexual functioning, criminal history) to SOR, but such studies do not account for possible period effects. The examination of period effects on recidivism rates has been a neglected area of investigation (Caldwell, 2016; REDACTED). The goal of this study is to examine the evolution of sexual recidivism rates in Canada over the 80-year period since the enactment of the first SOLs. To do this, all studies conducted between 1940 and 2019 that collected and reported sexual recidivism data on Canadian samples of “sex offenders” were analyzed using a quantitative meta-analytical framework. Weighted pooled estimates of rates of sexual recidivism were examined and the presence of possible study moderators (e.g., sample size, length of the follow-up period, type of sample, measure of recidivism) and their impact on sexual recidivism were considered. Canada is a good case study to examine changes in sexual recidivism base rates because it has produced over 20% of all SOR research (REDACTED) and has been at the forefront of key developments in risk assessment and management practices. It is not just the volume of the research that is important, but how it affected criminal justice practices over time, from intake assessment practices and the introduction of actuarial risk assessment instruments to the development of sex offender treatment programs and community re-entry programs. The study is useful in international contexts because actuarial risk assessment is a global practice and the approach used in the current study provides a template for researchers seeking to establish base rates of SOR in other countries.
The inspection of recidivism rates reported in Canadian studies has become almost inevitable against the backdrop of significant social, cultural, judicial, legal and policy changes (e.g., Knack et al., 2021; Lussier et al., 2021; Murphy et al., 2009). The tendency to pool sexual recidivism data from various countries and across periods (e.g., Hanson & Bussière, 1998; Lösel & Schmucker, 2005) may have masked significant trends. For instance, populist groups have suggested that the Canadian response model is overly lenient and not representative of best practices for the prevention of SOR (Lussier & Mathesius, 2019). Our findings make it possible to compare Canadian practices with practices in the United States and elsewhere that advocate a containment-focused rather than a treatment-oriented approach stressing the importance of rehabilitation (e.g., English, 1998). More fundamentally, SOR base rates are routinely used across Canada in risk assessment (e.g., actuarial assessment) and the relative absence of any sort of monitoring of the evolution of these rates is a shortcoming of such practices. Given the progress in scientific knowledge, training of criminal justice professionals (e.g., probation/parole officers), parole board decision-making, and risk management practices, it is reasonable to expect that sexual recidivism rates have declined over time. It could, however, be argued that, as a result of the Canadian rape law reform that led to several legal and procedural changes in the 1980s (e.g., Tang, 1998), it became easier to convict a person of a sex crime (see however, Gunn & Linden, 1997; Schissel, 1996). As well, measurement of recidivism has significantly improved over the years, most notably since 1972 due to the advent of the national Canadian Police Information Centre (CPIC), a centralized computerized system of criminal records operated by the Royal Canadian Mounted Police (Schellenberg, 1997). Using a centralized database avoids underestimating recidivism, for example, by failing to count offenses that occurred in different jurisdictions. Most Canadian researchers doing SOR research after the 1970s relied on CPIC data to measure sexual recidivism (charges/conviction for a sex crime).
To make sense of the growing body of SOR research in Canadian studies, the current study utilizes a meta-analytic framework which is a more rigorous alternative to narrative reviews (see, Glass, 1976). This method has been used in criminology, especially since the 1990s, to address policy-relevant questions, such as the identification of risk factors for criminal recidivism (e.g., Cottle et al., 2001; Gendreau et al., 1996), the impact of correctional treatment (e.g., Andrews et al., 1990; Lösel & Schmucker, 2005), the differential impact of sentencing practices (e.g., Villettaz et al., 2006), and the predictive validity of risk assessment instruments (e.g., Bonta et al., 2014). Interestingly, these well-known and oft-cited studies overlooked one key component of these effect sizes - recidivism rates. SOR research is dense, fragmented and contradictory (REDACTED) and the absence of systematic reviews on the topic has contributed to the presence of myths and misconceptions about SOR (see Soothill, 2010). Some researchers provided narrative reviews of SOR research (e.g., Furby et al., 1989; Fortune & Lambie, 2006; Greenberg, 1998; Proulx & Lussier, 2001), while others relied on the pooling of secondary data stemming from a very small subset of samples of offenders (e.g., Harris & Hanson, 2004). These methods provided an incomplete and at times conflicting picture of the phenomenon. A meta-analysis is preferred over these alternative review methods for several reasons (for a discussion, Schmucker & Lösel, 2011). First, a meta-analysis limits potential biases, such as researchers focusing on certain authors, samples, and studies or criticizing studies reaching conclusions that are not in line with their own biases. Second, meta-analysis can statistically control for various critical methodological aspects of studies (e.g., study setting, length of the follow-up period) that can account for the contrasting recidivism rates observed. It also avoids confirmation biases in which researchers actively search for studies that support their viewpoint. Third, meta-analysis allows for the pooling of findings from various studies while accounting for the sample size of each, which is critical in this field of research given the number of small-scale studies based on highly specific samples. Researchers have since moved to a meta-analytical approach of pooling recidivism rates, whether for justice-involved adolescents (e.g., Caldwell, 2010; Caldwell, 2016; McCann & Lussier, 2008; REDACTED), females (e.g., Cortoni, Hanson & Coache, 2010), or adults (e.g., Fazel & Wolf, 2015; Katsiyannis et al., 2018).
Data in this article are part of a larger study examining SOR worldwide since 1940. The search for relevant empirical studies initially involved two major terms: sex offender and recidivism. However, the scientific literature is complex, given its sociohistorical and multidisciplinary aspects, and a broader approach was needed. Our analysis relied on (a) researchers’ experience and knowledge about the topic, (b) examination of the scientific literature across academic disciplines, (c) consideration of studies across several decades (e.g., 1940s, 1950s, 1960s), (d) an examination of handbooks and encyclopedias dealing with related topics (e.g., Criminology, Corrections; Correctional Treatment and Intervention; Sexual Offending; Sexual Deviance; Sexual Disorders; Law and Mental Health; Abnormal Psychology), and (e) an examination of existing literature reviews and previously conducted meta-analyses4. A list of the search terms is in Appendix I. Preliminary examination of the scientific literature that reported SOR rates revealed several challenges. First, publications appear in scientific journals across multiple disciplines, mainly (a) law and legal studies, (b) medicine and psychiatry, (c) psychology and cognitive sciences, and (d) criminology, criminal justice, and sociology. While certain journal databases were prioritized in prior meta-analytic research on sexual offending (e.g., PsycINFO), after careful inspection it was decided that a broader approach was necessary. We reviewed previous meta-analyses of criminal and sexual recidivism (e.g., Hanson & Bussière, 1998; Lösel & Schmuker, 2005) and identified over 80 databases that could be useful. To avoid using overlapping databases, a list of all academic journals cited in quantitative meta-analyses of topics closely related to the current study (e.g., predictors of sexual recidivism; impact of sex offender treatment) was created. Gold Rush software, a library content comparison tool, was used to determine an optimal list of databases5 with minimal overlap. The literature search was expanded to include unpublished material (i.e., the “grey literature”) that is not captured by typical computerized retrieval systems (e.g., Conn et al., 2003) and includes, for example, government reports, dissertations, conference abstracts and proceedings, and technical or brief reports to funding agencies6.
---Insert Figure 1 about here---
The original search yielded more than 20,000 documents (Figure 1), about 20 % of which were part of the grey literature. A four-step procedure was used to identify documents relevant to this study. In the identification phase, relevant references from all sources and databases were inspected and then imported to an electronic document. In the screening phase, the scientific literature was examined based on general descriptive information about the study. Given the large number of documents to be screened, the focus was on examination of (a) study title, (b) abstract, executive summary, or study highlights, and (c) study keywords. Research assistants determined whether the study was relevant or potentially relevant and a training session was conducted during which research assistants and the lead author coded the same set of studies until acceptable inter-rater agreement was reached. Any document that was not a duplicate and was either directly or potentially relevant to the current study was then extracted. A total of 3,026 documents were selected for the next stage. The third stage, determining the eligibility of these documents, consisted of searching, accessing, reading, and analyzing the 3,026 documents. Studies were included if they met several criteria. First, the full report had to be available. Given the scope of this study, locating older studies and unpublished material was challenging. Efforts to avoid discarding research that was difficult to obtain, research assistants contacted lead authors and co-authors, searched websites such as Google Scholar and ResearchGate, used a reverse search approach to allow triangulation of the study content using other sources, and contacted researchers who might have a copy of the study. Second, research assistants determined whether the study was empirical (e.g., not a narrative review, critical review, meta-analysis, commentary, etc.). Third, research assistants determined whether the study included a measure of recidivism. Studies that did not report at least one sexual recidivism rate were excluded. The way sexual recidivism was measured (e.g., percentage, proportion, frequency) had no bearing on whether a study was included in the meta-analysis. However, the measure of recidivism used was coded to account for variability across studies. Fourth, based on a careful examination of the methodological characteristics of the study, research assistants determined whether the study design was longitudinal. The study had to be longitudinal in that recidivism was not based on the presence of a prior history of sexual offending, but rather whether they had sexually reoffended during some follow-up period. One of the reasons for this was to avoid studies that recruited individuals precisely because of their history of prior sexual recidivism. Measurements of recidivism based on a person’s past arrests/convictions were excluded. Fifth, research assistants determined whether the sample included participants who lived in Canada. The study only sampled persons in Canada at the time of recruitment and international studies were excluded. A total of 187 studies met these criteria. It was then determined whether the study included a quantitative measure of sexual recidivism. Twenty-one studies were excluded because they did not measure sexual recidivism or did not report the rate of sexual recidivism or the number of sexual recidivists. The current study is thus based on 165 studies. As these studies sometimes reported multiple sexual recidivism rates (e.g., for subgroups of the sample), there were 226 recidivism rates reported across these studies.
A critical assumption of meta-analyses is the independence of observations, which may be violated if multiple estimates of sexual recidivism stem from the same sample or overlapping samples (e.g., Cheung, 2019). It is not uncommon for a single study to report multiple recidivism rates or for multiple studies to stem from the same (or very similar) sample. The presence of non-independent estimates of recidivism raises critical issues, such as whether the scientific literature is biased toward findings stemming from a few samples. Independence of observation was mainly threatened in instances where the study included more than one group of offenders (e.g., “child molesters”, rapists, exhibitionists, treated offenders, treatment dropouts) and thus reported more than one sexual recidivism rate, and cases where researchers used the same sample (or a similar sample) more than once in different publications. A computerized database was created to organize and classify samples by province, institution/treatment program, etc. Each sample was given an identification number. Missing data or vague, uncertain, and imprecise reporting of methodological details made it difficult to draw firm conclusions about the independence of some samples.
