Taxometric Analysis of the Latent Structure of Pedophilic Interest

Abstract

The present study examined the latent structure of pedophilic interest. Using data from phallometric tests for pedophilic interest across four samples of offenders (ns = 805, 632, 531, 261), taxometric analyses were conducted to identify whether pedophilic interest is best characterized as taxonic or dimensional. Across the samples, the majority of analyses supported taxonic latent structure in pedophilic interest. Visual inspection of taxometric curves indicated trichotomous latent structure (i.e., three-ordered classes) may characterize pedophilic interest in these samples. In a second step of taxometric analysis, the results supported trichotomous latent structure, indicating the presence of a complement taxon and two pedophilic taxa. In comparison with the complement taxon, the men in the first pedophilic taxon were non-exclusively pedophilic and had similar rates of sexual recidivism and sexual compulsivity. The men in the second pedophilic taxon were exclusively pedophilic, had more child victims and total victims, sexually re-offended at a higher rate, and were more sexually compulsive. The finding of trichotomous latent structure in pedophilic interest is both consistent and inconsistent with previous taxometric studies and has implications for research, assessment, and treatment of pedophilic interest.

Introduction

Understanding the latent structure of psychological constructs requires empirical investigation (Beauchaine, 2003; Meehl, 1992, 1995a). Latent structure refers to the unobservable nature of a construct that is estimated using observable measurements on psychological tests. Two ways to conceptualize latent structure is as a single distribution or as two or more distinct classes (i.e., dimensional or taxonic; Meehl, 2004). Taxometric analyses are a family of analytic procedures that test whether the latent structure of a construct is best characterized as taxonic or dimensional (Ruscio, Haslam, & Ruscio, 2006; Waller & Meehl, 1998). The use of taxometric analyses has improved the understanding of the latent structure of a wide variety of psychological constructs (e.g., anxiety disorders, eating disorders, personality disorders; Haslam, Holland, & Kuppens, 2012).

Pedophilia, a sexual interest in prepubescent children, is an important construct in understanding, predicting, and preventing sexual offending against children (McPhail et al., 2017; Seto, 2008). Theoretical models predict that while most paraphilias in men are dimensional, pedophilic interest is likely to be taxonic and that men are either pedophilic or non-pedophilic (Hanson, 2010). The prediction for taxonic structure is based on the early onset of pedophilic interest and high stability over the life course (Bailey, Hsu, & Bernhard, 2016; Hanson, 2010; McPhail, 2018; Seto, 2012; Tozdan & Briken, 2015). Gradient models of erotic age interests can be though to conceptualize pedophilic interest as a dimension (Blanchard et al., 2012). However, gradient models can likely accommodate a finding that sexual interest in prepubescent children is taxonic, as most teleiophilic men may experience interest in physically mature adolescents, but little to no sexual interest in prepubescent children (Seto, 2017). Seto makes this hypothesis explicit, suggesting that it would be rare for an individual to have rather contrasting chronophilic attraction. Other theoretical work has conceptualized pedophilic interest as dimensional, suggesting men range from low levels of pedophilic interests to high levels of pedophilic interests (Finkelhor & Araji, 1986).

Without a robust empirical understanding of latent structure, researchers and clinicians must rely on untested assumptions regarding the structure of pedophilic interest to inform research and practice. For instance, past research examining neurological and neurodevelopmental correlates of pedophilic interest assume these interests are taxonic (Cantor & Blanchard, 2012; Cantor et al., 2008, 2015; Dyshniku, Murray, Fazio, Lykins, & Cantor, 2015; Fazio, Dyshniku, Lykins, & Cantor 2017; McPhail & Cantor, 2015). Certain clinical assessment procedures using phallometric testing also model pedophilic interest as dichotomous (i.e., two-ordered taxa; Blanchard, Klassen, Dickey, Kuban, & Blak, 2001; Cantor & McPhail, 2015), while other procedures assess pedophilic interest as a dimension (Marshall, O’Brien, & Marshall, 2009). The Pedophilic Disorder diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) represent an assumption of taxonic, but trichotomous structureFootnote 1 to the disorder (i.e., three-ordered taxa; American Psychiatric Association [APA], 2013). Research examining treatment outcomes assume measurement models of pedophilic interest that are dichotomous (Müller et al., 2014) or dimensional (Becker, Kaplan, & Kavoussi, 1988; Bradford & Pawlak, 1993; Marques, Nelson, West, & Day, 1994; Marshall, 1997; Ricci, Clayton, & Shapiro, 2006). Problems may arise when latent structure does not align with the measurement model used in research, assessment, and treatment due to the use of less than optimal research design, statistical procedures, diagnostic classifications, and treatment expectancies (Ruscio et al., 2006).

Taxometric Research on Pedophilic Interest

Schmidt, Mokros and Banse (2013) conducted a taxometric analysis of pedophilic interest using Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998), viewing time, and self-report measures of pedophilic interest. Their results supported a taxonic structure in pedophilic interest. Schmidt et al. also conducted a latent profile analysis, which found a two-class solution and a three-class solution modeled their data equally well. These findings suggest a dichotomous or trichotomous latent structure may characterize pedophilic interest. Other research examined the latent structure of pedophilic interests using phallometric, self-report, and behavioral measures in a large sample of sexual offenders (Stephens, Leroux, Skilling, Cantor, & Seto, 2017). Those investigators found support for dimensional structure in pedophilic interest. A third taxometric study examined latent structure in a large sample of sexual offenders who underwent phallometric assessment (Mackaronis, Byrne, Strassberg, Marcus, & Solari, 2011a). However, their findings were ambiguous and did not support taxonic or dimensional structure.Footnote 2

The discrepancy in findings may be partially explained by differences in the methods used and data conditions for taxometric analyses. The measures of pedophilic interest used by Schmidt et al. (2013) have an emerging body of empirical literature supporting their validity (Babchishin, Nunes, & Hermann, 2013; Banse, Schmidt, & Clarbour, 2010; Ó Ciardha, Attard-Johnson, & Bindemann, 2018; Schmidt, Babchishin, & Lehmann, 2017). In that study, the measures used were also within the range of data requirements for conducting taxometric analysis, suggesting the data conditions those authors were working with did little to skew the results. Stephens et al. (2017) used a mixture of validated measures of pedophilic interest along with measures with less well-established validity (i.e., the self-report measure used). In addition, one of the measures used by those authors was positively skewed beyond recommended limits, which can influence taxometric results (Waller & Meehl, 1998).

