The Characteristics of Persistent Sexual Offenders: A Meta-Analysis of Recidivism Studies

A meta-analysis of 82 recidivism studies (1,620 findings from 29,450 sexual offenders) identified deviant sexual preferences and antisocial orientation as the major predictors of sexual recidivism for both adult and adolescent sexual offenders. Antisocial orientation was the major predictor of violent recidivism and general (any) recidivism. The review also identified some dynamic risk factors that have the potential of being useful treatment targets (e.g

Sexual offenses are among the crimes that invoke the most public concern. Community surveys have found that 5% to 20% of men admit to at least one instance of sexual aggression (Grotpellier & Elliott, 2002;Koss, 1987;Lisak & Miller, 2002), and official records indicate that 1% to 2% of the adult male population will eventually be convicted of a sexual crime (California Office of the Attorney General, 2004; P. . The observed sexual recidivism rate is typically 10% to 15% after 5 years (Hanson & Bussière, 1998), but for some offenders, the rate is much higher . Identifying the characteristics of persistent sexual offenders is important for understanding this highly troubling behavioral disorder, as well as for the practical task of administrating policies directed toward high risk sexual offenders (e.g., treatment, civil commitment, community notification).
There is now a general consensus that sexual recidivism is associated with at least two broad factors: (a) deviant sexual interests and (b) antisocial orientation/lifestyle instability (Hanson & Bussière, 1998;Quinsey, Lalumière, Rice, & Harris, 1995;Roberts, Doren, & Thornton, 2002). Deviant sexual interests refer to enduring attractions to sexual acts that are illegal (e.g., sex with children, rape) or highly unusual (e.g., fetishism, autoerotic asphyxia). Although all sexual offending is socially deviant, men who commit such acts do not necessarily have enduring preferences for such behavior (Hudson & Ward, 1997; W. L. . Antisocial orientation refers to antisocial personality, antisocial traits (such as impulsivity, substance abuse, unemployment), and a history of rule violation. There is a strong association between rule violation and impulsive, reckless behavior, such as excessive drinking, frequent moves, fights, and unsafe work practices (Caspi et al., 1994;Gottfredson & Hirschi, 1990). Antisocial orientation facilitates sexual offending because individuals will not commit sexual crimes unless they are (a) willing to hurt others, (b) can convince themselves that they are not harming their victims, or (c) feel unable to stop themselves. Rapists are more likely than child molesters to have an antisocial orientation ; see review by West, 1983), but indicators of hostility and lifestyle instability are associated with sexual recidivism in both groups (Prentky, Knight, Lee, & Cerce, 1995;Rice, Quinsey, & Harris, 1991).
Contemporary theories posit that a variety of factors are associated with the development of sexual offending (Knight & Sims-Knight, 2003;Malamuth, 2003;Ward & Siegert, 2002). These models suggest that adverse family environments provide the breeding grounds for sexual offending. Lacking nurturance and guidance, the potential sexual offender develops problems in social functioning (e.g., mistrust, hostility, and insecure attachment) that, in turn, are associated with social rejection, loneliness, negative peer associations and delinquent behavior. The form of sexuality that develops in the context of pervasive intimacy deficits is likely to be impersonal and selfish and may even be adversarial. Further contributing to the risk of sexual offending are beliefs that permit nonconsenting sex. Attitudes allowing nonconsenting sex can develop through the individuals' trying to understand their own experiences and adopting the attitudes of their significant others (friends, family, abusers).
Such a model suggests that, apart from sexual deviancy and lifestyle instability, there may be three additional characteristics of persistent sexual offenders: (a) negative family background, (b) problems forming affectionate bonds with friends and lovers, and (c) attitudes tolerant of sexual assault. Examination of the most commonly used rating scales for sexual offenders indicate that these factors have considerable credibility among those conducting sexual offender evaluations (Beech, Fisher, & Thornton, 2003).
