Visual Reaction Time™ as a Predictor of Sexual Offense Recidivism

Abstract

In this study, we investigated the use of Visual Reaction Time™ (VRT™) for sexual interest in children to predict recidivism of sexual offenses among men who sexually abused children and men with other sexually deviant behaviors. The authors hypothesized that study participants with a higher VRT™ to stimuli of children would be more likely to sexually reoffend compared with those with a lower VRT™ to stimuli of children. Participants included 621 adult males on parole or probation for acting on a range of sexual paraphilias who sought outpatient treatment or evaluation at two separate therapists’ practices. Sample 1 consisted of 284 adult males followed up (by the lead author) during a 15-year period, while Sample 2 consisted of 337 adult males followed up (by the second author) during a 7-year period. A discrete-time hazard model found VRT™ to children to be significantly related to sexual recidivism. The researchers found that VRT™ to children measured at intake held up in its predictive ability over a 15-year period. When the participants were divided into three groups based on their VRT™, of the 97 participants who measured at least one standard deviation lower than the mean VRT™, 0% reoffended. The 432 participants in the medium-VRT™ group had an estimated recidivism rate of 7% after 15 years and the 92 participants who measured at least one standard deviation higher than the mean had an estimated recidivism rate of 27%.

Keywords

sexual recidivism recidivism prediction recidivism prediction risk assessment child molestation child sexual abuse Visual Reaction Time™VRT™sex offender

Introduction

Sexual interest in children as measured by phallometry has been recognized as a significant predictor of sexual recidivism in the meta-analyses by Hanson and Morton-Bourgon (2004) and Mann, Hanson, and Thornton (2010). Hanson and Morton-Bourgon wrote that among phallometric measures, sexual interest in children and having any deviant sexual interest helped predict the sexual recidivism of convicted sex offenders. Sexual interest in rape/violence and sexual interest in boys alone were not found to be significant in the 2004 article. However, sexual interest in violence was found to be significant in the updated meta-analysis by Mann et al. (2010). Hanson and Morton-Bourgon (2005) noted that the strongest predictors of the recidivism of sex offenders were those concerning sexual deviance and failure to self-regulate. They also noted that these sets of variables helped predict recidivism while seemingly useful variables such as neglect or abuse during childhood, sexual abuse during childhood, loneliness, low self-esteem, lack of victim empathy, denial of a sexual crime, low motivation for treatment at intake, and poor progress in treatment were not helpful in predicting sexual recidivism.

Like phallometry, Visual Reaction Time™ (VRT™) is an objective measure of sexual interest. Phallometry and VRT™ objectively measure very different things. While phallometry measures sexual arousal (increase in penis size), VRT™ measures sexual interest (increase in visual attention).

VRT™ as used with its specific scoring algorithm and its unique set of images is one of several commercially produced variations of viewing time.

Viewing time hypothesizes that people look longer at images of people who attract them. Researchers have investigated this phenomenon as early as 70 years ago (Brown, 1979; Glasgow, Osborne, & Croxen, 2003; Harris, Rice, Quinsey, & Chaplin, 1996; Quinsey, Ketsetzis, Earls, & Karamanoukian, 1996; Rosenzweig, 1942; Zamansky, 1956).

VRT™ is different from phallometry both in what it measures and in that it does not stand alone, but is combined with 15 other measures as part of the suite of tests that make up the Abel Assessment for sexual interest™ (AASI™).

Three questions need to be answered before considering the findings in the present study: (1) Can viewing time measure sexual interest? (2) Does the VRT™ version of viewing time, as an objective measure of sexual interest in children, produce similar results to phallometry? (3) Have studies been replicated by independent researchers?

Can viewing time measure sexual interest? A person’s sexual interests determine to who (or at times what) he or she is sexually attracted. B. Singer (1984) conceptualized three stages of sexual attraction: (a) an aesthetic response, such as increased visual attention to a stimulus; (b) an approach response, such as movement toward the stimulus; and (c) a genital response. Attentional measures of sexual interest such as VRT^™, from a conceptual basis, are concerned with the first of Singer’s three stages of attraction and are based on the assumption that the greater a person’s attraction to a stimulus, the more attentional resources will be devoted to attending to that stimulus.