A potential limitation to both studies, however, is the use of measures of sexual interest in children relative to sexual interest in adults. This method of assessing pedophilic interest is well-validated and widely used in research and clinical practice (McPhail et al., 2017). However, within the context of taxometric analysis, using relative measures of pedophilic interest may mix two distinct constructs within the analysis (i.e., pedophilic interest and teleiophilic interest). As a result, the use of relative measures of sexual interest may make it difficult to infer latent structure of sexual interest in children via taxometric analysis. Conceptualizing pedophilic interest as the sexual interest in children, without a comparison with sexual interest in adults, is consistent with relevant theory (e.g., Hanson, 2010; Seto, 2017) and may improve clarity of taxometric results.Footnote 3

An additional explanation for the discrepant results is that the latent structure of pedophilic interest is more complex than the possibilities considered in these studies (i.e., dichotomous vs. dimensional). For instance, pedophilic interest may be trichotomous (i.e., consist of three-ordered classes). Different classes of men may experience low, moderate, and high levels of pedophilic interest, with clear dividing boundaries between these three levels of sexual interest. Perhaps the clearest indication of a trichotomous structure in pedophilic interests available is the exclusive and non-exclusive subtypes (Cohen & Galynker, 2002; Finkelhor & Araji, 1986; Hall & Hall, 2007). Exclusively pedophilic sexual offenders have been found to be distinct in terms of rates of re-offending against children (Beier, 1998) and exclusive pedophilic interest predicts sexual recidivism (Eher, Olver, Heurix, Schilling, & Rettenberger, 2015; Eher, Rettenberger, Matthes, & Schilling, 2010). In contrast, studies treating pedophilia dichotomously failed to establish predictive validity in this diagnostic approach (Eher et al., 2010; Kingston, Firestone, Moulden, & Bradford, 2007; Moulden, Firestone, Kingston, & Bradford, 2009; Wilson, Abracen, Looman, Pichea, & Ferguson, 2011). Within the context of taxometric studies, this hypothesis suggests that within a putative pedophilic taxon, there may be a second taxon distinguishable by high levels of arousal to children. Alternatively, this third class may only be distinguishable by interest in children relative to adults.

Trichotomous structure is important to consider in taxometric research because simulation studies suggest taxometric curves and fit indices are more likely to be ambiguous or provide conflicting results in the presence of a third taxon (McGrath, 2008; Walters, McGrath, & Knight, 2010). Visual examination of the taxometric curves in Schmidt et al. (2013) (A. Mokros, personal communication, March 1, 2017) suggests that the latent structure in their sample may be trichotomous, as opposed to dichotomous. That is, their results may suggest pedophilic interest consists of three-ordered classes. The taxometric curves in Stephens and colleagues’ analysis are more ambiguous. This is not necessarily surprising, as trichotomous structures are known to produce conflicting results across different datasets (McGrath, 2008).

Present Study

Currently, the results of taxometric analyses of pedophilic interest have been equivocal. Thus, the present study aims to extend the understanding of the latent structure of pedophilia in the following ways. First, the present study examines the latent structure of pedophilic interest using phallometric testing. Second, given the divergent findings in previous studies, we have considered the hypothesis that pedophilic interest is neither dimensional nor dichotomous, but trichotomous. Third, given the absence of factor analytic studies examining whether female-oriented and male-oriented pedophilia load onto a single factor or separate factors, the present study examined the latent structure of pedophilic interest in three ways: both sexes combined, female-orientated, and male-orientated. Fourth, the present study conducted taxometric analyses in multiple datasets. Using multiple datasets may increase the confidence when latent structure is stable across samples. Last, using samples with different offense characteristics (e.g., non-sexual offenders, sexual offenders against adults, sexual offenders against children) and more theoretically and statistically justified phallometric indicators, the present study may resolve some of the validity issues present in past taxometric research using phallometric data (Mackaronis et al., 2011a).

Method

Samples

Precollected data from four samples of participants from three separate institutions in Canada were employed. Ethical approval for secondary analysis of these data was provided by the University of Saskatchewan Behavioural Research Ethics Board (certificate #Beh 16-167).

Institut Philippe-Pinel (IPP)

These participants are 632 sexual and non-sexual offenders who underwent phallometric testing at the Institut Philippe-Pinel de Montréal, in Montréal, Quebec, Canada, from 1984 to 2012. The sexual offenders are men who had committed sexual offenses against adults, children, or both. The sample consisted of men who (1) were serving a custodial sentence and participated in treatment at the institution, (2) were under community supervision and followed on an outpatient basis, or (3) were assessed as part of presentence hearings. The sample underwent phallometric assessments using a French translation of the child sexual violence auditory stimuli set (Barsetti, Earls, Lalumière, & Bélanger, 1998; Quinsey & Chaplin, 1988).

