Mental Disorder and Women’s Recidivism: A Meta-Analysis

This study quantitatively summarizes existing empirical research on the relationship between specific mental disorders and recidivism among justice-impacted women using meta-analysis. Eighteen studies were included following a comprehensive literature search. Results indicated that depression, PTSD, psychiatric history, and presence of any mental disorder (relative to no mental disorder) were independently and significantly associated with small increases in recidivism rates. Anxiety, psychosis-related and unspecified personality disorders, and self-harm/suicidality were not significantly related to recidivism. Findings support the gender-responsive position that some mental disorders are criminogenic and correctional practice should be guided by holistic, mental health- and trauma-informed approaches.


Introduction
Research continually demonstrates that a large proportion of women incarcerated in North America and Europe have mental health concerns at intake into correctional institutions and throughout serving their sentences (Al-Rousan et al., 2017;Bebbington et al., 2017;Fazel et al., 2016;Office of the Correctional Investigator, 2019;Prins, 2014).Mental health may constitute a specific responsivity issue, meaning that incarcerated individuals with untreated mental disorders may not be able to adequately engage in, or respond to, correctional programs aimed at reducing recidivism (McCormick et al., 2015).Although a large proportion of women involved in the criminal justice system have at least one mental health concern, the association between mental health and recidivism has only been examined in a limited number of studies (King et al., 2018;Van Voorhis et al., 2010).The results of these studies are equivocal, with some findings suggesting that severe mental health problems (e.g., mood disorders) are associated with increases in recidivism (Van Voorhis et al., 2010), while others have uncovered no relationship between mental health and reoffending (Hubbard & Pratt, 2002).Some researchers have also documented the opposite relationship -an inverse association between mental health and recidivism (Blanchette, 1996).Our study aims to address the limitations present in the literature by using metaanalytic techniques to summarize the relationship between specific mental health concerns and recidivism across a range of studies comprised specifically of justiceimpacted women.

Mental Health/Crime Link: Theoretical Basis
Although the potential direct associations between various mental disorders and recidivism have been investigated in a few empirical studies, comprehensive theoretical accounts attempting to explain the hypothesized mental health/crime link are relatively limited.However, gender-responsive paradigms that rely on trauma theory, holistic addictions theory, and relational theory (Brown et al., 2021) to explain criminal behavior among women, posit that trauma, childhood maltreatment, relational dysfunction, and poor mental health can lead to maladaptive behaviors such as substance use, which in turn manifests in dysfunctional survival strategies such as engagement in criminal behavior (see Daly, 1992;Harris & Fallot, 2001;Miller, 1988;Salisbury et al., 2017).Specifically, post-traumatic stress disorder (PTSD) has been theoretically linked to engagement in impulsive, risky, antisocial, self-destructive, non-violent (e.g., Jäggi et al., 2016), and violent behavior (e.g., Beckham & Moore, 2000;Komaroyskaya, 2009).
In addition, theoretical models and empirical evidence are emerging for specific symptom clusters.For example, theorists have proposed that the empirical association between borderline personality disorder (BPD) and violence (Moore et al., 2017) results from both dysregulated anger (Jackson et al., 2015), and elevated trait aggressiveness (Lowenstein et al., 2016;Moore et al., 2017).Schizophrenia has been theoretically and empirically linked with violent outcomes, but only when delusions and command hallucinations are present and left untreated (Keers et al., 2014;Lamsma & Harte, 2015).Finally, theoretical accounts and empirical findings have found a link between self-harm and both violence and recidivism (Power et al., 2013).
Overall, research has shown that justice-impacted women experience higher rates of mental health concerns than their male counterparts (Brown et al., 2018).As such, genderresponsive scholars argue that factors such as mental health, trauma, and emotional regulation should be prioritized in samples of justice-impacted women.However, these theoretical models have only proposed a limited number of symptom clusters and a minority of those symptoms are arguably, intuitively risk relevant.More research and theorizing regarding the link between mental disorders and crime is warranted.

Revisiting the Risk-Relevance of Mental Disorder in Treatment Contexts
Most studies examining the relationship between mental health and recidivism have been conducted with samples of men.One comprehensive study involving mostly men (Bonta et al., 2014) examined the predictive validity of eight clinical psychiatric indicators (psychosis, mood disorders, personality disorders, intellectual impairment, antisocial personality disorder/psychopathy, prior psychiatric admissions, length of inpatient stay in psychiatric hospital, and psychiatric treatment history) in a sample of individuals declared not criminally responsible on account of a mental health concern (NCRMD), and justice-impacted individuals who did not have this designation.General and violent re-arrests, re-convictions, or returns to a closed psychiatric facility for a new offense were assessed.Overall, findings indicated that only two clinical variables predicted general and violent recidivism: any (unspecified) personality disorder (PD) and antisocial personality disorder (APD)/Psychopathy were found to be moderate predictors of general recidivism (d = .44and d = .54,respectively) and violent recidivism (d = .41to d = .66,respectively).A comparison between the groups revealed no differences.
Although the Bonta et al. (2014) study had several strengths it was also limited in a few regards.First, when NCRMDs with psychosis were compared to NCRMDs without psychosis, it was not possible to ensure group equivalency on other mental disorders.Thus, either group could have had elevated rates of other disorders examined, making it impossible to decipher what disorder most influenced the resulting effect size and significance.Second, the comparison group consisted of justice-impacted men (and some women) who had not initially been labelled NCRMDs by the courts.However, within the comparison population, most individuals nevertheless have high rates of diagnosed mental disorders.This could certainly obscure the relationship between mental disorder and recidivism in this analysis.Third, the authors did not disaggregate findings by gender or utilize gender as a moderator in any analyses.Thus, it cannot be determined whether or how these results were affected by the inclusion of women, or whether they are replicable in all-women samples.