Due to questions about the independence of samples, sexual recidivism rates were analyzed in three different ways. First, the total number of data points (i.e., sexual recidivism rates) (k = 226) in each study was examined as a representation of the Canadian literature as a whole. These data points were not necessarily based on independent observations but made it possible to investigate whether the over-representation of certain studies in the literature might have created biases in the perception of sexual recidivism rates. Second, recidivism data were pooled within publications (i.e., one data point per publication; see analytical strategy for the pooling method)7. All data points stemming from the same institution/program (i.e., treatment program, penitentiary, etc.) were regrouped to account for dependence of samples. One data point was selected for each study sample8: at this stage recidivism rates were not pooled but were selected based on several criteria9. At the end of this process, there were 88 relatively independent data points. Third, it could not always be established if different publications were based on overlapping samples simply by reading each study’s methodology, so additional information was collected from other sources (e.g., internet, related publications, study authors), which sometimes included making inferences about authors’ affiliations or comparing descriptive sample statistics. After the final investigation, and using a very conservative approach, 53 data points were defined as independent. The different sampling strategies resulted in the following samples: (a) total number of non-independent observations (k = 226) representing the literature as a whole; (b) total number of relatively independent samples (k = 88); (c) total number of independent observations using a very conservative approach (k = 53).
Classifications of sex crime types have been inconsistent across time (see Appendix I) and therefore we did not examine SOR rates across sex crime types. Some researchers relied on legal terms, which have changed over time with respect to both definitions and terminology (e.g., rape, sexual assault); others used clinical terms (e.g., pedophiles), behavioral terms (e.g., hands-on), context of the abuse (e.g., intrafamilial) and victim characteristics (child molesters). These labels are not mutually exclusive, which further complicates distinguishing recidivism across crime type. For example, a single offense could be an intra-familial, hands-on, pedophilic offense against a child. Some researchers have relied on one specific subgroup, multiple subgroups and mixed samples, but not all researchers report this information, meaning that the composition of the study sample is unknown. The scientific literature most commonly distinguishes rapists (also sexual aggressors/sexual assaulters against women) and child molesters (also sexual aggressors against children, child sexual abusers and pedophiles) and review studies have not reported statistically significant differences between these two groups (see Hanson & Bussière, 1998). Categorization of perpetrators (e.g., rapists vs. child molesters) is typically based on the most recent sexual offense perpetrated, which does not recognize instances where the same individual perpetrated each crime-type (e.g., Ahlmeyer et al., 2000; Stephens et al., 2017). Examining sexual recidivism rates across types is a valuable avenue for research, but we had reservations about examining this line of analysis in the context of the current study. Canadian policymakers have not yet constructed SOLs that create specific sanctions for an explicitly targeted subgroup of perpetrators. When policymakers did target a specific group, it was based on perceived dangerousness (e.g., sexual psychopathy, dangerous sexual offenders). Therefore, we focused on factors more directly tied to past and current policies (e.g., age, proportion of sexual recidivists).
Sampling period. Publication year traditionally has been used to identify the period that a study is from. Because SOR research is based on longitudinal data, publication year may not reflect the timing of the study since follow-up time can extend for years or even decades. It is also well known that, although cumulative recidivism rates continue to increase over a long follow-up period, yearly recidivism rates are highest in the first year after the start of the follow-up period (e.g., Prentky et al., 1997). Therefore, the year marking the start of the sampling period was used to capture the time period of the study (range = 1939-2010). Average length of the sampling period was 6.8 years (SD = 6.3)10. For over 90% of cases where the information was available, the year marking the start of the sampling coincided with the year marking the start of the follow-up period (the period during which offenders were at risk of reoffending; n = 100; 92.6%)11. The study was conducted to detect period effects across decades that witnessed significant policy and legal changes in SOL (see, REDACTED). Trends in sexual recidivism rates were examined across the following periods: (1) 1940-1979; (2) 1980-1989; (3) 1990-1999; (4) 2000-2009; (5) 2010-2019. The early studies (i.e., 1940-1979) were not divided into ten-year intervals because there were relatively few independent studies during that period. Instances in which the sampling period was not clearly reported (n = 20; 10.7%) were coded to indicate this information was missing but were not excluded.
Study Moderators. Research assistants used a coding instrument developed by the research team. An initial version of the coding instrument was created by the senior researchers based on (a) their knowledge of the scientific literature, (b) experience conducting such research, and (c) an examination of sex offender recidivism research from various decades, starting with the 1940s. The initial version was pre-tested by three independent coders using 40 randomly selected studies. Adjustments to the coding instrument were made to solve coding issues (e.g., lack of clarity, statements too vague/precise, missing scoring categories, scoring sheet too precise given the information available). The coding form was tested a final time using 150 publications12. Each of these studies was coded twice by four raters. Kappa coefficients (Κ) and Intraclass Correlation Coefficients (ICC) are reported for the variables used. Given that the inter-rater agreements were conducted as part of the larger study, they do not reflect the coding of Canadian studies specifically. This could explain the lower inter-rater agreement for some moderators (e.g., type of study setting) that might have been more challenging to code for international studies.
---Insert Table 1 about here---
For this study, four categories of study moderators were examined (Table 1): (1) publication details, (2) author details, (3) sample details, and (4) recidivism details. Publication details included the type of publication (e.g., peer-reviewed journal; Κ = .94), year of publication (Κ = .96), and language (K = 1.00). Author details included the main affiliation of the lead author (Κ = .80). Sample details included the study setting (Κ = .67), sample size (Κ = .85), year the sampling started (Κ = .99), history of sexual offending prior to inclusion in the sample (ICC = .85), and study location (e.g., province where sample of offenders was collected)13. History of sexual offending refers to the percentage of the study sample with a prior conviction for any type of sexual offense. This variable was dichotomized to reflect instances where the prevalence of sexual recidivists recruited into the sample was 25% or less. This information was reported in only approximately 20% of studies. Information on recidivism included the average length of the follow-up period (Κ = .71) as well as the indicator used to measure recidivism (Κ = .81). A measure marking the start of the follow-up period would have been useful, but that information was rarely reported (missing for 158 publications). For those studies reporting both the year sampling started and the year the follow-up period started, the average gap between the two was less than one year (Mean = .88; SD = 2.4). This suggests that the sampling took place when offenders were released.
Sexual recidivism. The current study focused on changes in sexual recidivism rates over time. An aggregate measure was used to determine the sexual recidivism rate for a sample/subsample. If a study reported multiple rates using various definitions or sources of information, the highest rate reported was used. Situations where the criteria for measuring sexual recidivism included nonsexual violent offenses (e.g., assault homicide, threats) were excluded. Inter-rater agreement on the coding of sexual recidivism rates across publications (k = 141) was relatively good (ICC = .86; 95% C.I. = .81-.90).
Data analyses were conducted using IBM SPSS statistic 27 and Stata 16.1 (Stata Corp., College Station, TX, USA). The metaprop and metapreg commands were used to pool prevalence estimates of recidivism rates (e.g., Nyaga et al., 2014).
Weighted pooled base rate estimates. Meta-analysis methods can be used to get a more precise estimate of the prevalence of a phenomenon (e.g., Stoltenborgh et al., 2011). While pooled prevalence estimates using a meta-analytical framework have become increasingly popular in the past decade, there is no consensus about the best way to perform such analyses (Migliavaca et al., 2020) and various strategies have been proposed (e.g., invariance method, logit transformation). These methods can create measurement issues, especially with prevalence estimates that are very low (or very high) or are extremely sparse (e.g., see Newcombe, 1998). More specifically, these methods can produce confidence intervals that falls outside the 0-1 range while squeezing the variance towards 0 (Barendregt et a., 2013). For this study, the Freeman-Tukey double-arcsine transformation was used, which can address these limitations while stabilizing variance in the pooled prevalence estimates. The double arcsine transformation (see Nyaga et al., 2014) was obtained using the following formula:
where n refers to the number of cases in the category (recidivist) and N refers to the total sample size. The variance of t can be obtained with the following:
The values are then transformed back to the original unit of proportions using the following formula:
Heterogeneity assessment. Analyses of moderator effects were conducted using random effects models that allow for the possibility that random differences between studies (e.g., variation in settings, procedures, measurements, etc.) are not limited to sampling error. This assumption is reasonable given the variability embedded in the measurement of criminal recidivism rates across jurisdictions (e.g., reporting crimes to the police, police investigation, providing evidence, the courts, plea bargaining, etc.). Inconsistency of pooled estimates was examined using the Q statistic (Hedges & Olkin, 1985), tested whether studies reported the same recidivism rate.
The weighted pooled sexual recidivism rates for the 1940-2019 period are reported in Table 2. Using findings from non-independent samples that included 55,944 offenders, a total number of 226 estimations of sexual recidivism rates were pooled. Overall, the mean of the pooled base rate of sexual recidivism was .14 (95% C.I. = .13-.15). Rates were heterogenous across sampling periods (Q = 98.09, p < .001) and steadily decreased from the 1940-1979 period (Mean = .23; 95% C.I. = .18-.28) to the 2010-2019 period (Mean = .07; 95% C.I. = .05-.09), a drop of 69.6%. Pooled sexual recidivism rates for the independent samples were relatively similar. There were 88 relatively independent samples, with a total of 29,361 offenders. The mean recidivism rate for independent samples was .13 (95% C.I. = .11-.15). The rates reported for the independent samples were also heterogeneous (Q = 29.73, p < .001), which is reflected in a 66.7% drop in sexual recidivism rates between the 1940-1979 and the 2010-2019 periods. Findings were very similar for the smaller subset of independent samples (k = 53; n = 18,335). The pooled sexual recidivism rate was .12 (.10-.15) and dropped from .20 (.12-31) for the 1940-1979 period to .07 (.05-.11) for the 2010-2019 period, a 65% drop (Note that the 2010-2019 included only one study in the subsample of conservatively independent studies). Overall, these findings suggest that sexual recidivism rates dropped by 65-69% between the 1940-1979 and 2010-2019 periods.