The research evidence addressing some of these factors is surprisingly weak. Hanson and Bussière's (1998) meta-analytic review found that the average correlation between sexual recidivism and sexual abuse as a child was r ϭ Ϫ.01 (based on 5 studies), and the correlation between recividism and deviant sexual attitudes was r ϭ .09 (4 studies). The relationship between sexual recidivism and intimacy deficits (apart from marital status) has not been the target of a systemic review.
Recidivism studies are important because sexual offenders are likely to have many problems, not all of which are related to sexual offending. For example, sexual offenders are at increased risk for anxiety and depression (Raymond, Coleman, Ohlerking, Christensen, & Miner, 1999), but such problems have not been associated with sexual recidivism (Hanson & Bussière, 1998). Similarly, Hanson and Bussière's quantitative review found that sexual recidivism was unrelated to the seriousness of the index offense (e.g., victim injury, intercourse) and clinical presentation variables (e.g., denial, victim empathy, motivation for treatment). Hanson and Bussière's (1998) meta-analysis made an important contribution by summarizing the available evidence concerning recidivism risk factors for sexual offenders. Most of their findings, however, concerned static, historical factors. Static factors are useful for long-term recidivism prediction, but those interested in understanding (and changing) recidivism risk are interested in dynamic (changeable) risk factors, also called "criminogenic needs" (Andrews & Bonta, 2003), "stable dynamic risk factors" (Hanson & Harris, 2000b) or "causal psychological risk factors" (Beech & Ward, 2004).
The purpose of the present study was to update Hanson and Bussière's (1998) meta-analysis in light of the ongoing research on sexual-offender risk assessment. Rather than repeat all the variables from Hanson and Bussière, in the present study, we considered only findings that (a) were considered important to the understanding and management of sexual offenders and (b) were weak or controversial in the earlier review (e.g., denial, victim damage). The review focused on sexual recidivism but also considered nonsexual violent and general (any) recidivism. Sexual offenders are more likely to reoffend with a nonsexual offense than with a sexual offense (Hanson & Bussière), and an important question is whether the predictors of sexual recidivism are substantially different from the predictors of nonsexual recidivism.

Sample
Computer searches of PsycLIT, the National Criminal Justice Reference Service of the United States, and the library of Public Safety and Emergency Preparedness Canada were conducted by using the following key terms: child molester, exhibitionism, exhibitionist, failure, frotteur, incest, indecent exposure, paraphilias(c), pedophile, pedophilia, predict(ion), rape, rapist, recidivate, recidivism, recidivist, relapse, reoffend, reoffense, sex(ual) offender, sexual assault, sexual deviant. Additional sources included the reference lists of empirical studies and previous reviews, recent issues of relevant journals (e.g., Criminal Justice and Behavior, Sexual Abuse: A Journal of Research and Treatment), and letters sent to 34 established researchers in the field of sexual offender recidivism.
Included were studies that examined offenders who were released after an index sexual offense and for whom the recidivism rate (sexual, violent, or any) was reported after a follow-up period. Researchers assessing offender characteristics were blind to recidivism status. Studies needed to include sufficient statistical information, and at least 5 subjects for all marginal totals were required for dichotomous variables.
As of January, 2003, our search yielded 115 usable documents (e.g., published articles, books, government reports, conference presentations). In 15 cases, the analyses were based on raw data or analyses provided by the original researchers. When the same data set was reported in several articles, all the results from these articles were considered to come from the same study. Consequently, the 115 documents represented 82 different studies (country of origin: 35 United States; 26 Canada; 12 United Kingdom; 2 Austria; 2 Sweden; 2 Australia; and 1 each from France, the Netherlands, and Denmark). Of these studies, 41 (50%) were unpublished. The studies were produced between 1943 and 2003 (median ϭ 1996); the average sample size was 359 (median ϭ 174, range ϭ 12-1,407). Thirtyfive studies were the same as those included in Hanson and Bussière's (1998) review, 10 studies contained updated information (e.g., longer follow-up periods, new analyses), and 37 studies were new.
