Sexual Offender Recidivism Among a Population-Based Prison Sample

Abstract

The present study examines recidivism rates in sexual offenders using officially registered reconvictions in a representative data set of N = 1,115 male sexual offenders from Austria. In general, results indicate that most sexual offenders do not reoffend sexually after release from prison. More detailed, within the first 5 years after release, the sexual recidivism rate was 6% for the total sample, 4% for the rapist subgroup, and 8% for the child molester subgroup. The findings confirmed previous studies about sex offender recidivism which have shown that first-time sexual offenders are significantly less likely to sexually reoffend than those with previous sexual convictions. With regard to the relationship between age and sexual recidivism, the results challenged the traditional assumption of a clear linear function between age and recidivism. Taken together, compared with previous studies, the recidivism rates found in the present investigation are substantially lower than previous research has indicated.

Keywords

sexual offender recidivism child molester rapist prison

Sexual offending is an issue of high public concern and it is an undisputed fact that sexual violent behavior is a worldwide phenomenon that has long-lasting effects on the physical and mental health of victims (Craig, Browne, & Beech, 2008). Harris and Hanson (2004) pointed out that concerned politicians, the engaged media, and worried parents often assume that the recidivism risk of sexual offenders is extremely high and that “most sex offenders do it again” or at least “most sex offenders would do it again if you let them out.” Soothill (2010) concluded that the professional analysis of sexual (re-)offending is strongly affected by discrepancies between public and professional perceptions of sexual offending: On one hand, most research data indicate that only a minority of sex offenders is really dangerous with regard to recidivism. On the other hand, mass media tend to overreport sex offenses, focus on security aspects, and indicate an epidemic dimension of the problem which contributes to the appearance of a “sex crime panic” (Freedman, 1987). To reduce the distorted view of sex offender recidivism in the public, scientific knowledge is needed for a more objective consideration and public management of sexual aggression. Among others, the correct interpretation of recidivism rates of sex offenders derived from high-quality empirical studies is crucial in understanding the general risk for the public (Soothill, 2010). The present investigation about sex offender recidivism in Austria aims to contribute to a scientifically based and empirically founded discussion about sexually motivated (re-)offenses.

Previous meta-analytic research about sexual offender recidivism offered very heterogeneous results which were sometimes difficult to interpret. For example, Furby, Weinrott, and Blackshaw (1989) reported that because of the large variability between the studies, a meta-analytic combination of the studies was not possible at that time. The authors, instead, provided a comprehensive review and focused on the interpretation of qualitative trends and patterns across the included studies. However, even a qualitative analysis was difficult because of the large differences between the recidivism rates ranging from no relapse (e.g., Sturgeon & Taylor, 1980) up to rates indicating that almost every offender would recidivate (e.g., Gibbens, Way, & Soothill, 1977). Therefore, the authors concluded that “despite the relatively large number of studies on sex offender recidivism, we know very little about it” (Furby et al., 1989, p. 27).

About one decade later, Hanson and Bussière (1998) published until now by far the most cited meta-analytic review about sexual offender recidivism studies (Soothill, 2010). They identified 87 documents representing 61 different studies with the following countries of origin: 30 United States, 16 Canada, 10 United Kingdom, 2 Australia, 2 Denmark, and 1 Norway. The median of the publication date was 1989 which also indicates an increase of publications in the recent years. The 61 included studies provided information on a total of 28,972 sexual offenders. The average follow-up period was between 4 and 5 years. On average, the recidivism rate was 13.4% for the total sample for any sexual reoffense, 18.9% for the rapist subsample, and 12.7% for the child molester subsample, respectively (Hanson & Bussière, 1998). The recidivism rate for nonsexual violent reoffenses was 12.2% for the total (i.e., mixed) sample with a substantial difference between the child molester (9.9%) and rapist (22.1%) subgroups. For general recidivism (i.e., for any new criminal act) the recidivism rates were inevitably higher: For the total sample the recidivism rate was 36.3%, for the child molesters 36.9%, and for the rapists 46.2%.

In 2004, Harris and Hanson published an updated meta-analysis. They used data from only 10 Anglo-American follow-up studies (7 from Canada, 2 from the United States, and 1 from the United Kingdom) of adult male sexual offenders consisting of a combined sample size of 4,724 participants. Compared with the above-mentioned meta-analysis conducted by Hanson and Bussière (1998), there were longer follow-up periods up to 15 years and the researchers have explored the differential effects of several criminological variables on the reoffense risk like age, the relationship between offender and victim, previous offenses, and the time spent offense free in the community by the offender. For the combined total sample, the recidivism rate for any sexual reoffense was 14% within a 5-year, 20% within a 10-year, and 24% within a 15-year follow-up period. The recidivism rates for rapists and a mixed group of child molesters was similar. However, the authors reported significant differences within the child molester sample: The highest rates were observed among the extrafamilial boy-victim child molesters (35% after 15 years) and the lowest rates were found for the incest offenders (13% after 15 years). Harris and Hanson (2004) concluded their main results as follows: First, offenders with a prior sexual offense had recidivism rates about double the rate observed for first-time sexual offenders. Second, age was associated with a substantial decrease of recidivism. This was especially true for offenders older than 50 at release reoffending at half the rate of the younger (less than 50) offender group. And third, sexual offenders who remained offense free in the community for a certain period were at a substantially reduced risk for sexual recidivism.

The present study aims to provide a further contribution to the international status of research about sexual offender recidivism by investigating the recidivism rates of sexual offenders in a representative sample released from the Austrian Prison System. Until today, sex offender recidivism studies from German-speaking countries have received almost no consideration in the international state of research because of methodological shortcomings and weaknesses (Hanson & Bussière, 1998). For the present study, we applied a prospective-longitudinal research design which is regarded the best available study design for this purpose (Furby et al., 1989; Hanson & Bussière, 1998). Recidivism was defined as officially registered new convictions retrieved from the Federal Central Register of the Austrian Ministry of Internal Affairs. Reconviction rates were calculated for different follow-up periods to systematically examine the influence of time-at-risk on recidivism as demonstrated in previous studies (e.g., Prentky, Lee, Knight, & Cerce, 1997). Like in previous investigations about sexual offender recidivism (e.g., Hanson & Bussière, 1998), different outcome criteria in terms of different reconviction categories were used: general criminal reconviction (i.e., each new conviction for any kind of reoffense), violent reconviction (i.e., each new conviction because of nonsexual violent and sexual hands-on reoffenses), general sexual reconviction (i.e., each new conviction because of both sexual hands-on and hands-off reoffenses), and sexual hands-on reconviction. Because some previous studies (e.g., Rettenberger, Matthes, Boer, & Eher, 2010) have shown that there could exist substantial differences—at least for particular subgroups of sexual offenders—between general sexual recidivism which includes both hands-on and hands-off reoffenses and sexually motivated hands-on reoffenses only (i.e., sexual hands-on reconviction in the present study), we decided to differentiate between these two recidivism criteria, even if the majority of the existing research has used general sexual recidivism only.