--- Insert Table 2 about here ---
Next, we examined nine potential study moderators irrespective of the study period. These bivariate analyses were conducted to determine if sexual recidivism rates were statistically associated with publication, author, sample, or recidivism characteristics. These analyses were conducted using data from independent samples (k = 88; k =53) given that it is an assumption of the statistical tests conducted. Findings based on independent samples (k = 88) are reported in Table 3. Five of the nine potential study moderators were statistically significant: sexual recidivism rates did not vary in relation to the publication year or the study sample size but varied significantly for affiliation of the lead author, study setting, study location, length of the follow-up period, and measure of recidivism used. Lead authors affiliated with a mental health institution reported higher recidivism rates (Mean = .17, 95% CI = .10-.25) than those affiliated with other institutions, such as correctional (Mean = .11, 95% CI = .09-.13) and university (Mean = .14, 95% CI = .12-.17). Samples of offenders from correctional institutions demonstrated the lowest sexual recidivism rates (Mean = .11, 95% CI = .09-.13) compared to samples from other programs/institutions (Mean range = .15-.16). The analyses also revealed that samples with a higher proportion of offenders who at the time of inclusion in the sample would be considered a “sexual recidivist” (Mean = .15; 95% CI = .10-.21) showed higher recidivism rates than samples with a lower proportion of sexual recidivists (Mean = .08, 95% CI =.05-.13), but the difference was not statistically significant (p = .10). The lack of a significant statistical association should be interpreted with caution given the importance of missing information for this variable. There were significant variations across Canadian provinces, with studies conducted in Saskatchewan (Prairie region) showing higher recidivism rates (Mean = .24, 95% CI = .20-.28) compared to those from other provinces (Mean range = .05-.14). Not surprisingly, studies with a mean follow-up period of less than 48 months reported lower sexual recidivism rates (Mean = .10, 95% CI = .06-.15) compared to studies with a mean follow-up period of at least 120 months (Mean = .21, 95% CI = .17-.24). When we focused on the smaller subset of samples (k = 53) where we were more confident of their independence, other than observing that recidivism rates no longer varied significantly across study settings (Q = 3.20, p = .20), all other findings remained the same as reported above.
--- Insert Table 3 about here ---
The length of follow-up is inherently tied to the base rate of sexual recidivism observed. This could confound findings of the presence of a period effect as older studies are often characterized by a longer follow-up period14. Sexual recidivism rates are plotted against the mean follow-up time (in months) according to the different sampling periods (See Figures 2a-c). A regression line was fitted to each of the figures to determine whether sexual recidivism rates varied as a function of the length of the follow-up period. It showed that older studies (1940-1979 period) reported higher sexual recidivism rates than more recent studies with a comparable follow-up period. In fact, the fitted regression lines for the 1940-1979 cohort showed the highest intercept (0.20) among all the sampling periods. In other words, the 1940-1979 periods included samples that showed a higher risk of sexual recidivism. While the 1980-1989 period showed the steepest recidivism rate slope, that effect was limited to the non-independent samples (k = 226; see Figure 2a), meaning this effect could be attributable to a particular sample or a few samples whose recidivism rates were higher compared to other samples from the same period.
---Insert Figure 2a-b-c about here---
To further clarify the impact of follow-up length on sexual recidivism rates across sampling periods, a series of meta-regression analyses were conducted (Table 4), making it possible to examine the effect of the follow-up time across sampling periods while controlling for other potential moderators. All moderators that were significantly associated with sexual recidivism rates were considered (i.e., affiliation of the lead author, study setting, study location, length of the follow-up, measure of recidivism). Given the relatively small sample size, the number of moderators simultaneously analyzed in the meta-regression models was kept to a minimum15. To further inspect the drop, a series of dummy variables were created to represent study periods. The creation of dummy variables helped examine potentially non-linear trends in sexual recidivism rates across periods. The 1980-1989, 1990-1999, 2000-2019, and unknown sampling periods were each compared against the 1940-1979 reference category. Note that the 2000-2009 and 2010-2019 periods were merged due to the limited number of studies.
In the first meta-regression model (Model I), only sampling period was included. The unstandardized beta coefficients are significant for each time period compared and they also increase in magnitude with the passage of time, from -.060 (1980-1989) to -.137 (2000-2019). Three additional covariates were added in successive meta-regression models: length of the follow-up period, study setting, and sample size. Of these three covariates, only the length of the follow-up period was statistically significant (p < .001). Controlling for the length of the follow-up period, all three later study periods were significantly associated with lower sexual recidivism rates compared to the earliest study period. However, controlling for the average length of the follow-up period, which is also a variable measuring time, impacted the interpretation of the relationship between study period and SOR rates in two ways. First, the unstandardized beta coefficient for 1980-1989 decreased, suggesting that the drop was somewhat steeper when adjusting for the length of the follow-up period. This could be attributable to the fact that studies between 1980-1989 averaged somewhat longer follow-up periods than those from 1940-1979. Second, the beta coefficients for the 1990-1999 and the 2000-2019 periods became nearly identical, suggesting that the drop somewhat leveled off between 2000-2019. Inclusion of two additional moderators – affiliation of the lead author and study location – did not affect this finding (not shown). Findings remained unchanged when the smaller subset of independent samples (k = 53) was used.
---Insert Table 4 about here---
Our examination of the evolution of sexual recidivism rates in Canadian studies across eight decades reinforced previous observations that sexual recidivism is relatively rare. The general public’s perception that convicted offenders are all on a trajectory of life-course persistent sexual offending is incorrect (e.g., Mancini & Budd, 2016). On average, the weighted pooled sexual recidivism rate within studies of Canadian samples is slightly above 10%. However, this weighted pooled sexual recidivism rate varies across study characteristics. This observation is an important new contribution to the literature as prior studies tended to focus on how rates vary across individual-level risk factors without addressing study characteristics and measurement differences. Features of the study, such as the sampling period, lead author’s affiliation, and study setting, were associated with sexual recidivism rates. The interplay between offender and study characteristics, however, should be considered. For example, samples of offenders drawn from secured mental health hospitals tend to be overrepresented with individuals characterized by multiple risk factors of sexual recidivism, such as psychopathy, paraphilias, and a criminal history including past sexual offenses (e. g., Proulx et al., 1997; Quinsey et al., 1995). Importantly, sexual recidivism rates reported in Canadian studies have significantly declined since the 1970s. The nature of the data and the methodological considerations associated with recidivism studies make it difficult to understand why the drop occurred. It seems, however, that the most significant drop occurred for offenders sampled during the 1980s; the drop seems to have been slower and more gradual thereafter reaching historically low rates in more recent years. The evolution of sexual recidivism rates is not a function of publication year, which could explain why this change has been missed by researchers.
The current study was not designed to identify the factors responsible for the observed drop in sexual recidivism rates reported in Canadian studies over the years. Moreover, the current study was not designed to distinguish cohort from period effects (Glenn, 2003); whether doing so is even possible is a point of contention (see, Bell & Jones, 2013; Palmore, 1978; Robertson et al., 1999). Both cohort and period effects seek to explain changes associated with time. On the one hand, a cohort effect implies that recidivism rates are different for individuals having experienced an initial event the same year/years (e.g., birth, first arrest, admission to a treatment program) as opposed to those having experienced it during some other year/years. A period effect, on the other hand, would impact recidivism rates of all individuals the same way, irrespective of their age/cohort as a result of cultural, social, legal or political factors (see, Yang & Land, 2006). The possible presence of cohort/period effects reiterates the importance of interpreting these findings with caution. It is unclear from the data presented whether the 60% drop in sexual recidivism rates occurred because of period or cohort effects.
Several factors could explain the sexual recidivism drop in Canada between 1940 and 2019. The decline could reflect context-dependent and evolving social problems and reactions to these problems (e.g., Fabio et al., 2006; Neil & Sampson, 2021). The 1970s to the mid 1980s in Canada saw a significant rise in crime (e.g., Ouimet, 2002), including sexual offenses (Lussier et al., 2021). After this period, the crime rate dropped (e.g., Blumstein & Farrington, 2000; Farrell et al., 2014) in a manner similar to the drop in sexual recidivism observed in the current study. In fact, in Canada, between 1991 and 2010, per Uniform Crime Reporting data, sexual assaults dropped by 45% (Farrell & Brantingham, 2013). This could reflect a period effect as similar crime drops were reported in the same period in other geographic locations (e.g., Tseloni et al., 2010). Having said that, criminologists, especially criminal career researchers, have long stressed the importance of distinguishing between prevalence and incidence of offending (e.g., Blumstein & Graddy, 1981). Prevalence is more closely tied to the crime rate within a jurisdiction whereas incidence relates more to the concept of recidivism as examined in the current study. Change in prevalence does not necessarily go hand-in-hand with change in incidence (see, Berg et al., 2016). A drop in crime rates could be due to fewer people participating in crime, the incidence of crime dropping for active offenders, or a combination of both. Some argued through the inspection of individual offending careers that the general crime drop that occurred in the 1990s was more reflective of persons who were at a low-risk of offending abstaining from crime rather than a change in incidence of offending for high-risk offenders (Payne & Piquero, 2020). Indeed, at the individual level, declines in rates of offending appeared primarily due to persons abstaining from crime as opposed to active offenders engaging in fewer criminal offenses (Nagin et al., 1995). It is unclear whether such observations apply to sexual offending and perpetrators of sexual offenses.
The sexual recidivism drop could reflect scientific progress on training and practices in the criminal justice system16, especially within corrections. The translation of research and scientific knowledge to actual changes in practice and intervention is gradual and may not occur at the same time in different settings (e.g., Morris et al., 2011). One major change in correctional practices is adherence to RNR principles (Andrews & Bonta, 1990). RNR principles are used at each stage of an offender’s trajectory within the criminal justice system, including: (a) assessment at intake in prison for all individuals convicted of a crime; (b) treatment intensity adjusted to offenders’ risk and needs; (c) development of treatment for sex offenders based on cognitive-behavioral principles; (d) development of knowledge and tools to assist therapists in determining treatment progress, changes, and other effects of therapeutic processes; (e) development of risk assessment instruments specifically designed for individuals convicted of a sex crime to assist parole boards in decision-making; (f) development of knowledge and tools to assist in designing and evaluating correctional practices and monitoring offenders in the community; and (g) community-based programs and interventions aligned with inpatient/prison-based programs. The possibly significant variations in correctional practices across Canadian provinces is a reminder that the evolution of practices has been multilayered and has affected the trajectories of numerous offenders. The way knowledge and practices are diffused across jurisdictions may reflect the attitudes, philosophies, and training experiences of those responsible for administering and promoting different correctional principles (e.g., Charette et al., 2021). While the decline in recidivism coincides with significant and gradual changes in correctional practices, the decline was not necessarily caused by these advancements, making it important to consider other factors as well.