Women's Mental Health and Recidivism
There is a dearth of research examining the potential relationship between mental health and recidivism among justice-impacted women (King et al., 2018).The available individual empirical studies have yielded inconsistent results, with some finding that mental disorders are associated with increased recidivism (e.g., Benda, 2005;King et al., 2018), and others finding little or no support for such a link (e.g., Blanchette & Motiuk, 1996;Du et al., 2013).
Between the years 2008 and 2010, across four distinct studies, Van Voorhis and colleagues have arguably undertaken the most extensive and systematic investigation of risk factors for recidivism among women, including mental health variables.These studies included one sample of probationers in Maui (N = 158) and five independent Missouri samples: probationers (two partially overlapping (N = 272-298) and one independent sample (N = 304)), two pre-release (N = 150; N = 272), and one prison sample (N = 244).Recidivism was measured as breaches, any returns to custody, and rearrests, and follow up periods consisted of 6-, 12-, and 24-months.The mental health variables examined included any lifetime diagnosis of mental disorder, current depression and anxiety, and current psychosis and suicidality.Overall, having a lifetime diagnosis of any mental disorder did not significantly predict recidivism, whereas depression/anxiety were predictive in some samples but not others.For example, across the probation samples, elevated anxiety/depression significantly but modestly predicted breaches and returns at all follow-up times (r = .12to r = .20).Conversely, in the prison (N = 244) and pre-release samples (N = 150; N = 272), there was no relationship between depression/anxiety and recidivism.The psychosis/suicidality predictor was assessed in the probation sample of 298 women and only for breaches at six and 12-month follow-ups.It was found that current psychosis/suicidality predicted recidivism (r = .11to r = .12).
There are several reasons for the inconsistencies in the findings.First, the averages, range, and distributions of scores may have differed across sites (e.g., depression/ anxiety scores may have been uniformly high in prison relative to in probation samples).Second, there was no attempt to match samples on other important variables a-priori.Thus, samples could have differed systematically in terms of mental health profiles, correctional profiles, estimated risk, and criminogenic needs, as well as other relevant factors.Finally, it is unclear whether the scales and scoring procedures remained static across sites; the measures may not be directly comparable.King and colleagues (2018) investigated the relationship between severe mental disorders (SMD) and recidivism in a sample of 2,311 incarcerated women.SMD included depressive disorders, bipolar disorder, schizoaffective disorder, and dissociative disorder.Twenty percent of the sample (n = 453) were diagnosed with SMD.Recidivism was defined as a return to custody and the follow-up was eight years.Logistic regression analyses were used, and risk relevant covariates were controlled for including age, race, marital status, number of children, prior incarcerations, and substance use.Overall, SMD did not significantly predict recidivism (OR = 1.2, 95% CI = [.96,1.51] p = .11).In addition, a time-to-recidivism cox regression analysis was conducted.The hazard ratio indicated that women with SMD were at 16% greater hazard to recidivate than women without SMD (p < .05).There were two major limitations to the study.First, SMD did not include other conditions that can be severe, such as BPD, PTSD, or other psychotic disorders such as schizophrenia.If a meaningful proportion of women in the comparison group experienced any of these disorders, the relationship between SMD and recidivism would likely be attenuated.

The Current Study
The degree of variability in findings both within and between studies points to the need for additional research wherein attempts are made to address the limitations of existing studies.As such, the association between mental health concerns and recidivism for justice-impacted women requires further research (Becker et al., 2011;King et al., 2018).The aim of this meta-analysis was to quantitatively synthesize the existing literature regarding the potential association between specific mental disorders (e.g., depression, anxiety) and recidivism, as well as other mental health indicators (e.g., selfharm, psychiatric history) and recidivism among justice-impacted women.This metaanalysis differs from previous studies and contributes uniquely to scientific literature in several ways.It is the first quantitative synthesis of the relationship between mental health and recidivism specific to justice-impacted women.Second, the analysis includes both prospective and retrospective longitudinal studies.Retrospective studies were included only when the temporal order of the mental health assessment and the subsequent time at risk could be absolutely ensured (i.e., cross-sectional studies were excluded).Third, rather than restricting analyses to women who were incarcerated at the time of assessment or documentation, this study also included samples of women in the community.Finally, every effort was made to examine one mental disorder at a time, rather than combining various disorders into clusters such as "mood and anxiety disorders"safeguarding against the potential that the distinct disorders could be differently (e.g., one positively and another inversely associated) related to recidivism. 1  Overall, study hypotheses were derived from existing theory and evidence germane to how certain diagnostic symptom clusters (associated with specific mental disorders) may be related to behavioral outcomes.As such, there were four goals of the current study.The first goal was to examine whether various mental disorders were predictive of recidivism among justice-impacted women.This included an examination of whether the presence of any mental disorder was predictive of general recidivism.It was hypothesized that: (1) Depressive disorders, PTSD, BPD, and mixed internalizing and externalizing disorders (such as bipolar disorder) would significantly predict recidivism; (2) Non-PTSD disorders would negatively predict recidivism; and, (3) The presence of any mental disorder would not be associated with recidivism.
Second, this study examined whether externalizing behaviors indicative of mental health problems, such as self-harm and suicide attempts, predict recidivism.It was hypothesized that: (4) Externalizing behaviors would be associated with moderate increases in recidivism rates, particularly violent offenses.
Third, this study examined whether general mental health indicators predict recidivism.It was hypothesized that: (5) Non-diagnostic indicators of mental health, such as a history of psychiatric hospitalizations, would not be significantly related to recidivism.
Finally, the fourth goal of this study was to examine whether there were other factors that moderated the relationship between mental health disorder and indicators and recidivism.This goal was exploratory in nature; no hypotheses were made.