Viewing time has been found to be a reliable and valid measure of sexual interest in men. Rosenzweig (1942) found that patients who the staff had rated as being highly interested in sexual matters evidenced longer viewing times to sexual stimuli than those patients who were rated as having low interest in sexual matters. Early studies found that as the erotic content of heterosexual images increases, so does viewing time by men (Brown, Amoroso, Ware, Pruesse, & Pilkey, 1973). Landolt, Lalumière, and Quinsey (1995) found viewing time to increase linearly with attractiveness ratings of images that depicted the head and shoulders of adults. Based on samples of heterosexual men, viewing time correlated significantly with ratings of image attractiveness, sexual arousal, and sexual stimulation (Lang, Searles, Lauerman, & Adesso, 1980; Quinsey et al., 1996; Quinsey, Rice, Harris, & Reid, 1993). Based on the results of four independent samples, Quinsey et al. (1996) calculated the mean Pearson correlation (r) between image sexual attractiveness ratings and viewing time to be r = .72, range = .54 to .91.

Support for the validity of viewing time as a specific measure of sexual interest in children comes from convergent validity analyses between viewing time and phallometry assessments. Based on a sample of heterosexual, university-aged men (N = 24) and heterosexual university-aged women (N = 24) viewing images of adult, pubescent, and young children, the correlations between viewing time and phallometry for men ranged from r = −.05 to r = .84, with a mean correlation of r = .42, SD = 0.27 (Quinsey et al., 1996).

In another study by Quinsey et al. (1996) in the same article, the correlations between viewing time and rating of sexual attractiveness for adults, pubescent, and young children was .70 for all subjects, .80 for males, and .60 for females.

Does the VRT™ version of viewing time, as an objective measure of sexual interest, produce similar results to phallometry? VRT™ was found by researchers to be similar to phallometry in its ability to discriminate between known groups of child abusers categorized by the sex and age of their victims (Abel, Huffman, Warberg, & Holland, 1998; Abel, Jordan, Hand, Holland, & Phipps, 2001; Abel, Lawry, Karlstrom, Osborn, & Gillespie, 1994; Abel, Phipps, Hand, & Jordan, 1999; Abel & Wiegel, 2009; Gray & Plaud, 2005; Johnson & Listiak, 1999; Letourneau, 2002; Maram, 2002; Seghorn & Wiegel, 1999).

As VRT™ is a proprietary objective measure and early studies comparing VRT™ with phallometry were conducted by researchers connected with the inventor (Gene G. Abel, the second author), have those studies been replicated by independent researchers?

Replication is important. Independent researchers in five centers compared VRT™ with phallometry. Their results showed VRT™ to be similar to phallometry in discriminating between known groups (Gray & Plaud, 2005; Johnson & Listiak, 1999; Letourneau, 2002; Maram, 2002; Seghorn & Wiegel, 1999).

VRT™ has been widely accepted as an objective measure. U.S. Probation recommends its use for federal contracts. It has been accepted in federal and state courts as scientifically valid. It is used by criminal justice facilities and sex-specific therapeutic practices in 50 states, Puerto Rico, Canada, Australia, and Ireland with its initial use starting 17 years ago.

This article reports the results of a sexual recidivism study conducted on 621 men using the VRT™ for sexual interest in children from the AASI™. The authors tested the hypothesis that VRT™ for sexual interest in children was related to sexual offense recidivism. As the sample included men with a variety of sexual deviations that led to probation or parole, the authors also tested whether VRT™ to children was more effective at predicting recidivism in men who admitted to sexually abusing children than for men with other deviant sexual behaviors.

Method

Sample

There were two sample groups of participants in this study.

Sample 1 consisted of 284 adult males who sought treatment or evaluation for deviant sexual behaviors at the clinical practice of the lead author during the 15-year period from 1995 to 2010. Their ages ranged between 18 and 75 years at first evaluation (M = 38.6, median = 37, SD = 12.6). The participants in Sample 1 were 73% Caucasian (207), 15% Hispanic/Latino (43), 5% Native American (14), 5% African American (13), and 1% Asian American (3); 1% classified themselves as Biracial/Other (3), and 1 participant did not answer the question. To ensure compatibility with the data structure of the second sample, reoffense dates were grouped into 6-month intervals starting from the participants’ first assessment. The number of 6-month follow-up periods ranged from 1 to 30 (M = 11.2, median = 9, SD = 7.3) with the longest follow-up being 15 years. Fourteen of the 284 participants sexually reoffended during follow-up.

Sample 2 consisted of 337 adult males who sought treatment or evaluation for deviant sexual behaviors at the clinical practice of the second author during the 7-year period from 1996 to 2003. Their ages ranged between 17 and 95 years at their first evaluation (M = 39.6, median = 38, SD = 12.8). The participants in Sample 2 were 78% Caucasian (264), 15% African American (50), 1% Hispanic/Latino (5), 1% Native American (5), and 0% Asian American (1); 3% described themselves as Biracial/Other (10) and 2 participants did not answer the question. The differences in ethnicity between Samples 1 and 2 were statistically significant according to a chi-square test of independence (χ² = 63.8, df = 5). Sample 1 had a higher percentage of Hispanic/Latinos (χ² = 40.3, df = 1) and a lower percentage of African Americans (χ² = 17.8, df = 1).