There are at least three additional explanations, not mutually exclusive, for the drop in sexual recidivism rates in Canada. Media coverage of sexual assault cases increased significantly in Canada during the 1980s and 1990s, possibly increasing social awareness of this issue and creating a deterrent effect (e.g., Boudreau & Ouimet, 2010). It is reasonable to assume sexual offending and associated societal response to it (e.g., type of offenses, prevalence, public perceptions, prosecution) changed during that period (see Zatkin et al., 2021). In fact, this was assumed to be true as past meta-analytic work purposefully excluded earlier studies (e.g., Caldwell, 2016). However, by excluding these studies, changes in recidivism rates over time remained largely undocumented. As well, evolution in the understanding of and responses to sexual offending may have led to different challenges for the criminal justice system and the decline in the sexual recidivism rate could reflect changes in both kinds of offenders and kinds of offenses. In the 1950s and 1960s, individuals involved in homosexual acts between consenting adults made up a significant proportion of those admitted to treatment programs (e.g., Turner et al., 1958). A closer look at offenders sampled across periods, their offending history and individual characteristics, could help shed some lights on this issue. It could help to determine, for example, whether changes to the criminal code, changes to the reporting to the police and the handling of these cases by police investigators might have led to the prosecution of a greater number of low-risk individuals. For example, it has been suggested that the Canadian rape law reform unintentionally created ambiguities in the definition of what is a sexual assault (as opposed to the notion of rape and attempted rape, both of which were abandoned in 1983) (e.g., Roberts, 1990; Roberts & Pires, 1992). This ambiguity occurred during a period marked by a shift toward greater legal intervention (Schissel, 1996), which might have created a spillover effect on cohorts of offenders being sampled to study recidivism rates. It is reasonable to suggest that sexual recidivism rates have dropped as a result of changes to attrition rates in the criminal justice system that, in turn, create a distinct composition of individuals convicted for sexual offenses. Over time, these changes could have led to an increased number of individuals being convicted for sexual offenses who represented a lower risk for sexual recidivism. However, research suggests that from 1970 to 2006, Canada observed a decline in the conviction rate for sexual assault cases (e.g., Daly & Bouhours, 2010). It is unclear whether such decline also impacted individuals returning to the criminal justice system for sexual offenses. That said, reporting patterns to the police, the handling of such cases by law enforcement, new investigation techniques and technology, clearance rates and other factors related to justice system processing and survivors’ perceptions of the handling of their case (e.g., Lapsey et al., 2021; Murphy et al., 2022) all could contribute to changes in the risk factor profile of samples of persons adjudicated for a sex offense, which in turn could influence recidivism rates. Changes in detected rates of sexual recidivism may also reflect crime displacement: sexual reoffenses may be committed elsewhere (e.g., “sexual tourism”; Thomas, 2013) or have taken forms that are more difficult to detect, investigate, or prosecute (e.g., online sex crimes, child pornography).
Alternatively, the change may reflect the fact that researchers working in the 1950s and 1960s did not have access to the knowledge, methods, computerized information, funding, and resources available to today’s researchers. It could be reasonably argued that with time, with more resources and computerized correctional files available to researchers, larger scale studies became more prominent giving a more representative outlook on offenders’ risk. In other words, earlier studies could have been based on smaller samples which may be less representative of the population of justice-involved individuals than larger samples. Our study findings, however, did not show heterogeneity of sexual recidivism rates across sample size. Sample size has been an issue in this area of research with approximately 25 % of studies relying on sample sizes of less than 100. Given the creation of large justice system databases, we suspected to observe smaller sample sizes in earlier study periods. However, there was no association between a sample size smaller than 100 and study period [X2(10) = 10.3, p = .413]. We also did not find mean differences in sample size across study period [F(5, 82) = 0.55, p = .734]. Alternatively, while pioneers in the field (e.g., Atcheson & Williams, 1954; Mohr et al., 1962) provided essential directions for contemporary research, methodological rigor and practices have evolved in ways that may have impacted observed sexual recidivism rates (e.g., Hanson et al., 1993; Proulx et al., 1997; Soothill et Gibbens, 1978). Indeed, Lösel and Schmuker’s (2005) meta-analysis on treatment strategies and sexual recidivism indicated that most of what is referred to as the “treatment effect” could be attributed to a study’s methodology rather than to treatment characteristics. Said differently, these recidivism rates are intrinsically tied to their measurement and future studies should examine the methodological rigor and research practices across decades that could account for the possible decline in recidivism rates.
Finally, the gradual drop could be an artefact of selection bias in earlier research. In individual studies, researchers’ reliance on official data from police, corrections, and other agencies will underestimate the true rate of sexual reoffending. However, when a large number of studies are looked at through meta-analytical lenses, it is reasonable to hypothesize that there is a selection bias toward publication of studies with a higher sexual recidivism rate17. The quest to describe the risk and associated risk factors for sexual recidivism might have played a role in selecting samples with higher base rates of sexual recidivism. It is important to investigate whether individual-level characteristics of those involved in sexual offenses has changed over time. Our findings show that studies with a higher proportion of sexual recidivists tended to be associated with higher rates of sexual recidivism, although this association was not statistically significant, most probably due to the high prevalence of missing data. It could be that earlier studies had a greater prevalence of sexual recidivists (higher risk offenders) compared to those included in more recent samples although we did not find empirical evidence supporting this hypothesis18. Finally, a case could then be made that pooled sexual recidivism rates are artificially inflated because studies reporting very low sexual recidivism rates remain unpublished (i.e., the file drawer problem). In fact, researchers have sometimes failed to report sexual recidivism rates because rates were too low for data analyses (e.g., Lussier & Gress, 2014). Furthermore, some subgroups have consistently shown very low sexual recidivism rates (e.g., incest offenders; Firestone et al., 1999) and could be underrepresented in the scientific literature partly because of their lower risk of sexual recidivism.
Based on our findings, we think it is important to raise a number of questions about measuring recidivism, interpreting recidivism rates, estimating risk probabilities, and communicating the risk of sexual recidivism over time. Heterogeneity of sexual recidivism rates across studies raise fundamental policy questions about how the criminal justice system makes use of recidivism data. An underlying assumption of actuarial tools is that recidivism risk probabilities are stable across time and jurisdiction. These fixed and stable risk probabilities estimated from a developmental sample are then used to determine risk probabilities of other justice-involved individuals. Actuarial risk assessors use group data to infer the likelihood that a given individual will recidivate. The accuracy of the assessment depends on the extent to which the developmental sample group data reflect the individual of interest19. Importantly, if individuals come from different decades, and if base rates vary across these decades, then the accuracy of the assessment is threatened, even if the case in question has a level of risk similar to their reference group. Actuarialists are generally aware of possible “cohort effects” and some researchers make a point of going beyond the developmental sample by testing the reliability and validity of an instrument in the setting where it is being used by criminal justice practitioners (e.g., Hanson, 2022). In more recent years, however, researchers have noted that beyond possible “cohort effects”, actuarial instruments were not sensitive to “age effects”. More specifically, they tend to overestimate the risk of older adult offenders. For example, items composing actuarial instruments are correlated with a person’s age at release, suggesting that statistical analyses that identified those items as “predictors” of sexual recidivism were disproportionally focused on characteristics of younger adult offenders because their recidivism rates are higher (Barbaree et al., 2007; Lussier & Healey, 2009). Items included in actuarial risk assessment tools are typically static and therefore cannot account for aging and the changing nature of risk (Lussier & Davies, 2011).
The current study findings, along with those of Caldwell (2016), highlight the need for future research to more directly evaluate the role of age, period, and cohort effects (APC) on sexual recidivism rates. From a policy standpoint, there is cumulative evidence that actuarial risk assessment instruments should be calibrated to account for APC effects (e.g., see Yang & Land, 2013). If APC effects are impacting sexual recidivism rate estimates, then these rates need to be calibrated and updated regularly to be properly communicated (Abbott, 2017). If base rates vary across samples and over time, risk estimations for individuals with the same score on a given risk assessment tool may differ, leading to significant errors in the estimation of risk (Mossman, 2006). Our finding that the risk of sexual recidivism varies across settings and through time raise questions about the use of risk assessment instruments that are not calibrated to account for such variations. Variability in base rates over time is only one factor leading some to question whether actuarial risk assessment instruments should be used at all, given the difficulties in estimating risk using this approach (Hart et al., 2007).
Beyond the examination of APC effects, whether changes in sexual recidivism rates are occurring in other countries should be examined, especially jurisdictions with different social and political contexts. That said, it was necessary for the current study to focus on Canada given that combining recidivism rates reported in various countries to establish a base rate for sexual recidivism, to calibrate actuarial instruments, or to determine the effect of treatment fails to consider particularities and differences in criminal justice systems. These particularities may include sociocultural and sociohistorical factors ranging from citizens’ views of the legitimacy of the criminal justice system to the way data about arrests and convictions is recorded (e.g., Ivković, 2009; Newman, 1977). It could be argued that official data on sexual recidivism is more reflective of how the criminal justice system functions and underestimates actual criminal activity. Recidivism rates based on official data can also be influenced by historical, political, cultural, social, legal, individual, and correctional factors (e.g., Cohen, 1986; Maltz, 1984; Mosher et al., 2010). Recognizing this raises questions about the extent to which traditional measures of sexual recidivism using official data fail to capture all instances of sexual reoffending (e.g., Abbott, 2020; Scurich & John, 2019) and are influenced by offenders’ ability to avoid detection (Lussier et al., 2011). One possibility that could extend the work conducted here is to establish a national database measuring SOR rates (see Hargreaves & Francis, 2014). This is certainly feasible given that various countries (e.g., Canada, USA) have established sex offender registries (e.g., REDACTED) that could be used as large, centralized databases for examining SOR rates. However, these registries were enacted no earlier than the 1990s (albeit some exceptions) and primarily in the 2000s and thus would not provide answers to questions about the social, legal, and cultural changes that impacted sexual recidivism rates over several decades.
Additionally, critical questions can be raised about whether it is even possible to make conclusions about a ‘global’ indicator of sexual recidivism given (a) there is wide variability in the methodological characteristics of studies and (b) such methodological variation has implications for the observed rate of sexual recidivism (REDACTED). The current study shows that the search for factors associated with sexual recidivism rates must be extended well beyond offender characteristics. In our study, recidivism rates varied across study settings, author affiliations, and region. For example, studies conducted in Saskatchewan reported significantly higher recidivism rates. Numerous hypotheses could be raised to explain the discrepancies observed across provinces but, more important, these findings point to disparities that need to be considered in any national strategy aimed at preventing sexual recidivism. They also raise critical questions about risk assessment evaluation using actuarial tools that do not take such variations into account, which may lead to underestimating risk for some offenders and overestimating it for others.