Method Inclusion and Exclusion Criteria
To be considered for inclusion the studies had to include: a sample of justice-impacted women, a minimum total sample size of 10 women, a measure of a mental health disorder or indicator used to assess mental health status in the relevant sample, an outcome measure of recidivism, and an effect size or data from which an effect size could be calculated.Excluded from this review were: studies utilizing exclusively inappropriately mixed mental health indicators, studies examining exclusively APD, psychopathy, or substance abuse/dependence disorders, studies where univariate effect sizes could not be obtained, and studies limited to continuous recidivism outcomes.Additionally, there were an insufficient number of studies reporting effects for violent recidivism, thus violent recidivism was not examined in the current meta-analysis.

Study Selection
A thorough search of published and unpublished literature was conducted.The end date of the literature search was March 31st, 2020.Published journal articles, book chapters, unpublished manuscripts, dissertations and theses, and government and university reports were considered for inclusion.The following academic databases were searched: Criminal Justice Abstracts, ProQuest Dissertations & Theses Global, Foreign Doctoral Dissertations, Medline, National Criminal Justice Reference Service (U.S.), Open Access Theses and Dissertations, ProQuest Databases, PsycArticles, PsycBooks, PsycINFO, PubMed, Sociological Abstracts, Social Sciences Fulltext, Theses Canada, Web of Science, University of Nevada Las Vegas (UNLV), and the Center for Crime and Justice Policy.In addition, multidisciplinary databases such as Academic OneFile, Cambridge Journals Online, Cambridge University eBooks, Sage Journals Online, ScienceDirect Journals, SpringerLink Journals, Taylor and Francis Journals Online, Wiley Online Journals, Worldcat, and the Criminal Behavior and Mental Health journal were also included.Finally, a number of non-academic and alternative databases were also searched (see Pettersen, 2022 for information on the specific databases).
A number of search terms were used within the various databases. 2The searches resulted in the identification of 751 records from academic sources and 8,751 records from alternative sources, including academic, government, and professional networking websites, Google Scholar, and manual searches.Based on titles and abstracts, all but 140 records were excluded.In some cases, the primary investigator or the corresponding author were contacted via email for additional information.In total, 47 authors were contacted and roughly 50% responded.Inquiries included attempts to obtain unpublished univariate data, gender-aggregated data, or additional contextual information (e.g., base rates, method of measurement, recidivism definition).Screening resulted in the rejection of 111 more studies.Reasons for exclusions were problematic designs (e.g., temporal order of measurement was reversed; cross-sectional design; k = 6), lack of data for women (k = 65), lack of a recidivism outcome (k = 10), absence of mental health predictors (k = 21), inappropriate non-criminal justice sample (k = 6), inappropriate youth sample (k = 1), inadequate data to derive an effect size (k = 9), inadequate comparison (k = 5) and studies that contained a sample of less than 10 women (k = 5).Twenty-nine studies remained eligible for inclusion, but eleven were rejected at the coding stage for: unreliable reporting (k = 1), containing an all-recidivist sample (k = 1), reporting only multivariate findings (k = 4), or for having an inadequate comparison (k = 5).Thus, the exhaustive database search resulted in eighteen studies suitable for inclusion.See Figure 1 for a graphic representation of the database search from start to finish.

Coding Protocol
The coding manual is available from the first author upon request.Study characteristics were coded, including publication status (published vs. unpublished), peer review status, (yes or no), and study design (retrospective vs. prospective).Sample characteristics for the overall samples before and after attrition, as well as any number of subsamples were also coded.Example items include (sub)sample size(s), the predominant mental disorder present in the sample, and mean age.Effect size data for group comparisons were also coded, which included mental health diagnosis or indicator, measurement of the mental health predictor, current versus lifetime diagnosis/indicator assessment, mental health assessment method and protocol, name of assessment measure, recidivism definition and information (e.g., re-arrests for a new offense vs. any returns to custody), source of recidivism data (e.g., official records vs. self-report), measurement of outcome (i.e., continuous, such as time-to-recidivism vs. dichotomous), length of follow up, and type of analysis.Finally, variables assessing independence of subsamples and recidivism outcomes, the number (%) of women who recidivated, and effect sizes were coded.General indicators of mental health (e.g., use of psychotropic medication, previous psychiatric treatment) and diagnoses presented in the DSM-IV and DSM-5 (APA, 2000(APA, , 2013) ) were coded.The study was not limited to assessment tools specifically adapted for any version of the DSM however, nor to dichotomous measures of mental disorder (i.e., disordered vs. non-disordered).
Risk of Bias.Risk of bias variables assess the degree of confidence with the study findings' representativeness and generalizability.To assess risk of bias, the following variables were coded: (1) the representativeness of the overall sample, the nondisordered cohort(s), and the disordered cohort(s), (2) cohort recruitment (i.e., whether the disordered and non-disordered cohorts were recruited from the same underlying population), and (3) whether the overall attrition rate at follow-up exceeded 20%.Notably, although these variables were coded for each study, risk of bias analyses could not be conducted due to three main reasons.First, there were challenges coding these items in each individual study as there was a lack of adequate reporting and definitional challenges.Second, there were considerable inconsistencies with how the same item was coded across studies.Finally, there was a lack of utility for these items in analyses (e.g., limited variability in scores, difficulty assessing comparability across studies, etc.).However, Table 1 presents a summary of the individual study findings for risk of bias variables.
Inter-Rater Reliability.To assess inter-rater reliability, 22% of studies (k = 4) were coded by both the primary and secondary coder.These studies were randomly selected and coded independently in a double-blind fashion.For categorical variables, inter-rater reliability was indexed by Cohen's Kappa, where values exceeding .40 are considered indicative of moderate levels of agreement.Inter-rater reliability for continuous variables was assessed using a two-way mixed model, Interclass Correlation Coefficient (ICC), with higher values indicating greater reliability.
Overall, perfect inter-rater agreement was achieved for 90% of all sample-and study-related variables coded (18/20).The setting in which the mental health assessment was conducted (e.g., community, local jail, state/federal prison, or forensic mental health unit/hospital) achieved moderate to high agreement, as indicated by a Kappa of .64.Risk assessment type was the only other variable for which perfect agreement was not obtained (Kappa = .43).Inter-rater reliability for effect sizes indicated that perfect agreement was achieved for 93% of these variables (14/15).Low inter-rater agreement was obtained for the mental health assessment method variable (Kappa = .27).This was because in a single study, eight effect sizes were reported, and one coder's misidentification of assessment method resulted in complete disagreement for all eight effects, having a strong negative effect on the overall agreement rating.However, agreement was perfect on this variable in three of four studies, indicating that coders generally understood how to assess the variable and follow the coding protocol guidelines.Notably, a coding meeting between first and secondary coders resulted in complete consensus on all items, and these consensus ratings were included in the analyses.For more details on the inter-rater reliability coding process and findings from this study, see Pettersen (2022).

