Do Sanctions Affect Undetected Sexual Offending?

Abstract

Undetected sexual offending creates challenges for risk assessment since estimated sexual recidivism rates are based on documented charges or convictions. Courts and other stakeholders may be primarily interested in the true risk for sexual reoffense and not simply risk for detected sexual offenses. Attempts to study and quantify the rate of undetected sexual offending have resulted in a wide variety of estimates. In this study, we explore whether sanctions imposed for detected sexual offenses increase the detection rate of subsequent offenses, and thereby suppress undetected sexual offending in an exceptionally high-risk sample who were ultimately committed as Sexually Violent Persons. Results indicate the detection rate of sexual offenses increased following an initial sanction, subsequently decreasing the proportion of undetected to detected offending. This effect only occurred after the first sanction. Overall, the sample had a high detection rate and spent little time in the community before subsequent arrests. These results differ from other reports that high rates of sexual offenses go undetected.

Keywords

risk assessment undetected sexual offending sanctions high risk sexually violent person

Undetected offenses can be understood as illegal acts that could have led to a criminal charge or conviction had the offense been detected. Reasons why offenses may go undetected in sexual offending cases are varied (Drury et al., 2020; Lovell et al., 2020). In a nationally representative U.S. sample of females aged 12 and over, reasons that victims choose not to report sexual offenses to police (between 2005 and 2010) included fear of retaliation, belief that police would not do anything to help, believing offenses were a personal matter or not important enough to report, and/or a desire to avoid getting their abuser into trouble (Lauritsen & Rezey, 2013).

It is recognized that detected criminal offense recidivism rates may under-estimate an individual’s actual offense risk, and this may be especially true for sexual offenses (e.g., DeLisi et al., 2016; Falshaw et al., 2003). It is estimated that 63% of non-sexual assaults are reported to police, whereas only about 31% of sexual assault victims report the offense to police (https://www.rainn.org/statistics/criminal-justice-system). A more precise figure of this under-estimate will be important for stake holders and when providing education to the community at large.

Quantifying the Rate of Undetected Sexual Offenses

Attempts to quantify the rate of undetected sexual offending have been difficult. Beyond the inherent difficulty of measuring unknown events, there are differences in defining what undetected sexual offending is, methods for studying frequency rates, and how to account for individual differences in offending behavior. Previous studies have relied on surveys or interviews of potential victims (Koss et al., 1987) and potential offenders (Abel et al., 1987; Koss et al., 1987; Lauritsen & Rezey, 2013; Lisak & Miller, 2002; Sinozich & Langton, 2014; Stephens et al., 2021) in the community. Other studies have used self-report interviews or questionnaires of those in custody for other sexual offenses (Bourke & Hernandez, 2009; DeLisi et al., 2016; Drury et al., 2020; Groth et al., 1982; Weinrott & Saylor, 1991), and reports of clinicians or community supervising agents (Falshaw et al., 2003). One potential difficulty arising from survey research is that researchers, as opposed to police or courts, determine what reported sexual behaviors meet the criteria of a criminal charge had the perpetrating individual been caught. This can sometimes be difficult. For example, the age of sexual consent varies across jurisdictions. The self-report of an 18-year-old male in a sexual relationship with a 17-year-old girlfriend could count as an undetected sexual offense in one state, but not in another. Additionally, counting undetected offenses versus undetected victims can produce divergent results (e.g., an individual may have one detected victim and four undetected offenses involving that one victim). Consequently, estimated rates of undetected sexual offending can vary considerably based on the methodologies used.

One area where prominent differences emerge is using reports from persons in community samples versus reports provided by incarcerated populations. The ratio of detected to true sexual offending reported in victim and community member surveys is typically higher than those reported by detected offenders. Additionally, individuals with prior convictions for sexual offenses tend to have fewer reported victims compared to individuals who have never been arrested, such as college students (Lovell et al., 2020). Statistics provided by the anti-sexual violence organization RAINN (Rape, Abuse & Incest National Network) primarily uses data from the National Crime Victimization Survey, conducted by the U.S. Department of Justice, and data from other federal agencies. They estimate the ratio of arrests for sexual assaults to true sexual assaults that occurred to be 50:1000 (https://www.rainn.org/statistics/criminal-justice-system). This would suggest the true number of sexual offenses is 20 times larger than number of reports leading to an arrest.

Abel et al. (1987) found a similarly high ratio of total reported violent sexual crimes (e.g., rape and child molestation) to arrests based on interviews of individuals (n = 561) in the community who admitted committing sexual offenses. This methodology led to an estimate of one arrest for every 30 offenses. In other words, self-reported sexual offending was 30 times higher than what was documented in criminal records. Similarly, a national sample of college students (n = 6159) in the U.S. surveying both victims and perpetrators found reports of rape that were 10 to 15 times higher than nationally reported arrest rates at the time (Koss et al., 1987). A smaller study (n = 1882) interviewing male college students who self-reported engaging in rape or sexual assault found relatively high rates of undetected sexual offending, with 120 men responsible for 483 rapes, or 4 rapes each on average (Lisak & Miller, 2002). Yet, more than two-thirds of these men were considered repeat rapists and accounted for 439 of the reported rapes, averaging 5.8 rapes each. Taken together, these studies suggest true sexual offending may range from 15 to 30 times higher than what is known by official reports.

Only a few studies have examined undetected sexual offending among individuals with prior sexual offense charges/convictions. One study utilized a confidential questionnaire with 137 men in secure custody who had been convicted of rape or child molestation (Groth et al., 1982). They found undetected rapes were 1.85 times higher than convictions for rape, while undetected child molestation was 2.35 times higher than convictions for child molestation. In a study utilizing a computer-administered self-report questionnaire with men being evaluated for civil commitment as a sexually violent person (SVP) in Washington, those with a history of rape charges (n = 37) self-reported 6.5 times more victims than officially detected, whereas those with child offense charges (n = 67) self-reported 7.0 times more child victims than officially known (Weinrott & Saylor, 1991). A study in the U.K. examined undetected sexual offending by individuals (n = 173) released from prison following a prior sexual offense (Falshaw et al., 2003). Undetected sexual offenses were obtained from community supervision agents and the outpatient clinic and compared to official records of conviction in national databases. They found actual reoffending was 1.25 times higher than convictions for reoffending.

Several studies have utilized polygraphs to validate completed self-report questionnaires of undetected offending (Bourke & Hernandez, 2009; DeLisi et al., 2016; Drury et al., 2020) and found undetected rates within the range of other research findings. For example, DeLisi and colleagues’ (2016) study revealed the sample of 119 federal sex offenders reported an average of 3.7 contact sexual offense victims per offender (Med = 1.0). While the methodology of polygraph validation appears to increase the credibility of the self-report, these studies included offenders with only known prior histories of non-contact sexual offending (e.g., pornography cases) in their samples, making it difficult to generalize to individuals with detected contact sexual offenses who would be eligible for scoring on actuarial static and dynamic risk tools.

Sanctions, Offense Rates, and Time at Risk in the Community

Another important distinction when considering undetected offending rates, may be whether an individual has previously been sanctioned for sexual offenses. The current study tests whether prior sanctions for sexual offenses increase detection rates of subsequent offenses, and thereby suppresses the subsequent number of undetected sexual assault victims. Formal sanctions by the justice system are intended to deter further criminal behavior (Mears & Cochran, 2018), but research on the effectiveness of sanctions to achieve this goal has been mixed (Villettaz et al., 2015). There are few studies examining the effectiveness of sanctions in reducing future sexual offenses. Actuarial instruments like the Static-99R (Hanson & Thornton, 2000; Helmus et al., 2012) and Static-2002R (Hanson & Thornton, 2003; Helmus et al., 2012) include items measuring prior sentencing occasions for criminal charges and prior sexual offense charges/convictions. Both items have been found to predict sexual reoffense risk (Helmus & Thornton, 2015). Their predictive validity may simply reflect a persistence of sexual offending and chronic antisociality irrespective of any reduced likelihood of future sexual offending due to prior sanctions (Hanson & Thornton, 2003). Their validity could also reflect a reduced likelihood of any subsequent reoffending going undetected.