While risk assessment and risk prediction tools that place a heavy emphasis on establishing a relatively stable base of sexual recidivism are now routinely used, there have been criticisms about the measurement of sexual recidivism (e.g., Langevin et al., 2004; Lussier & Cale, 2016; Webster et al. 2006). This meta-analysis of the weighted pooled prevalence of SOR in Canadian studies does not suggest how to resolve methodological issues common to most empirical studies. For example, the vast majority of studies were based on samples of convicted offenders and on official data on offending. Also, the current study did not examine how research might have evolved over the years in ways that impacted the composition of study samples, such as differences over time in the prevalence of “high-risk” offenders, the prevalence of offenders having perpetrated certain sexual offenses, or how perpetrators were sampled. There is a lack of clear guidelines as to what information should be analyzed and reported in recidivism studies (REDACTED) and this impacted the number of study moderators examined in the current meta-analyses. For the studies included in this meta-analysis, rarely were both the start and end date of the sampling period reported. It is unclear how this could have impacted the study findings given the different types of samples in the scientific literature (e.g., prison populations, individuals admitted to a hospital or to a treatment program, offenders released in a specific year). When coding these studies, there was a lack of detail about the characteristics of the sexual recidivism being analyzed, which is generally limited to the number of recidivists and the legal designation of the offenses. Information about the type of sexual offense committed by recidivists is rarely reported. Sexual recidivism is thus an aggregate measure that includes criminal offenses that can vary significantly along various dimensions (e.g., the number of crime events included in a single charge; the level of violence, the level of sexual intrusiveness, the crime type). This type of aggregate measure of sexual recidivism may also mask other trends over time (e.g., de-escalation, increase in severity of sexual recidivism).
Also, the current meta-analysis does not report the proportion of sexual recidivists for specific follow-up periods (as this is rarely reported) but rather the pooled proportion of sexual recidivists according to the mean follow-up period of each study/sample. The information included in the studies examined did not allow for an examination of the functional form of survival curves using a meta-analytic framework. Thus, it is not possible to examine, for example, whether the time until sexual recidivism has increased over the years. These Canadian studies also showed an overrepresentation of research based on samples from Ontario and an underrepresentation of other regions (e.g., Canadian territories, Maritimes). Precautions were taken to determine study independence, but it is difficult to determine how fully independent samples were based on the information provided: offenders can move across provinces, take part in different treatment/intervention programs, serve time in provincial and federal institutions where independent studies are conducted, etc. Finally, in Canada, there are federally-run custody sentences of at least two years and provincially-run sentences (generally community-based or custody sentences of less than two years). Unfortunately, because the data were not consistently reported, we could not statistically control for whether justice-involved offenders were serving their sentences in federally-run or provincially-run institutions to determine whether the recidivism drop was specific to offenders serving longer custodial sentences (i.e., 2 years and more), shorter custodial sentences (i.e., less than 2 years) or community-based sentences. Because sentencing has evolved over time, the recidivism drop could partly reflect an evolution in sentencing for individuals convicted of sexual offenses. While sentencing has evolved over time, it is also incredibly difficult to answer questions about the impact of sentencing on recidivism (e.g., Nagin et al., 2009) and rarely has research specifically focused on the impact of sentences on sexual recidivism.
It has been a long road since the Canadian enactment of laws regarding sexual psychopathy in the 1940s and their subsequent revocation in the 1970s. The Canadian approach to the problem of SOR, considered by some to be more cautious than approaches in the United States (Petrunik, 2003), has focused less on restrictive SOL, such as SORN laws, and more on evidence-based practices that promote risk management and rehabilitation. The 60% drop in reported sexual recidivism rates observed in Canada between 1940 and 2019 may provide some justification for the different path taken by the Canadian government. To the best of our knowledge, this drop has never been scientifically documented and the factors responsible for this drop remain relatively unclear. Importantly, the sexual recidivism rate drop occurred around the same time as a well-documented crime drop in Canada. It is unclear whether the drop in sexual recidivism is simply a function of drop in general offending, but the fact that the crime drop occurred in several countries (Farrell & Brantingham, 2013; Tseloni et al., 2010), stresses the importance of examining cohort/period effects further. In fact, the sexual recidivism drop could have been occurring in countries other than Canada. This change may be more than just government decision-making. Several hypotheses were raised to explain the recidivism drop observed in Canadian studies that should be examined in future research. One hypothesis is that this drop reflects relatively slow and gradual improvement in training and practices in correctional institutions based on an accumulation of empirical evidence about sexual offending, risk, clinical/risk assessment practices, risk factors, and treatment/intervention programs. Detecting period effects and the factors responsible for such effects is a difficult endeavour, which could explain, not only the relative absence of studies on the issue, but why studies actively avoided addressing such questions (see e.g., Caldwell, 2016). Comparative criminological research could be helpful in examining whether the drop is specific to Canada, putting the findings from the current study into a wider perspective. Of importance, we also discovered a decline over time in the number of SOR studies in Canada using more recent cohorts of offenders. This may limit researchers’ ability to continue to address questions about the evolution of sexual recidivism over time as well as the ability to detect possible APC effects.
Abbott, B. R. (2017). Sexually violent predator risk assessments with the violence risk appraisal guide-revised: A shaky practice. International Journal of Law and Psychiatry, 52, 62-73.
Abbott, B. R. (2020). Illuminating the dark figure of sexual recidivism. Behavioral Sciences & the Law, 38(6), 543-558.
Abracen, J., & Looman, J. (2004). Issues in the treatment of sexual offenders: Recent developments and directions for future research. Aggression and Violent Behavior, 9(3), 229-246.
Ahlmeyer, S., Heil, P., McKee, B., & English, K. (2000). The impact of polygraphy on admissions of victims and offenses in adult sexual offenders. Sexual Abuse, 12, 123-138.
Andrews, D. A., Zinger, I., Hoge, R. D., Bonta, J., Gendreau, P., & Cullen, F. T. (1990). Does correctional treatment work? A clinically relevant and psychologically informed meta‐analysis. Criminology, 28(3), 369-404.
Andrews, D. A., Bonta, J., & Hoge, R. D. (1990). Classification for effective rehabilitation: Rediscovering psychology. Criminal justice and Behavior, 17(1), 19-52.
Andrews, D. A., Bonta, J., & Wormith, J. S. (2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52(1), 7-27.
Andrews, D. A., Bonta, J., & Wormith, J. S. (2011). The risk-need-responsivity (RNR) model: Does adding the good lives model contribute to effective crime prevention? Criminal Justice and Behavior, 38(7), 735-755.
Atcheson, J. D., & Williams, D. C. (1954). A study of juvenile sex offenders. American Journal of Psychiatry, 111(5), 366-370.
Barendregt, J. J., Doi, S. A., Lee, Y. Y., Norman, R. E., & Vos, T. (2013). Meta-analysis of prevalence. Journal of Epidemiology & Community Health, 67(11), 974-978.
Bell, A., & Jones, K. (2013). The impossibility of separating age, period and cohort effects. Social Science & Medicine, 93, 163-165.
Berg, M. T., Baumer, E., Rosenfeld, R., & Loeber, R. (2016). Dissecting the prevalence and incidence of offending during the crime drop of the 1990s. Journal of Quantitative Criminology, 32(3), 377-396.
Blumstein, A., & Graddy, E. (1981). Prevalence and recidivism in index arrests: A feedback model. Law and Society Review, 265-290.
Blumstein, A., & Farrington, D. (2000). The crime drop in America (pp. 13-44). A. Blumstein, & J. Wallman (Eds.). New York: Cambridge University Press.
Boer, D.P., Wilson, R.J., Gauthier, C.M., & Hart, S.D. (1997). Assessing risk for sexual violence: Guidelines for clinical practice. In C. D. Webster & M. A. Jackson (Eds.), Impulsivity: Theory, Assessment, and Treatment (pp. 326-342). Guilford.
Bonta, J., Blais, J., & Wilson, H. A. (2014). A theoretically informed meta-analysis of the risk for general and violent recidivism for mentally disordered offenders. Aggression and Violent Behavior, 19(3), 278-287.
Boudreau, M.E. & Ouimet, M. (2010). L'impact de la couverture médiatique des violences sexuelles sur les taux d'agressions sexuelles au Québec entre 1974 et 2006. Canadian Journal of Criminology and Criminal Justice, 52, 497-525.
Caldwell, M. F. (2016). Quantifying the decline in juvenile sexual recidivism rates. Psychology, Public Policy, and Law, 22(4), 414-426.
Charette, Y., Goossens, I., Seto, M. C., Nicholls, T. L., & Crocker, A. G. (2021). Is knowledge contagious? diffusion of violence-risk-reporting practices across clinicians’ professional networks. Clinical Psychological Science, 9(2), 284-293.
Chenier, E. (2003). The criminal sexual psychopath in Canada: Sex, psychiatry and the law at mid-century. Canadian Bulletin of Medical History, 20(1), 75-101.
Cheung, M. W. L. (2019). A guide to conducting a meta-analysis with non-independent effect sizes. Neuropsychology review, 29(4), 387-396. https://doi.org/10.1007/s11065-019-09415-6
Cohen, J. (1986). Research on criminal careers: individual frequency rates and offense seriousness. In A. Blumstein, J. Cohen, J. A. Roth, & C. A. Visher (Eds.), Criminal careers and career criminals (pp. 292-418). National Academy Press.
Cohen, M., & Jeglic, E. L. (2007). Sex offender legislation in the United States: What do we know? International Journal of Offender Therapy and Comparative Criminology, 51(4), 369-383.
Conn, V. S., Valentine, J. C., Cooper, H. M., & Rantz, M. J. (2003). Grey literature in meta-analyses. Nursing Research, 52(4), 256-261.
Cormier, B. M., & Simons, S. P. (1969). The problem of the dangerous sexual offender. Canadian Psychiatric Association Journal, 14(4), 329-335.
Cortoni, F., Hanson, R. K., & Coache, M. È. (2010). The recidivism rates of female sexual offenders are low: A meta-analysis. Sexual Abuse, 22(4), 387-401.
Cottle, C. C., Lee, R. J., & Heilbrun, K. (2001). The prediction of criminal recidivism in juveniles: A meta-analysis. Criminal Justice and Behavior, 28(3), 367-394.
Daly, K., & Bouhours, B. (2010). Rape and attrition in the legal process: A comparative analysis of five countries. Crime and Justice, 39(1), 565-650.
Ewert v. Canada, FC 1093 (2015).
Fabio, A., Loeber, R., Balasubramani, G. K., Roth, J., Fu, W., & Farrington, D. P. (2006). Why some generations are more violent than others: Assessment of age, period, and cohort effects. American Journal of Epidemiology, 164(2), 151-160.
Farrell, G., & Brantingham, P. J. (2013). The crime drop and the General Social Survey. Canadian Public Policy, 39(4), 559-580.
Farrell, G., Tilley, N., & Tseloni, A. (2014). Why the crime drop? Crime and Justice, 43(1), 421-490.
Fazel, S., & Wolf, A. (2015). A systematic review of criminal recidivism rates worldwide: Current difficulties and recommendations for best practice. PloS One, 10(6), e0130390.
Firestone, P., Bradford, J. M., McCoy, M., Greenberg, D. M., Larose, M. R., & Curry, S. (1999). Prediction of recidivism in incest offenders. Journal of Interpersonal Violence, 14(5), 511-531.