Regional Psychiatric Centre (RPC)

These participants are 261 federally incarcerated sexual offenders who underwent phallometric testing at the RPC in Saskatoon, Saskatchewan, Canada. The sexual offenders are men who had committed sexual offenses against adults, children, or both. Phallometric testing was conducted for the purposes of risk and treatment need assessment. The phallometric test used by this lab employed slide-based stimuli (for description of procedure, see Canales, Olver, & Wong, 2009).

Regional Treatment Centre (RTC)

These participants are federally incarcerated sexual and non-sexual offenders who underwent phallometric testing at the RTC in Kingston, Ontario, Canada. The sexual offenders are men who had committed sexual offenses against adults, children, or both. Phallometric testing was conducted for the purposes of risk and treatment need assessment. Three samples of offenders were assessed at this institution. The first sample consisted of 531 offenders who underwent phallometric assessment after 1993 (hereafter referred to as RTC 1). Within the RTC 1 sample, 382 offenders underwent phallometric assessment using audio-based stimuli (RTC 1: Audio) and 377 offenders underwent phallometric assessment using slide-based stimuli (RTC 1: Slide; Looman & Marshall, 2001). Within the RTC 1 dataset, 228 men received both audio and slide assessments. However, given the recommended sample size of 300 to perform taxometric analyses (Ruscio et al., 2006), we included these men in both the audio and slide datasets. The audio stimuli are the English version of the child sexual violence auditory stimuli set (Quinsey & Chaplin, 1988). Because a distinct slide-based stimuli set was used prior to 1993, 805 offenders who underwent phallometric prior to 1993 were used as a standalone sample (RTC 2; Baxter, Marshall, Barbaree, Davidson, & Malcolm, 1984).

Phallometric Measures and Procedure

Phallometric tests measure changes in penile circumference while stimuli depicting different ages, sexes, and sexual activities are presented (Laws, 2009). Sexual interest in a certain age–sex category is indicated by increases in penile tumescence while attending to stimuli reflective of that age and sex. Much of the available literature shows that sexual offenders against children can be differentiated from other groups based on phallometric tests for pedophilic interest (e.g., Blanchard et al., 2001; Cantor & McPhail, 2015; McPhail et al., 2017). Phallometric testing is also a robust predictor of sexual recidivism by sexual offenders against children (Hanson & Morton-Bourgon, 2005; McPhail et al., 2017) and has been shown to have somewhat adequate reliability (test–retest r = .51, 95% confidence interval [CI] .47, .55, k = 6, N = 1265; McPhail & Olver, 2018).

The phallometric tests used either audio-based or slide-based stimuli. The audio-based stimuli involve an assessee listening to sexual interactions between two people, narrated in the first-person. The sexual interactions differ according to the age of the sexual partner (prepubescent child/adult), the sex of the partner (female/male), and the use of force involved in the sexual interaction (non-coerced/coerced) (Barsetti et al., 1998; Quinsey & Chaplin, 1988). In the IPP sample, there are four stimulus trials for which data are available: prepubescent female (non-coerced), prepubescent female (coerced), prepubescent male (non-coerced), and prepubescent male (coerced). In the RTC 1: Audio sample, there are 2 stimulus trials across 4 stimulus categories for which data are available: prepubescent female (non-coerced), prepubescent female (coerced), prepubescent male (non-coerced), and prepubescent male (coerced).Footnote 4 The slide-based stimuli involve projected photographs of a single nude or partially clothed person onto a screen in front of the assessee. The individuals in the slides vary by age, from early childhood to young adulthood, and sex (Baxter et al., 1984). For the slide-based phallometric tests, stimuli trials depicting female and male children aged 5, 8, and 10 were used in the present study. In the RTC 1: Slide, RTC 2, and RPC samples, there are 6 stimulus trials for which data are available for analysis.

Phallometric data for the samples were transformed into percent full erection scores. This data processing method involves dividing the maximum penile change during a stimulus trial by an estimate of full erection (i.e., 30 mm of penile tumescence change; Becker, Stein, Kaplan, & Cunningham-Rathner, 1992b; Howes, 1995; Hunter & Goodwin, 1992) and multiplying the product by 100.

Data Analyses

Three taxometric procedures were used to examine the latent structure of pedophilic interest. In each of the analyses, percent full erection scores during prepubescent child stimulus trials were used as the indicators of pedophilic interests.

Mean Above Minus Below a Cut (MAMBAC)

The MAMBAC procedure is based on the logic that if an indicator separates two latent taxa, then there is an optimal cutting score to distinguish the taxa (Meehl & Yonce, 1994; Ruscio et al., 2006). Using two indicators of pedophilic interest, this procedure uses multiple cutting scores on one indicator to sort cases into two groups. Means scores on the second indicator for the group falling below the cut score are then subtracted from the mean scores for the group falling above the cut score. This subtraction is then repeated across many cut scores on the first indicator and these mean differences on the second indicator are plotted on a graph. Under ideal data conditions, taxonic latent structure results in peaked curves, while dimensional latent structure produces a concave curve. Trichotomous latent structure will result in twin-peaked curves (McGrath, 2008).