Analytic Strategy and Effect Size Conversion
Choice of Effect Size Measures.Because of the need for an effect size measure suitable to analyses involving predictor variables sharing measurement scales (e.g., both binary and continuous predictors), Cohen's d, was selected to represent distinct, individual effects reported in each study included in the aggregate analyses.In addition, the Common Language Effect size (CLES), a universal standardized effect size measure was also included (McGraw & Wong, 1992).The CLES is the probability of obtaining a difference score greater than zero (0) in the distribution of interest.Following established standards (Borenstein et al., 2009), each derived effect size was weighted by the inverse of its variance.Weighting studies by the inverse of their variance is preferable to weighting by sample size alone.Thus, the smaller the variance of an individual effect size, the bigger the weight assigned to it, giving more weight to more precise effect sizes than broad, uncertain estimates (Helmus & Babchishin, 2013).
To aggregate effect sizes, a minimum of three wholly independent effect sizes are required for each predictor (Helmus & Babchishin, 2013).The following mental health predictors had less than three independent effect sizes: psychotropic medication use, compliance with psychotropic medication, suicide attempts as separate from non-lethal self-harm, and outpatient psychiatric treatment history.Thus, these variables were not examined in the meta-analysis.Additionally, given that the number of available studies was relatively small, some mental health indicators were combined and analyzed together.For example, effect sizes for self-harming behaviors and effect sizes for suicidal thoughts and attempts were combined into a single self-harm predictor.
Study Heterogeneity and Outliers.To assess between-study heterogeneity and the presence of outliers, both Cochran's Q and I 2 statistics were used.Cochran's Q provides information on the statistical significance of between-study variability, whereas the I 2 is the percentage of error above chance.Unlike Q, I 2 is not dependent on the number of included studies and can be compared across analyses.Higgins et al., (2003) suggest that 25%, 50%, and 75% can be considered "low", "moderate", and "high" degrees of variability, respectively.Finally, although there are no universally agreed-upon conventions outlining best practices for managing outliers in meta-analyses, Hanson and Bussière (1998) suggest that if the outlier is the most extreme value (i.e., either the largest or the smallest among the effect sizes), the overall Q is significant, and it accounts for more than 50% of the total variability in the aggregate effect, removing the effect size may be warranted.Helmus and Babchishin (2013) further note that an outlying effect size, in addition to having the largest or smallest value of those included in the specific analysis, should also have the most extreme weighted squared deviation.Analyses were conducted with and without the outlier(s) and both fixedand random-effects models were reported.Independence of Observations.Effect sizes are non-independent if the relevant sample included in analyses overlap partially or completely with another sample for which an effect size is available, whether the same or different outcome measures were used.Importantly, a number of studies reported a large number of effect sizes pertaining to identical predictors and outcomes.When this occurred, the included effect size was selected to maximize sample size, length of follow-up, and the comprehensiveness of the recidivism outcome.
Moderators.Four moderators were examined in the current study.First, a relatively limited number of studies and effect sizes were included, resulting in moderator variables that did not have a sufficient number of independent effects for each level of the moderator (categorical moderators).Second, moderator analyses require significant variability, requiring at minimum, a moderate I 2 value (Helmus & Babchishin, 2013).
As such, only moderators for depression and anxiety were examined.The moderators examined included: race (majority White vs. non-White majority), lifetime versus current assessment, peer reviewed versus non-peer reviewed, and study design (retrospective vs. prospective).

Depression and General Recidivism
There were 11 independent effect sizes available for depression.Depression included diagnosis of major depressive disorder (MDD; k = 2), dysthymia (DY; k = 1), depressive disorder not otherwise specified (DD-NOS; k = 4), scores on a measure of depression (k = 3), and scores on a measure of risk of developing depression (k = 1).
Table 3 presents descriptive study information including recidivism information, sample sizes, and attrition rates.It also details each study's unique effect size contributing to the meta-analysis.As outlined in this table, only two of the eleven contributing effect sizes were significant, indicating that depression was predictive of recidivism for these two studies.Table 4 displays the meta-analytic results which includes the weighted average Cohen's d effect size and associated 95% confidence intervals, and CLES effect sizes for all predictors of general recidivism.When looking explicitly at depression (with one outlying study removed), it was found that depression was a significant, albeit small predictor of general recidivism (Cohen's d = .13,95% CI [.06, .21]).However, when breaking the analyses down further, it was found that when only studies that used a binary diagnosis of depression were included in the analyses,   Van Voorhis et al., 2007, 2008, 2009, 2010).
depression significantly predicted general recidivism.In contrast, when including studies that measured depression using continuous scores, depression was not a significant predictor.
Depression Moderation Analyses.Results of all supplementary categorical moderator analyses are included in Table 5.In total, four categorical moderator analyses were examined; race (majority White vs. non-White majority), lifetime versus current assessment, peer reviewed versus non-peer reviewed, and study design (retrospective vs. prospective).Only race was a significant moderator, with results suggesting that depression significantly predicted recidivism only for samples in which most women were White (>60%).Data allowed for examination of length of follow-up and publication year as continuous moderators; neither were significant, t(df = 8) = 1.67, p > .05,ns; t(df = 8) = .89,p > .05,ns, respectively.