Data were collected in 6-month intervals starting from the participants’ first assessment. The number of 6-month follow-up periods ranged from 1 to 15 (M = 4.3, median = 4, SD = 3.2) with the longest follow-up period being seven and a half years. Eight of the 337 participants sexually reoffended during follow-up.

All participants were receiving outpatient evaluation or cognitive-behavioral treatment. The opportunity time for reoffense was from initial assessment to the end of probation or parole.

Table 1 shows the deviant sexual behaviors in which participants admitted involvement prior to their initial assessments. Participants in Sample 1 engaged in a wider range of deviant sexual behaviors (M = 1.4, median = 1, SD = 1.5) than participants in Sample 2 (M = 0.8, median = 0, SD = 1.3). The means were significantly different according to a pooled variances t test (t = 4.8, df = 619, p ≤ .0001).

Table 1.

Deviant Sexual Behaviors of Participants by Sample.

	Sample 1		Sample 2		Combined
	n = 284		n = 337		n = 621
	Frequency	%	Frequency	%	Frequency	%
Child sex abuse	124	44	98	29	222	36
Voyeurism	67	24	48	14	115	19
Public exposure	47	17	31	9	78	13
Fetishism	36	13	20	6	56	9
Bestiality	36	13	14	4	50	8
Public masturbation	33	12	34	10	67	11
Frottage	22	8	18	5	40	6
Obscene phone calls/letters	20	7	18	5	38	6
Rape	11	4	5	1	16	3

Twenty-two of the 621 participants in the combined sample reoffended during follow-up.

Materials

Materials included the VRT™ measure from the AASI™. The assessments were conducted according to the protocols previously published for them (Abel et al., 1998; Abel et al., 2001). Each participant received the entire suite of tests in the AASI™.

The visual stimuli for VRT™ contain four age categories: adults, teenagers, grade-school children, and preschool children. Caucasians and African Americans are represented equally in all four categories as are males and females. The adult category includes men and women 21 years of age and older. The teenage category represents teens 14 to 17 years of age. The grade-school category represents children 6 to 13 years of age. The preschool category represents children 5 years of age and younger. All images were age rated by a pediatrician, a pediatric surgeon, and five community members of different ethnicities.

Participants viewed two sets of 80 images. Each age category was seven pictures deep and included seven pictures of males and females of a specific age and race. All images showed a frontal view of an adult, teenager, or child in a bathing suit against a neutral background. None of the images depicted sexual content or sexually aroused individuals. In addition, participants viewed images representing paraphilias such as voyeurism, exhibitionism, frottage, sadism/masochism, and fetishism.

Procedures

Participants were asked to sign consent forms authorizing their data for research prior to beginning the AASI™. Participants completed the AASI™ sex-specific questionnaire. Next, they completed the VRT™ assessment, during which they were seated in front of a computer that was linked to a slide projector. While they looked at a group of practice slides, participants were shown how to advance slides by pressing a key on the computer keyboard and how to use the keyboard to rate their sexual arousal to each slide on a 7-point Likert-type scale (1 = highly sexually disgusting, 4 = sexually neutral, and 7 = highly sexually arousing). During the actual VRT™ assessment, the computer measured the time each participant took to advance the slides as well as the time the participants took to rate the slides to record their sexual arousal. The most recent participants viewed the identical VRT™ stimuli on the screen of a laptop computer, rather than on a slide projector. Both the questionnaire and VRT™ were completed prior to the beginning of treatment.

Recidivism was defined by arrests and charges for sexual offenses. Probation violations were not counted as sexual recidivism. Information included the therapist’s knowledge and information collected from parole and probation officers. In Sample 1, information was collected post hoc from the clients’ files. In Sample 2, information was collected from a questionnaire sent to parole and probation officers and returned by them every 6 months.

Sixteen VRT™ stimuli categories of age, gender, and race were used to assess the participants’ relative sexual interest in children and their sexual interest in adults (including adolescents). For relative sexual interest in children, the mean VRT™ of eight categories of children was divided by the mean VRT™ of eight categories of adults and adolescents.

Results

Hypotheses Tests

The authors tested two hypotheses. Hypothesis 1 proposed that the participants with a higher VRT™ to children would reoffend at a greater rate than participants with a lower VRT™ to children. Hypothesis 2 proposed that VRT™ to children would be better at predicting the recidivism of men who admitted to having sexually abused children than predicting recidivism for men who presented with other sexual deviations.