Maximum Eigenvalue (MAXEIG)

The MAXEIG procedure requires three or more indicators of pedophilic interest. In this procedure, one variable serves as an input indicator, which is used to sort cases from lowest to highest according to score on the variable, and all other variables are used as output indicators (Meehl & Yonce, 1996; Ruscio et al., 2006; Waller & Meehl, 1998). In MAXEIG analyses, there are separate trials conducted in which each variable is used as the input indicator and the other variables are used as the output indicators. For example, if five variables are available, a MAXEIG analysis will include five trials in which each variable is selected to be the input variable while the other four variables serve as output variables. The input indicators are used to construct multiple subsamples of participants. In the present analyses, the sample was divided into multiple, overlapping subsamples of 100. In each subsample, the MAXEIG procedure computes an eigenvalue to determine the association between the 2 or more output indicators. This eigenvalue is derived from a variance–covariance matrix in which the diagonal is replaced with zeros, leaving only covariances in the matrix, and which represents the shared variance accounted for by the linear combination of indicators (Ruscio et al., 2006; Tabachnick & Fidell, 2013). The first and largest eigenvalue is plotted on a graph. In the presence of a latent taxon, a subsample composed of all or mostly taxon or complement members will result in small eigenvalues because it is anticipated that variables will not covary within the taxon or complement class. Subsamples that are composed of a mixture of complement and taxon members will result in higher eigenvalues since taxon members tend to score high on the output indicators and complement members will tend to score low on the indicators output. Taxonic latent structure will produce peaked curves under ideal data conditions. Dimensional latent structure will produce eigenvalues across subsamples that are relatively consistent because subsamples are mixtures of individuals with varying levels of the trait. Trichotomous latent structure will result in twin-peaked curves (McGrath, 2008).

Latent Mode Factor Analysis (L-Mode)

The L-Mode procedure involves conducting a factor analysis using multiple indicators of pedophilic interest that is constrained to a single factor solution (Ruscio et al., 2006; Waller & Meehl, 1998). The logic of this procedure is that scores on a factor will more validly separate taxon and complement members than scores on single indicators. The factor scores for each participant are plotted in a graph that has factor score on the x-axis and the relative frequency of cases at different factor scores on the y-axis. Taxonic latent structure will produce an L-Mode curve with a bimodal distribution (i.e., one peak represents complement member scores on the latent factor, the other peak represents taxon member scores) and dimensional latent structure will produce an L-Mode curve with a unimodal distribution. Trichotomous latent structure will result in a trimodal L-Mode curve (McGrath, 2008).

Comparison Curve Fit Index (CCFI)

While examining graphical output is a feature of taxometric analysis interpretation, the interpretability of taxometric curves is susceptible to the data conditions of the variables included in the analysis. To ameliorate some of the difficulty in interpreting taxometric curves, a comparison curve fit index is computed. The CCFI is a fit index that measures the similarity between the research data being used in a taxometric analysis (i.e., phallometric scores) and boot-strap simulated dimensional or taxonic comparison data. This method simulates two sets of comparison data that have the same sample distributions, correlations between indicators, and indicator skew as the original research data (Ruscio et al., 2006). One set of comparison data are simulated assuming the underlying latent structure is taxonic, while the other set of comparison data assumes dimensional latent structure. These samples of simulated data undergo the same taxometric procedures as the original research data. The taxometric curves produced by the original research data are compared with the curves produced by the simulated dimensional data and simulated taxonic data, with a root mean square residual (RMSR) being computed for the distance of the research data curve from both simulated curves. A CCFI is computed during each taxometric analysis using the following formula: RMSRdimensional/(RMSRdimensional + RMSRtaxonic). The CCFI quantifies whether the research data is more similar to dimensional or taxonic latent structure, represented by the simulated data. CCFIs range from 0 to 1, with values from 0 to 0.399 indicating better fit for dimensional structure, values from 0.600 to 1 indicating better fit for taxonic structure, and values around .50 indicating ambiguous fit (Ruscio et al., 2006). All taxometric analyses were performed using the R taxometric program by Ruscio (2014).

Treatment of Phallometric Data

Phallometric data represent percent full erection, which quantifies, from 0 to 100, the maximum percent of full erection a man exhibited during each phallometric stimulus trial. Each stimulus trial was entered as an indicator into the taxometric analyses. For instance, there were six different stimulus trials assessing pedophilic interest in the RTC 2 sample and the six trials were treated as separate indicators in the analyses. Taxometric analyses were run in three ways: (1) using all stimulus trials, (2) restricted to stimulus trials depicting female children, and (3) restricted to stimulus trials depicting male children. As there is currently little understanding whether phallometric stimuli load on separate age–sex factors or onto age factors, there appears to be little justification for not assessing the latent structure of the two sexes separately.

Analytic Plan

As the validity of variables used as indicators in taxometric analysis is a key consideration in interpretation of results, validity indices for the phallometric indicators are reported. In taxometric analysis, calculation of validity indices requires participants to be identified as putative taxon and complement class members. In order to establish putative taxon and complement class membership, the base rate classification technique was used because this method appears more accurate (Ruscio & Kaczetow, 2009; Ruscio, Ruscio, & Meron, 2007). Past taxometric studies of pedophilic interest have produced somewhat divergent taxon base rates (14.2% in Schmidt et al., 2013; 28.7% in Stephens et al., 2017), not to mention the differing conclusions regarding latent structure these studies resulted in. Other research estimates that the proportion of sex offenders classified as pedophilic is around 40% (Blanchard et al., 2001). This amount of ambiguity in the estimates of pedophilic base rates appears to have the potential to skew taxometric results. As a result, we chose not to use a specific base rate for a putative pedophilic taxon due to these divergent findings. Under these circumstances, the taxometric software program uses the taxon base rate estimated by the analyses to classify individuals into the putative taxon and complement class. The average estimated taxon base rate in the present study was between 18.6 and 25.4% across datasets, which coheres closely with previous taxometric studies.