To examine differential effects in recidivism, we calculated reconviction rates separately for different offender subgroups (e.g., rapists vs. child molesters). In the last step of the data analysis, the influence of two core variables in criminological research—age and previous delinquency—was examined by comparing subgroups of the total sample with different parameter values on the dimension of interest (e.g., by comparing young vs. old and first-time vs. previously convicted sex offenders). Given the previous literature about differential effects in recidivism of sexual offenders (e.g., Furby et al., 1989; Hanson & Bussière, 1998; Harris & Hanson, 2004; Hirschi & Gottfredson, 1983; Sampson & Laub, 2003), we expected that, first, child molesters show higher sexual recidivism rates than rapists for sexual recidivism categories, whereas for general and violent recidivism rapists exhibit higher recidivism rates; second, recidivism rates decline with increased age; third, first-time sexual offenders are significantly less likely to sexually reoffend than those with previous sexual convictions; and fourth, age and previous delinquency are generally robust predictors of future criminal behavior.

Method

Database and Collection Procedure

All included sex offenders were registered between 2001 and 2009 at the Federal Evaluation Centre for Violent and Sexual Offenders (FECVSO) of the Austrian Prison System, a department subordinated to the Austrian Ministry of Justice (Eher, Matthes, Schilling, Haubner-MacLean, & Rettenberger, 2012). The obligation of the FECVSO is to register every sex offender convicted of an unconditional prison sentence and to routinely evaluate sexual offenders from the whole country for risk assessment and treatment planning purposes. Every sexual offender sentenced to an unconditional prison term by an Austrian court because of a sexually motivated offense has to be reported to the FECVSO by the prison institution which is responsible for the inmate. Thus, the evaluation centre collects and evaluates data about all imprisoned sexual offenders in the Austrian Prison System. Data of sexual offenders who were convicted for a suspended sentence or just a fine (e.g., relatively young first-time offenders with minor hands-off offenses) are not reported to the FECVSO and, therefore, are not included in the database of the present study. Most of the participants of the present sample had been allocated to at least one treatment regime either during their prison sentence, or after conditional release, or both. Treatment variables were not considered for further statistical analyses in the present study, because no systematically collected data about the quality and quantity of treatment were available at the time of data analyses. A time period of 2.5 years was defined as the minimum follow-up period as previous research has shown that—using reconvictions as recidivism criterion—shorter periods provide no meaningful results (Prentky et al., 1997).

Statistical Analysis

The present study used fixed as well as unequal follow-up periods, that is, the actual exposure time can considerably vary between the participants, depending on their date of release from prison. The most commonly used method of estimating officially registered recidivism is to calculate the percentage or proportion of individuals who were reconvicted during the study period (Prentky et al., 1997). Of course, this method will inevitably underestimate the actual rate of recidivism, because some participants who are “on risk” for a shorter period of time may still relapse if they would have had been followed up for a longer period. This major methodological problem is addressed by using survival analyses (Helmus, Thornton, Hanson, & Babchishin, 2012). This method controls for unequal follow-up periods (Prentky et al., 1997). Therefore, survival analyses allow the inclusion of all cases of a sample into one single analysis (Helmus et al., 2012). Differences between cumulative survival curves (e.g., differences in the recidivism rates between rapists and child molesters) were tested using Wilcoxon (Gehan) statistics (Gehan, 1965; Pyke & Thompson, 1986).

To examine the relationship between age and previous delinquency, on one hand, and recidivism, on the other hand, Cox regression analyses were used for unequal follow-up periods and logistic regression analyses for fixed follow-up periods. Cox regression analyses generate hazard ratios (HRs; Exp[B]), which are similar to rate ratios (RRs; comparing two probabilities), but representing the relative risk at any given point of time (Babchishin, Hanson, & Helmus, 2012) associated with one or more predictor variables from data (Hanson, 2006). Similar to an odds ratio (OR), the HR is an indicator of the strength of the association between predictor and outcome. More specifically, “hazard functions assess the risk, at a particular moment, that individuals will fail if they have not already done so” (Babchishin et al., 2012, p. 4).

To compare the contribution of age and previous delinquency to the risk of reoffense with other studies, area under the receiver operating characteristic curves (AUC values of ROC curves; see, for instance, Hanley & McNeil, 1982) were calculated. The ROC curve is produced by plotting the hit and false alarm rates across all possible cutoff values. This statistical procedure is commonly used to examine the predictive accuracy of binary decisions, such as “release” or “do not release” (Mossman, 1994). The AUC values lie between 0 and 1, with an AUC value of .5 indicating prediction at the level of chance and an AUC value of 1 indicating perfect prediction. In the context of recidivism prediction, the AUC is commonly interpreted as the probability that a randomly selected recidivist will have a higher score on a risk variable than will a randomly selected nonrecidivist. Referring to Cohen (1992), Rice and Harris (2005) formulated criteria for the interpretation of the predictive accuracy of risk assessment tools: AUC values of .71 or above (r ≥ .37) are classified as “good” and AUC values between .64 and .70 (r ≥ .24) are classified as “moderate.” Significant AUC values that are below the value of .64 (r < .24) are classified as “small.”

Although ROC analyses have some important improvements in comparison with correlation coefficients, they also have some methodological shortcomings (Eher et al., 2012). Therefore, some authors propose the use of ORs calculated by using logistic regression analyses for reporting predictive accuracy (Hosmer & Lemeshow, 2000). Logistic regression analyses estimate the strength of the association between one or more predictor variables and a dichotomous outcome variable (e.g., recidivism vs. nonrecidivism). The most important indicator of the relationship between the variables is the OR which is not biased by a restriction of the range in either the predictor or the outcome variable.