Most of the studies examined mixed groups of adult sexual offenders (72 mixed-offense types, 7 child molesters, 2 rapists, 1 exhibitionist; 67 predominantly adults, 15 adolescents; all male). All the offenders had committed offenses that meet contemporary definitions of sexual crimes (i.e., old studies containing homosexuals were excluded). Most of the offenders were released from institutions (41 institution only, 17 community only, 22 institution and community, and 2 unknown). The offenders in 31 studies came from treatment programs. When demographic information was presented, the offenders were predominantly Caucasian (in 40 of 43 studies).
The most common sources of recidivism information were national criminal justice records (48), provincial or state records (32), records from treatment programs (18) and self-reports (13). Other sources (e.g., childprotection records, parole files) were used in 20 studies. In 34 studies, multiple sources were used. The source of the recidivism information for 10 studies was unknown. The recidivism criterion was arrest in 25 studies, reconviction in 24 studies, and reincarceration in 3 studies. Twenty-six studies used multiple criteria (e.g., arrest, parole violations, noncriminal justice-system reports). Recidivism was assessed solely from self-reports in 2 studies, and, in 2 studies, the recidivism criterion was unknown. The follow-up period ranged from 12 months to 330 months, with a mean of 76.0 months (SD ϭ 56.8).

Coding Procedure
Each study was coded separately by the two authors, who used a standard list of variables and explicit coding rules (available on request). Variables were first coded into specific individual variables (e.g., sexual interest in boys), which were then subsumed into general categories (e.g., sexual deviancy). The general categories were as follows: 1. Sexual deviancy: deviant sexual interests, such as children, rape, and other paraphilias, as well as sexual preoccupations and gender dysphoria); 2. Antisocial orientation: antisocial personality (e.g., antisocial personality disorder, psychopathy, Minnesota Multiphasic Personality Inventory Scale 4), antisocial traits (e.g., lifestyle instability, substance abuse, hostility), and a history of rule violation (e.g., childhood criminality, history of nonsexual crime, violation of conditional release); 3. Sexual attitudes: tolerance of sexual crime, support for adultchild sex, and low sex knowledge; 4. Intimacy deficits: poor social skills, negative social influences, conflicts in intimate relationships, emotional identification with children, and loneliness; 5. Adverse childhood environment: conflicts with and separation from parents, neglect, and physical and sexual abuse; 6. General psychological problems: internalization of psychological problems (e.g., anxiety and low self-esteem, as well as major mental illness); and 7. Clinical presentation: denial, minimization, lack of victim empathy, low motivation for treatment, and poor progress.
Only one finding per individual variable was coded per sample on the basis of (a) sample size and (b) completeness of information. If the sample sizes and descriptive detail were equivalent, the median value was used. Because many of the variables were intended to be dynamic (changeable), posttreatment findings were privileged over pretreatment findings except when the posttreatment finding was based on an insufficient number of cases.
Interrater reliability was calculated for approximately 10% of the sample (n ϭ 10), by using kappa for dichotomous and categorical variables and two-way random effects model intraclass correlation coefficients, or ICC (type absolute agreement) for ordinal and interval variables (Design 2 in Orwin, 1994). The agreement for the sample characteristics (e.g., adults or adolescents, treated or not) was perfect for 14 variables, excellent ( Ͼ .80; ICC Ͼ .85) for 6 variables, and fair ( Ͼ .50) for 3 variables: Did the recidivism criteria include parole violations, arrests, or nonjustice system reports? Kappa was low (-.11) for 1 variable (did the recidivism criteria include reconvictions?); for 8 of the 10 studies, the raters agreed that the researchers counted convictions, but each rater identified a different study where he or she thought conviction records had not been consulted.