The validity indices reported here are skewness of phallometric indicators, the average correlation among phallometric indicators in the full sample, the average correlation among phallometric indicators in the putative taxon class and putative complement class (i.e., nuisance covariance), and the average standardized mean difference (i.e., Cohen’s d) on the phallometric indicators between the putative taxon and complement class. As a general set of heuristics, for taxometric analyses to be optimally valid skew should be < 1, correlations in the taxon and complement should be lower than r = .30, and between class validity should be greater than d = 1.25 (Ruscio et al., 2006). To aid interpretation of the results, when indicator validity estimates are not in these specified ranges, taxometric analyses result in higher rates of ambiguous findings (i.e., CCFIs between 0.40 and 0.60; Ruscio & Kaczetow, 2009). However, when indicator validity estimates are not within these specified ranges, there is a negligible increase in inaccurate results (i.e., a taxonic result when data are actually dimensional and vice versa; Ruscio & Kaczetow, 2009). These validity estimates will be reported in each dataset separately and the combined male and female stimuli analyses and the analyses for female stimuli and male stimuli separately.

Next, the CCFIs resulting from the MAMBAC, MAXEIG, and L-Mode analyses will be reported for each of the datasets separately. The CCFI values are reported for the combined male and female stimuli analyses and the analyses examining the female stimuli and male stimuli separately. The graphical output from the three taxometric procedures is also provided. Given the large number of analyses run across datasets, most of the graphical output is presented as supplemental material and the interested reader is encouraged to visually inspect these taxometric graphs.

Because we considered latent structure beyond dimensional and dichotomous, we relied on simulation studies that provide taxometric graphical output for trichotomous structure to identify trichotomous structure in the graphical output in the present study (McGrath, 2008; Walter et al., 2010). Unfortunately, CCFIs do not provide an indication of whether trichotomous structure is present in the data. If trichotomous latent structure is indicated in the graphical output of the taxometric analyses, we pursued the possibility of trichotomous structure with post hoc taxometric analyses using the procedure described in Ruscio and Ruscio (2004). To test for trichotomous structure, after the first set of taxometric analyses are run, those participants identified as belonging to the pedophilic taxon are used in a second set of taxometric analyses and the participants identified as belonging to the complement are removed from further analyses. This second set of analyses proceeds in a similar manner as the first: using taxon members, MAMBAC, MAXEIG, and L-Mode analyses are conducted. If these results indicate taxonic latent structure (i.e., graphical output look taxonic and CCFIs are > 0.60), this suggests the presence of a third class. If these results indicate ambiguous or dimensional latent structure, this suggests there is only the taxon and the complement identified in the first step of analysis (i.e., dichotomous structure). If dichotomous or trichotomous latent structure is indicated by these analyses, the identified taxa will be characterized using variables available in the datasets.

Results

Descriptive statistics and validity indices for the variables in the datasets are presented in Table 1. Given the elevated level of positive skew present in the data, scores on all indicators above the 99th percentile were winsorized (Tabachnick & Fidell, 2013; Wilcox, 2005). Because elevated skew (i.e., skew > 1.00) and nuisance covariance (i.e., r in putative taxon or complement > .30) represent violations of taxometric assumptions, the following procedures were used to increase confidence and interpretability of the results: (1) a dual threshold for interpreting CCFIs (i.e., CCFIs < .400 are interpreted as indicating dimensional structure, CCFIs > .600 are interpreted as indicating taxonic structure, and CCFIs between .400 and .600 are interpreted as ambiguous) and (2) imposing a multiple hurdles approach to interpreting taxometric results. A multiple hurdles approach involves interpreting taxometric results when the majority of CCFIs (i.e., 2 out of the 3 CCFIs produced by the MAMBAC, MAXEIG, and L-Mode procedures) or the mean of the three CCFIs exceed the dual threshold (Ahmed, 2010; Ruscio & Kaczetow, 2009; Ruscio et al., 2007). We recommend readers interpret the present findings using a conservative approach (i.e., using dual thresholds and a multiple methods approach).

Table 1 Descriptive statistics and validity estimates for indicators

CCFIs for the taxometric analyses of pedophilic interest, female-oriented pedophilic interest, and male-oriented pedophilic interest are presented in Table 2. The results were most consistent with taxonic latent structure. Pedophilic interest was indicated to be taxonic across 66.7% of the CCFIs (i.e., CCFI ≥ .60; 10/15 CCFIs), 20% were ambiguous (i.e., CCFI between .40 and .60; 3/15 CCFIs), and 13.3% supported dimensional structure (i.e., CCFI ≤ .40; 2/15 CCFIs). When using the majority method for identifying latent structure, 80% of the analyses indicated taxonic structure to pedophilic interest. For the averaged CCFIs, 60% supported taxonic structure and 40% were ambiguous. The weighted mean CCFI for the five datasets was .610 (N = 2482). The taxometric curves for the RTC 1: Audio dataset are presented in Fig. 1. We report the curves for this dataset because these illustrate the potential for trichotomous structure most clearly. While a similar pattern is also detectable for other analyses, it is less clear (see Figures S1 to S12).

Table 2 Comparative Curve Fit Index across the full samples
Fig. 1
figure1

L-Mode, MAMBAC, and MAXEIG graphs comparing research data to simulated categorical (left) and dimensional (right) data. The dark line represents the average data curve, the gray line represents the middle 50% of the simulated data and the light lines show the minimum and maximum values of the simulated data. Curves taken from the analysis of indicators of interest in both sexes in the RTC 1: Audio dataset

For female-oriented pedophilic interest, taxonic structure was indicated across 50% of the CCFIs (i.e., 6/12 CCFIs), while 41.6% were ambiguous (i.e., 5/12 CCFIs) and 8.3% supported dimensional structure (i.e., 1/12 CCFIs). When using the majority method for identifying latent structure, 50% of the datasets indicated taxonic structure to pedophilic interest. For the averaged CCFIs, 25% of the datasets supported taxonic structure and 75% were ambiguous. The weighted mean CCFI for the four datasets was .580 (N = 1850). For male-oriented pedophilic interest, taxonic structure was indicated across 75% of the CCFIs (i.e., 9/12 CCFIs), while 8.3% were ambiguous (i.e., 1/12 CCFIs) and 16.6% supported dimensional structure (i.e., 2/12 CCFIs). When using the majority method for identifying latent structure, 100% of the datasets indicated taxonic structure to pedophilic interest. For the averaged CCFIs, 50% of the datasets supported taxonic structure and 50% were ambiguous. The weighted mean CCFI for the four datasets was .603 (N = 1850). Notably, the IPP dataset did not contain enough phallometric trials to allow for analyses of sex orientation.