Sample Characteristics

A total of 1,115 male¹ sexual offenders were registered between 2001 and 2009 at the FECVSO and were followed up after prison release until September 14, 2011. For the present study, a minimum follow-up period of at least 30 months was used. All offenders were convicted by an Austrian criminal court because of at least one sexually motivated crime and were subsequently detained in an Austrian prison. For registration purposes, the FECVSO collected data on demographics (e.g., age), criminological characteristics (e.g., number of previous offenses), and information about the current sex offense (e.g., data about the relationship between offender and victim). Data collection is based on an internal FECVSO coding manual which provided detailed definitions of all relevant variables for the data collection process. To warrant high data quality, the FECVSO staff involved in data entry was trained in the application of the FECVSO coding manual. Information about the age at the time of release as well as variables pertaining to the index offense, criminal history, and follow-up time is presented in Table 1. To separately examine the recidivism rates for different subgroups, a child molester and a rapist subgroup was defined due to their predominant victim type. Child molesters were defined as offenders having committed sexual crimes against victims younger than 14 years and rapists as having committed sexual crimes against victims 14 years or older at the time of the offense. Forty-one participants of the total sample were sexual hands-off offenders (e.g., exhibitionism or child pornography), sexual murderers, or offenders with sexually motivated crimes who were sentenced for a nonsexual offense (e.g., in cases of sexual burglary). Thus, those participants could not be allocated to either of the subgroups and therefore were excluded from further subsample analyses.

Table 1.

Characteristics of the Total Sample (N = 1,115).

	M	SD	Range
Index offense subgroups
Child molesters (n = 537)	48.2^a
Rapists (n = 532)	47.7
Other (n = 46)	4.1
Criminal history
No prior offenses	51.7
No prior violent offenses	68.3
No prior sexual offenses	88.3
Number of prior offenses	3.03	5.86	0-75
Number of prior violent offenses	1.04	2.59	0-40
Number of prior sexual offenses	0.27	1.44	0-32
Number of prior sexual offenses against a child	0.11	0.55	0-8
Number of prior sexual offenses against an adult victim	0.06	0.29	0-3
Number of prior sexual hands-off offenses	0.12	1.25	0-31
Age and follow-up
Age at time of release (years)	40.75	13.12	15.83-83.74
Follow-up (years)	6.39	1.70	2.52-10.45
Follow-up (months)	76.64	20.42	30.23-125.37

Values are percentages unless otherwise noted.

Results

Recidivism Rates of the Total Sample, Rapists, and Child Molesters

In the first step of data analysis, we used survival tables to examine the recidivism rates of the total sample (N = 1,115) as well as for both subgroups, child molesters (n = 537) and rapists (n = 532). Therefore, cumulative failure rates (FRs) as a function of exposure time were calculated where rates represent the proportion of the sample which did not “survive” (were reconvicted within a given time frame). Using annual steps, 10 time gates were generated to calculate recidivism rates. Table 2 shows the reconviction rates for the total sample (N = 1,115) for general, violent, sexual, and sexual hands-on recidivism.

Table 2.

Cumulative Recidivism Rates for General, Violent, Sexual, and Sexual Hands-On Reconvictions for the Total Sample (N = 1,115).

	Recidivism rates at given time gate
	1 year	2 years	3 years	4 years	5 years	6 years	7 years	8 years	9 years	10 years
General recidivism	.09 (SE = .01; n = 1,115)	.18 (SE = .01; n = 1,009)	.24 (SE = .01; n = 916)	.29 (SE = .01; n = 849)	.31 (SE = .33; n = 716)	.33 (SE = .01; n = 567)	.35 (SE = .02; n = 417)	.35 (SE = .02; n = 293)	.37 (SE = .02; n = 143)	.41 (SE = .05; n = 27)
Violent recidivism	.04 (SE = .01; n = 1,115)	.08 (SE = .01; n = 1,069)	.12 (SE = .01; n = 1,017)	.15 (SE = .01; n = 972)	.17 (SE = .01; n = 839)	.19 (SE = .01; n = 677)	.20 (SE = .01; n = 501)	.21 (SE = .01; n = 353)	.33 (SE = .02; n = 174)	.33 (SE = .02; n = 34)
Sexual recidivism	.01 (SE = .00; n = 1,115)	.02 (SE = .00; n = 1,098)	.03 (SE = .01; n = 1,079)	.05 (SE = .01; n = 1,054)	.06 (SE = .01; n = 921)	.06 (SE = .01; n = 750)	.07 (SE = .01; n = 563)	.08 (SE = .01; n = 379)	.08 (SE = 01; n = 186)	.08 (SE = .01; n = 36)
Sexual hands-on recidivism	.00 (SE = .00; n = 1,115)	.01 (SE = .00; n = 1,102)	.02 (SE = .00; n = 1,085)	.04 (SE = .01; n = 1,065)	.04 (SE = .01; n = 934)	.05 (SE = .01; n = 760)	.05 (SE = .01; n = 570)	.06 (SE = .01; n = 385)	.06 (SE = 01; n = 188)	.06 (SE = .01; n = 37)

Note. Recidivism rates are cumulative failure rates derived from survival analysis representing the proportion of the total sample which do not survive. n = number entering interval; SE = standard error of cumulative proportion nonsurviving at end of interval.

Not surprisingly, the highest recidivism rates could be found for general recidivism (any further reconviction), starting with a FR of 9% within 1 year’s follow-up up to 41% FR after 10 years. For violent recidivism, we found a stable recidivism rate of 3% to 4% per year throughout the first 4 years, dropping to 1% to 2% per year until the 8th year of follow-up. Thereafter, there is a relatively high increase between the 8th year and 9th year from 21% up to 33% FR. For sexual and sexual hands-on recidivism, a stable increase of 1% per year throughout the first 3 years could be observed, dropping to 1% for every 2 years until the 8th year. For the last 3 years no further increase was observable. However, especially the values of the last 2 years have to be interpreted with caution due to the small numbers of participants left for this long follow-up period.

Although the reconviction rates for general and violent recidivism are inevitably higher than for sexual recidivism, given the fact that the latter are subcategories of the former, it is, nevertheless, remarkable that both general sexual and sexual hands-on recidivism are lower to such an extent than the recidivism rates for general and violent reoffense. For example, taking a look at the 5 years reconviction rates, there is a substantial difference between general (31%) and violent recidivism (17%) inasmuch as the general recidivism rates are about 2 times higher. The difference, however, between violent and sexual recidivism is even higher: The 5 years violent recidivism rate (17%) is about 3 times higher than the rate for sexual recidivism (6%). Furthermore, both rates—sexual and sexual hands-on reconvictions—do not exceed the 10% FR limit within a 10 years follow-up period. Also, there are only marginal differences between general sexual and sexual hands-on recidivism.