Fortune, C. A., & Lambie, I. (2006). Sexually abusive youth: A review of recidivism studies and methodological issues for future research. Clinical Psychology Review, 26(8), 1078-1095.
Furby, L., Weinrott, M. R., & Blackshaw, L. (1989). Sex offender recidivism: a review. Psychological Bulletin, 105(1), 3-30.
Frechette, J., & Lussier, P. (2021). Betting Against the Odds: The Mysterious Case of the Clinical Override in Risk Assessment of Adult Convicted Offenders. International Journal of Offender Therapy and Comparative Criminology.
Freund, K., Seeley, H. R., Marshall, W. E., & Glinfort, E. K. (1972). Sexual offenders needing special assessment and/or therapy. Canadian Journal of Criminology and Corrections, 14(4), 345-365.
Furby, L., Weinrott, M. R., & Blackshaw, L. (1989). Sex offender recidivism: A review. Psychological Bulletin, 105(1), 3–30.
Gendreau, P., Little, T., & Goggin, C. (1996). A meta‐analysis of the predictors of adult offender recidivism: What works! Criminology, 34(4), 575-608.
Glenn, N. D. (2003). Distinguishing age, period, and cohort effects. In Handbook of the life course (pp. 465-476). Springer, Boston, MA.
Gunn, R., & Linden, R. (1997). The impact of law reform on the processing of sexual assault cases. Canadian Review of Sociology/Revue Canadienne de Sociologie, 34(2), 155-174.
Greenberg, D. M. (1998). Sexual recidivism in sex offenders. The Canadian Journal of Psychiatry, 43(5), 459-465.
Greenland, C. (1972). Dangerous sexual offenders in Canada. Canadian Journal of Criminology and Corrections, 14(1), 44-54. https://doi.org/10.3138/cjcc.14.1.44
Greenland, C. (1984). Dangerous Sexual Offender Legislation in Canada, 1948-1977: An Experiment that Failed. Canadian Journal of Criminology, 26(1), 1-12.
Hall, G. C. N., & Hirschman, R. (1991). Toward a theory of sexual aggression: A quadripartite model. Journal of Consulting and Clinical Psychology, 59(5), 662.
Hanson, R. K. (2022). Prediction statistics for psychological assessment. American Psychological Association.
Hanson, R. K., & Bussiere, M. T. (1998). Predicting relapse: a meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66(2), 348-362.
Hanson, R. K., & Harris, A. J. (2000). Where should we intervene? Dynamic predictors of sexual offense recidivism. Criminal Justice and behavior, 27(1), 6-35.
Hanson, R. K., & Harris, A. J. (2001). A structured approach to evaluating change among sexual offenders. Sexual Abuse, 13(2), 105-122. https://doi.org/10.1023/A:1026600304489
Hanson, R. K., & Thornton, D. (1999). Static 99: Improving actuarial risk assessments for sex offenders (User Report 99–02). Ottawa, Ontario, Canada: Department of the Solicitor General of Canada.
Hanson, R. K., Steffy, R. A., & Gauthier, R. (1993). Long-term recidivism of child molesters. Journal of Consulting and Clinical Psychology, 61, 646-652.
Hargreaves, C., & Francis, B. (2014). The long term recidivism risk of young sexual offenders in England and Wales–enduring risk or redemption? Journal of Criminal Justice, 42(2), 164-172.
Hart, S., Michie, C., & Cooke, D. (2007). Precision of actuarial risk assessment instruments: Evaluating the ‘margins of error’ of group v. individual predictions of violence. British Journal of Psychiatry, 190(S49), S60-S65. doi:10.1192/bjp.190.5.s60
Hedges, L.V. & Olkin, 1. (1985). Statistical Methods for Meta-Analysis. Academic Press.
Helmus, LM, Hanson, RK, Murrie, DC, & Zabarauckas, CL. (2021). Field validity of Static-99R and STABLE-2007 with 4,433 men serving sentences for sexual offences in British Columbia: New findings and meta-analysis. Psychological Assessment, 33(7), 581-595.
Ivković, S. (2009). The Croatian police, police integrity, and transition toward democratic policing. Policing: An International Journal, 32(3), 459-488. https://doi.org/10.1108/13639510910981617
Jakimiec, J., Porporino, F., Addario, S., & Webster, C.D. (1986). Dangerous offenders in Canada, 1977-1985. International Journal of Law and Psychiatry, 9(4), 479-489.
Katsiyannis, A., Whitford, D.K., Zhang, D. et al. (2018). Adult recidivism in United States: A meta-analysis 1994–2015. Journal of Child and Family Studies, 27, 686-696.
Knack, N., Blais, J., & Fedoroff, J. P. (2021). Exploring Inconsistencies in the Interpretation of Canada's Section 161 Order for Sexual Offending. Canadian Journal of Criminology and Criminal Justice, 63(1), 52-88.
La Fond, J. Q. (2005). Preventing sexual violence: How society should cope with sex offenders. Washington, DC: American Psychological Association.
Langevin, R., & Lang, R. A. (1985). Psychological treatment of pedophiles. Behavioral Sciences & the Law, 3(4), 403-419.
Langevin, R., Curnoe, S., Fedoroff, P., & Bennett, R. (2004). Lifetime sex offender recidivism: 25-year follow-up study. Canadian Journal of Criminology and Criminal Justice, 46(5), 531-552.
Lapsey Jr, D. S., Campbell, B. A., & Plumlee, B. T. (2021). Focal concerns and police decision making in sexual assault cases: A systematic review and meta-analysis. Trauma, Violence, & Abuse, 1524838021991285.
Laws, D. (1989). Relapse prevention with sex offenders. The Guilford Press.
Lee, S. C., & Hanson, R. K. (2021). Updated 5-year and new 10-year sexual recidivism rate norms for Static-99R with routine/complete samples. Law and Human Behavior, 45(1), 24–38.
Lieb, R., Quinsey, V., & Berliner, L. (1998). Sexual predators and social policy. Crime and justice, 23, 43-114.
Lösel, F., & Schmucker, M. (2005). The effectiveness of treatment for sexual offenders: A comprehensive meta-analysis. Journal of Experimental Criminology, 1(1), 117-146.
Lussier, P., McCuish, E. C., & Cale, J. (2021). Understanding sexual offending: An evidence-based response to myths and misconceptions. Springer.
Lussier, P., & Cale, J. (2016). Understanding the origins and the development of rape and sexual aggression against women: Four generations of research and theorizing. Aggression and Violent Behavior, 31, 66-81.
Lussier, P., & Gress, C. L. (2014). Community re-entry and the path toward desistance: A quasi-experimental longitudinal study of dynamic factors and community risk management of adult sex offenders. Journal of Criminal Justice, 42(2), 111-122.
Lussier, P., & Mathesius, J. (2019). Trojan horse policies: Sexual predators, SORN laws and the American experience. Psychology, Crime & Law, 25(2), 133-156.
Lussier, P., Bouchard, M., & Beauregard, E. (2011). Patterns of criminal achievement in sexual offending: Unravelling the “successful” sex offender. Journal of Criminal Justice, 39(5), 433-444.
Malamuth, N. M. (1986). Predictors of naturalistic sexual aggression. Journal of Personality and Social Psychology, 50(5), 953–962.
Maltz, M. (1984). Recidivism. Academic Press.
Mancini, C., & Budd, K. M. (2016). Is the public convinced that “nothing works?” Predictors of treatment support for sex offenders among Americans. Crime & Delinquency, 62(6), 777-799.
Marcus, A. M. (1966). A multi-disciplinary two part study of those individuals designated dangerous sexual offenders held in federal custody in British Columbia, Canada. Canadian Journal of Corrections, 8(2), 90-103.
Marcus, A. M., & Conway, C. (1971). A Canadian group approach study of dangerous sexual offenders. International Journal of Offender Therapy, 15(1), 59-66.
Marcus, D. A., & Conway, C. (1969). Dangerous sexual offender project. Canadian Journal of Corrections, 11(3), 198-205.
Marshall, W. L. (1971). A combined treatment method for certain sexual deviations. Behaviour Research and Therapy, 9, 292-294.
Marshall, W. L. (1973). The modification of sexual fantasies: A combined treatment approach to the reduction of deviant sexual behavior. Behaviour Research and Therapy, 11(4), 557-564.
Marshall, W. L., & Barbaree, H. E. (1984). A behavioral view of rape. International Journal of Law and Psychiatry, 7(1), 51-77.
Marshall, W. L., & Barbaree, H. E. (1990). An integrated theory of the etiology of sexual offending. In W. L. Marshall, D. R. Laws, & H. E. Barbaree (Eds.), Handbook of sexual assault: Issues, theories, and treatment of the offender (pp. 257–275). Plenum Press.
Marshall, W. L., & Laws, D. R. (2003). A brief history of behavioral and cognitive behavioral approaches to sexual offender treatment: Part 2. The modern era. Sexual Abuse, 15(2), 93-120.
Marshall, W. L., Laws, D. R., & Barbaree, H. E. (1990). Handbook of sexual assault: Issues, theories, and treatment of the offender. Plenum Press.
Migliavaca, C. B., Stein, C., Colpani, V., Munn, Z., & Falavigna, M. (2020). Quality assessment of prevalence studies: a systematic review. Journal of Clinical Epidemiology, 127, 59-68.
Mohr, J. W., Turner, E. R., Edward, R., & Ball, R. B. (1962). Exhibitionism and pedophilia. Corrective Psychiatry & Journal of Social Therapy, 8, 172-186.
Morris, Z. S., Wooding, S., & Grant, J. (2011). The answer is 17 years, what is the question: understanding time lags in translational research. Journal of the Royal Society of Medicine, 104(12), 510-520.
Mosher, C. J., Miethe, T. D., & Hart, T. C. (2010). The mismeasure of crime. Sage Publications.
Mossman D. (2006). The imperfection of protection through detection and intervention. Lessons from three decades of research on the psychiatric assessment of violence risk. Journal of Legal Medicine, 30, 109–140.
Murphy, L., Fedoroff, J. and Martineau, M. (2009). Canada’s sex offender registries: Background, implementation, and social policy considerations. The Canadian Journal of Human Sexuality, 18 (1-2), 61.
Murphy, A., Hine, B., Yesberg, J. A., Wunsch, D., & Charleton, B. (2022). Lessons from London: a contemporary examination of the factors affecting attrition among rape complaints. Psychology, Crime & Law, 28(1), 82-114.
Nagin, D. S., Cullen, F. T., & Jonson, C. L. (2009). Imprisonment and reoffending. Crime and Justice, 38(1), 115-200.