Post hoc Analysis

Visual inspection of the taxometric curves suggests trichotomous latent structure may also be present (see Fig. 1; McGrath, 2008; Walters et al., 2010). Most strikingly, multiple L-Mode curves have trimodal distributions.Footnote 5 Following the procedure to assess for trichotomous latent structure outlined by Ruscio and Ruscio (2004), we conducted a second set of taxometric analyses using men classified as taxon members in the first round of analysis. To identify taxon members in each sample, averaged taxon base rates produced by the three analyses were used, which were: IPP = 22.6%, RTC 1: Audio = 23.4%, RPC = 20.6%, RTC 1: Slide = 18.6%, RTC 2 = 25.4%. For example, in the IPP sample, the 22.6% men in the sample that showed the highest level of responding to child stimuli were classified as taxon members. Because sample sizes were not large enough to conduct the taxometric analyses in each dataset separately, the second step of analyses combined the RTC 2, RPC, and RTC1: Slides datasets (N = 332) and the RTC 1: Audio and IPP datasets (N = 228)Footnote 6.

Table 3 provides descriptive and validity estimates for the second step of taxometric analysis. The majority of validity estimates are within the expected ranges, suggesting skew and nuisance covariance was less problematic compared with the first step of analysis. CCFIs from the second step of analyses for pedophilic interest, female-oriented pedophilic interest, and male-oriented pedophilic interest are presented in Table 4. Pedophilic interest was indicated to be taxonic across 83.3% of the CCFIs (i.e., 5/6 CCFIs) while 16.7% of the CCFIs were ambiguous (i.e., 1/6 CCFIs) and 0% supported dimensional structure. When using the majority method for identifying latent structure, 100% of the analyses indicated taxonic structure to pedophilic interest; for the averaged CCFIs, 100% supported taxonic structure. The weighted mean CCFI for the five datasets was .610 (N = 2482). For female-oriented pedophilic interest, taxonic structure was indicated across 100% of the CCFIs (i.e., 3/3 CCFIs) and the averaged CCFIs supported taxonic structure. For male-oriented pedophilic interest, taxonic structure was indicated across 100% of the CCFIs (i.e., 3/3 CCFIs) and the average CCFI supported taxonic structure. Notably, the combined audio datasets did not contain enough phallometric trials to allow for analyses of sex orientation. Figure 2 presents the taxometric curves for the combined audio datasets. The curves from the audio datasets most clearly suggest taxonic structure. While the curves from the other datasets suggest a similar pattern, it is less clear (see Figures S13 to S20). Taken together, the taxometric analyses indicate that the pedophilia taxon is itself taxonic, suggesting a trichotomous latent structure to pedophilic interest.Footnote 7

Table 3 Descriptive statistics and validity estimates for indicators in the taxon
Table 4 Comparative curve fit indices in the pedophilia taxon
Fig. 2
figure2

L-Mode, MAMBAC, and MAXEIG graphs comparing research data to simulated categorical (left) and dimensional (right) data. The dark line represents the average data curve, the gray line represents the middle 50% of the simulated data and the light lines show the minimum and maximum values of the simulated data. Curves taken from the analysis of indicators of interest in both sexes in the combined audio dataset

In order to describe the three taxa, phallometric indices and demographic and offense characteristics were examined (see Table 5). It is important to note that these analyses proceeded in a post hoc manner, while not every variable available in the datasets is analyzed here, variables that are descriptively important, of theoretical interest, available across datasets, or available for every individual in a dataset were pursued. When the same variable was available across multiple datasets, a weighted mean and pooled standard deviation were computed.

Table 5 Phallometric, demographic, and offense characteristics in the three taxa

When compared with the other taxa, men in the second pedophilic taxon were younger, had a greater number of total and child victims, showed greater arousal to children of both sexes and female children by relative and absolute phallometric indices, and showed greater arousal to male children by an absolute phallometric index. A surprising result was that the second pedophilic taxon showed greater absolute arousal to adults when compared with the two other taxa (ds= 0.66 and 1.30) and the first pedophilic taxon showed somewhat higher arousal to adults than the non-pedophilic taxon (d = 0.49). When compared with the non-pedophilic taxon, the first pedophilic taxon was younger and showed greater arousal to children by absolute and relative phallometric indices. To characterize the relative arousal patterns of the three taxa, the non-pedophilic taxon was teleiophilic, the first pedophilic taxon was non-exclusively pedophilic, and the second taxon was exclusively pedophilic. These descriptive terms will be used for the remainder of the article.