There have been a few studies that examined the effect of sentence length on future sexual reoffending. Some find longer sentences to be associated with decreased sexual recidivism (Hsieh et al., 2018), others find it associated with increased sexual offense reconvictions (Budd & Desmond, 2014), and others find no association with sexual offense recidivism (Moss et al., 2022). When examining sentences leading to incarceration versus placement in the community, sentence type has not been found to be related to sexual offense recidivism (Moss et al., 2022; Nunes et al., 2007). Having undetected sexual offenses and prior sanctions for sexual offenses also do not appear to be associated with increased deviant sexual interests. Stephens et al. (2021) found no differences in phallometric results amongst those who had detected child victims (but denied undetected child victims), reported both detected and undetected child victims, and only had undetected child victims per self-report (i.e., no detected/sanctioned sexual offenses). However, those who were willing to report undetected child victims/offending had higher scores on the Screening Scale for Pedophilic Interests, Version 2 (SSPI-2; Seto et al., 2017). Thus, while their willingness to disclose may seem like they have more deviant interests on the SSPI-2, sexual interests to children were similar across groups regardless of the willingness to report additional child victims or having prior sanctions for child sexual offenses.

Most individuals arrested for a sexual offense have no prior history of detected sexual offending (Helmus, 2021). Those with prior detected sexual offenses may subsequently have fewer undetected victims due to increased community supervision, lifetime supervision laws, and increased scrutiny by police (Abbott, 2020; Helmus, 2021). Lussier and Davies (2011) examined crime offending trajectories in a sample of individuals with sexual offense convictions (n = 246) and found those with low rates of detected offending remained in the community longer than those with high rates. It may be that the high-rate group simply had more criminal offenses, possibly increasing their chances of detection. However, the study did not consider undetected offending; individuals who commit offenses more frequently may simply be more likely to be caught sooner. Another study identified 408 individuals suspected of sexual assault via previously unprocessed sexual assault kits (Lovell et al., 2020). Those with the lowest rates of prior arrests for both sexual and general offending comprised 43.6% of the potentially undetected offenders, whereas those with the highest number of previously detected sexual offenses comprised notably fewer of the potentially undetected offenders (18.6%) (Lovell et al., 2020). However, this study was unable to examine whether there were differences in time at risk in the community for these two groups that may account for differences in the number of prior arrests.

Taken together, it remains unclear whether imposed sanctions reduce risk for further sexual offending. The question cannot be answered by only considering detected sexual offending, since those with a prior offense history are likely to be monitored more closely in the community. Consequently, subsequent offenses may be more likely to be identified (detected) sooner thereby reducing time at risk in the community. The current study seeks to examine the rate of detected and undetected victims of sexual offenses during community release periods between successive sanctions for sexual offenses for individuals ultimately committed under an SVP law.

Study Aims and Hypothesis

We believe this is the first study to examine the proportion of detected and undetected sexual offense victims following formal criminal sanctions for a detected sexual offense. Determining whether receiving a sanction changes the rate of detected and undetected sexual offending is relevant for risk assessment practices. Specifically, knowing whether being sanctioned changes offending behavior (detected or not) can inform evaluators how to account for undetected sexual offending when completing risk assessments on individuals being considered for SVP commitment. In the current study, we wanted to examine whether the proportion of detected and undetected victims is stable during release periods following previous sanctions. We hypothesized the following:

1. The proportion of detected victims prior to the first sanction for a sexual offense will be lower than the proportion of victims detected after the first sanction.

2. The proportion of victims detected following each successive sanction will increase across each successive occasion.

3. Time at risk in the community will decrease following each successive sanction.

4. When time at risk in the community is accounted for, the average number of total victims per year will be constant across successive release periods.

Method

Participants

Records of 200 adult males civilly committed under Wisconsin’s SVP law and placed at Sand Ridge Secure Treatment Center (SRSTC) were reviewed and analyzed. About a quarter of the sample (25.5%) had a non-sexual offense conviction as a juvenile prior to their first detected sexual offense. Over a third (36.5%) had a non-sexual offense conviction as an adult prior to their first detected sexual offense. Their ages when last in the community ranged from 14 to 64 years old (M = 31.95, SD = 9.54). The entire sample continuously spent time in a detention, jail, prison, or a forensic psychiatric hospital setting immediately prior to being admitted to SRSTC. In this sample, 4.5% were below the age of 18 when last in the community of which two individuals were 14 and one was 15 years old.

Most of the sample was White (69.5%). Other ethnic/racial groups identified as Black (25.5%), Native American (3.0%), or Hispanic (2.0%). The following mental disorders were coded: Pedophilic Disorder (49.5%), Other Specified Paraphilic Disorder with hebephilic features (10.5%), Other Specified Paraphilic Disorder with coercive features (13.5%), Sexual Sadism Disorder (14.0%), Antisocial Personality Disorder (55.5%), any Substance Use Disorder (49.5%), any major mental illness including Bipolar Disorder Type I and psychotic spectrum disorders (8.5%), and Intellectual Disability (7.5%). An “Other” category captured diagnoses not otherwise coded, such as depressive or anxiety disorders (42.5%). As demonstrated by their Static-99R scores (M = 6.52, SD = 1.92) at the time last taken into custody, this was a well above average risk sample. Regarding offense categories, 56.0% had detected offenses exclusively against children, 9.5% had detected offenses exclusively against adults, and 34.5% had detected offenses against both children and adults.

Measure

Sexual History Disclosure Questionnaires

The Sexual History Disclosure Questionnaire is an unpublished form developed by Anna Salter and Eric Holden for the Wisconsin Department of Corrections (https://manualzz.com/doc/10546895/sex-offender-disclosure-questionnaire-form-doc-1867). It was adapted for use at SRSTC, with the authors’ permission, when the facility opened in 2001 and closely corresponded with the original form. Respondents document their detected and undetected sexual contact with children including their age at the time of the contact, the ages and gender of the children, their relationship to the children (i.e., family, acquaintance, stranger, or children of live-in relationship), the type of sexual acts, and the type of force used if any. They also document their history of detected and undetected adult victims of sexual abuse including the victim’s age, gender, relationship to victim, type of sexual act, and type of force used. Over time, the questionnaire was divided into sections that were administered separately: the Sexual Disclosure Questionnaire #1: Sex Offenses against Adults, and the Sexual Disclosure Questionnaire #2: Sexual Contact with Children. The essential elements of the original questionnaire remained.

The Sexual History Disclosure Questionnaires were used routinely at SRSTC as a clinical tool to help determine an individual’s true sexual offense history and identify treatment needs that may not have been known based only on the official criminal history (e.g., sexual compulsivity; cross-over offending). Individuals presented their completed questionnaires within treatment groups, revised them according to feedback, and worked towards feeling confident about their full disclosure. Once confident, the individuals participated in polygraph testing to validate their self-reported offense histories. If the polygraph examiner identified them as deceptive, the individual was invited to talk about the result in treatment, revise the form if necessary, and complete polygraph re-evaluation. Polygraph testing was frequently completed as part of treatment at SRSTC until it grew out of favor sometime around 2016.

Procedures

This study used archival record review of case files available at SRSTC. No contact with participants was made. Approval of the study was obtained through the Institutional Review Board at the institution.