The next two tables show the reconviction rates for both subgroups for each reoffense category, Table 3 for child molesters and Table 4 for rapists. Child molesters obviously showed lower FRs than rapists for general and violent recidivism (e.g., looking again at the 5-year rates: 38% vs. 23% and 24% vs. 10%, respectively). However, the general sexual reconviction rate was twice as high for the child molester subgroup as for the rapists (8% vs. 4%), whereas the difference between sexual hands-on recidivism was only marginal (5% vs. 4%). Statistical calculations by means of survival analyses demonstrated highly significant differences between child molesters and rapists for general (Wilcoxon [Gehan] test = 35.79, df = 1, p < .001) and violent recidivism (Wilcoxon [Gehan] test = 35.22; df = 1, p < .001). Only small differences could be found for sexual and sexual hands-on recidivism between child molesters and rapists. However, survival analysis shows that the difference between both subgroups reached statistical significance for sexual (Wilcoxon [Gehan] test = 4.89, df = 1, p < .05) but not for sexual hands-on recidivism (Wilcoxon [Gehan] test = 0.34; df = 1, ns).

Table 3.

Cumulative Recidivism Rates for General, Violent, Sexual, and Sexual Hands-On Reconvictions for the Child Molester Subgroup (n = 537).

	Recidivism rates at given time gate
	1 year	2 years	3 years	4 years	5 years	6 years	7 years	8 years	9 years	10 years
General recidivism	.05 (SE = .01; n = 537)	.11 (SE = .01; n = 508)	.16 (SE = .02; n = 477)	.21 (SE = .02; n = 448)	.23 (SE = .02; n = 378)	.24 (SE = .02; n = 316)	.26 (SE = .02; n = 226)	.26 (SE = .02; n = 163)	.27 (SE = .02; n = 89)	.36 (SE = .08; n = 17)
Violent recidivism	.02 (SE = .01; n = 537)	.04 (SE = .01; n = 527)	.07 (SE = .01; n = 510)	.09 (SE = .01; n = 499)	.10 (SE = .01; n = 434)	.12 (SE = .02; n = 366)	.12 (SE = .02; n = 262)	.12 (SE = .02; n = 190)	.14 (SE = .02; n = 105)	.14 (SE = .02; n = 22)
Sexual recidivism	.02 (SE = .01; n = 537)	.03 (SE = .01; n = 527)	.04 (SE = .01; n = 518)	.07 (SE = .01; n = 507)	.08 (SE = .01; n = 440)	.08 (SE = .01; n = 374)	.08 (SE = .01; n = 270)	.08 (SE = .01; n = 191)	.08 (SE = .01; n = 107)	.08 (SE = .01; n = 23)
Sexual hands-on recidivism	.01 (SE = .00; n = 537)	.01 (SE = .00; n = 531)	.02 (SE = .01; n = 523)	.05 (SE = .01; n = 517)	.05 (SE = .01; n = 451)	.05 (SE = .01; n = 382)	.05 (SE = .01; n = 276)	.05 (SE = .01; n = 196)	.05 (SE = .01; n = 109)	.05 (SE = .01; n = 24)

Table 4.

Cumulative Recidivism Rates for General, Violent, Sexual, and Sexual Hands-On Reconvictions for the Rapist Subgroup (n = 532).

	Recidivism rates at given time gate
	1 year	2 years	3 years	4 years	5 years	6 years	7 years	8 years	9 years	10 years
General recidivism	.14 (SE = .01; n = 532)	.24 (SE = .02; n = 458)	.30 (SE = .02; n = 404)	.35 (SE = .02; n = 370)	.38 (SE = .02; n = 312)	.40 (SE = .02; n = 233)	.42 (SE = .02; n = 178)	.43 (SE = .02; n = 121)	.43 (SE = .02; n = 50)	.43 (SE = .02; n = 9)
Violent recidivism	.06 (SE = .01; n = 532)	.12 (SE = .01; n = 497)	.17 (SE = .02; n = 465)	.21 (SE = .02; n = 436)	.24 (SE = .02; n = 372)	.25 (SE = .02; n = 283)	.27 (SE = .02; n = 219)	.28 (SE = .02; n = 149)	.30 (SE = .03; n = 63)	.30 (SE = .03; n = 11)
Sexual recidivism	.00 (SE = .00; n = 532)	.01 (SE = .00; n = 525)	.02 (SE = .01; n = 516)	.03 (SE = .01; n = 505)	.04 (SE = .01; n = 443)	.05 (SE = .01; n = 345)	.06 (SE = .01; n = 268)	.07 (SE = .02; n = 172)	.07 (SE = .02; n = 73)	.07 (SE = .02; n = 12)
Sexual hands-on recidivism	.00 (SE = .00; n = 532)	.01 (SE = .00; n = 525)	.02 (SE = .01; n = 517)	.03 (SE = .01; n = 506)	.04 (SE = .01; n = 445)	.05 (SE = .01; n = 347)	.06 (SE = .01; n = 269)	.07 (SE = .01; n = 172)	.07 (SE = .01; n = 73)	.07 (SE = .01; n = 12)

The Influence of Former Delinquency on Recidivism

In the next step, the impact of former delinquency on recidivism rates was investigated. First, reconviction rates were calculated using survival tables for first-time sexual offenders and offenders with a previous sexual conviction, respectively. For this purpose, only violent and sexual recidivism rates were calculated because general recidivism is of minor relevance for forensic practice (Furby et al., 1989), and differences between sexual and sexual hands-on recidivism were marginal in previous analyses. Of the total sample, 11.7% (n = 131) had at least one prior conviction because of a sexual offense, whereas the majority—88.3% (n = 984)—of the sample consisted of so-called first-time sexual offenders, that is, offenders without a sexual conviction prior to the current offense (Table 5). Within the first 5 years after prison release, 5% of the first-time sexual offender subgroup recidivated sexually, whereas 13% of the sexual offenders with at least one prior sexual offense were reconvicted for a new sexual offense. A similar difference, slightly smaller but still statistically significant, was observed for violent recidivism (16% vs. 25%). The difference of the FRs of both groups was highly significant for both reoffense categories (Wilcoxon [Gehan] test = 15.17, df = 1, p < .001). The Wilcoxon (Gehan) test for violent recidivism was also significant but there was more difference between both groups for sexual recidivism (Wilcoxon [Gehan] test = 4.26, df = 1, p < .05).

Table 5.

Cumulative Recidivism Rates for Violent and Sexual Reconvictions for First-Time Sexual Offenders (n = 984) Versus Offenders With Previous Sexual Convictions (n = 131).