Hypothesis 1

To test the first hypothesis, the authors used two methods of analysis: (a) a VRT™ means or t test and then (b) a discrete-time hazard model to create an idealized recidivism prediction model. Harrell’s c index (Harrell, Lee, & Mark, 1996) and Cohen’s d (Cohen, 1988) were used as measures of effect size.

First, to demonstrate a difference in VRT™ means, the authors used an independent group’s t test to determine if those participants who reoffended had a higher VRT^™ than those who survived (did not reoffend). Table 2 shows the summary statistics for VRT^™. The mean VRT™ for the 22 reoffenders was 0.80 (SD = 0.19, median = 0.80) while the mean for the non-re-offenders was 0.66 (SD = 0.20, median = 0.66). The difference between means was statistically significant, t(619) = 3.3, p ≤ .0011.

Table 2.

Summary Statistics for VRT™ to children by Reoffense Status.

	n	Minimum	Maximum	M	Median	SD
Reoffended	22	0.52	1.28	0.80	0.80	0.19
Did not reoffend	599	0.15	1.41	0.66	0.66	0.20
Total	621	0.15	1.41	0.67	0.66	0.20

Second, the authors used a discrete-time hazard model with complementary log-log link (Hosmer & Lemeshow, 1999; Prentice & Gloeckler, 1978; J. D. Singer & Willett, 2003) to test whether VRT™ predicted reoffense.

Discrete-Time Hazard Model

The authors used a discrete-time hazard model with complementary log-log link developed by Prentice and Gloeckler (1978). Prentice and Gloeckler developed this model specifically to handle survival analysis when the researcher only knows that an event has occurred at some point within an interval (6-month intervals in this study), but does not know the exact time. In this case, the data are said to be interval censored. The discrete-time hazard model with complementary log-log link is similar to the Cox proportional hazard model (Cox, 1972). Like the Cox proportional hazard model, the discrete-time hazard model with complementary log-log link produces a coefficient whose antilog is a hazard ratio (ratio of the probability of an event happening at that time). Unlike the Cox proportional hazard model, time is considered to be measured in discrete intervals rather than continuously. The discrete-time hazard model requires a term to specify the baseline hazard function, not specifying a term or terms for time means that an event is equally likely to happen at any time during the specified interval.

Table 3 shows the results of the discrete-time hazard model with complementary log-log link where the authors tested the ability of VRT™ to children to predict recidivism (Model 1). To do this, we entered the effect for VRT™ and a dummy variable for whether the time period was the first 6 months or not to specify the baseline hazard model. The authors attempted a variety of specifications of the baseline hazard function such as linear, quadratic, logarithmic, and a spline. The dummy variable for the first 6-month period was by far the best fitting for the data.

Table 3.

Discrete-Time Hazard Model 1.

	Coefficient	SE	Chi-square	Significance	Hazard ratio
Intercept	−7.67	0.81	89.7	.0001	—
VRT^™	2.81	0.95	8.8	.0031	16.7
First 6 months	1.15	0.46	6.3	.0118	3.2

Note. VRT^™ = Visual Reaction Time™.

Entering VRT™ to children into the equation alongside the first period dummy demonstrates that VRT™ is statistically significantly related to recidivism (coefficient = 2.81, χ² = 8.8, df = 1, p = .0031, 95% confidence interval [CI] = [0.91, 4.71]). The positive coefficient means that as VRT™ to children increases, so does the probability of recidivism. The hazard ratio is the antilog of the regression coefficient; therefore, the hazard ratio for a one unit change in VRT™ is 16.7, 95% CI = [2.5, 111.4]. Because a change in VRT™ of one unit represents most of the range (0.15-1.41), the authors decided to estimate the coefficient with VRT™ standardized so that the new VRT™ variable would represent the change in log hazard due to a one standard deviation change in VRT™. The standardized coefficient for VRT™ is 0.56 and the standardized hazard ratio is 1.8, meaning that the hazard rate increases 80% for each one standard deviation change in VRT™.

The term for the first period dummy was also statistically significant (coefficient = 1.15, χ² = 6.3, df = 1, p ≤ .0118, 95% CI = [0.24, 1.61]). Because the coefficient is positive, the model indicates that reoffense is more likely to happen in the first 6 months than in any other individual 6-month interval. The hazard ratio for the first 6 months is 3.2, indicating that participants are 3.2 times as likely to reoffend in the first 6 months.

Survival Plot: Low-, Medium-, and High-VRT™ to Children Groups

Figure 1 shows the survival rates based on estimates according to Kaplan and Meier (1958). Figure 1 plots the predicted survival rates (participants not reoffending) by grouped sexual interest in children as measured by VRT™ across 15 years. Because the longest follow-up period in this study was 15 years, the authors estimated the survival curve out to 15 years.

Figure 1.