All men who had been admitted to SRSTC at any point between 2001 and 2017 were potentially eligible for inclusion in the study. Inclusion criteria required a legible and completed sexual history disclosure to be available in the individual’s clinical record. Inclusion criteria further limited the sample to men who had at least one truthful finding on either the original Sexual History Disclosure Questionnaire or one of the split sections (Sexual Disclosure Questionnaire #1: Sex Offenses Against Adults or Sexual Disclosure Questionnaire #2: Sexual Contact with Children) provided that the questionnaire with the truthful finding was consistent with their primary offense profile. A truthful polygraph finding was required for study inclusion to support the legitimacy of the self-reports; however, the limited accuracy and validity of polygraph testing is recognized (Iacono & Ben-Shakhar, 2018).

Multiple records were used to code relevant variables including completed sexual history disclosure questionnaire(s), associated polygraph test reports, the most recent SVP annual report, the original Department of Corrections (DOC) report recommending civil commitment, Pre-Sentence Investigation (PSI) report(s), and the individual’s criminal history available through a public database (Wisconsin Consolidated Court Access Program; CCAP). Missing records or those not providing adequate detail prompted additional searches for records. Other records reviewed most frequently included additional SVP reports and revocation summaries. Sufficient information to code the study variables were identified for 200 individuals.

Coding focused on each period the individual was in the community prior to their initial arrest for a sexual offense and following each such arrest thereafter. This included the date and age at the time of arrest for a sexual offense; the number and demographic characteristics of the victims for which they were arrested (contact sexual offenses only); the sentencing date if convicted; the date and their age when released back into the community following each sexual offense sanction (accounting for any non-sexual offense custodial time); length of release periods before the next sexual offense sanction; how many undetected contact sexual offense victims they had during release periods; and the demographic characteristics of these undetected victims.

Detected victims listed on the sexual history questionnaires were matched with descriptions in the official reports and removed from the list of undetected victims. Detected sexual offenses were defined to be consistent with Category A and B sexual offenses in the Static-99R coding rules. As such, the first detected sexual offense had to be a Category A sexual offense charge or conviction. Although non-contact sexual offenses were recorded, information about victim characteristics was not coded as the focus of this study was on undetected contact sexual offending. Consistent with the Static-99R scoring rules, only detected sexual offenses when the individual committing the offense was older than 11-years-old were counted. An undetected victim was defined as a victim of a sexual offense that would have counted towards the Static-99R scoring had the individual been detected and arrested, charged, or convicted for the offense. The victim/offense may have been “detected” or known by police, social services, or others but was considered undetected if the individual was never formally sanctioned (e.g., detained, arrested, or charged). Determination as being undetected is because an illegal sexual act occurred, but since the individual were not detained, arrested, or charged, the offense could not be counted by the Static-99R.

Occasionally, individuals identified potential undetected victims who were not victims of a sexual offense as defined by legal criteria. For example, responses on a sexual history questionnaire might identify a peer-aged girlfriend or boyfriend (within two years of each other) with no indication of any type of nonconsenting act as a victim. Otherwise, reported undetected victims that would have qualified as a statutory sexual offense in Wisconsin were counted. Undetected victims were also not counted if the reported “victim” was appreciably older than the individual and there was no indication that force or coercion was used. For example, some individuals reported having an undetected “victim” who was in their adulthood when the individual was in their early teens. In these cases, it was judged that the individual was more likely the actual victim. In cases when the individual reported undetected sexual offending against the same victim across multiple release periods, we coded the age and release period when the sexual offending began.

Training and Reliability

Detailed coding instructions were created for the current study (see Supplemental Materials). Coding for the study was completed by the first author. Two additional coders not associated with this study were trained and coded ten cases independently (n = 20) to assess the replicability and reliability of the first author’s coding. One of these coders was a supervisor at SRSTC and the other was a researcher who has never worked at SRSTC. Both coders have excellent knowledge of sexual risk assessments and the Static-99R coding instructions.

Intraclass correlation coefficients (ICC₁) were used to determine the interrater reliability of the number of undetected victims and Static-99R scores. Cicchetti’s (1994) guidelines indicate that ICC values of .75 and above are considered excellent. Comparing the independent coder ratings (n = 20) with the first author’s ratings yield ICC₁ values between .81 and .99 for total undetected victims, total undetected victims prior to first sexual offense arrest, and total undetected victims following first sanction. Additionally, there was excellent interrater reliability found between the independent coders Static-99R scores and the first author’s Static-99R scores for the first release (ICC₁ = .88) and at the time of last sanction (ICC₁ = .87).

Planned Statistical Analysis

Descriptive statistics examined the frequencies of detected victims at each arrest ending a release period and the number of undetected victims during the release period prior to being arrested. Chi-square analyses were conducted to test differences in the relative proportions of detected and undetected sexual offense victims across respective release periods. Proportions of detected and undetected offenses based on total victim counts were calculated and compared using z-tests to determine statistical significance (Bruning & Kintz, 1977; Equation (1) below).

z = \frac{P_{1} - P_{2}}{\sqrt{\frac{p (1 - p)}{N_{1}} + \frac{p (1 - p)}{N_{2}}}}

(1)

where P_i = Proportion i; N_i = Sample size i; and

p = \frac{N_{1} P_{1} + N_{2} P_{2}}{N_{1} + N_{2}}

Z-scores exceeding ±1.96 are produced when two proportions are significantly different from one another at α =.05. Effect sizes for testing differences between proportions ( P ) are indexed by h, where h = ϕ₁ - ϕ₂ and ϕ = (2 arcsin $\sqrt{P}$ ). Cohen (1988) identifies small, medium, and large effect sizes for h as 0.2, 0.5, and 0.8, and respective sample sizes of 400, 64, and 25 are needed to identify these effect sizes at α =.05 with statistical power of .80. Paired sample t-tests were used to evaluate mean time differences between release periods for days in the community.

Univariate outliers were identified by visual inspection of histograms and boxplots, and by computing standardized scores to identify extreme values (e.g., z-scores > ±3.5). Identified outliers were assigned a score value one unit larger than the next most extreme score within that variable’s distribution (Tabachnick & Fidell, 2019). In this sample, two outliers were identified for undetected frequency count victim variables and seven outliers were identified for the release time variables. Analyses were conducted in SPSS Version 26 and Excel 2016.

Results

For the complete sample (N = 200), there was a total of 1045 undetected victims reported prior to the first arrest for a sexual offense (M = 5.23, SD = 9.43, Med = 2.00). The distribution of reported undetected victims ranged from 0 to 69 but was positively skewed such that most individuals reported having no undetected victims (36.0%) or between one and five undetected victims (38.5%). Individuals reporting between six and ten undetected victims became less common (11.0%) as did those reporting between 11 and 15 undetected victims (5.0%). Reports of 16 or more undetected victims were infrequent (9.5%).

We identified 841 officially detected victims that led to one or more sanctions within the total sample (M = 4.20, SD = 2.11, Med = 4.00, range 1 – 14). Out of the 200 cases, seven individuals were civilly committed as an SVP following their first sexual offense sanction and were never released back to the community. This was predominately due to extraordinary details about the offense (e.g., sexually motivated homicide) or continued contact sexual offending while in secure confinement. At the time of the first arrest, the 200 cases had a total of 284 detected victims (M = 1.42, SD = 1.33, Med = 1.00). Of the 193 cases who had one or more release periods, there were a combined total of 1780 victims (M = 9.22, SD = 16.86, Med = 5.00). Given the large variance in the number of total victims following first release as well as the skewed distribution of the sample, the median of 5.00 is likely a better representation than the average. Of these, a total of 557 victims were detected following their first sanction for a sexual offense (M = 2.89, SD = 1.97, Med = 2.00) whereas 1223 undetected victims were reported following their first sanction for a sexual offense (M = 6.34, SD = 16.59, Med = 2.00). Again, this distribution was positively skewed with only four cases reporting an extraordinary number of victims, ranging between 55 and 143. Most cases reported no undetected victims (34.0%) or between one and five undetected victims (40.7%). Reports of six to ten undetected victims were uncommon (10.3%) as were reports of 11 to 15 undetected victims (5.7%). Reports of 16 or more undetected victims were similarly uncommon (9.3%). When outliers were adjusted, the variance reduced somewhat, although the distribution remained highly skewed (M = 5.37, SD = 11.01, Med = 2.00).