Nagin, D. S., Farrington, D. P., & Moffitt, T. E. (1995). Life‐course trajectories of different types of offenders. Criminology, 33(1), 111-139.
Neil, R., & Sampson, R. J. (2021). The Birth lottery of history: Arrest over the life course of multiple cohorts coming of age, 1995–2018. American Journal of Sociology, 126(5), 1127-1178.
Newcombe, R. G. (1998). Interval estimation for the difference between independent proportions: comparison of eleven methods. Statistics in medicine, 17(8), 873-890.
Newman, G. R. (1977). Problems of method in comparative criminology. International Journal of Comparative and Applied Criminal Justice, 1(1-2), 17-31.
Nyaga, V. N., Arbyn, M., & Aerts, M. (2014). Metaprop: a Stata command to perform meta-analysis of binomial data. Archives of Public Health, 72(1), 1-10.
Olver, M. E., Nicholaichuk, T. P., Gu, D., & Wong, S. C. (2013). Sex offender treatment outcome, actuarial risk, and the aging sex offender in Canadian corrections: A long-term follow-up. Sexual Abuse, 25(4), 396-422.
Ouimet, M. (2002). Explaining the American and Canadian crime "drop" in the 1990's. Canadian Journal of Criminology, 44(1), 33-50.
Palmore, E. (1978). When can age, period, and cohort be separated. Social Forces, 57(1), 282-295.
Pascoe, H. (1961). Deviant sexual behaviour and the sex criminal. Canadian Medical Association Journal, 84(4), 206-211.
Payne, J. L., & Piquero, A. R. (2020). Developmental criminology and the crime decline: a comparative analysis of the criminal careers of two New South Wales Birth cohorts. Cambridge University Press.
Petrunik, M. (2003). The hare and the tortoise: Dangerousness and sex offender policy in the United States and Canada. Canadian Journal of Criminology and Criminal Justice, 45(1), 43-72.
Petrunik, M. G. (2002). Managing unacceptable risk: Sex offenders, community response, and social policy in the United States and Canada. International Journal of Offender Therapy and Comparative Criminology, 46(4), 483-511. https://doi.org/10.1177/0306624X02464009
Pithers, W. D. (1990). Relapse prevention with sexual aggressors: A method for maintaining therapeutic gain and enhancing external supervision. In W. L. Marshall, D. R. Laws, & H. E. Barbaree (Eds.), Handbook of sexual assault: Issues, theories, and treatment of the offender (pp. 343−361). Plenum.
Porporino, F. J., Fabiano, E. A. & Robinson, D. (1991). Focusing on successful reintegration: Cognitive skills training for offenders (Research Rep. No. 19). Ottawa, Canada: Correctional Service of Canada.
Prentky, R. A., Lee, A. F., Knight, R. A., & Cerce, D. (1997). Recidivism rates among child molesters and rapists: A methodological analysis. Law and Human Behavior, 21(6), 635-659.
Price, R. (1970). Psychiatry, criminal law reform and the mythophilic impulse. On Canadian proposals for the control of the dangerous offender. University of Ottawa Law Review, 4, 1-61.
Proulx, J., McKibben, A., & Lusignan, R. (1996). Relationships between affective components and sexual behaviors in sexual aggressors. Sexual Abuse, 8(4), 279-289.
Proulx, J., Pellerin, B., Paradis, Y., McKibben, A., Aubut, J., & Ouimet, M. (1997). Static and dynamic predictors of recidivism in sexual aggressor. Sexual Abuse, 9(1), 7-27.
Proulx, J., & Lussier, P. (2001). La prédiction de la récidive chez les agresseurs sexuels. Criminologie, 34(1), 9-29.
Quinsey, V. L. (1984). Sexual aggression: Studies of offenders against women. In D. N. Weisstub (Ed.), Law and mental health: International perspectives (Vol. I, pp. 84-121). Pergamon Press.
Quinsey, V. L., Chaplin, T. C., & Varney, G. (1981). A comparison of rapists' and non-sex offenders' sexual preferences for mutually consenting sex, rape, and physical abuse of women. Behavioral Assessment, 3(1), 27-135.
Quinsey, V. L., Khanna, A., & Malcolm, P. B. (1998). A retrospective evaluation of the Regional Treatment Centre sex offender treatment program. Journal of Interpersonal Violence, 13(5), 621-644.
Quinsey, V. L., Rice, M. E., & Harris, G. T. (1995). Actuarial prediction of sexual recidivism. Journal of Interpersonal Violence, 10(1), 85-105.
Reid, A. A. (2020). Reducing the use of imprisonment: Lessons from 20 years’ experience in Canada. The British Journal of Criminology, 60(6), 1480-1501.
Roberts, J. (1990). Sexual assault legislation in Canada: An evaluation. Report no. 3. Ottawa, ON: Minister of Justice and Attorney General of Canada.
Roberts, J. & Pires, A. (1992). Le renvoi et la classification des infractions d’agression sexuelle. Criminologie, 25, 27–63.
Robertson, C., Gandini, S., and Boyle, P. (1999). Age-period-cohort models: a comparative study of available methodologies. Journal of Clinical Epidemiology 52, 569-583.
Schellenberg, K. (1997). Police information systems, information practices and individual privacy. Canadian Public Policy/Analyse de Politiques, 23-39.
Schissel, B. (1996). Law reform and social change: A time-series analysis of sexual assault in Canada. Journal of Criminal Justice, 24(2), 123-138.
Schmucker, M., & Lösel, F. (2011). Meta-analysis as a method of systematic reviews. The Sage Handbook of Criminological Research Methods, 425-443.
Scurich, N., & John, R. S. (2019). The dark figure of sexual recidivism. Behavioral Sciences & the Law, 37(2), 158-175.
Seto, M. C., & Barbaree, H. E. (1999). Psychopathy, treatment behavior, and sex offender recidivism. Journal of Interpersonal Violence, 14(12), 1235-1248.
Soothill, K. (2010). Sex offender recidivism. Crime and Justice, 39(1), 145-211.
Soothill, K. L., & Gibbens, T. C. N. (1978). Recidivism of sexual offenders: A Re-Appraisal. British Journal of Criminology, 18(3), 267-276.
Stephens, S., Reale, K. S., Goodwill, A. M., & Beauregard, E. (2017). Examining the role of opportunity in the offense behavior of victim age polymorphic sex offenders. Journal of Criminal Justice, 52, 41-48.
Stoltenborgh, M., van IJzendoorn, M. H., Euser, E. M., & Bakermans-Kranenburg, M. J. (2011). A Global Perspective on Child Sexual Abuse: Meta-Analysis of Prevalence Around the World. Child Maltreatment, 16(2), 79–101. https://doi.org/10.1177/1077559511403920
Tang, K. L. (1998). Rape law reform in Canada: The success and limits of legislation. International Journal of Offender Therapy and Comparative Criminology, 42(3), 258-270.
Thomas, T. (2013). The traveling sex offender: monitoring movements across international borders. The Wiley‐Blackwell Handbook of Legal and Ethical Aspects of Sex Offender Treatment and Management, 445-461.
Tseloni, A., Mailley, J., Farrell, G., & Tilley, N. (2010). Exploring the international decline in crime rates. European Journal of Criminology, 7(5), 375-394.
Turner, R. E., Hutchison, H. C., & Williams, L. O. D. (1958). The forensic clinic of the Toronto Psychiatric Hospital. Canadian Journal of Corrections, 1(1), 15-20.
Villettaz, P., Killias, M., & Zoder, I. (2006). The effects of custodial vs. non‐custodial sentences on re‐offending: A systematic review of the state of knowledge. Campbell Systematic Reviews, 2(1), 1-69.
Webster, C., Gartner, R., & Doob, A. (2006). Results by design: The artefactual construction of high recidivism rates for sex offenders. Canadian Journal of Criminology and Criminal Justice, 48(1), 79-93.
Williams, S.M. (1996). A national strategy for managing sex offenders. Forum on Corrections Research, 8.
Wormith, J. S. (1986). Assessing deviant sexual arousal: Physiological and cognitive aspects. Advances in Behaviour Research and Therapy, 8(3), 101-137.
Wormith, J. S., & Olver, M. E. (2002). Offender treatment attrition and its relationship with risk, responsivity, and recidivism. Criminal Justice and Behavior, 29(4), 447-471.
Yang, Y., & Land, K. C. (2006). A mixed models approach to the age‐period‐cohort analysis of repeated cross‐section surveys, with an application to data on trends in verbal test scores. Sociological Methodology, 36(1), 75-97.
Yang, Y., & Land, K. C. (2013). Age-period-cohort analysis: New models, methods, and empirical applications. Taylor & Francis.
Zamble, E., & Quinsey, V. L. (1997). The criminal recidivism process. Cambridge University Press.
Zimring, F. E. (2009). An American Travesty. University of Chicago Press.
Zatkin, J., Sitney, M., & Kaufman, K. (2021). The relationship between policy, media, and perceptions of sexual offenders between 2007 and 2017: A review of the literature. Trauma, Violence, & Abuse, 1524838020985568
Table 1. Descriptive information about Canadian studies on Sex Offender Recidivism (SOR)
Year of publication
Type of publication
Language of the publication
Affiliation of lead author
Mental health institution
Sample size (total)
Less than 100
Between 100 and 299
300 and over
Sampling period (start)
At least one prior conviction for sexual offenses (%)
25% or less of the sample
More than 25% of the sample
Offender age group
Length of follow-up period
Less than 48 months
Between 48 and less 120 months
At least 120 months
Measure of recidivism
Note. Based on 165 studies.
Table 2. Weighted pooled sexual recidivism rates across periods
Sexual recidivism rates
Year sampling started
All study samples
(independent samples – conservative approach)
Note. These pooled estimates do not take into consideration the average length of the follow-up period. K refers to the total number of samples. Double arcsine transformation used to estimate pooled prevalence. Also, n refers to the total number of offenders included in the study. *** p<.001
Table 3. Study moderators of sexual recidivism rates observed across independent samples (k = 88)
Weighted pooled rate
Year of publication
Type of publication
Mental health institution
Less than 100
Between 100 and 299
300 and over
At least one prior conviction for sexual offenses (%)
25% or less of the sample
More than 25% of the sample
Treatment program study
Offender age group
Less than 48
Between 48 and less 120
At least 120
Measure of recidivism
Table 4. Covariates of sexual recidivism rates using a series of meta-regressions
k = 88
k = 80
k = 80
k = 80
Study setting (non-correctional)
Sample size (>300)
Note. Random-effects model using the DerSimonian-Laird method. Sample size varies across models due to missing data. Reference category for the variable Study period is the category “1940-1979”. The “2000-2009” and “2010-2019” periods were aggregated due to the small number of studies during each period.