The interrater reliability of the effect sizes was .83 for a single rater and .90 for the average of two raters. In the 10 reliability studies, Rater 1 identified 134 findings, and Rater 2 identified 131 findings, with agreement on 245 of the 265 findings (92.5%). The actual agreement would be greater because judges conferred on their final rating. Most differences involved simple omissions or clerical errors.

Index of Predictive Accuracy
The effect size indicator was the standardized mean difference, d, defined as follows: d ϭ (M 1 -M 2 )/S w , where M 1 is the mean of the deviant group, M 2 is the mean of the nondeviant group, and S w is the pooled-within standard deviation (Hasselblad & Hedges, 1995). In other words, d measures the average difference between the recidivists and the nonrecidivists, and compares this difference to how much recidivists differ from each other, and how much nonrecidivists differ from each other. The formula for calculating d can be found in Hanson and Morton-Bourgon (2004).
The d statistic was selected because it is less influenced by recidivism base rates than correlation coefficients-the other statistic commonly used in meta-analyses. According to Cohen (1988), d values of .20 are considered "small;" those of .50, "medium;" and .those of .80, "large." The value of d is approximately twice as large as the correlation coefficient calculated from the same data. When the 95% confidence interval for d does not contain zero, it can be considered statistically significant at p Ͻ .05. When the confidence intervals for two predictors do not overlap, they can be considered significantly different from each other ( p Ͻ .05).

Aggregation of Findings
Two methods were used to summarize the findings: median values (Slavin, 1995) and weighted mean values (Hedges & Olkin, 1985). The averaged d value, d., was calculated by weighing each d i by the inverse of its variance: where k is the number of findings, w i ϭ 1/v i , and v i is the variance of the individual d i (fixed effect model). The variance of the weighted mean was used to calculate 95% confidence intervals: Weighting d values by the inverse of their variance means that findings from small samples are given less weight than findings from large samples.
When d i was calculated from 2 ϫ 2 tables, the variance of d i was estimated by using Formula 6 from Hasselblad and Hedges (1995): When d i was calculated from other statistics (t, ROC areas, means, etc.), the variance of d i was estimated by using Formula 3 from Hasselblad and Hedges (1995): To test the generalizability of effects across studies, Hedges and Olkin's (1985) Q statistic was used: The Q statistic is distributed as a 2 with k -1 degrees of freedom (k is the number of studies). A significant Q statistic indicates that there is more variability across studies than would be expected by chance. Outliers were excluded from each category if the single extreme value accounted for more than 50% of the total variance (Q).

Results
On average, the observed sexual recidivism rate was 13.7% (n ϭ 19,267; 73 studies), the violent nonsexual recidivism rate was 14.3% (n ϭ 6,928; 24 studies), the violent recidivism rate (including sexual and nonsexual violence) was 14.3% (n ϭ 11,361; 29 studies) and the general (any) recidivism rate was 36.2% (n ϭ 12,708; 56 studies). Studies that specified in advance the number of recidivists and nonrecidivists were excluded from the rate calculations (e.g., Dempster, 1998). The average follow-up time was 5-6 years. These figures should be considered underestimates because not all offenses are detected.
The 82 studies produced 1,620 effect sizes (total of 29,450 sexual offenders). The average effect size for the published studies (M ϭ .23, SD ϭ .40, n ϭ 555) was the same as in the unpublished studies (M ϭ .22, SD ϭ .35, n ϭ 1,065), t(1618) ϭ .50, p ϭ .62. The correlation between sample size and effect size was positive (r ϭ .12, n ϭ 1,620, p Ͻ .001). (A negative correlation is expected when there is selective reporting of significant findings.) The correlation remained significantly positive after removing 103 d values estimated as zero because they were reported as nonsignificant.