Except for one individual who had 11 release periods, the rest of the sample had seven or fewer release periods. The victim characteristics for each of the release/sanction periods is provided in Table 1. There were 189 individuals released to the community following their first arrest, charge, or conviction for a sexual offense (average age of 21.08, SD = 6.63) who contributed data to the analyses reported below. In addition to the seven cases who were never released, four cases were not released until after their second sanction for a sexual offense. This was due to having committed a new sexual offense while on escape status (not considered release per Static-99R coding) and/or receiving new criminal charges for a sexual offense committed while in a custodial setting. This occurred during other release periods as well. The number of individuals who experienced subsequent release periods diminished as the number of release periods experienced increased, primarily due to SVP commitment. Of the 189 individuals released to the community following their first arrest, 145 were sanctioned and released a second time (average age = 25.43, SD = 8.19), 79 of whom were sanctioned and released a third time (average age = 27.06, SD = 8.36). Of these 79 individuals, 38 were sanctioned and released a fourth time (average age = 28.97, SD = 9.37) and 19 were sanctioned and released a fifth time (average age = 31.84, SD = 9.71). The number of individuals experiencing additional sanctions and release periods dropped below ten. For individuals experiencing one or more release periods in the community (n = 193), an average of 2095.92 (SD = 1897.83, Med = 1508.00) days were spent in the community between their first sexual offense arrest and their civil commitment as an SVP – an average of 5.74 years (Med = 4.13 years).

Table 1.

Frequency of total victim characteristics for each release period for cases with at least one release period (n = 193).

Release Period	N	Male, %	Female, %	Child, %	Teen, %	Adult, %	Stranger, %	Acquaintance, %	Family, %
1	744	21.4	77.6	53.0	23.7	22.4	28.9	54.2	12.9
2	570	28.8	71.2	49.3	28.8	21.9	37.2	53.2	6.3
3	302	39.7	59.9	20.5	41.7	37.4	17.2	65.6	5.3
4	99	35.4	63.6	19.2	33.3	47.5	17.2	63.6	8.1
5	42	28.6	71.4	28.6	28.6	38.1	19.0	35.7	9.5
6	13	38.5	61.5	23.0	38.5	38.5	38.5	46.2	0
7	7	42.9	57.1	57.1	0	42.9	0	42.9	28.6
8	1	0	100	100	0	0	100	0	0
9	1	0	100	100	0	0	100	0	0
10*	1	0	100	0	0	100	100	0	0

Note. = One case had an 11^th release but was revoked for technical violation with no further victims. When proportions in categories total less than 100%, this indicates the remaining proportion of victims were coded as unknown for that category.

Hypothesis 1

The proportion of detected victims prior to the first sanction for a sexual offense will be lower than the proportion of victims detected after the first sanction.

A chi-square test of independence examining the relation between victim type (undetected versus detected) prior to and after the first sanction was significant, χ² (1) = 45.70, p < .001, Cramer’s V = .130. Across the 189 evaluable individuals, the proportion of victims detected at the first arrest period was 22.57% (253 of 1121 total victims). After the first arrest, 548 victims were detected across all subsequent arrests/sanctions, which is 34.62% of all victims identified after first arrest (1583). The proportional difference of detected victims prior to and after first arrest was statistically significant but reflects a small effect size (z = −6.76, p < .001, h = .266).

Hypothesis 2

The proportion of victims detected following each successive sanction will increase across each successive occasion.

To determine whether the proportion of victims detected was the same following increasing numbers of sanctions, individuals with one (n = 189), two (n = 142), three (n = 75), and four (n = 35) release periods were identified and analyzed to ensure comparisons across release periods included the same individuals. Comparisons across five or more release periods were not conducted because the sample sizes of these subsamples (n ≤ 18) would be too small to detect even large effect sizes. For each subsample of participants with one, two, three, or four release periods, chi-square tests examined whether the frequencies of detected to undetected victims were independent of the release periods following each sanction. Results are presented in Table 2.

For the 189 participants with a release period following first arrest the chi-square test of independence was significant, χ²(1) = 37.95; p <.001. The proportion of victims detected during the first release period (36.2%) was significantly higher than the proportion detected prior to the first arrest (22.6%), z = −6.16, p < .001, h = .287. For individuals with at least two release periods (n = 142), the chi-square test was also significant, χ²(2) = 29.86, p <.001. Pairwise comparisons found the proportion of victims detected during first and second release periods (36.0% and 33.0%, respectively) were significantly higher than the proportion of detected victims prior to the first arrest (23.0%), z = −5.12, p < .001, h = .287 and z = −3.99, p < .001, h = .224, respectively. The proportion of victims detected during the first and second release periods did not significantly differ from one another. For individuals with three or more release periods (n = 75), the chi-square test was not significant. Pairwise comparisons between the proportion of victims detected prior to the first arrest (23.7%) was significantly less than the proportion of victims detected during the first release period (31.0%; z = −2.00, p = .023, h = .157), but no other pairwise comparisons were significant. Finally, for the 35 participants with at least four release periods, the chi-square test was significant (χ²(4) = 19.34; p <.001) and pairwise comparisons revealed a similar pattern to the other subsets. The proportion detected prior to the first arrest (23.0%) was significantly less than the proportion detected at the first (44.4%; z = −3.31, p < .001, h = .451), second (38.9%, z = −2.72, p = .003, h = .349), third (45.8%; z = −3.65, p < .001, h = .491), and fourth (43.0%: z = −3.27, p = .001, h = .430) release period. No other pairwise comparisons were significant.

Table 2.

Chi-square tests of distributional independence and z-tests of proportion equivalence of detected victims across release periods.

Releases	N	Type of Victim	Before 1^st Arrest	1^st Release Period	2^nd Release Period	3^rd Release Period	4^th Release Period	Chi square test
1	189	Not detected	868	409
1	189	Detected	253 (22.6%)^a	232 (36.2%)^b				X²(1) = 37.95, p < .001, Cramer’s V = .147
2	142	Not detected	589	304	311
2	142	Detected	172 (23.0%)^a	171 (36.0%)^b	153 (33.0%)^b			X²(2) = 29.86, p < .001, Cramer’s V = .133
3	75	Not detected	315	168	204	208
3	75	Detected	98 (23.7%)^a	75 (31.0%)^b	82 (28.7%)^ab	65 (23.8%)^ab		X²(3) = 5.72, p = .13, Cramer’s V = .069
4	35	Not detected	121	40	58	45	49
4	35	Detected	36 (23.0%)^a	32 (44.4%)^b	37 (38.9%)^b	38 (45.8%)^b	37 (43.0%)^b	X²(4) = 19.34, p = .001, Cramer’s V = .198

Note. Only cases who had one, two, three, or four release periods were included for analysis. Four cases were excluded because they were not released until after their second sanction for a sexual offense. Proportions with different superscripts differ significantly (p <.05) in pairwise comparisons. Proportion of detected victims is provided in percentages enclosed in parentheses. Two outliers were detected in the first and second release periods for undetected victim counts and assigned a score that was one unit larger than the next most extreme score within that variable distribution.

Hypothesis 3

Time at risk in the community will decrease following each successive sanction.

To examine whether increasing numbers of sanctions resulted in decreased time in the community following each sanction, the number of days in the community following each release period was compared to the number of days in the community during subsequent release periods. Given the directional nature of our hypothesis that additional sanctions reduced time in the community, single-tailed tests of significance were used to determine whether the reductions in days were statistically significant. The mean and standard deviation of days in the community across the first four release periods are shown in Table 3.