Survival rate by the VRT™ measure of sexual interest in children showing predicted probability of not having reoffended over a 15-year period.

Participants were divided into three VRT™ groups. The 97 men whose VRT™ measures were at least one standard deviation lower than the mean had an estimated survival rate of 100% at 15 years (0% reoffense). The 92 men whose measures were at least one standard deviation higher than the mean had an estimated survival rate of 73% (27% reoffense). The reoffense rate for this high group increased as their VRT™ to children increased. The medium group (between the two standard deviations) was the largest; 432 men in that group had an estimated 93% survival rate (7% chance of reoffending).

Predicted Reoffense Rates

In linear models, the coefficients demonstrate the effect of a unit change in the independent variable on the dependent variable. In nonlinear models such as those that are used in survival analysis, it is often easier to demonstrate the effect by showing the predicted values at various levels of the independent variables. Figure 2 shows selected predicted reoffense rates for Years 1, 5, 10, and 15. These values for VRT™ to children from the discrete-time hazard model are shown at various standard deviations from the mean. At 1 year out, the model predicts only a 0.4% (95% CI = [0.1%, 1.3%]) risk of reoffense for clients with a VRT™ at the low end of two standard deviations below the mean, a 1.3% (95% CI [0.7%, 2.4%]) risk at the mean, and a 3.8% (95% CI = [1.8%, 8.2%]) risk at the high end of two standard deviations above the mean. At 10 years out, the model predicts an increase in reoffense risk to 2.2% (95% CI = [0.7%, 6.5%]) for clients with a VRT™ at the low end (two standard deviations below the mean), a 6.5% (95% CI = [4.0%, 10.8%]) risk at the mean, and at the high end (two standard deviations above the mean), an increase in reoffense risk to 18.6% (95% CI = [10.2%, 33.4%]). At 15 years out, the model predicts an increase in reoffense risk to 3.1% (95% CI = [1.1%, 9.3%]) for clients with a VRT™ at the low end (two standard deviations below the mean), a 9.3% (95% CI = [5.6%, 15.5%]) risk at the mean, and at the high end (two standard deviations above the mean) an increase in reoffense risk to 25.8% (95% CI = [14.3%, 44.7%]).

Figure 2.

Predicted reoffense rate based on VRT™ to children.

These are the estimated probabilities that a client would reoffend by that period of time. For example, for two standard deviations above the mean at 15 years out, the 25.8% risk is the estimated probability that he would have reoffended at some point by that time. However, should he have survived (not reoffended) through all but the last interval, his risk of reoffense in the 6-month period right before the 15th year is only 1.6%.

Measures of Effect Size

Harrell’s concordance index for censored data (c index) is a nonparametric measure of association that estimates the probability that an observation with the larger value of the dependent variable has the larger predicted value (Harrell et al., 1996). In the case of a binary dependent variable, Harrell’s c index (for noncensored data) is equivalent to the area under the receiver operator characteristics curve (AUC) that is commonly used as a measure of effect size in logistic regression (Hanley & McNeil, 1982). While the interpretations for Harrell’s c index for censored and noncensored data are equivalent, the formulas are not. For censored data, pairs of observations, where the censored times exceed those of noncensored, were omitted from the calculation. Harrell’s c index was 0.68.

Cohen’s d (Cohen, 1988) is a standardized measure of effect size that compares the difference between two means relative to the standard deviation. The meta-analyses of Hanson and Morton-Bourgon (2004) and Mann et al. (2010) attempted to translate the effect sizes of studies conducted with various statistical procedures into a common metric. The metric Hanson and Morton-Bourgon chose was Cohen’s d. Cohen’s d calculated from the means of VRT™ to children for reoffenders and non-re-offenders divided by the pooled standard deviation was 0.71.

Hypothesis 2

This hypothesis proposed that VRT™ to children would be more effective as a predictor of recidivism for the men who admitted to child sexual abuse (36% of the sample) than for the men who were arrested for other sexual deviations (74% of the sample).

The authors tested Hypothesis 2 by adding terms to the model for child sexual abuser status and the interaction of child sexual abuser status and VRT™ and testing whether this new model (Model 2) fits significantly better. Neither of the terms for child sexual abuse status (coefficient = −0.19, χ² = 0.1, df = 1, p ≤ .91, 95% CI = [−3.53, 3.16]) and the interaction term for VRT™ and child sexual abuse status (coefficient = 0.21, χ² = 2.0, df = 1, p ≤ .02, 95% CI = [−3.83, 4.25]) were statistically significant (see Table 4). The likelihood ratio test comparing Model 2 with Model 1 was not statistically significant (χ² = 0.03, df = 2, p ≤ .985). Surprisingly, the authors found no evidence of a different relapse prediction effect for VRT™ to children between those participants who admitted to child sexual abuse and those who admitted to other deviant behaviors.