Anxiety and General Recidivism
There were 11 independent effect sizes available for anxiety.Anxiety predictors included all diagnoses of anxiety disorders: generalized anxiety disorder (GAD; k = 1), PTSD (k = 4), anxiety disorders NOS (k = 4), continuous scores on general measures of anxiety (k = 1), and anxiety and fear symptoms (k = 1).Table 3 presents descriptive study information including recidivism information, sample sizes, and attrition rates.It also details each study's unique effect size contributing to the meta-analysis.As outlined, only two of the eleven contributing effect sizes were significant, indicating that anxiety was predictive of recidivism for these two studies.Anxiety was not significantly associated with general recidivism in the main analysis (see Table 4).However, Q indicated significant betweenstudy variability, but no outliers were identified.Excluding PTSD from analyses did not affect the resulting effect size.The effect did not depend on how anxiety was measured; binary and continuous measurement yielded comparable effect sizes.
Anxiety Moderation Analyses.Results of all supplementary categorical moderator analyses are reported in Table 5. Results suggested that anxiety significantly predicted recidivism only in samples consisting mainly of White women (>60%).No other significant categorical moderation effects were identified.Nevertheless, small, significant effects were uncovered when analyses were restricted to effects involving only current mental health assessments and when studies were not peer reviewed, contradicting results of main analyses (d = .08-.20).Data allowed for examination of length of follow-up and publication year as continuous moderators; neither were significant, t(df = 8) = 1.67, p > .05,ns; t(df = 8) = .89,p > .05,ns, respectively.

Post-Traumatic Stress Disorder and General Recidivism
Post-Traumatic Stress Disorder was examined in isolation.The reasons for its separation from anxiety as a whole are as followed: (1) the etiology of typical anxiety disorders such as generalized anxiety disorder and PTSD are believed to be different -PTSD is caused by exposure to severe trauma, (2) the symptoms of PTSD are profoundly different from other anxiety disorders and it would be difficult to justify hypotheses suggesting their behavioral and life outcomes would be the comparable, (3) risk/needs correctional assessments measures and addresses PTSD as a condition wholly distinct from other anxiety disorders, and (4) due to the tremendous differences between cause, expression of, experience of, and reported outcomes of anxiety disorders in general relative to PTSD, it is possible that their potential relationships to criminal behavior may be juxtaposed (i.e., if combined into a single predictor, null findings, or completely unreliable conclusions may possibly occur).The initial hypothesis pertaining to PTSD specifically was supported; effect sizes were significant and indicated that PTSD is associated with increased recidivism (Cohen's d = .18,95% CI [.01, .35]).

Psychosis and General Recidivism
There were seven independent effect sizes available for psychosis, which included psychotic disorders NOS (k = 3) and symptoms associated with psychoses (k = 4).As outlined in Table 3, only one of the seven contributing effect sizes were significant.As seen in Table 4, the aggregate effect size for psychosis fell below threshold for statistical significance in the random effects model (Cohen's d = .14,95% CI [-.02, .29]).Due to inadequate independent effect sizes moderators for psychosis could not be examined.

Personality Disorder and General Recidivism
There were four independent effect sizes available for personality disorder (PD), including PD NOS (k = 2) and BPD symptoms (k = 2).As outlined in Table 3, three of the  b Includes overlapping study (Blanchette & Motiuk, 1996).four contributing effect sizes were significant.Treating PD as a unitary construct may not be optimal, given that authors in the two studies examining PD NOS did not describe the nature of these disorders.That is, diagnoses may or may not relate meaningfully to criminal conduct.Nevertheless, given that three of the four effect sizes were significant, the decision was made to make full use of the sparse data available and conduct exploratory analyses.Further, given that a clear outlier was detected, the analyses were run excluding the outlier (see Table 4).Results without the outlier resulted in a non-significant positive aggregate effect (Cohen's d = .20,95% CI [-.05, .44]).Due to inadequate independent effect sizes moderators could not be examined.
Overall, no firm conclusion can be drawn regarding the effect of PD on general recidivism.

Any Mental Disorder and General Recidivism
There were six independent effect sizes available for the any mental disorder including diagnosis of any major mental disorder (MMD; k = 2), 3 diagnosis of an axis I disorder (k = 1), 4 any history of mental disorder (MD; k = 1), and any mental disorder (undefined; k = 2).As outlined in Table 3, only one of the six contributing effect sizes were significant.As seen in Table 4, the aggregate effect size for any mental disorder was significant (Random effects: Cohen's d = .33,95% CI [.12, .53]),indicating that it significantly predicted general recidivism.

Self-Harm and Suicidality and General Recidivism
Self-harm was indexed by only four independent effect sizes and included suicidal ideation (k = 1), suicide attempts (k = 1), history of self-harm (k = 1), and suicidal or homicidal ideation (k = 1).As outlined in Table 3, only one of the four contributing effect sizes were significant.As seen in Table 4, the aggregate Cohen's d effect size was not significant in the random effects model (Cohen's d = .05,95% CI [-.06, .15]).

Psychiatric History and General Recidivism
Psychiatric history was indexed by three independent effect sizes, which included variables such as "registered with any public mental health service while at risk in the community".As outlined in Table 3, two of the three contributing effect sizes were significant.As seen in Table 4, the aggregate Cohen's d effect size was significant in both models (Cohen's d = .36,95% CI [.04, .68]),and was the largest aggregate effect size obtained in the current meta-analysis.Although psychiatric history may be a significant predictor of general recidivism, no firm conclusion can be drawn based on the limited data available in this study (k = 3; N = 667).