The mean number of days in the community across all 189 individuals with an initial release period was 1002.33 days between their first arrest for a sexual offense and their final SVP commitment after subtracting their time in custody. Across the 142 individuals with both a first and second release period, the mean time in the community during the second release period was 219.5 days less than during the first release period which is statistically significant, t(141) = 1.77, p = .039, d = .21. Across the 75 individuals who had a first, second, and third release period, time in the community during the first release period (978.95 days) was significantly greater than during the second release period (685.15 days), t(74) = 1.88, p = .032, d = .27. Time in the community during the third release period (709.55 days) did not differ significantly from either the first or second release periods, t(74) = 1.58, p = .059, d = .25 and t(74) = −.16, p = .438, d = .03, respectively. Finally, for the 35 individuals who had a first, second, third, and fourth release period, the times in the community between the release periods did not significantly differ from one another (all p’s >.298).

Table 3.

Comparison of time in the community during release periods.

	1st Release Period	2nd Release Period	3rd Release Period	4th Release Period
N	M (SD) days in community
189	1002.33 (1185.21)
142	988.65 (1152.28)^a	769.15 (982.07)^b
75	978.95 (1233.10)^a	685.15 (889.83)^b	709.55 (882.89)^ab
35	669.09 (913.39)^a	571.57 (960.80)^a	692.71 (916.53)^a	636.83 (817.32)^a

Note. Means with different superscripts differ significantly (p < .05) in pairwise comparisons. Single-tailed tests were used to evaluate whether the time spent in the community during subsequent release periods were significantly less than time in community during preceding release periods as hypothesized. Outliers detected at the first and second release period were assigned a score one unit larger than the next most extreme score within that variable distribution.

Hypothesis 4

When time at risk in the community is accounted for, the average number of total victims per year will be constant.

The frequencies of detected and undetected victims during the first four release periods, shown in Table 2, were used to compute the expected number of victims per person per release period. The numbers of victims in each release period (by type and total) were divided by the number of individuals in each subsample. These expected numbers of detected, undetected, and total victims, per person per period, were then divided by the mean duration of the release periods (see Table 3) and multiplied by 365.25. This calculation provides an annualized rate of the expected number of detected, undetected, and total victims per person for each of the first four release periods. As seen in Table 4, more victims per year are expected at release periods 2, 3 and 4 than during the first release period. Overall, the average expected numbers of victims per person per release period ranged from 1.12 to 2.03 per year.

Table 4.

Expected number of undetected, detected, and total contact victims per person per year across release periods.

		1st Release Period	2nd Release Period	3rd Release Period	4th Release Period
N		Expected marginal mean victim Count/Person/Year
189	UndetectedDetectedTotal	0.790.451.24
142	UndetectedDetectedTotal	0.790.451.24	1.040.511.55
75	UndetectedDetectedTotal	0.840.371.21	1.450.582.03	1.430.451.87
35	UndetectedDetectedTotal	0.620.501.12	1.060.681.74	0.680.571.25	0.800.611.41

Note. Expected means based on average number of undetected and detected victims per offender per release period divided by the average number of days in the community during each release period multiplied by 365.25 to derive annualized rates.

Discussion

The current paper sought to understand the rate of continued undetected sexual offending after previous sanctions in a high-risk sample who were eventually civilly committed under the state’s SVP law. Undetected offending creates challenges in risk assessments and policymaking since estimated recidivism rates are based on charges or convictions for offenses documented in official criminal histories. For sexual offense risk assessments, actuarial static and dynamic risk measures are most frequently used (Kelley et al., 2020), and derive their estimated recidivism rates primarily from sexual offenses appearing on official Records of Arrests and Prosecutions (RAP). As such, it is widely recognized that these estimated sexual offense recidivism rates are at best an under-estimation of the individual’s true sexual offense risk (DeLisi et al., 2016; Falshaw et al., 2003). Indeed, developers of actuarial risk scales have recognized the issue of undetected offending when estimating risk projections as well (Hanson et al., 2003; Helmus, 2021; Thornton et al., 2021). To determine how much the actuarial risk scales underestimate risk, a credible estimate of the undetected sexual offenses is needed that is relevant to the population for whom the risk assessments are being completed.

Previous estimates for undetected sexual offending in community and custodial samples have varied widely, creating challenges for policy makers, evaluators, and courts who may be aware of the observed rates of sexual offending from published studies, but recognize that these rates may not reflect offenses that have gone unreported and undetected. The community may also be unaware of true sexual offending rates and rely on general education through advocacy groups and online sources. For example, a faith-based group called Talk About Abuse to Liberate Kids (TAALK) indicates that 86% of sexual assaults go unreported to authorities (https://www.taalk.org/taalk-with-god/offender-discipleship/undetected-sex-offenders). One of the statistics found on RAINN’s website is that out of every 1000 sexual assaults, 975 offending individuals (97.5%) “will walk free” (https://www.rainn.org/statistics/criminal-justice-system). These statistics are likely to produce alarm for community members and law makers. Given such statistics are easily found online, it is not surprising that fact finders may be skeptical of actuarial risk results derived from empirically observed sexual recidivism rates. As such, policy makers may call for stricter sentencing guidelines and residency restrictions, and community notification meetings for anticipated prison releases may be met with protest and outrage.

Even when advocacy groups such as RAAIN rely on solid information, like the National Crime Victimization Survey completed by the U.S. Department of Justice, the reported statistics remain concerning. If only 50 out of 1000 sexual assaults result in an arrest, this suggests a detection rate of only 5%. However, this detection rate appears to fluctuate based on victim characteristics (e.g., female college students report sexual offenses to the police less frequently than their non-college involved female peers). Further, as described earlier, a review of empirical sources indicates the estimated rate of undetected sexual offenses varies dramatically depending on differences in samples and methodologies. Statistics put forth by the U.S. Department of Justice may, in fact, be quite accurate for routine community samples but are likely not appropriate to use with highly selected high-risk samples. The current study focused on determining whether individuals in an SVP sample would be increasingly more likely to be detected for sexual offending following prior detections/sanctions, thereby resulting in proportionally less undetected sexual offending and less time at risk in the community. SVP samples can be quite different from routine community samples, and we wanted to better understand the rate of undetected offending in this sample. The results are likely not representative of routine community samples and first-time offenders.

Overall, we found that the individuals in this study continued to commit detected and undetected sexual offenses despite prior sanctions for such behavior. However, the proportion of victims detected significantly increased after an initial sanction, resulting in a lower proportion of undetected victims and fewer total victims overall. Of note, this effect only appears present following the first sanction. Second and third sanctions for sexual offenses did not appear to further suppress sexual offending or significantly increase the detection rate. The time free in the community was significantly less during 2^nd release periods compared to the first release periods. However, time free in the community did not continue to decrease significantly across subsequent release periods in this sample, likely because the length of time in the community became less than two years. Thus, the first sanction may have a suppressing effect for some individuals, but those who persist may reflect more chronic antisociality (Hanson & Thornton, 2003). This high-risk sample was detected rapidly and spent limited time in the community before their next arrest.

As compared to the detection rates offered by RAAIN and the U.S. Department of Justice, our SVP sample had a high rate of detection even before their first conviction for a sexual offense. Across the complete sample (N = 200), there were 284 detected and 1045 undetected victims at the time of first arrest for a sexual offense, indicating that true sexual offending was 4.7 times higher than detected offenses. This also indicates a detection rate of 21.4% for what would be considered “first time offenders” in the prison system. This detection rate is materially higher than rates reported by the U.S. Department of Justice and other studies focusing on community samples (Abel et al., 1987; Koss et al., 1987). What might account for the difference in reported detection rates? The National Crime Victimization Survey asks participants about contact sexual assaults and verbal threats of sexual assault (https://bjs.ojp.gov/content/pub/pdf/cv20sst.pdf). Their survey methodology allows for an extensive reach nationwide of potential victims who are likely able to report experiences of sexual assault by people who may never end up being detected and sanctioned. Indeed, it appears that out of every 1000 sexual assaults, 69% are never reported to police (https://www.rainn.org/statistics/criminal-justice-system). As such, these data provide a better estimation of the detection rate for the general community but may be quite limited when applied to specific samples (e.g., those who are already known to the criminal justice system). The current sample was notably high risk for sexual offending and likely general criminal offending as well. Even prior to their first arrest for a sexual offense, 25.5% of the current sample had been convicted of a non-sexual offense as a juvenile and 36.5% already had a non-sexual offense conviction as an adult. In addition, the higher rate of detection in this sample may also reflect their tendency to commit the type of sexual offenses that are more likely to be reported by victims.