Table 4.

Discrete-Time Hazard Model 2.

	Coefficient	SE	Chi-square	Significance	Hazard ratio
Intercept	−7.6	1.19	40.6	.0001	—
VRT^™	2.72	1.54	3.1	.0779	15.2
First 6 months	1.16	0.46	6.3	.0120	3.2
CSA status	−0.19	1.67	0.0	.9116	0.8
VRT™ * CSA status	0.21	2.02	0.0	.9169	1.2

Note. VRT™ = Visual Reaction Time™; CSA = child sexual abuser.

Discussion

Sexual Recidivism Rates

Rates for sexual recidivism are much lower than the rates given by the media. Mann et al. (2010) found in an analysis of 10 studies that sexual recidivism was in the range of 10% to 15% after 5 years. Sexual recidivism in our study was about 4% after 5 years. What may account for this low reoffense rate finding are two factors affecting the men in the study. Participants were receiving treatment and were also monitored continuously as they were on probation or parole.

VRT™ Effect Size Compared With Phallometry

The effect size of VRT™ for sexual interest in children appears to be of moderate to large size. The Cohen’s d for VRT^™ was 0.71. By comparison, Hanson and Morton-Bourgon (2004) in their analysis using phallometry as a predictor of recidivism reported a mean Cohen’s d of 0.32. Their meta-analysis included 13 studies with 2,180 separate phallometric measurements and an average follow-up time of 5 to 6 years. Cohen divided effect sizes for d into three categories: small, medium, and large. Although Cohen did not define ranges, 0.2 is often interpreted as the middle of the small range, 0.5 as the middle of the medium range, and 0.8 as the middle of the large range. Therefore, according to Cohen, the effect size found for VRT™ (0.71) would be labeled as moderate to large.

Predicting Recidivism Across Deviations

Having men in this study limited to those who presented with a range of sexual deviations (including child sexual abuse) afforded the opportunity for some interesting findings. VRT™ to children predicted sexual recidivism across deviant behaviors. Previous research suggests an important possibility for this phenomenon: Men who presented with deviations other than child sexual abuse also crossed into child sexual abuse. Abel, Becker, Cunningham-Rathner, Mittelman, and Rouleau (1988) showed that among men who presented with any sexual paraphilia, there was a considerable crossing of diagnosis. Looking at men who presented with deviant behavior, other than child sexual abuse, they found that a high percentage of these men also sexually abused children. For instance, 46% of exhibitionists, 52% of voyeurs, and 71% of men involved in bestiality also sexually abused little girls outside the home. Boys were also sexually abused by all categories, though in lower percentages. In addition, the men presenting with other deviant behaviors also crossed diagnostically into incest. This crossing of diagnosis from other deviant behaviors into child sexual abuse might well account for the finding that high VRT™ to children predicted recidivism across paraphilias.

Limitations

The major limitation of this study is the small number of participants who reoffended. The precision of estimates of recidivism suffers accordingly.

Although 621 participants appear to be an adequate sample size, only in 22 of those cases did a reoffense occur. Why so few? As both authors were providing a cognitive-behavioral treatment with a strong relapse prevention module and closely supervised follow-up, one could speculate that these interventions were effective. Another possibility might be that the probation and parole officers may have been unaware of reoffenses that occurred out of their jurisdiction.

When examining this study, readers should be aware of how the low number of reoffenses might affect the results. This made modeling the baseline hazard rate of recidivism difficult. Although we observed no reduction in the risk of recidivism after the first 6 months time period, there were insufficient numbers of recidivists to allow the full 15-year survival function to be modeled with confidence.

When the Kaplan–Meier estimation was applied to the three VRT™ groups—low, medium, and high—there were no known reoffenses among the 97 men whose VRT™ to children was at least one standard deviation lower than the mean. However, in a much larger sample, one would expect a few reoffenses among this group. Similarly, the discrete-time hazard model is an idealized statistical predictive model with few participants in the outer ranges.

Clinical Applications and Future Directions

While acknowledging this study’s limitations, the results are indicative of the ability of VRT™ to predict a sexual abuser’s recidivism at intake and may well offer clinicians the opportunity to triage clients. In an economy where funding cutbacks are common and resources reduced, cost-effective treatment—that protects the public, especially the children—becomes increasingly important. The clinician might incorporate VRT™ recidivism measures into decision making. A clinician can divide clients into those with a low probability and those with a high probability to reoffend. A clinician might then concentrate on clients with the greatest likelihood of reoffending, giving them a more focused treatment and increased surveillance.