Discussion
The main goal of the study was to examine the potential relationship between various mental health indices and recidivism among justice-impacted women through quantitative syntheses of the existing literature.Diagnoses of specific mental disorders as well as general indicators of mental health status (e.g., self-harming behaviors) were examined.Results for the following mental health conditions are presented below: depression, anxiety, PTSD, psychosis, personality disorder, and any mental health disorder.Results for two general indicators: self-harm and suicidality, and psychiatric history are also presented.First, depression emerged as a significant but small predictor of recidivism.The relationship was evident when current or lifetime diagnoses were assessed, and when predictors were binary diagnoses.However, depression did not predict recidivism when continuous measures of depression were used.Although the extreme variability across effect sizes (when depression was measured continuously) may have accounted for the nil finding.The small relationship that did emerge between most measures of depression and recidivism can be explained, however.Consider a woman whose motivation is low, feelings of hopelessness intense, and whose future appears beyond her own control.In the absence of treatment, asking this woman to abandon dysfunctional or violent romantic connections (likely her only known source of connection), to seek and maintain the discipline necessary to complete vocational training or any other prescribed programming, or to abandon antisocial systems of belief seems unfathomable.Changing one's established behavioral patterns towards improved well-being, overall functioning, and better life outcomes, requires adequate motivation, energy, and an internal locus of control (i.e., change in one's external reality depends on one's own actions)-factors that characterize people with depression.
The results did not support any association between anxiety and recidivism.Thus, consistent with results of some previous empirical research (e.g., Hubbard & Pratt, 2002), the current meta-analysis suggests that anxiety may be among the less riskrelevant mental health conditions.In contrast, PTSD evidenced a small, but positive relationship with recidivism.Individuals with PTSD experience stress-induced negative affect, including clinically significant anger and hostility, chronic hypervigilance, paranoid ideation, acute fear, flashbacks to traumatic events that in many cases involved acts of extreme violence and other personal attacks such as sexual assault (Donley et al., 2012;Jäggi et al., 2016).Traumatic event(s) can alter systems of meaning and world views which further entrench beliefs about the world as unjust, dangerous, and acutely threatening, necessitating a proactive, self-protective, and sometimes aggressive approach (Jäggi et al., 2016;Jernigan, 2000;Moore & Elkavich, 2008;Parker et al., 2010).Moreover, it has been found that a proportion of trauma-victims and individuals diagnosed with PTSD develop an addiction-like response to trauma causing them to become involved in reenactments of the traumatic experience and risky, antisocial behaviors in general (see Levy, 1998 for further elaborations of several established theoretical accounts that seek to explain this effect).This psychiatric presentation is further combined with a chronic sense of a foreshortened future, meaning that the client perceives her life expectancy to be short, which naturally encourages impulsive behavior and short-term life strategies, including criminal behaviors (Herman, 1992;van der Kolk, 2001).The combination of these symptoms resulting from traumatic experiences and evidenced in PTSD and other trauma-related conditions are theoretically linked to engagement in impulsive, risky, antisocial, self-destructive, and non-violent (e.g., drug offenses; Jäggi et al., 2016) as well as violent crime (e.g., Komaroyskaya, 2009).
An interesting finding that emerged was that race moderated the relationships between two mental health conditions: anxiety and depression, with recidivism.That is, these findings suggest that both depression and anxiety, as measured in the included studies, predicted recidivism only in samples where most women were White (>60%).
It seems plausible that the potential causes underlying these effects may be due to the systematic racial bias that has been consistently found in assessment and diagnoses of psychiatric disorders, including in correctional populations (e.g., Baglivio et al., 2016), sentencing disparities along racial lines (Skeem & Lowenkamp, 2016), or the intersection between the two.However, the degree to which observed differences in diagnostic rates relate to the accuracy of the psychiatric assessments conducted in forensic contexts or the validity of sentencing decisions in terms of adherence to current legislation and legal precedence based on case law, remain unknown.Practically, from an assessment standpoint, we know that psychiatric diagnoses are descriptive terms and not scientifically derived constructs; the etiology of most mental conditions remains largely unknown as do optimal treatment approaches.To address questions regarding explanatory factors involved in racial differences of the relevance of various mental disorders to the prediction of recidivism, a number of steps are required.First, diagnostic criteria better informed by etiological factors that are likely to uncover the true prevalence rates in the population must be established.Second, the overall accuracy of diagnostic systems must be examined against the backdrop of these prevalence rates.Third, the reasons for sentencing disparities across race must be understood, and the interaction of these elements must be thoroughly examined.
From a theoretical standpoint, the finding that race moderates the relationships between anxiety and depression, and recidivism, may be further explained through an intersectional criminological lens.Research focusing on understanding the relationship between mental health and criminality among women has typically focused on individual deficiencies, using a gender-needs lens alone, which fails to consider other areas of oppression and control that impact women in the criminal justice system, as well as the health care system (Bunn, 2019;Gueta, 2020).Limited examination of how other need areas (e.g., experiencing homelessness) and axes of marginalization may interact with health care needs further ignores the intersection of barriers that women face and leads to partial and inadequate services for women-particularly racialized women (Gueta, 2020).Although it was beyond the scope of the current meta-analysis to explore how the interaction of needs predicts recidivism outcomes, the findings from this metaanalysis further denotes the importance of incorporating an intersectional approach to future research focusing on the relationship between mental health and recidivism among justice-impacted women.
Psychosis was not supported as a predictor of general recidivism.Unfortunately, only two studies reported effects for BPD, making it impossible to examine the riskrelevance of this condition, despite the link, as outlined in the literature, to antisocial, interpersonally aggressive, and criminal behavior (e.g., persons experiencing BPD are more likely to commit serious violent offenses including acts of intimate partner violence; Lawson et al., 2010).
The unexpected finding that any mental disorder predicted recidivism, producing one of the largest significant effect size (albeit still modest in size) obtained in the study, deserves mention.There are multiple potential explanations for this finding.It may simply reflect that a disproportionate number of the women included in these samples were diagnosed with the most risk-relevant disorders included in each study's definition of "any mental disorder".Conversely, it could mean that many women experienced the symptom clusters common to many diagnoses typically included in these studies: (1) lack of motivation, hopelessness and suicidality, an external locus of control, and low self-efficacy (e.g., MDD, PTSD), (2) emotional instability, particularly dysregulated anger (MDD, BPD, Bipolar disorder), fear (PTSD), hostility and feelings of dissociation and social alienation (MDD; PTSD), lack of trust in themselves and others, devaluation of one's own life, and dangerous world beliefs that together support impulsive, short-term life strategies including substance abuse and risk-taking (PTSD; BPD), or (3) impaired empathy (ADHD; BPD; Bipolar disorders), impulsiveness (ADHD; BPD; PTSD; Bipolar disorders), and a sense of entitlement (Bipolar disorders; BPD).An alternative explanation is that simply being afflicted with any mental disorder increases a woman's risk of recidivism through a mechanism that is not yet understood.
Personality disorder was examined as a unitary construct (i.e., BPD symptomology and personality disorders not otherwise specified) but did not emerge as a significant predictor of recidivism.In terms of non-diagnostic indicators, contrary to the hypothesis, existing theoretical accounts, and previous empirical findings, self-harm and suicidality did not significantly predict recidivism and the effect size was so small it is unlikely to have any clinical relevance.
Psychiatric history was the strongest predictor of recidivism and was in the range of the average effect reported in the field of psychology (d = .40;Cumming & Calin-Jageman, 2017).Individuals with more severe psychiatric disorders, particularly those that involve externalizing symptoms and other behavioral indicators, are more likely to make contact with psychiatric services both in the community and while incarcerated.For example, a person experiencing acute psychosis or a manic episode who is behaving irrationally or in a threatening way in public is more likely to be identified as having a mental health problem and be addressed psychiatrically relative to an individual with mild generalized anxiety without overt, externalizing symptomology.Thus, this finding appears to support the contention that severe mental health challenges, perhaps particularly those with externalizing features, may be at increased risk of recidivating.