Our data suggest the first formal sanction for a sexual offense makes a material difference on the detection rate, which would not be evident in the National Crime Victimization Survey and other data sources. When calculating the detection rate for individuals who were released following their first sanction, (n = 189), the detection rate increased from 22.6% to 36.2% after the first sanction, which is a significant difference. Detection rates remained similarly high but did not continue to increase with subsequent sanctions. Using this information, as opposed to information that may predominately include first-time, never detected offenders, is particularly relevant for providing education to interested parties in the community, the courts, and parole boards regarding individuals who are being examined as potential candidates under SVP laws. While individuals with prior detected sexual offenses will be identified as greater risk on actuarial instruments, their risk for engaging in future undetected sexual offending does not appear to be as dramatically high as general community crime reports have indicated. This is likely because high risk individuals appear to sexually reoffend rapidly and are caught quickly. Individuals with prior sanctions for sexual offenses spend little free time in the community between arrest periods ranging between about 1.57 years to 2.74 years, although there was a large degree of variability. When accounting for their limited time at risk, they tended to offend against one to two victims per year.

Overall, it is probably most beneficial for those interested in estimating the detection rate for contact sexual offenses in a population previously convicted of sexual offenses to consider only detected and undetected sexual offenses that occur following the first formal sanction for a sexual offense. Of the cases who had one or more releases following their first sanction (n = 193), the total number of victims was 1780. Of these, 557 victims (31.2%) were detected. This indicates that the true sexual offense rate was about 3.2 times higher than the reported rate of criminal offenses and within the range found in similar samples who had been incarcerated for a sexual offense (i.e., previous studies report rates from 1.25 to 7.0 times higher; Falshaw et al., 2003; Groth et al., 1982; Weinrott & Saylor, 1991). However, this ratio used in isolation would be misleading. Although the range of total victims reported by each case was quite large, the distribution was positively skewed with most having reported smaller numbers of victims. Accurate estimations of undetected sexual offending within a high-risk sample would need to account for both the detection rate and the median number of detected and undetected victims.

Implications for policy and practice

Those interested in sexual reoffending risk will want to understand the magnitude of the difference between the probabilistic estimates provided by actuarial static and dynamic instruments and true sexual reoffense rates accounting for undetected sexual offending. Thornton et al. (2021) note that evaluators need to account for possible undetected offending separately from estimates of long-term risk. Some researchers proposed multipliers to account for undetected sexual offending as well as long-term risk (Doren, 2010). Others have criticized the use of constant multipliers because it does not account for differences in risk levels (Wollert & Cramer, 2012). Indeed, the current study suggests that different risk groups (e.g., community vs. those with prior sanctions) appear to have different detection rates. In this study, those who had one prior detected sexual offense and one formal criminal sanction spent less time in the community due to early sexual reoffense detection and, as a result, subsequently offended against less total victims following their first sanction. Further, using a simple multiplier of 3.2 (the ratio of detected to total true victims) would be misleading because it does not account for the positively skewed distribution of the sample. In other words, most of the individuals in the sample had zero or very few undetected sexual offense victims.

As noted by Thornton et al. (2021), a probabilistic model that accounts for undetected offending, based on victim reporting rates and criminal detection rates, was posed by Hanson et al. (2003). Based on a review of research studies and official crime rates, this model identified a range of five to 15 victims per each individual who engages in rape and child molestation, and a variable detection rate for sexual offenses, believed to range from 10% to 20% per victim. The model assumes that offending individuals have a relatively constant victimization rate in terms of victims per year. Individuals from high-risk samples are believed to have one new victim per year while individuals from low-risk samples are believed to have one new victim every two years. Individuals who have higher victimization rates are more likely to get caught sooner based on a stable detection rate of offenses. Thus, individuals with one new victim per year will be more likely to be identified than individuals with one victim every other year. Hanson et al. (2003) provided the following equation to account for real (i.e., true) sexual reoffense risk:

R R R = O R R \div D R I

RRR: Real (i.e., true) Recidivism Rate

ORR: Observed Recidivism Rate (e.g., probabilistic estimates from actuarial scales)

DRI: Detection Rate Per Individual (i.e., proportion caught after calculating the number of individuals who are caught after each new victim using the detection rate per victim until the average number of victims per offending individual is met)

The detection rate per individual (DRI) can be obtained by using the parameters based on Hanson et al.‘s (2003) review of victim reporting rates and criminal detection rates (i.e., 10% to 20% per victim). Imagine a sample of 100 offending individuals who have a constant rate of one victim each per year. If we use the midpoints from Hanson et al.‘s (2003) model by applying a detection rate per victim of 15% and an average number of victims per offending individual of ten, we can calculate the detection rate per individual (DRI) for ten victims as follows:

Victim 1 Out of 100 offending individuals, 15 are caught (100-15 =) 85 are undetected

Victim 2 Out of 85 offending individuals, 12.75 are caught 72.25 are undetected

Victim 3 Out of 72.25 offending individuals, 10.84 are caught 61.41 are undetected

And so forth until the tenth victim:

Victim 10 Out of 23.16 offending individuals, 3.47 are caught 19.69 are undetected.

Using these parameters, we would conclude that out of 100 offending individuals persisting to offend once per year, about 20 individuals would remain undetected after having offended against 10 victims each, and 80 would have been identified/sanctioned after having offended against 1 to 10 victims over a 10-year span. This rate of 80% is the detection rate per individual based on the parameters we set for this probabilistic model. To apply this in practice using the above Hanson et al. (2003) equation and a DRI of 80%:

R R R = O R R \div . 80

For simplicities sake, assume an evaluation case in which the individual was estimated to have a risk probability of 25% based on the results of an actuarial risk scale. The estimated true sexual reoffense rate would be about 31.25% as seen here:

31.25 % = 25 % \div . 80

The problem with the probabilistic model put forth by Hanson et al. (2003) as well as another model by Scurich and John (2019), is that these models are based on statistical assumptions and small changes can make very large differences in the final estimates. Further, the statistical assumptions were primarily obtained from data provided by sources like the U.S. Department of Justice and are likely to be a better reflection of a routine community rate as opposed to rates by high-risk individuals with prior sex offense convictions and sanctions. This does not mean the structure of the models are problematic, but it reflects that different parameters are likely best when applying these models to different types of sex offending individuals.