The widespread use of VRT™ may provide many important research possibilities. Future recidivism studies might include female sexual abusers as well as men. As research on recidivism is continually hampered by low sexual offense recidivism rates, one could bring a large number of treatment centers together to confirm this study’s findings with a sample size of thousands of abusers. One-half of this proposed study has been completed as VRT™ assessment at intake is commonly performed throughout the United States and the data have been gathered and stored over many years.

In article after article on sexual recidivism, researchers have expressed their desire to find dynamic variables. Because, in this study, each participant’s VRT™ was measured only at intake, VRT™ to children produced limited data. This study suggests that VRT™ holds up well over 15 years as a static predictor. It is possible that a future study could show that VRT™ responds to treatment and therefore could become a dynamic variable. It is especially difficult to do treatment outcome studies because few treatment centers can accumulate a large enough sample to examine treatment’s relation to recidivism. However, once again, we suggest that by using VRT™ at multiple treatment centers, as the measure is standardized, it would be possible to seamlessly combine results into a large sample to test whether treatment can lower an abuser’s VRT™ to children and concomitantly reduce recidivism. Such a study would not only help establish the utility of VRT™ for sexual interest in children as a dynamic variable, but could also help establish the efficacy of cognitive-behavioral therapy as important in protecting children from victimization.

To advance research in sexual recidivism, future studies might also attempt to combine VRT™ with other predictive variables and examine the possibility of incorporating VRT™ for sexual interest in children into existing actuarial recidivism indices.

Conclusion

Previous researchers have proposed that sexual interest in children is a major predictor of sexual recidivism. This study not only corroborates the previous research based on an objective measure, phallometry but also adds evidence to this recidivism hypothesis with an alternative objective measure of sexual interest in children, VRT™. These results add strength to the previous findings outlined by Hanson and Morton-Bourgon (2005) and Mann et al. (2010).

The authors found no evidence of a different relapse prediction effect for VRT™ to children between men who admitted to child sexual abuse and men who admitted to other sexually deviant behaviors.

VRT™ to children was more powerful than expected in one sense: The effect of an abuser’s high VRT™ to child stimuli assessed once (before treatment) as a predictor of recidivism persisted over many years. When the participants were divided into three groups based on their VRT™, among participants who measured at least one standard deviation lower than the mean VRT™, 0% reoffended. Participants in the medium-VRT™ group had an estimated recidivism rate of 7% after 15 years and participants, who measured at least one standard deviation higher than the mean, had an estimated recidivism rate of 27%.

Footnotes

Acknowledgements

The authors would like to acknowledge all the probation and parole officers who cooperated in the study. The authors would also like to acknowledge the assistance of Candice Osborn, MA, and R. Karl Hanson, PhD.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Gene G. Abel is the owner and the inventor of the Abel Assessment for sexual interest™, the instrument used in the study.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Partial funding for the study was provided by Abel Screening, Inc.

References

Abel

G. G.

Becker

J. V.

Cunningham-Rathner

Mittelman

Rouleau

J. L.

(1988). Multiple paraphilic diagnoses among sex offenders. Bulletin of the American Academy of Psychiatry and the Law, 16, 153-168.

Abel

G. G.

Huffman

Warberg

Holland

C. L.

(1998). Visual reaction time and plethysmography as measures of sexual interest in child molesters. Sexual Abuse: A Journal of Research and Treatment, 10, 81-95.

Abel

G. G.

Jordan

Hand

C. G.

Holland

L. A.

Phipps

(2001). Classification models of child molesters utilizing the Abel Assessment for sexual interest™. Child Abuse & Neglect, 25, 703-718.

Abel

G. G.

Lawry

S. S.

Karlstrom

Osborn

Gillespie

C. F.

(1994). Screening tests for pedophilia. Criminal Justice and Behavior, 21, 115-131.

Abel

G. G.

Phipps

Hand

Jordan

(1999, September). The reliability and validity of visual reaction time as a measure of sexual interest in children. Paper presented at the 18th annual meeting of the Association for the Treatment of Sexual Abusers, Orlando, FL.

Abel

G. G.

Wiegel

(2009). Visual reaction time: Development, theory, empirical evidence and beyond. In Saleh

Grudzinskas

Bradford

J. M.

Brodsky

(Eds.), Sex offenders: Identification, risk assessment, treatment, and legal issues (pp. 101-118). Oxford, UK: Oxford University Press.

Brown

(1979). Viewing time of pornography. Journal of Psychology, 102, 83-95.

Brown

Amoroso

D. M.

Ware

E. E.

Pruesse

Pilkey

D. W.

(1973). Factors affecting viewing time of pornography. Journal of Social Psychology, 90, 125-135.

Cohen

(1988). Statistical power analysis for the behavioral sciences (2nd ed.). Mahwah, NJ: Erlbaum.