Limitations and Considerations
Taken together, findings suggest that some aspects of women's mental health may be relevant to the prediction of recidivism.However, a crucial point to consider is that the validity and reliability of a meta-analysis depends on the amount of data available and the quality of the included studies in terms of overall scientific and methodological rigor (Helmus & Babchishin, 2013).Thus, the current findings are best considered tentative and exploratory in nature.
First, to ensure the current study provided meaningful findings, it was imperative that the sample, predictors, comparison group, and measured outcomes were precisely defined (Bown & Sutton, 2010).As such, in order to report accurate and precise estimates, this led to a stringent inclusion criteria, resulting in a small number of studies incorporated in the current analyses.As such, conclusions are deemed tentative and findings are in need of replication in future research.The most striking among the limitations uncovered, was the fact that no studies accounted for diagnostic comorbidity.Thus, in any given study, the inclusion of a woman in one disorder group (e.g., general anxiety) did not preclude her from inclusion in another disorder group (e.g., depression).Thus, resulting comparison groups likely consisted of a large proportion of women with another disorder other than the one being examined as a predictor of recidivism.Depending on the nature of the predictor diagnosis as well as the nature of the diagnoses of the women in the comparison group, this could cause attenuated or artificially inflated effect sizes.In short, no studies ensured that the comparison group consisted of women without mental disorder.Considering that recent estimations of the proportion of federally incarcerated women in Canada diagnosed with at least one mental disorder, which was approximately 80% of the total population in 2018 (OCI, 2019), adequately addressing this limitation would be challenging at the recruitment level.Nevertheless, a potentially achievable standard would be to exclude women who have disorders suspected of increasing the risk of recidivism (e.g., BPD) from the non-disordered comparison group.
Additionally, none of the included studies ensured group equivalency by using techniques such as matching procedures or conducted post-hoc analyses examining differences among disordered groups in terms of extraneous risk-relevant constructs.Thus, the disordered and non-disordered groups could have differed significantly on important factors including risk factors (i.e., antisocial history, antisocial associates, substance abuse) and overall estimated risk of recidivism.Only one study reported average or majority risk classifications for the overall sample, making investigations into the risk-relevance of mental health status by risk level impossible.Similarly, several variables included in the coding manual were never reported.This includes: the proportion of the women who had received mental health treatment specifically suited to their disorder, the proportion of women prescribed psychopharmaceuticals for their mental health condition and associated medication compliance rates, and the proportion of the overall sample who met diagnostic criteria for the disorder under investigation in recidivism analyses.
In terms of methodology and measurement limitations, it is worth noting that definitions of both predictors and outcome variables were wholly inconsistent across studies.For example, for the "any mental disorder" predictor, three studies failed to report the diagnoses included in their "catch all" variable, and two studies reported disparaging and narrow definitions.Similarly, length of follow-up and most importantly, base rates, varied immensely.Recidivism rates ranged from 7% to 59%.Also, the definition of recidivism was heterogeneous.The definition of recidivism included women returned to custody for any reason (such as a technical violation) or women officially reconvicted for a new crime.Thus, the extent to which we captured genuine criminal re-offending versus an inability to follow the rules is unknown.Similarly, it is difficult to disentangle to what extent re-involvement with the justice system reflects true criminal relapses versus being over-policed due to confounding factors such as racial biases or intensive supervision practices that many newly released justiceimpacted women will inherently find themselves subjected to (e.g., parole, intensive probation programs).In summary, the variability in all aspects of methodology and measurement made it difficult to determine whether the sampling reflect the same population of effect sizes or even the effect sizes for the same population of women.