Based on the current study, if one was to use a probabilistic model like Hanson et al. (2003) to determine the true sexual offense risk in a high-risk sample (e.g., SVP evaluations), it would be prudent to use Table 4 to estimate the number of victims per year per offending individual. Using only the first release period provides the largest number of cases (n = 189), and the cell sizes drop dramatically when considering cases with multiple release periods, although there is some slight variation with number of victims per year when considering multiple release periods. Regardless of whether one only uses the total expected victims per year for the first release period (1.24) or the mean of marginal means (1.38) for the four release periods combined, it appears that this sample had about one new victim per offending individual per year. Given the large variance in the total number of victims following the first release as well as the skewed distribution of the sample, we identify using the median as a better representation of the total number of victims per offending individual (Med = 5.00). Table 2 provides us with the detection rate per victim (36% for the first release period (n = 189)). We can apply these parameters to the Hanson et al. (2003) model to obtain the detection rate per individual:

Victim 1 Out of 100 offending individuals, 36 are caught 64 are undetected

Victim 2 Out of 64 offending individuals, 23.04 are caught 40.96 are undetected

Victim 3 Out of 40.96 offending individuals, 14.75 are caught 26.21 are undetected

Victim 4 Out of 26.21 offending individuals, 9.44 are caught 16.78 are undetected

Victim 5 Out of 16.78 offending individuals, 6.04 are caught 10.74 are undetected

Thus, out of 100 offending individuals, about 11 go undetected while 89 are detected. This gives us a detection rate per individual (DRI) of 89%. We can now fit this into our equation using our previous example of a case in which the actuarial instrument suggested a risk probability of 25%:

T r u e S e x u a l R e o f f e n s e R i s k = 25 % \div . 89

This provides us with a true sexual reoffense risk estimate of 28.1%. Note that this probabilistic model continues to have the same potential weakness as noted earlier. That is, small changes can change the outcome. For example, if instead of using the detection rate per victim for only those with more than one release period (n = 189), we use the detection rate for all the cases with at least one release period (n = 193). Those with at least one release period had a detection rate per victim of 31% (557 detected victims divided by 1780 total victims). This changes the detection rate per individual (DRI) to 84% and our true sexual reoffense risk for the previous example would change to 29.8%. The important part of using such models is to apply credible parameters that are empirically associated with the sample being assessed. While the parameters offered by the current study would be appropriate for similarly high-risk cases with prior sanctions for sexual offending and being examined as potential candidates under dangerousness commitments, it would be inappropriate to apply them to community samples and routine incarcerated samples. Likewise, it would be inappropriate to apply data on community, first-time offending individuals to such cases as SVP samples. While the probabilistic models may be sound, it is the data applied to the model that needs to be credible and appropriate to the sample of interest.

Anyone wanting to apply data from the current study to probabilistic models also needs to be aware that there remains a large variance, albeit positively skewed, distribution in the number of victims per offending individual. There are some individuals who were able to bypass detection and accumulate vast numbers of victims, although these were fortunately less common. With further analysis of these data, including potential co-variates, we can begin to develop more complex working models to extrapolate risk of undetected sexual offending from observed rates.

Limitations and Future Directions

The current dataset is based on a highly select sample of those who were ultimately civilly committed as SVPs in Wisconsin. By nature of the sample, all these individuals continued to sexually recidivate until they were committed. We did not have access to a comparison group of individuals with a history of sexual offending who desisted from further sexual offending. We also cannot know the offense trajectory of individuals who are never caught or convicted for sexual offending. Thus, any recommendations based on these data are limited to similar high-risk populations. We believe the reported results are generalizable to other SVP samples. These results may not be as generalizable to those in correctional “routine/complete” samples or those in the community. Future research is needed to assess the comparability of these results with such groups.

The average age of our sample when last in the community was 31.95 (SD = 9.54). This is fairly consistent with the median age of the U.S. prison population (https://bjs.ojp.gov/content/pub/pdf/aspp9313.pdf) but is much younger than the typical SVP population. Given that the average age of patients discharged from SVP commitments is around 56 years old (e.g., Azizian et al., 2021), it is difficult to determine whether the offense patterns they previously exhibited in their younger years should be assumed to continue after discharge from their SVP commitments. Indeed, results from Bouchard and Lussier’s (2015) estimation methods suggested that risk of being reconvicted (or caught) are not equally distributed among age groups, with a risk of conviction twice as high for younger offenders. This may be partially due to older offenders being better able to avoid or delay detection. Using the current age weights in the Static-99R to account for reduced reoffense risk associated with older age will help with this confound of age. However, exploring the rate of undetected sexual offending in both older and previously discharged SVP patients will be an important avenue for future research.

There are potential covariates we have yet to explore that may allow for more precise estimates of undetected offending. These include psychiatric diagnosis, ethnicity, child vs. adult offenses, stranger vs. adult offenses, familial versus extra-familial offenses, cross-over offending, non-sexual criminal offending, Static-99R scores, age at first offense, and age at release from a sanction for a sexual offense. These variables will need to be considered in other analytic and distributional models. Thus, we recognize the potential importance of these variables and methods, and plan to focus on such in future research.

The results in this study apply well when estimating the number of undetected victims of contact sexual offenses in a high-risk population. We were unable to code the number of undetected offenses that occurred to each victim and did not include victims of non-contact offenses (e.g., exhibitionism). Thus, these data may underestimate the total rate of undetected sexual offending. On the other hand, high frequency sexual offense behaviors like exhibitionism increase the likelihood of being caught, thereby raising the likelihood of detecting continued sexual offending. Despite the limitations of this research, this study is an important first step to quantify the rate of undetected sexual offending by examining individual reporting of sexual offenses across their lifespan, sanctions, and time at risk in a high-risk sample.

Supplemental Material

Supplemental Material - Do Sanctions Affect Undetected Sexual Offending?

Supplemental Material for Do Sanctions Affect Undetected Sexual Offending? by Sharon M. Kelley, Rachel E. Kahn, James C. Mundt, Robert M. Barahal in Sexual Abuse

Footnotes

Author’s Note

The results from this study were presented at the Western Society of Criminology 48^th Annual Conference on February 4, 2022, and the annual MASOC/MATSA virtual conference on April 7, 2022. The authors take responsibility for the integrity of the data, the accuracy of the data analyses, and have made every effort to avoid inflating statistically significant results. We would like to thank Maaike Helmus and Lakshmi Subramanian for their assistance with inter-rater reliability coding and David Thornton for his suggestions and critique of the manuscript.

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.

Funding

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

Disclaimer

The views expressed are those of the authors and not necessarily those of the Wisconsin Department of Health Services – Division of Care and Treatment Services

ORCID iDs

Sharon M. Kelley

James C. Mundt

Supplemental Material

Supplemental material for this article is available online.

References

Abbott

B. R.

(2020). Illuminating the dark figure of sexual recidivism. Behavioral Sciences & the Law, 38(1), 543–558. https://doi.org/10.1002/bsl.2494

Abel

G. G.

Becker

J. V.

Mittelman

Cunningham-Rathner

Rouleau

Murphy

W. D.

(1987). Self-reported sex crimes of nonincarcerated paraphiliacs. Journal of Interpersonal Violence, 2(1), 3–25. https://doi.org/10.1177/088626087002001001

Azizian

Olver

M. E.

Rokop

D’Orazio

D. M.

(2021). A preliminary analysis of sexual recidivism and predictive validity of the Static-99R in men discharged from state hospitals pursuant to California’s Sexually Violent Predator Act. Sexual Abuse. Online first publication. https://doi.org/10.1177/1079063221109726

Bouchard

Lussier

(2015). Estimating the size of the sexual aggressor population. In Blokland

Lussier

(Eds), Sex offenders: A criminal career approach (1st ed., pp. 351–371). John Wiley & Sons, Ltd.

Bourke

M. L.

Hernandez

A. E.

(2009). The “Butner study” redux: A report of the incidence of hands-on child victimization by child pornography offenders. Journal of Family Violence, 24(3), 183–191. https://doi.org/10.1007/s10896-008-9219-y

Bruning

J. L.

Kintz

B. L.

(1977). Computational Handbook of statistics (2nd ed.). Scott, Foresman and Company.

Budd

Desmond

S. A.

(2014). Sex offenders and sex crime recidivism: Investigating role of sentence length and time served. International Journal of Offender Therapy and Comparative Criminology, 58(12), 1481–1499. https://doi.org/10.1177/0306624X13507421

Cicchetti

D. V.

(1994). Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychological Assessment, 6(4), 284–290. https://doi.org/10.1037/1040-3590.6.4.284

Cohen

(1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates, Inc.

10.