	Recidivism rates at given time gate
	1 year	2 years	3 years	4 years	5 years	6 years	7 years	8 years	9 years	10 years
Sexual recidivism
First-time sexual offenders	.01 (SE = .00; n = 984)	.01 (SE = .00; n = 974)	.02 (SE = .00; n = 959)	.04 (SE = .01; n = 938)	.05 (SE = .01; n = 821)	.05 (SE = .01; n = 672)	.06 (SE = .01; n = 505)	.06 (SE = .01; n = 342)	.06 (SE = .01; n = 168)	.06 (SE = .01; n = 36)
Previous sexual convictions	.05 (SE = .02; n = 131)	.06 (SE = .02; n = 124)	.09 (SE = .02; n = 120)	.12 (SE = .03; n = 116)	.13 (SE = .03; n = 100)	.13 (SE = .03; n = 78)	.16 (SE = .04; n = 58)	.16 (SE = .04; n = 37)	.16 (SE = .04; n = 18)	—^a
Violent recidivism
First-time sexual offenders	.04 (SE = .01; n = 984)	.08 (SE = .01; n = 945)	.11 (SE = .01; n = 902)	.14 (SE = .01; n = 865)	.16 (SE = .01; n = 749)	.19 (SE = .01; n = 605)	.19 (SE = .01; n = 444)	.20 (SE = .01; n = 317)	.22 (SE = .02; n = 156)	.22 (SE = .02; n = 34)
Previous sexual convictions	.05 (SE = .02; n = 131)	.11 (SE = .03; n = 124)	.18 (SE = .03; n = 115)	.23 (SE = .04; n = 107)	.23 (SE = .04; n = 90)	.23 (SE = .04; n = 72)	.27 (SE = .04; n = 57)	.27 (SE = .04; n = 36)	.27 (SE = .04; n = 18)	—^a

Note. Recidivism rates are cumulative failure rates derived from survival analysis representing the proportion of the total sample which do not survive.

Not sufficient cases to calculate numbers; n = number entering interval; SE = standard error of cumulative proportion nonsurviving at end of interval.

In the next step, the influence of previous delinquency on violent and sexual recidivism was investigated by calculating ORs and RRs. This procedure allows examining the quantitative relationship between previous delinquency (i.e., the predictor) and sexual and violent recidivism (i.e., the predicted outcome). For calculating the ORs, a subsample of the total sample (n = 836) with fixed 5-year follow-up periods was used. Previous research about the validity of risk assessment instruments has indicated 5-year periods to be an appropriate follow-up time (Rettenberger, Boer, & Eher, 2011). The observed fixed 5 years recidivism rates were 6.2% (n = 52) for sexual and 17.6% (n = 147) for violent recidivism within the first 5 years after release from prison, respectively. Thus, the observed reconviction rates were similar to those calculated by survival tables presented in Table 2. For Cox regression analyses (RR) the total sample (N = 1,115) was used because it controls for unequal follow-up periods (Helmus et al., 2012). The results of both regression analyses and the AUC values resulting from ROC analyses are presented in Table 6. For ROC analyses, the number of prior convictions was used as the independent variable and violent and sexual recidivism as the dependent variable.

Table 6.

The Predictive Relationship Between Prior Delinquency and Sexual and Violent Recidivism Using ROC Curves as Well as Logistic and Cox Regression Analyses.

	Recidivism criterion
	Violent recidivism		Sexual recidivism
	AUC/OR/RR	95% CI	AUC/OR/RR	95% CI
ROC curves (n = 1,115)
Prior general offenses	0.69***	[0.63, 0.73]	0.66***	[0.58, 0.74]
Prior violent offenses	0.67***	[0.62, 0.73]	0.62**	[0.54, 0.70]
Prior sexual offenses	0.53 (ns)	[0.48, 0.59]	0.61**	[0.52, 0.70]
Cox regression (n = 1,115)
Prior general offenses	1.04***	[1.02, 1.06]	1.04***	[1.03, 1.05]
Prior violent offenses	1.05 (p = .08)	[0.99, 1.11]	1.08**	[1.06, 1.11]
Prior sexual offenses	1.28***	[1.21, 1.35]	1.07*	[1.02, 1.13]
Logistic regression (n = 836)
Prior general offenses	1.06**	[1.02, 1.09]	1.06***	[1.03, 1.09]
Prior violent offenses	1.06 (p = .16)	[0.98, 1.14]	1.17***	[1.09, 1.25]
Prior sexual offenses	1.45***	[1.21, 1.74]	1.11*	[1.01, 1.23]

Note. Cox regression analyses consider unequal follow-up periods. Logistic regression analyses use fixed 5-year follow-up periods. ROC = receiver operating characteristic; AUC = area under the ROC curves; ORs = odds ratios derived from logistic regression analyses; RRs = rate ratios derived from Cox regression analyses; CI = confidence interval; ns = not significant.

p < .05. **p < .01. ***p < .001.

The Influence of Age on Recidivism

To investigate changes in reconviction rates as a function of different age stages, a manageable number of age bands have to be defined. To warrant comparability with previous investigations about age and recidivism in sexual offenders, the allocation of age bands proposed by Thornton (2006) was used as a point of reference. Furthermore, there are a number of additional relevant methodological similarities between the present study and the investigation reported by Thornton which allow comparisons between both studies (e.g., the use of a representative sample of prison-released male sex offenders). However, Thornton’s first two age bands with offenders younger than the age of 18 and offenders aged between 18 and 24, in our study, were combined into one age band due to few cases in the age band younger than 18. Thus, four age bands were defined for the subsample with fixed 5-year follow-up periods (n = 836; see Table 7). Table 8 shows the violent and sexual reconviction rates for each age band for the total sample with fixed 5-year follow-up periods (n = 836) as well as for the child molester subgroup (n = 414) and the rapist subgroup (n = 388). In a next step, logistic regression models and ROC analyses were calculated to examine the predictive relationship between age and recidivism (see Table 9). For logistic regression, the age at time of release was used as independent variable, whereas for ROC analyses, age bands were used.

Table 7.

Absolute and Relative Frequencies by Age Bands Using the Subsample With Fixed 5-Year Follow-Up Periods (n = 836).

	Absolute frequency	Relative frequency
<25 years old	102	12.2
25-39.99 years old	329	39.4
40-59.99 years old	328	39.2
>60 years old	77	9.2

Note. The indication of age refers to age at the time of release from prison.

Table 8.

Sexual and Violent Reconviction Rates by Age Bands Using the Subsample With Fixed 5-Year Follow-Up Periods (n = 836).