Effect sizes were not related to publication date (r ϭ .001) or to the thoroughness with which researchers tracked recidivism information (r ϭ .026). The thoroughness of the recidivism search was rated on a 7-point scale (1 ϭ self-report only, 4 ϭ one method searches, and 7 ϭ multiple sources including national criminal records). The effect sizes were the same whether recidivism was measured by arrests (M ϭ .23, SD ϭ .34, n ϭ 1,001) or convictions (M ϭ .21, SD ϭ .39, n ϭ 542; t(1541) ϭ 1.14, p ϭ .25). The effect sizes for adult offenders (M ϭ .23, SD ϭ .36, n ϭ 1,319) were similar to effect sizes for adolescents (M ϭ .21, SD ϭ .37, n ϭ 301; t(1618) ϭ 1.11, p ϭ .27). The coding did not permit comparisons between rapists and child molesters. Table 1 presents the association with recidivism for the seven broad categories of risk factors. These broad comparisons were based on one individual finding per study (see coding manual and Hanson and Morton-Bourgon (2004) for further information about the individual predictors). The strongest predictors of sexual recidivism were those related to sexual deviancy (d. ϭ .30) and antisocial orientation (d. ϭ .23). The general category of sexual attitudes was also significantly related to sexual recidivism, but the effect was small (d. ϭ .17, 95% confidence interval of .04 to .28) and contained significant variability. The effects were not significant for child molester attitudes, low sex knowledge, or other deviant sexual attitudes (e.g., prudish attitudes toward masturbation).

Comparisons Across Categories of Risk Predictors
The general category of intimacy deficits showed a small, significant relationship to sexual recidivism (d. ϭ .15), with substantial variation among its subcomponents. Of the subcomponents of intimacy deficits, relatively larger effects were found for conflicts in intimate relationships (d. ϭ .36, 4 studies) and emotional identification with children (d. ϭ .42, 3 studies) than for social skills deficits (d. ϭ Ϫ.07, 6 studies) or loneliness (d. ϭ .03, 6 studies). The general categories of adverse childhood environment (d. ϭ .09), general psychological problems (d. ϭ .02), and clinical presentation (d. ϭ Ϫ.02) had little or no relationship with sexual recidivism.
Antisocial orientation (antisocial personality, antisocial traits, history of rule violation) was the major predictor of violent nonsexual recidivism (d. ϭ .51), violent (including sexual) recidivism (d. ϭ .54) and any recidivism (d. ϭ .52). Almost all of the individual indicators of antisocial orientation were significantly related to nonsexual violent, violent, and general recidivism; most of the relationships were moderate to large. Among the strongest individual predictors of any recidivism were general problems with self-regulation (d. ϭ .75, 6 studies, which included measures of impulsivity, lifestyle instability, and Factor 2 of the Psychopathy Checklist Revised, or PCL-R (Hare et al., 1990), a history of nonviolent crime (d. ϭ .68, 9 studies), psychopathy (PCL-R total scores; d. ϭ .67, 9 studies), and a history of nonsexual crime (d. ϭ .63, 8 studies).
Few variables other than antisocial orientation were predictive of nonsexual violent or general (any) recidivism. Sexual attitudes showed a small relationship with general recidivism (d. ϭ .24). Sexual deviancy was unrelated to violent nonsexual recidivism (d. ϭ Ϫ.05) and general (any) recidivism (d. ϭ .04).