Directions for Future Research
First and foremost, future research should be conducted to accumulate a larger number of viable effect sizes for both the predictors included in the current study as well as predictors for which virtually no data yet exist and that were excluded from the current syntheses.Researchers should further investigate the risk-relevance of PTSD, especially as it pertains to violent recidivism, ADHD, and bipolar disorders.Perhaps most relevant is BPD, which has substantial theoretical and empirical evidence supporting its relevance to recidivism risk.Although no meaningful relationship between selfinjurious behavior, suicide attempts and recidivism were uncovered, future investigations should continue to investigate the correlates and distal outcomes related to these externalizing behaviors.
Researchers attempting to study the relationship between mental health and correctional outcomes among adult women in future studies should make every effort to address the limitations of existing research and adopt higher standards in terms of sampling procedures, methodology, analytical approaches, and reporting practices.For example, the proportion of women with other potentially relevant disorders and the proportion of the disordered subsample's mental health treatment status and medication compliance rates must be reported.These variables are among the most crucial in this research context and could extinguish any effect associated with the disorder of interest, if found effective.A requirement of future research is to assess the estimated risk of the sample and the potential presence and effect of risk factors.Without addressing established risk-relevant constructs alongside mental health, the utility of assessing mental disorder and indicators of mental health, and using these to predict recidivism, will remain unclear.
Empirical research suggests that women with severe mental health challenges are involved disproportionately in serious institutional infractions, including violence, and represent a continuous challenge to institutional management and control, and to both staff and clients' safety and well-being (Lord, 2008).Additionally, punitive measures in correctional settings may undermine therapeutic effects of correctional programming.Further empirical investigations are required to examine institutional outcomes for women who are mentally ill, in terms of prevalence rates, mental health needs profiles, and the viability and efficacy of developing alternative management strategies that minimize the use of punitive measures that exacerbate existing mental health conditions while maintaining institutional order.
The current meta-analysis should be interpreted as a first, exploratory step in a longer-term program of research dedicated to the mental health of justice-impacted women.Thus, making definite practice and policy recommendations based on this study alone would be inappropriate.Nevertheless, correctional agencies routinely conduct their own treatment outcome studies.Given that findings suggest, across several analyses and across moderators, that some mental health conditions are related to recidivism, these agencies and their clients would likely benefit from methodologically rigorous treatment change and examination of correctional outcomes for both disordered and non-disordered women separately.This would constitute a starting point for the development of pre-treatment mental health interventions for those so acutely affected that their participation in programming is unlikely to elicit any change or develop treatment modules that specifically target symptom clusters linked directly with increased recidivism.

Conclusion
The current study is the first quantitative synthesis of existing empirical findings regarding the relevance of mental health to justice-impacted women's correctional outcomes.The findings provide crucial support for the construction of a scientifically sound, evidence-based understanding of the needs of this understudied population, provide rich data that can help guide and improve the quality of future studies, and demonstrates the relevance of mental health in the management and treatment of women in conflict with the law.Despite the need for replication and additional research on mental health indicators and various outcomes, the study lends support for the recommendation that correctional practice should be guided by holistic, mental healthand trauma-informed approaches to the management and treatment of justice-impacted women.Lastly, given that PTSD evidenced a significant, albeit small association with recidivism suggests that incarcerated women may benefit from interventions that directly target trauma as a core treatment target.These interventions are often called trauma-specific interventions.They differ from trauma-informed interventions.Trauma-informed approaches typically target the sequalae of trauma (e.g., substance use, emotional dysregulation) in a way that empowers women, respects women and makes them feel safe.In contrast, a trauma-specific approach focuses specifically on treating the trauma itself (Covington, 2022;Gueta et al., 2022).

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Figure 1 .
Figure 1.Literature Review Flow Chart: Academic Databases and Alternative sources.Note.The search engines, number of terms, and multitude of sources did not permit for derivation of the number of overlapping and/or unique studies.MH = mental health.
Note. '-' indicates no special sample features or missing data.P = Prospective; R = Retrospective.Peer Review = Minimum one document per study was peer-reviewed.CA = Canada; AU = Australia; U.S. = United States; U.K. = United Kingdom; SE = Sweden.CF = Correctional facility.COM = Community.FS = All fire-setters; LR = all low risk; SA = all substance abusers; Race = majority (≥60%); Mix = No majority race.a overlapping document (Blanchette & Motiuk, 1996).b Recidivism = fire-related offenses only.c Included overlapping document (Ministry of Justice, 2013).d Included four partially overlapping studies/documents ( Note.RCUS = return to custody; NC = New charge; REA = Re-arrest; REC = Re-conviction; RCJS = Return to criminal justice system; FREC = Felony re-conviction; AF = Any fail (any REA, REC, RCUS, and any technical breaches); OF = Offense fail (any re-arrest, re-conviction, or technical violations).x = missing data.Follow-up is presented in months and has been rounded to a single integer.% attrition has been rounded to one whole integer.n = sample in analyses.k = number of studies.a Recidivism = fire-related offenses (e.g., arson).

Table 1 .
Risk of Bias Variables by Individual Study.

Table 2 .
Study and Sample Characteristics.

Table 3 .
Mental Health, Recidivism, Attrition, Sample Size, and Effect Size by Study.

Table 4 .
Mental Health Predictors and General Recidivism: Aggregate Effects and Between Study Variability.Note.The primary model is the random effects model; for significant effects, bold font indicates that the effect size is interpretable, while standard font indicates that d from the fixed effects model is disregarded due to excessive between-study variability, regardless of significance.When d is not significant, bold font is not used.k = number of studies; N = sample size; CLES = Common Language Effect Size; Q = between study variability; I 2 = percentage of error above chance.PTSD = post-traumatic stress disorder.

Table 5 .
Supplementary Categorical Moderator Analyses: Aggregate Effects and Between Study Variability.Note.Only depression and anxiety had an adequate number of independent effect sizes for each level of the moderator and therefore the only predictors presented in this table.Depression analyses excluded the outlying effect size.k = number of studies; N = sample size; CL = Common language effect size; Q = between study variability; I 2 = percentage of error above chance.*Significant at p < .05.