	Reconviction rate (%)
	<25 years old	25-39.99 years old	40-59.99 years old	>60 years old
Total sample (n = 836)
Violent recidivism	38.2 (n = 39)	19.5 (n = 64)	12.2 (n = 40)	5.2 (n = 4)
Sexual recidivism	6.9 (n = 7)	4.3 (n = 14)	8.5 (n = 28)	3.9 (n = 3)
Child molesters (n = 414)
Violent recidivism	14.8 (n = 4)	12.4 (n = 16)	10.9 (n = 21)	6.2 (n = 4)
Sexual recidivism	7.4 (n = 2)	4.7 (n = 6)	10.9 (n = 21)	4.6 (n = 3)
Rapists (n = 388)
Violent recidivism	45.8 (n = 33)	23.9 (n = 45)	16.2 (n = 19)	0.0 (n = 0)
Sexual recidivism	6.9 (n = 5)	3.7 (n = 7)	6.0 (n = 7)	0.0 (n = 0)

Table 9.

The Predictive Relationship Between Age and Sexual and Violent Recidivism Using Logistic Regression Analyses and AUC Values (n = 836).

	Recidivism criterion
	Violent recidivism		Sexual recidivism
	OR/AUC	95% CI	OR/AUC	95% CI
Total sample (n = 836)
OR	0.95***	[0.93, 0.97]	1.01	[0.99, 1.03]
AUC	0.65***	[0.61, 0.70]	0.47	[0.39, 0.55]
Child molesters (n = 414)
OR	0.98 (p = .09)	[0.96, 1.00]	1.01	[0.98, 1.04]
AUC	0.56	[0.47, 0.65]	0.47	[0.38, 0.56]
Rapists (n = 388)
OR	0.94***	[0.92, 0.97]	0.98	[0.94, 1.02]
AUC	0.65***	[0.59, 0.71]	0.52	[0.39, 0.66]

Note. AUC = area under the receiver operating characteristic curves; ORs = odds ratios derived from logistic regression analyses; CI = confidence interval.

p < .05. **p < .01. ***p < .001.

As the reconviction rates in Table 8 indicate, there was a significant negative linear function between age at time of release and violent recidivism for the total sample (B = −.051, SE = .008, Wald = 38.27, df = 1, p < .001) and for the rapist subgroup (B = −.058, SE = .013, Wald = 21.18, df = 1, p < .001) as well as a trend for the child molester subgroup (B = −.021, SE = .012, Wald = 2.97, df = 1, p = .085). As indicated by the reconviction rates in Table 8, the relationship between age at time of release and sexual reconviction revealed to be more complex. For the total mixed sample (n = 836), there was no simple linear function between age and recidivism (B = .006, SE = .011, Wald = 0.31, df = 1, ns). To check for a possible nonlinear relationship between age and recidivism exhibiting one bend in it, the term “age at release squared” was included into the model.

The inclusion of age at release squared did not lead to a significant improvement of the model: log likelihood ratio χ²(1, n = 836) = 0.31, ns. However, adding a cubic term exhibited a clear tendency of statistical significance for the improvement of the model: log likelihood ratio χ²(1, n = 836) = 3.82, p = .051. This result indicates a cubic trend within the relationship between age on release and recidivism with two bends in it; in other words, the curvilinear relationship between age and sexual reconvictions is best described as a bimodal function.

Similar results could be observed in the child molester subgroup. First, there was also no linear function between age on release and sexual recidivism (B = .010, SE = .014, Wald = 0.56, df = 1, ns). The consideration of a quadratic term did not result in a significant improvement, log likelihood ratio χ²(1, n = 414) = 0.97, ns, but the addition of a cubic term again showed a tendency of improvement, log likelihood ratio χ²(1, n = 414) = 2.99, p = .084. For the rapist subgroup, again there was no linear relationship between age on release and sexual recidivism (B = −.022, SE = .023, Wald = 0.91, df = 1, ns). The inclusion of age at release squared did not lead to an improvement, log likelihood ratio χ²(1, n = 388) = 0.08, ns, whereas the inclusion of a cubic term exhibited a significant improvement of the model fit, log likelihood ratio χ²(1, n = 388) = 4.46, p < .05, indicating a curvilinear relationship between age on release and sexual reconvictions in rapists.

Discussion

The present study explores the recidivism rates of sexual offenders released from the Austrian Prison System between 2001 and 2009 by using a large and representative sample of male sexual offenders. These results about the total sample may justify two preliminary conclusions with regard to sexual recidivism: First, the sexual recidivism (reconviction) rates for prison-released sexual offenders are in general comparatively low indicating evidence that most sexual offenders do not recidivate and only a minority—less than 10%—of the sexual offender population in the prison system reoffends. This result is in line with a number of previous investigations and current meta-analytic research about sex offender recidivism (e.g., Hanson & Bussière, 1998; Harris & Hanson, 2004; Soothill, 2010). Second, there is obviously no big difference between general sexual recidivism (which includes hands-on and hands-off reoffenses) and sexual recidivism by hands-on offenses only, at least in our sample. Also this finding is not surprising given the fact that our sample consisted of hands-on sexual offenders in more than 95% of all cases.

In contrast to previous studies reporting higher sexual recidivism rates for rapists compared with child molesters (e.g., Serin, Mailloux, & Malcolm, 2001) or at least similar rates for both groups (Harris & Hanson, 2004), in our study the general sexual reconviction rate (hands-on and hands-off reoffenses) was twice as high for the child molester subgroup as for the rapists (8% vs. 4%), whereas the difference between sexual hands-on recidivism was only marginal (5% vs. 4%). However, rapists exhibited higher reconviction rates for general and violent recidivism than child molesters (38% vs. 23% and 24% vs. 10%, respectively). Higher relapse rates for general and violent offenses in the rapist subgroup can be linked to differences between child molesters and rapists in their degree of psychopathic personality traits (Porter et al., 2000). Research on typologies for sexual offenders has also shown that constructs related to general criminality like lifestyle impulsivity and pervasive anger are more relevant for rapists, whereas constructs related to sexual deviance like degree and fixation of paraphilic interests seem to be of major importance for child molesters (Eher, Neuwirth, Fruehwald, & Frottier, 2003; Prentky et al., 1997). Taken together these research findings and the results of the present study, the observed differences in reconviction rates may be due to the fact that, at least in our sample of prison-released sexual offenders, rapists are more generally criminal, whereas child molesters have more sexual problems.

Next we investigated the influence of prior delinquency and age on the recidivism rates or our sexual offender group. We could find a significant difference between the reconviction rates of first-time sexual offenders and offenders with previous sexual convictions. Again, this result is in line with previous findings about the influence of prior sex offenses on the recidivism risk: Harris and Hanson (2004) compared offenders with and without previous sexual convictions using a large sexual offender sample (n = 4,724). Within a follow-up period of 5 years, they found the sexual recidivism rate for offenders previously convicted for sexual offenses about 3 times higher than for first-time sexual offenders (10% vs. 25%). However, the recidivism rates reported by Harris and Hanson were substantially higher compared with our rates, which might at least partly be due to the fact that five subsamples of their meta-analysis had used more sensible recidivism criteria like charges or police information.