Figure 1. Flowchart for empirical studies on sex offender recidivism
Figure 2a. Fitted linear regression lines on all reported sexual recidivism rates across periods (k = 226)
Figure 2b. Fitted linear regression lines on pooled sexual recidivism rates across periods (k = 88)
Figure 2c. Fitted linear regression lines on pooled sexual recidivism rates across periods (k = 53)
Note. No regression lines were fitted for the 2010-2019 period in Figure 2b-c due to the small number of observations.
Appendix I. List of keywords used to identify SOR research
"sex* offender" OR "sex* aggressor" OR "sex* criminal" OR "child molester*" OR "rapist" OR “sex* assaulter” OR “sex* abuser” OR "sex* murderer" OR “child molester” OR "child molestation" OR "sex* offen?e" OR "perverted sex* behavio?r" OR "incest*" OR "sex with minor" OR “intercourse with minor” OR “sex* aggression” OR "sex* crime" OR "sex* abuse" OR "sex* exploitation" OR "sex* harass*" OR "sex* homicide" OR "sex* murder" OR "sex* batter*" OR "rape" OR "child* pornography" OR sexual abnormalit*” OR "sex* pervert" OR "sex-pervert" OR “sexual perver*” OR “dangerous offenders” OR "sex* deviant" OR "sex* devianc*" OR "sex* perversion" OR "sex* sadism" OR "sex* interest in child*" OR "courtship disorder" OR "paraphilia" OR "hebephilia" OR "teleiophilia" OR "exhibitionism" OR "voyeurism" OR "frotteurism*" OR "pedophilia" OR "sexual psychopath" OR "sexual criminal psychopath" OR "sex* predator" OR " sexually violent predator" OR "hebephile" OR "teleiophile*" OR "exhibitionist" OR "voyeur" OR "frotteur" OR “paraphiliac” OR "sexual psychopath" OR "sexual criminal psychopath" OR OR "sex* assault" OR "sex* charge" OR "sex* delinquency" OR "indecent exposure" OR "gross indecency" OR “indecent assault” OR "carnal knowledge" OR "child* luring" OR "indecent behavior" OR “sexual felony” OR “unlawful intercourse” OR “unlawful sexual intercourse” OR “sex* delinquent” OR “sex* coercion" OR “sex* coercive” OR "sex* violence" OR "sexually violent" OR "sex* misconduct" OR “sexual harm” OR “inappropriate sex* behavio*” OR “atypical sexual behavio*”
"recidivism" OR "recidivist*" OR "recidivate" OR "recidivation" OR "rearrest*” OR "re-arrest*" OR "new arrest" OR "reconviction*" OR "re-conviction*" OR "new conviction" OR "reincarceration*" OR "re-incarceration*" OR "new incarceration" OR "parole violation* OR "new charge" or "new sex* charge" OR "new police contact" OR “lapse” OR "relapse*" OR "re-lapse*" OR "repeat offender*" OR "repeat offending" OR "repeat rape” OR “repeat sex* abuse” OR “repeat sex* offending” OR “discharged” OR “reoccurrence” OR “re-occurrence” OR “prognosis” OR “prognostic” OR “rehospitaliz*” OR “rehospitalis*” OR “re-hospitaliz*” OR “re-hospitalis*” OR “treatment outcome” OR “treatment impact” OR “treatment efficacy” OR “treatment effectiveness” OR “effectiveness of treatment” OR “efficacy of treatment” OR “impact of treatment” OR "community failure" OR "supervision failure" OR "parole revocation" OR “parole revoked” OR "parole discharged" OR “parole violation” OR "technical violation" OR “follow-up” OR “prison release” OR “parole suspension” OR “return to prison” OR “program evaluation” OR “program efficacy” OR “program effectiveness” OR “program outcome” OR “risk prediction” OR “risk assessment” OR “risk management” OR “desistance” OR “persistence” OR “offending trajector*” OR “continuity” OR “longitudinal study” OR “longitudinal data” OR “prisoner re-entry” OR “community re-entry” OR “community re-entry” OR “prisoner reentry” OR "reoffend*" OR "re-offend*" OR "re-offen?e*" OR "reoffen?e*" OR “criminal career”
 Greenland (1984) argued that the average time spent incarcerated by an individual convicted of a nonviolent sexual offense (e.g., homosexual acts between consenting adults) and designated as a dangerous sexual offender was longer than for those who perpetrated serious violent crimes (e.g., homicide).
 Following public outrage over a series of homicides involving children, a citizen’s initiative in the 1950s eventually led to the development of one of the first Canadian clinics specializing in the assessment and treatment of sexual deviance (Pascoe, 1961; Turner et al., 1958).
 The Canadian government provided the courts with a number of legal dispositions to prevent SOR (e.g., Dangerous offender legislation; long-term offender disposition; peace bonds). The new wave of SOL culminated with the enactment of a federal, non-public, sex offender registry that followed the example of Ontario‘s non-public registry. Historically, these SOLs, in particular preventive detention measures, have received very little attention from Canadian scholars given the courts’ relative reluctance to use them as well as the fact that, after the most recent version of Dangerous Offender legislation was initiated, very few designated DO return to the community.
 The search for empirical studies was initially focused on a series of terms. The terms that best capture the topic are: “sex offender recidivism”, “sexual recidivism”, “sexual reoffending”, and “sex crime recidivism”. While these terms may seem straightforward, focusing the search for those who are most likely to reoffend led to a number of problems. For example, the study is not limited to examination of individuals designated as convicted “sex offenders” but also includes subgroups of individuals who may have been convicted for specific sex offenses (e.g., rapists, sexual assaulters, etc.) as well as individuals who may present some form of sexual deviance (e.g., pedophiles, exhibitionists). The search for relevant publications was therefore significantly broadened to attempt to capture the whole extent of the scientific literature. In that context, a key “element” of systematic and meta-analytic reviews is identifying search terms and breaking them into key components.
 The following databases were ultimately selected for our search of academic publications: Academic search premier (EBSCO); Criminal Justice Abstracts (EBSCO/Proquest); Embase (Elsevier); ERIC (EBSCO); International Bibliography of the Social Sciences (ProQuest); Law Journal Library (HeinOnline); Legal Source (EBSCO). Medline (Ovid); NCJRS (ProQuest); PsycInfo (Ovid); Sociological Asbtract (ProQuest), and; Web of science.
 To complement the search of published studies in academic journals, the following sources were consulted: Grey Matters: a practical tool for searching health-related grey literature; Open grey; Publications numériques du Québec (BAnQ); Réseau Santécom; Réseau informatisé des bibliothèques gouvernementales; RIBG; Institut de recherche et documentation en économie de la santé: IRDES; Banque de données en santé publique: BDSP; RESSAC; Érudit; WHO-IRIS; Collection mémoires et thèses électroniques; ProQuest Dissertations & Theses Global; Thèses Canada; Theses.fr; DART-Europe; WHO MiNDbank; CrimDoc (Criminology Library Grey Literature); Conference Papers Index; EThOS (Electronic Theses Online Service); GreyNet International; Networked Digital Library of Theses & Dissertations; Google Scholar, and; ResearchGate.
 Recidivism rates were first pooled within a publication as several studies reported recidivism rates for different offender groups (e.g., child molesters, sexual aggressors of women, treated and untreated offenders). Pooling these rates was done by weighing the size of each sample.
 We considered using multilevel analyses by nesting rates per study location but there were too many variations in methodological features across publications.
 It sometimes made little sense to pool all findings from the same location/institution (e.g., the same 500 cases analyzed after 1 year, after 3 years, after 10 years). The strategy was then to identify and select one recidivism rate per sample, based on a number of criteria (in order of importance): (a) sample size (priority was given to larger samples); (b) length of the follow-up period (priority given to longer follow-up periods); (c) how complete and detailed the methodological information was.
 While using data from the period over which the follow-up took place would have provided complementary information, this information was rarely reported. The percentage of missing information was 83.4% for the variable marking the start (year) of the follow-up period and 63.7% for the period (year) marking the end of the follow-up period.
 Publication year may occur several years after the end of the follow-up period. For the current study, on average, the study was published 5.4 years (SD = 3.8) after the end of the follow-up period.
 This study is part of a larger study examining sex offender recidivism research (k > 850) around the world. The coding sheet was developed and tested using a random sample of 150 studies from this pool of international studies, including Canadian studies.
 No inter-rater agreement was computed for the variable reflecting the province where the study was collected. Surprisingly, perhaps, the information was not always easily available and required examining additional sources of information (websites, government resources, other publications, etc.). The data was collected by three researchers and cross-examined multiple times to identify errors, resolve missing data, and score the item.
 Indeed, the average length of follow-up in the older studies between 1940-1979 (Mean = 86.7; SD = 70.1) and 1980-1989 (Mean = 112.0 months; SD = 56.4) was longer compared to studies conducted between 1990-1999 (Mean = 84.0; SD = 44.9) and 2000-2009 (Mean = 53.7; SD = 28.5).
 Due to space limitations, not all results are reported but are available upon request from the first author.
 Earlier research on DSOs showed that about 40% returned to the community at some point during the follow-up period. Today, the percentage of DO who return to the community is closer to zero, a reflection of both changing practices and changes in the population designated as “dangerous” offenders from sexually deviant offenders (i.e., defined in terms of sexual orientation, interests, and preferences) to sexually violent recidivists.
 Although not reported here, our inspection of the recidivism rates using a trim-and-fill method suggests a selection bias toward reporting the study findings with higher sexual recidivism rates. This will be addressed in a future study.
 Based on ANOVA, across periods, there was a significant difference in the proportion of a sample that, at the time of recruitment, include persons who would be considered sexual recidivists [F(4, 5257) = 67.4, p < .001). The proportion of sexual recidivists being sampled peaked during the 1990s (Mean = 41.1%), which was significantly higher than all other study periods (Scheffé test, all p values < .001). In comparison, during the 1940-1979 period, the weighted mean prevalence of sexual recidivists was 34.8% and down to 30.3% for the 1980-1989 study period. In other words, the nature of samples studies across decades changed, but the change was not linear, and the highest proportion of sexual recidivists occurred when sexual recidivism rates were not at their highest and thus is unlikely to explain the drop observed in the current study. These findings should be interpreted with caution given that studies rarely reported information about the proportion of their sample that, at the time of the inclusion in the study, were considered sexual recidivists.
This is what actuarial risk assessment tools do: using a number of items, extrapolate the risk probabilities of an individual using data from another cohort (developmental sample). These actuarial tools typically present adjusted risk probabilities for some follow-up period (after 3 years, 5 years, 10 years, and so on). But it does not adjust the risk probabilities for other aspects that were shown to be statistically associated with recidivism (e.g., study setting, location, lead author affiliation).