10.

Cox

D. R.

(1972). Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Statistical Methodological), 34, 187-220.

11.

Glasgow

D. V.

Osborne

Croxen

(2003). An assessment tool for investigating paedophile sexual interest using viewing time: An application of single case methodology. British Journal of Learning Disabilities, 31, 96-102.

12.

Gray

S. R.

Plaud

J. J.

(2005). A comparison of Abel Assessment for sexual interest™ and penile plethysmography in an outpatient sample of sex offenders. Journal of Sexual Offender Civil Commitment, 1, 1-10.

13.

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.

14.

Hanson

R. K.

Morton-Bourgon

(2004). Predictors of sexual recidivism: An updated meta-analysis (Corrections User Report No. 2004-02). Ottawa, Canada: Public Safety and Emergency Preparedness Canada.

15.

Hanson

R. K.

Morton-Bourgon

(2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73, 1154-1163.

16.

Harrell

F. E.

Lee

K. L.

Mark

D. B.

(1996). Multivariate prognostic models issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Statistics in Medicine, 15, 361-387.

17.

Harris

G. T.

Rice

M. E.

Quinsey

V. L.

Chaplin

T. C.

(1996). Viewing time as a measure of sexual interest among child molesters and heterosexual men. Behaviour Research and Therapy, 34, 389-394.

18.

Hosmer

D. W.

Lemeshow

(1999). Applied survival analysis. New York, NY: Wiley.

19.

Johnson

S. A.

Listiak

(1999). The measurement of sexual preference: A preliminary comparison of phallometry and the Abel Assessment. In Schwartz

B. K.

(Ed.), The sex offender (Vol. III, Chapter 26, pp. 2-18). Kingston, NJ: Civic Research Institute.

20.

Kaplan

E. L.

Meier

(1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53, 457-481.

21.

Landolt

M. A.

Lalumière

M. L.

Quinsey

V. L.

(1995). Sex differences and intra-sex variations in human mating tactics: An evolutionary approach. Ethology and Sociobiology, 16, 3-23.

22.

Lang

A. R.

Searles

Lauerman

Adesso

(1980). Expectancy, alcohol, and sex guilt as determinants of interest in and reaction to sexual stimuli. Journal of Abnormal Psychology, 89, 644-653.

23.

Letourneau

E. J.

(2002). A comparison of objective measures of sexual arousal and interest: Visual reaction time and penile plethysmography. Sexual Abuse: A Journal of Research and Treatment, 10, 81-95.

24.

Mann

R. E.

Hanson

R. K.

Thornton

(2010). Assessing risk for sexual recidivism: Some proposals on the nature of psychologically meaningful risk factors. Sexual Abuse: A Journal of Research and Treatment, 22, 191-217.

25.

Maram

(2002, May). Penile plethysmograph & Abel Assessment effectiveness: Research outcome. Paper presented at the 5th Annual Conference of the California Coalition on Sexual Offending, San Diego, CA.

26.

Prentice

P. L.

Gloeckler

L. A.

(1978). Regression analysis of grouped survival data with applications to breast cancer data. Biometrics, 34, 57-67.

27.

Quinsey

V. L.

Ketsetzis

Earls

Karamanoukian

(1996). Viewing time as a measure of sexual Interest. Ethology and Sociobiology, 17, 341-354.

28.

Quinsey

V. L.

Rice

M. E.

Harris

G. T.

Reid

K. S.

(1993). The phylogenetic and ontogenetic development of sexual age preferences in males: Conceptual and measurement issues. In Barbaree

H. E.

Marshal

W. L.

Hudson

S. M.

(Eds.), The juvenile sex offender (pp. 143-163). New York, NY: Guilford Press.

29.

Rosenzweig

(1942). The photoscope as an objective device for evaluating sexual interest. Psychosomatic Medicine, 4, 150-157.

30.

Seghorn

T. K.

Weigel

(1999, September). Comparative use of Abel Assessments and penile plethysmograph laboratory assessments in an outpatient forensic practice. Paper presented at the 18th Annual Research and Treatment Conference of the Association for the Treatment of Sexual Abusers, Lake Buena Vista, FL.

31.

Singer

(1984). Conceptualizing sexual arousal and attraction. Journal of Sex Research, 20, 230-240.

32.

Singer

J. D.

Willett

J. B.

(2003). Applied longitudinal data analysis: Modeling change and event occurrence [Kindle Version 1.10.5]. Oxford, UK: Oxford University Press. doi:10.1093/acprof:oso/9780195152968.001.0001

33.

Zamansky

H. S.

(1956). A technique for measuring homosexual tendencies. Journal of Personality, 24, 436-448.