The impact of previous offenses on the predictability of future crimes was also confirmed by the results of our regression and ROC analyses. With reference to ORs and RRs, the results indicated a stable and significant association between the number of previous general, violent, and sexual offenses, on one hand, and violent and sexual reconviction rates, on the other hand, meaning that an increase in the number of previous convictions was associated with a significant increase in violent and sexual recidivism rates. For example, concerning the number of prior violent convictions, an OR of 1.17 indicates that the odds for sexual recidivism (relationship between recidivists and nonrecidivists) increase by 17% for each increase by 1 of the total number of previous violent offenses. Furthermore, in the ROC analyses, the number of previous convictions as the independent variable and reconviction as dependent variable exhibited AUC values from moderate too good (Rice & Harris, 2005), indicating that previous delinquency has a relatively high predictive validity in sexual offenders. As another example, violent recidivism was predicted simply by the total number of previous general offenses at the level of AUC = .69, which is remarkable given the fact that most established risk assessment instruments for sexual offenders have comparable AUCs (Helmus et al., 2012; Rettenberger et al., 2011). These findings once more support one of the oldest assumptions in psychological prediction research already formulated, for example, in 1911 by Edward L. Thorndike: The best predictor of future behavior is past behavior.

Similar to the relationship between previous delinquency and recidivism, the relationship between age and recidivism has produced a huge amount of research in the last decades (e.g., Hirschi & Gottfredson, 1983; Sampson & Laub, 2003). In a nutshell, most researchers in criminology and psychology have described the aging effect as a linear function in terms of a stable decrease in crime rates when offenders are getting older. In simple words, the older the offender, the lower the probability of criminal behavior (Hirschi & Gottfredson, 1983). In the present study, such an “aging effect” could also be observed for the total sample as well as for child molesters and rapists separately, but only for violent recidivism. The total sample and both subgroups showed a consistent decrease in the proportion of recidivists across the four age bands for violent recidivism, with a stronger aging effect for the rapist subgroup than for the child molesters. However, no such effect could be found between age and sexual recidivism. Regression analyses exhibited no significant linear relationship between age at the time of release and sexual recidivism meaning that the simple formula of the aging effect—the older the sex offender, the lower the recidivism risk for sexual violence—could not be applied.

Concerning the total sample and especially the rapist subgroup, however, the inclusion of a cubic term led to a significant (for the child molesters significance was failed marginally) improvement of the model indicating a curvilinear relationship between age on release and sexual reconvictions in this sample. In other words, the relationship between age on release and recidivism may best be described not by a linear function (like for violent recidivism) but rather by a curve with two bends in it. Table 8 illustrates this relationship: Between the first and the second age band, a decrease can be observed, which becomes a striking increase between the second and third age group. Thereafter, the reconviction rates dropped down again in the last age band.

The results of the present study are in line with other current findings about the relationship between age and recidivism which has challenged the traditional assumption of a clear linear function between age and recidivism in sex offenders: Thornton (2006), for example, postulated “a variable connection” (p. 123) between age and sexual recidivism and also reported a cubic trend for a subsample with prior sexual offenses in his data set. He provided a hypothesis which might explain this more variable connection between age and recidivism compared with other offender populations and suggested that a possible explanation for the lack of a further decrease during the middle years of life—that is, in the present study from the second to the third age band—might be that in these age groups sex had come to serve as the primary means of gratifying not only sexual motives but rather a diverse range of motives like expression of anger, reduction in stress, or the gratification of intimacy needs. Thus, the reduction in one motivational force (e.g., sexual hormones) could still leave sexual delinquent behavior motivated by a range of other forces (Thornton, 2006). The decrease in recidivism for the oldest offenders in this sample (released at age 60 or older) might then reflect new variables beginning to act (e.g., loss of physical health; see Thornton, 2006). Given the results of the present study showing that child molesters have consistently higher recidivism rates throughout the life span, it could be further hypothesized that these additional motivational forces might be more relevant and present in child molesters than in rapists. Of course, these assumptions are beyond the scope of the present study but might provide starting points for future research.

There are some relevant limitations of the present study which should be addressed. First of all, there are relevant subgroups of sexual offenders who were not considered in the present sample. For example, for the present study there was no differentiation between intra- and extrafamilial child molesters despite the fact that previous research indicated substantial differences between these groups (Eher & Ross, 2006; Harris & Hanson, 2004). Future research with the present data set might identify additional differential effects in the child molester subgroup (a further differentiation relevant to recidivism would be, for example, “girl-victim” vs. “boy-victim” child molesters). Also, the small but important group of forensic patients detained in forensic psychiatric hospitals was not part of this study. Another risk-relevant group exhibiting different relapse rates might be sexual offenders in outpatient settings. Therefore, interpretation of the present study results may not be transferred to any of these sex offender subgroups because previous investigations indicated risk-relevant differences between these groups and “routine” prison-released sex offender samples (Phenix, Helmus, & Hanson, 2009). The international generalizability of the present findings might also be reduced because of potential differences between prison-released samples of sexual offenders of European and non-European countries.

Another possible limitation concerns the source of recidivism data which were retrieved from only one data source: For the present study, we have used official reconviction data from the computerized database of the Austrian Ministry of Internal Affairs as the only recidivism criterion. As a consequence, the reconviction rate is inevitably an underestimation of the actual recidivism rate because it has to be assumed that a substantial amount of sexual (re-)offenses remains still undetected and some of the detected (re-)offenses do not lead to official convictions or at least not to official sexual (re-)convictions² (e.g., Craig et al., 2008). Furthermore, with the data set of the present study, we were not be able to investigate if offenders recidivated in another country than Austria because we had no access to foreign recidivism databases. This potentially biasing effect could be especially due to offenders with a foreign nationality (i.e., non-Austrian offenders). The next methodological issue refers to the follow-up periods. The informative value of the present study is limited to follow-up periods which do not exceed 10 years after release from prison. Furthermore, due to comparatively small sample sizes in the higher time gates, especially the recidivism rates for the longer follow-up periods have to be interpreted cautiously. However, the prospective-longitudinal research design of the present study precluded the opportunity to extend the follow-up periods but at the same time provided more reliable results, given the fact that a prospective-longitudinal research design is regarded as the best available study design for recidivism studies (Furby et al., 1989; Hanson & Bussière, 1998). Nevertheless, despite of all these limitations, the present study provides insights which could be relevant for criminal justice policy as well as for future criminological research.