The same major predictors were found for adolescent sex offenders as for adult sexual offenders. For adolescent sex offenders, sexual recidivism was predicted by sexual deviance (d. ϭ .36 Ϯ .24, 95% confidence interval of .12 to .60, 7 studies; n ϭ 734) and antisocial orientation (d. ϭ .19 Ϯ .17, 14 studies; n ϭ 1,958). Antisocial orientation also predicted violent nonsexual recidivism (d. ϭ .33 Ϯ .19, 5 studies; n ϭ 825), any violent recidivism (d. ϭ .46 Ϯ .26, 3 studies; n ϭ 559), and any recidivism (d. ϭ .41 Ϯ .13, 10 studies; n ϭ 1,400) among adolescent sex offenders. Table 2 presents some of the most promising targets for intervention. All of these findings were based on at least 5 studies with a combined sample of at least 1,000, with no significant variability between studies. The most confidence can be placed in those findings with narrow confidence intervals (large sample size). Readers should be cautioned, however, that the findings based on the minimum inclusion criteria (5 studies, 1,000 subjects) could still be substantially changed by new, large studies with divergent results. The promising dynamic risk factors included variables related to sexual deviancy (any deviant sexual interest, sexual preoccupations), antisocial personality (antisocial personality disorder, psychopathy as measured by PCL-R), and antisocial traits (general self-regulation problems, employment instability, hostility). The potentially misleading risk factors were negative family background, internalization of psychological problems, and poor clinical presentation (e.g., denial, low motivation for treatment).

Discussion
Most sexual offenders were not caught for another sexual offense (13.7%); on average, they were more likely to recidivate with a nonsexual offense than a sexual offense (overall recidivism rate of 36.2%). The major predictors of general (any) and violent recidivism were variables related to antisocial orientation, such as antisocial personality, antisocial traits, and a history of rule viola-tion. These are the same risk factors that predict general and violent recidivism among mentally disordered offenders (Bonta, Law, & Hanson, 1998) and unselected groups of offenders (Gendreau, Little, & Goggin, 1996).
The variables that predicted sexual recidivism were similar, but not identical, to the predictors of nonsexual recidivism. Sexual deviancy and antisocial orientation were the major predictors of sexual recidivism for both adult and adolescent sexual offenders. Sexual deviancy was unrelated to nonsexual recidivism. For the general categories of deviant sexual attitudes and intimacy deficits, some of the individual variables were related to sexual recidivism (e.g., emotional identification with children, conflicts in intimate relationships) and some were not (e.g., loneliness). Such variability suggests that further research is needed to uncover those aspects of attitudes and social functioning most associated with persistent sexual offending.
The present results also suggest that the factors that initiate sexual offending may not be the same as the factors associated with persistence. Negative family backgrounds and internalization of psychological problems are common among sexual offenders (Lee, Jackson, Pattison, & Ward, 2002;Raymond et al., 1999;Smallbone & Dadds, 1998), but these factors were unrelated to sexual recidivism. The prototypic sexual recidivist is not upset or lonely; instead, he leads an unstable, antisocial lifestyle and ruminates on sexually deviant themes. There is some evidence, however, that sexual offenders are more likely than other groups to respond to stress through sexual acts and fantasies (deviant or otherwise; Cortoni & Marshall, 2001;McKibben, Proulx, & Lusignan, 1994) thereby creating discrete time periods where they are at increased risk of sexual recidivism (Hanson & Harris, 2000b).
The distinction between sexual recidivists and nonrecidivists invites comparisons with Moffitt's (1993) distinction between adolescence-limited and life-course persistent delinquents. In Moffitt's typology, the life-course persistent offenders have behavior problems in childhood, engage in interpersonal violence, and have .14 Ϯ .17 .11 9.35 1,118 (7) Note. Each mean is followed by its 95% confidence interval. Total is number of subjects; k is the number of studies. All Q values were not significant, p Ͼ .05. many sexual partners (Sluyter et al., 2003)-all characteristics that predicted sexual and nonsexual recidivism in the present study. These characteristics can be considered common manifestations of low self-control (e.g., Gottfredson & Hirschi, 1990), but they have also been considered as an evolutionary adaptation to stressful childhood environments (Belsky, Steinberg, & Draper, 1991;Moffit, Caspi, Belsky, & Silva, 1992). The substantial overlap in the characteristics of persistent sexual and persistent nonsexual offenders suggests that those concerned with the assessment and management of sexual offenders could profit from the substantial literature on the assessment and treatment of general criminal offenders (e.g., Andrews & Bonta, 2003;Gendreau, French, & Gionet, 2004). For those involved in applied risk assessments with sexual offenders, the review confirms sexual deviancy and antisocial orientation as major predictors of sexual recidivism and extends the range of relevant variables to include some potentially changeable characteristics: sexual preoccupations, lifestyle instability/ impulsivity, pro-offending attitudes, and intimacy deficits. Readers will notice, however, that the predictive accuracy of most of the characteristics was small. Consequently, prudent evaluators need to consider a range of potential risk factors in an overall evaluation. The best methods for combining risk factors into an overall evaluation remain an active topic of scientific debate (Berlin, Galbreath, Geary, & McGlone, 2003;Hanson, Morton, & Harris, 2003).