The increasing level of arousal to children and adults displayed by the two pedophilic taxa presents an interpretive puzzle and may be indicative of problems of sexual compulsivity/hypersexuality in the pedophilic taxa. Previous research has suggested a link between hypersexuality and paraphilic interests generally (Bouchard, Dawson, & Lalumière, 2017; Cantor et al., 2013; Davis, 2017; Dyer & Olver, 2016; Kafka & Hennen, 2003; Långström & Hanson, 2006; Långström & Seto, 2006; Långström & Zucker, 2005; Sutton, Sratton, Pytyck, Kolla, & Cantor, 2015), and pedophilic interest specifically (Davis, 2017; Klein, Schmidt, Turner, & Briken, 2015; Sutton et al., 2015). The RPC dataset contained a measure of sexual compulsivity (i.e., the Sexual Compulsivity item of the Violence Risk Scale-Sexual Offender versionFootnote 8; Olver, Wong, Nicholaichuk, & Gordon, 2007), and sexual recidivism rates, which can serve as a proxy measure of hypersexuality. These variables, in tandem, provide further opportunity to examine the level of hypersexuality in the three taxa. There was a significant difference in sexual compulsivity among the three taxa (Mantel–Haenszel χ2[1] = 17.83, p < .001). The odds of sexual compulsivity problems in the exclusively pedophilic taxon was greater when compared with the teleiophilic (odds ratio [OR] = 13.34) and non-exclusively pedophilic taxa (OR 4.89). The sexual recidivism rates did not differ between the non-pedophilic and non-exclusively pedophilic taxa (OR 0.93); however, the odds of sexual compulsivity problems did differ between these two taxa (OR 2.73). The exclusively pedophilic taxon had a sexual recidivism rate higher than the non-pedophilic (OR 2.44) and non-exclusively pedophilic taxa (OR 2.64). There was not a significant difference in sexual recidivism rate among the three taxa (χ2[2] = 2.86, p = .24, φ = .11); however, there was a significant difference in sexual recidivism rates in the exclusively pedophilic taxon compared with non-pedophilic taxon (Fisher’s exact p = .008).

Discussion

The present research used phallometric test data from multiple samples of mixed sexual and non-sexual offenders to examine latent structure in pedophilic interest. The first step of taxometric analyses indicated that, across datasets, pedophilic interest was taxonic. Given the shape of the taxometric curves, we further considered that pedophilic interests may be trichotomous in structure, rather than dichotomous. In particular, the curves produced by L-Mode analyses suggested a third mode to the right of the larger complement class mode, which is consistent with previous simulation studies examining the effects of trichotomous structure on taxometric curves (see Fig. 6 in McGrath, 2008). When restricting taxometric analyses to putative taxon members, CCFI values and taxometric curve shapes supported taxonic structure. In particular, the L-Mode curves were more clearly bimodal, which is anticipated if two taxa are present in the data. These results indicate that the structure of pedophilic interest is trichotomous. This finding held when female-oriented and male-oriented pedophilic interest were examined separately.

To further understand the finding of a three-class structure in pedophilic interest, we conducted a series of post hoc analyses to characterize the three taxa. The taxa were characterized by different levels of arousal to children, arousal to children relative to adults, and number of child victims. One main interpretation of these results is that the three taxa appear teleiophilic, non-exclusively pedophilic, and exclusively pedophilic. The exclusively pedophilic taxon also displayed higher arousal to adults compared with the other taxa. To us, this suggested the presence of problems with hypersexuality/sexual compulsivity in the exclusively pedophilic taxon. We found support for this post hoc hypothesis.

Trichotomous structure in pedophilic interest has implications for interpreting the results of previous taxometric studies. The findings reported by Schmidt et al. (2013) support taxonic structure. However, the Bayesian classification probabilities and the L-Mode curve from that study (A. Mokros, personal communication, March 1, 2017) are relatively similar to the curves in the present research. The latent profile analysis conducted by those authors showed a two-class structure was the most parsimonious model; however, a three-class model provided fit statistics equivalent to the two-class model. These two findings suggest that trichotomous latent structure may have been present in their data. Those investigators did not consider the trichotomous hypothesis, but they would not have had a large enough sample size to test for the presence of a second pedophilic taxon. Given these previous findings by Schmidt et al., a trichotomous result in our data is not completely unexpected.

The present findings are at odds with the dimensional result reported by Stephens et al. (2017). Those authors had a large sample and used a more diverse set of measures of pedophilic interest than what was available in the present research. These aspects of their study speak strongly to the potential correctness of their result. One main explanation for the divergent results is that the measures used by Stephens and colleagues were relative measures that compared pedophilic interest to teleiophilic interest. At this point, it is unclear what effect considering pedophilic interest and teleiophilic interest simultaneously has on taxometric results. However, Stephens et al. did not appear to consider an alternative beyond dimensional or dichotomous latent structure, which may have limited their pursuit of novel findings present in those data. The shape of the curves presented in their study is relatively ambiguous and are not suggestive of trichotomous structure. Simulation studies suggest that CCFIs and taxometric curves become more ambiguous, or even misleading, when a third class is present in the data and provides a possible alternative explanation for the conflicting findings across studies (McGrath, 2008; Walters et al., 2010). Despite three taxometric studies conducted to date, there remains uncertainty which latent structure best characterizes pedophilic interest. This state of affairs demands further studies examining the latent structure of pedophilic interest.

Paraphilias have been conceptualized as involving three inter-related aspects: sexual self-regulation (e.g., ability to manage one’s sexual behavior), atypical sexual interests (e.g., interest in prepubescent children), and intensity of sexual interest (Hanson, 2010). To varying degrees, the present research provides support for this conceptualization. The main results indicate that the taxa were distinguishable in terms of the level of atypical sexual interest. In addition, the taxa displayed differing rates of sexual compulsivity and levels of overall arousal, suggesting that sexual self-regulation and intensity of sexuality were present to differing degrees in the taxa. To a lesser degree, the rate of prior offending increased across the taxa. Taken together, we found three taxa that were separable in terms of pedophilic interest and this distinction was associated with intensity of sexual interests and sexual self-regulation problems. These findings cohere well with previous empirical literature suggesting an association between hypersexuality and paraphilic interests, including pedophilia (see Kafka [2010] for a review; Bouchard et al., 2017; Cantor et al., 2013; Davis, 2017; Dyer & Olver, 2016; Klein et al., 2015; Sutton et al., 2015; Walton, Cantor, Bhullar, & Lykins, 2017). Given that the present research used phallometric measures of sexual interest, the finding that arousal to adult stimuli increased across the three taxa may be partially explained by the increasing rate of sexual compulsivity in the taxa. This, however, would not account for the differences in differing levels of relative interest in children found in the taxa.