Footnotes

Declaration of Conflicting Interests

Peer Briken is consultant at Dr. Pfleger GmbH (Bamberg, Germany). The authors declared no other potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Babchishin

K. M.

Hanson

R. K.

Helmus

(2012). Communicating risk for sex offenders: Risk ratios for Static-2002R. Sexual Offender Treatment, 7, 1-12. Retrieved from http://www.sexual-offender-treatment.org/111.html

Cohen

(1992). A power primer. Psychological Bulletin, 112, 155-159.

Craig

L. A.

Browne

K. D.

Beech

A. R.

(2008). Assessing risk in sex offenders: A practitioner’s guide. Chichester, UK: John Wiley & Sons.

Eher

Matthes

Schilling

Haubner-MacLean

Rettenberger

(2012). Dynamic risk assessment in sexual offenders using STABLE-2000 and the STABLE-2007: An investigation of predictive and incremental validity. Sexual Abuse: A Journal of Research and Treatment, 24, 5-28. doi:10.1177/1079063211403164

Eher

Neuwirth

Fruehwald

Frottier

(2003). Sexualization and lifestyle impulsivity: Clinically valid discriminators in sexual offenders. International Journal of Offender Therapy and Comparative Criminology, 47, 452-467. doi:10.1177/0306624x03253844

Eher

Ross

(2006). Reconsidering risk for reoffense in intrafamilial child molesters: New aspects on clinical and criminological issues. Sexual Offender Treatment, 1, 1-9. Retrieved from http://www.sexual-offender-treatment.org/39.html

Freedman

E. B.

(1987). Uncontrolled desires: The response to the sexual psychopath, 1920-1960. Journal of American History, 74, 83-106. doi:10.2307/1908506

Furby

Weinrott

M. R.

Blackshaw

(1989). Sex offender recidivism: A review. Psychological Bulletin, 105, 3-30. doi:10.1037//0033-2909.105.1.3

Gehan

E. A.

(1965). A generalized Wilcoxon test for comparing arbitrarily singly-censored samples. Biometrika, 52, 203-223. doi:10.2307/2333825

10.

Gibbens

T. C. N.

Way

C. K.

Soothill

K. L.

(1977). Behavioral types of rape. British Journal of Psychiatry, 130, 32-42. doi:10.1192/Bjp.130.1.32

11.

Hanley

J. A.

McNeil

B. J.

(1982). The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology, 143, 29-36.

12.

Hanson

R. K.

(2006). Does Static-99 predict recidivism among older sexual offenders? Sexual Abuse: A Journal of Research and Treatment, 18, 343-355. doi:10.1007/s11194-006-9027-y

13.

Hanson

R. K.

Bussière

M. T.

(1998). Predicting relapse: A meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348-362. doi:10.1037//0022-006x.66.2.348

14.

Harris

A. J. R.

Hanson

R. K.

(2004). Sex offender recidivism: A simple question (User Report No. 2004-03). Ottawa, Ontario: Public Safety and Emergency Preparedness Canada.

15.

Helmus

Thornton

Hanson

R. K.

Babchishin

(2012). Improving the predictive accuracy of Static-99 and Static-2002 with older sex offenders: Revised age weights. Sexual Abuse: A Journal of Research and Treatment, 24, 64-101. doi:10.1177/1079063211409951

16.

Hirschi

Gottfredson

(1983). Age and the explanation of crime. American Journal of Sociology, 89, 552-584.

17.

Hosmer

D. W.

Lemeshow

(2000). Applied logistic regression. New York, NY: Wiley.

18.

Mossman

(1994). Assessing predictions of violence: Being accurate about accuracy. Journal of Consulting and Clinical Psychology, 62, 783-792. doi:10.1037/0022-006X.62.4.783

19.

Phenix

Helmus

Hanson

R. K.

(2009). Static-99R: Evaluator’s workbook. Retrieved from http://static99.org/pdfdocs/st-99rworkbookwithsamplesandsummaries.pdf

20.

Porter

Fairweather

Drugge

Hervé

Birt

Boer

D. P.

(2000). Profiles of psychopathy in incarcerated sexual offenders. Criminal Justice and Behavior, 27, 216-233. doi:10.1177/0093854800027002005

21.

Prentky

R. A.

Lee

A. F. S.

Knight

R. A.

Cerce

(1997). Recidivism rates among child molesters and rapists: A methodological analysis. Law and Human Behavior, 21, 635-659. doi:10.1023/A:1024860714738

22.

Pyke

D. A.

Thompson

J. N.

(1986). Statistical analysis of survival and removal rate experiments. Ecology, 67, 240-245. doi:10.2307/1938523

23.

Rettenberger

Boer

D. P.

Eher

(2011). The predictive accuracy of risk factors in the Sexual Violence Risk-20 (SVR-20). Criminal Justice and Behavior, 38, 1009-1027. doi:10.1177/0093854811416908

24.

Rettenberger

Matthes

Boer

D. P.

Eher

(2010). Prospective actuarial risk assessment: A comparison of five risk assessment instruments in different sexual offender subtypes. International Journal of Offender Therapy and Comparative Criminology, 54, 169-186. doi:10.1177/0306624X08328755

25.

Rice

M. E.

Harris

G. T.

(2005). Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law and Human Behavior, 29, 615-620. doi:10.1007/s10979-005-6832-7

26.

Sampson

R. J.

Laub

J. H.

(2003). Life course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology, 41, 301-339. doi:10.1111/j.1745-9125.2003.tb00997.x

27.

Serin

R. C.

Mailloux

D. L.

Malcolm

P. B.

(2001). Psychopathy, deviant sexual arousal, and recidivism among sexual offenders. Journal of Interpersonal Violence, 16, 234-246. doi:10.1177/088626001016003004

28.

Soothill

K. L.

(2010). Sex offender recidivism. Crime and Justice, 39, 145-211.

29.

Sturgeon

V. H.

Taylor

(1980). Report of a 5-year follow-up study of mentally disordered sex offenders released from Atascadero State Hospital in 1973. Criminal Justice Journal, 4, 31-63.

30.

Thorndike

E. L.

(1911). Animal intelligence. New York, NY: Macmillan.

31.

Thornton

(2006). Age and sexual recidivism: A variable connection. Sexual Abuse: A Journal of Research and Treatment, 18, 123-135. doi:10.1177/107906320601800202