DeLisi

Caropreso

D. E.

Drury

A. J.

Elbert

M. J.

Evans

J. L.

Heinrichs

Tahja

K. M.

(2016). The dark figure of sexual offending: New evidence from federal sex offenders. Journal of Criminal Psychology, 6(1), 3–15. https://doi.org/10.1108/JCP-12-2015-0030

11.

Doren

D. M.

(2010). Empirically based risk assessment estimate extrapolations across time and outcome measure. In Schlank

(Ed.), The Sexual Predator: Legal Issues, Assessment, & Treatment (IV, pp. 18-118). New York, NY: Civic Research Institute.

12.

Drury

A. J.

Elbert

M. J.

DeLisi

(2020). The dark figure of sexual offending: A replication and extension. Behavioral Sciences & the Law, 38(1), 559–570. https://doi.org/10.1002/bsl.2488

13.

Falshaw

Bates

Patel

Corbett

Friendship

(2003). Assessing reconviction, reoffending and recidivism in a sample of UK sexual offenders. Legal and Criminological Psychology, 8(2), 207–215. https://doi.org/10.1348/135532503322362979

14.

Groth

A. N.

Longo

R. E.

McFadin

J. B.

(1982). Undetected recidivism among rapists and child molesters. Crime & Delinquency, 28(3), 450–458. https://doi.org/10.1177/001112878202800305

15.

Hanson

R. K.

Thornton

(2000). Improving risk assessments for sex offenders: A comparison of three actuarial scales. Law and Human Behavior, 24(1), 119–136. https://doi.org/10.1023/A:1005482921333

16.

Hanson

R. K.

Thornton

(2003). Solicitor general Canada. Notes on the development of the static-2002 (user report 2003-01). Available from www.static99.org

17.

Hanson

R. K.

Thornton

Price

(2003). Estimating sexual recidivism rates: Observed and undetected [conference session]. In: Association for the Treatment of Sexual Abusers (ATSA) Annual Research and Treatment Conference, St. Louis, MO, United States, October 9, 2003.

18.

Helmus

Thornton

Hanson

R. K.

Babchishin

K. M.

(2012). Improving the predictive accuracy of Static-99 and Static-2002 with older sex offenders: Revised age weights. Sexual Abuse, 24(1), 64–101. https://doi.org/10.1177/1079063211409951

19.

Helmus

L. M.

(2021). Estimating the probability of sexual recidivism among men charged or convicted of sexual offences: Evidence-based guidance for applied evaluators. Sexual Offending: Theory, Research, and Prevention, 16(1), 1–24. https://doi.org/10.5964/sotrap.4283

20.

Helmus

L. M.

Thornton

(2015). Stability and predictive and incremental accuracy of the individual items of static-99R and static-200R in predicting sexual recidivism: A meta-analysis. Criminal Justice and Behavior, 42(9), 917–937. https://doi.org/10.1177/0093854814568891

21.

Hsieh

M.-L.

Hamilton

Zgoba

K. M.

(2018). Prison experience and reoffending: Exploring the relationship between prison terms, institutional treatment, infractions, and recidivism for sex offenders. Sexual Abuse, 30(5), 55–575. https://doi.org/10.1177/1079063216681562

22.

Iacono

W. G.

Ben-Shakhar

(2018). Current status of forensic lie detection with the comparison question technique: An update of the 2003 National Academy of Sciences report on polygraph testing. Law and Human Behavior, 43(1), 86–98. https://doi.org/10.1037/lhb0000307

23.

Kelley

S. M.

Ambroziak

Thornton

Barahal

R. M.

(2020). How do professionals assess sexual recidivism risk? An updated survey of practices. Sexual Abuse, 32(1), 3–29. https://doi.org/10.1177/1079063218800474

24.

Koss

M. P.

Gidycz

C. A.

Wisniewski

(1987). The scope of rape: Incidence and prevalence of sexual aggression and victimization in a national sample of higher education students. Journal of Consulting and Clinical Psychology, 55(2), 162–170. https://doi.org/10.1037/0022-006X.55.2.162

25.

Lauritsen

J. L.

Rezey

M. L.

(2013). Measuring the prevalence of crime with the national crime victimization survey. Bureau of Justice Statistics, United States Department of Justice. https://bjs.ojp.gov/library/publications/measuring-prevalence-crime-national-crime-victimization-survey

26.

Lisak

Miller

P. M.

(2002). Repeat rape and multiple offending among undetected rapists. Violence and Victims, 17(1), 73–84. https:/doi.org/10.1891/vivi.17.1.73.33638

27.

Lovell

Huang

Overman

Flannery

Klingenstein

(2020). Offending histories and typologies of suspected sexual offenders identified via untested sexual assault kits. Criminal Justice and Behavior, 47(4), 470–486. https://doi.org/10.1177/0093854819896385

28.

Lussier

Davies

(2011). A person-oriented perspective on sexual offenders, offending trajectories, and risk of recidivism: A new challenge for policymakers, risk assessors, and actuarial prediction? Psychology, Public Policy, and Law, 17(4), 530–561. https://doi.org/10.1037/a0024388

29.

Mears

D. P.

Cochran

J. C.

(2018). Progressively tougher sanctioning and recidivism: Assessing the different types of sanctions. Journal of Research in Crime and Delinquency, 55(2), 194–241. https://doi.org/10.1177/0022427817739338

30.

Moss

Stephens

Seto

M. C.

(2022). The association between sentencing and recidivism in men who have committed a sexual offence. Journal of Sexual Aggression. online first publication https://doi.org/10.1080/13552600.2021.2015467

31.

Nunes

K. L.

Firestone

Wexler

A. F.

Jensen

T. L.

Bradford

J. M.

(2007). Incarceration and recidivism among sexual offenders. Law and Human Behavior, 31(3), 305–318. https://doi.org/10.1007/s10979-006-9065-5

32.

Scurich

John

R. S.

(2019). The dark figure of sexual recidivism. Behavioral Sciences & the Law, 37, 158–175. https://doi.org/10.1002/bsl.2400

33.

Seto

M. C.

Stephens

Lalumière

M. L.

Cantor

J. M.

(2017). The revised Screening scale for Pedophilic interests (SSPI-2): Development and criterion-related validation. Sexual Abuse, 29(1), 619–635. https://doi.org/10.1177/1079063215612444

34.

Sinozich

Langton

(2014). Rape and sexual assault among college-age females, 1995-2013. Bureau of Justice Statistics, United States Department of Justice. https://bjs.ojp.gov/library/publications/rape-and-sexual-assault-among-college-age-females-1995-2013

35.

Stephens

Klein

L. K.

Seto

M. C.

(2021). Differences in sexual interest in children between men who commit undetected versus detected sexual offenses. Sexual Abuse, 33(3), 255–273. https://doi.org/10.1177/1079063219889056

36.

Tabachnick

B. G.

Fidell

L. S.

(2019). Using Multivariate statistics (7th edition). Pearson.

37.

Thornton

Hanson

R. K.

Kelley

S. M.

Mundt

J. C.

(2021). Estimating lifetime and residual risk for individuals who remain sexual offense free in the community: Practical applications. Sexual Abuse, 33(1), 3–33. https://doi.org/10.1177/1079063219871573

38.

Villettaz

Gillieron

Killias

(2015). The effects on re-offending of custodial vs. non-custodial sanctions: An updated systematic review of the state of knowledge. Campbell Systematic Reviews, 1(1), 1–92. https://doi.org/10.4073/csr.2015.1

39.

Weinrott

M. R.

Saylor

(1991). Self-report of crimes committed by sex offenders. Journal of Interpersonal Violence, 6(3), 286–300. https://doi.org/10.1177/088626091006003002

40.

Wollert

Cramer

(2012). The constant multiplier assumption misestimates long-term sex offender recidivism rates. Law and Human Behavior, 36(5), 390–393. https://doi.org/10.1037/h0093924

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.33 MB