Another outstanding research question is whether changes on the potentially dynamic factors are actually associated with reductions in recidivism risk. In general, evaluations of treatment progress showed little relationship to recidivism, with an average d. of .14. Nevertheless, there were some recent examples in which ratings of progress in treatment were significantly related to recidivism (Beech, Erikson, Friendship, & Ditchfield, 2001, d ϭ .50;Marques, Day, Wiederanders, & Nelson, 2002, d ϭ .55). Both of these studies used highly structured approaches to evaluating treatment gains and were informed by recent empirical research.
Evaluators may also note that many of the variables used in clinical assessments had little or no relationship with recidivism (e.g., denial, low victim empathy, low motivation for treatment). The lack of relationship may be linked to the difficulty of assessing sincere remorse in criminal justice settings. It is also possible that evaluators looking for risk factors have little to gain from listening to offenders' attempts to justify their transgressions. Psychotherapists often consider full disclosure desirable, and courts are lenient toward those who show remorse; few people, however, are inclined to completely reveal their faults and transgressions. Research has even suggested that full disclosure of negative personal characteristics is associated with negative social outcomes, including poor progress in psychotherapy (Kelly, 2000). Consequently, resistance to being labeled a sexual offender may not be associated with increased recidivism risk, even though it does create barriers to engagement in treatment. Offenders who minimize their crimes are at least indicating that sexual offending is wrong.
The present findings may also be useful to those wishing to improve treatment programs for sexual offenders. On average, sexual offenders who attend treatment are less likely to recidivate than are comparison groups (Hall, 1995;Hanson et al., 2002), but it is easy to locate well-controlled studies that find no effect for sexual offender treatment (e.g., Marques, Wiederanders, Day, Nel-son, & van Ommeren, 2005). For general offenders, treatment is effective only when it targets criminogenic needs (i.e., characteristics associated with offending; Dowden & Andrews, 2000). A review of the core treatment targets of sexual offender treatment programs (McGrath, Cumming, & Burchard, 2003, Table 9.1) suggests that most programs direct considerable resources toward characteristics that have little or no relationship with recidivism (e.g., offense responsibility, victim awareness, and empathy). An important question is whether programs that target the major predictors of sexual offense recidivism (e.g., lifestyle instability, deviant sexual interests, sexual preoccupations) are more effective than programs that target other factors.
Meta-analyses provide broad overviews and can easily neglect potentially important differences between studies. The definitions of constructs varied across studies, as did the samples. The present study focused on mixed groups of sexual offenders, and there was no effort to identify distinct predictors for specific subgroups (e.g., rapists, exhibitionists). Nevertheless, the findings were remarkably consistent. For 70% of the individual findings, the amount of variability across studies was no more than would be expected by chance ( p Ͻ .05). Furthermore, there was substantial consistency within many of the categories of predictors, with almost all of the variables being significant (e.g., sexual deviancy, history of rule violation) or nonsignificant (e.g., general psychological problems, clinical presentation; see Hanson & Morton-Bourgon, 2004). There is still substantial variability across studies that remains to be explained, but it appears that research is getting closer to identifying the constructs that are, and are not, related to recidivism among sexual offenders. Hopefully these research gains can be used to promote effective interventions and just social policies for sexual offenders.