Examining the potential interaction between these aspects of sexuality is an interesting and important avenue for future research. For instance, disentangling whether pedophilic interest is associated with high sex drive (e.g., frequency of sex acts, sexual preoccupation, impersonal sexual behavior; Carvalho, Stulhofer, Vieira, & Jurin, 2015; Knight & Graham, 2017; Stulhofer, Jurin, & Briken, 2016), problematic sexuality (e.g., use sex to cope with negative affective states, compulsive sexual behavior; Knight & Graham, 2017), or both may improve our understanding of this relationship and inform intervention efforts with these men.

When arousal to children relative to adults was examined in the three taxa, the two pedophilic taxa were non-exclusive and exclusive in their arousal to prepubescent children. The distinction between exclusive and non-exclusive pedophilic interest is an important consideration in clinical forensic practice. Research examining rates of sexual recidivism by pedophilic and non-pedophilic sexual offenders found that there is little difference between these two groups (Eher et al., 2010; Kingston et al., 2007; Moulden et al., 2009; Stephens et al., 2017; Wilson et al., 2011). In contrast, exclusivity of pedophilic interest has a strong relationship with sexual recidivism (Beier, 1998; Eher et al., 2010) and has been found to predict sexual recidivism over and above well-validated measures of sexual recidivism risk (Eher et al., 2015). For instance, Eher et al. (2015) found exclusively pedophilic sexual offenders to have a sexual recidivism rate approximately five times that of non-exclusively pedophilic sexual offenders (21.2 vs. 3.9%). In a non-clinical sample of pedophilic men, exclusivity of pedophilic interest was associated with having committed a sexual offense (Bailey, Bernhard, & Hsu, 2016).

In the present research, the two pedophilic taxa displayed different rates of sexual recidivism (50 vs. 27.5%) and the non-exclusively pedophilic taxon had a sexual recidivism rate equivalent to the teleiophilic taxon (27.5 vs. 29.1%). The elevated rate of sexual recidivism in the exclusively pedophilic taxon may also be partially explained by the finding that these men had co-occurring risk factors for acting on their sexual interest: high levels of pedophilic interest, exclusive pedophilic interests, and sexual compulsivity. Future research may disentangle the influences of intensity of sexual interest, sexual compulsivity, and exclusivity of pedophilic interest when examining associations between pedophilic interest, sexual behavior, and other constructs.

Further implications of a trichotomous structure in pedophilic interest remain to be examined. While theoretically interesting, research in multiple domains will help elaborate whether the two pedophilic taxa are meaningfully different in terms of developmental and neurobiological correlates, sexual behavior, co-occurring mental health issues, and treatment response. For instance, sexual offenders in the two pedophilic taxa may be followed in longitudinal research to establish base rates of recidivism and the environmental, dynamic risk, and personality factors that are related with sexual recidivism and desistance in each taxon. For research evaluating the effectiveness of treatment, membership in one of the pedophilic taxa may moderate treatment response and using a trichotomous classification may help identify those who respond differentially to treatment (Beauchaine, 2003).

Limitations

The samples and measures used in the current study limit our ability to generalize the results. All samples included were comprised of men who were involved in the criminal justice system, with the majority of men being incarcerated. Whether a trichotomous structure will be replicated in a sample more representative of the male population remains to be seen.

The present research relied on phallometric data, limiting our analyses to a single method of measuring pedophilic interest, whereas more conceptually distinct measures increase construct coverage (Ruscio & Ruscio, 2004). A separate issue that likely resulted from using conceptually similar indicators of pedophilic interest was the elevated level of nuisance covariance present in these data (we discuss this limitation in more detail below). While examining latent structure across multiple datasets improves confidence in the findings, trichotomous latent structure in pedophilic interest awaits replication, both with non-clinical samples and with measures other than, or in addition to, phallometric testing.

The data conditions in the samples were not ideal for taxometric analysis. The moderate level of positive skew (i.e., skew > 1.00) and elevated level of nuisance covariance (i.e., r in taxon or complement > .30) present in these data likely affected the findings. Simulation studies do allow us to identify, with some confidence, the effects. Under data conditions like those in the present study, curves from the taxometric procedures become increasingly distorted and ambiguous (Meehl, 1995b; Ruscio et al., 2006). Specifically, the peaks in taxonic MAMBAC and MAXEIG curves shift to the right when nuisance covariance in the taxon class is higher than in the complement class, while the taxon mode in L-Mode curves becomes more ambiguous (Ahmed, 2010; Meehl & Yonce, 1994; Ruscio et al., 2006). MAXEIG appears to be the analysis most affected by high nuisance covariance (Ahmed, 2010), which appears to be true in our results. Positive skew imparts a similar effect on taxonic curves, flattening MAMBAC curve peaks and reducing differentiation of a right mode in L-Mode curves when a taxon is present in the data (Ahmed, 2010; Meehl & Yonce, 1994; Ruscio et al., 2006). The curves in Fig. 1 and supplemental material tend to follow this pattern, with most taxonic peaks shifted to the right of the graphs. Under these conditions, taxometric curves derived from taxonic data are more likely to be judged as dimensional (Ahmed, 2010).

In data conditions similar to those found in the present study, CCFIs tend to be robust to violations of multiple statistical assumptions. For instance, CCFIs continue to be relatively accurate when skew raises above 1 (Ahmed, 2010; Ruscio et al., 2007). However, as nuisance covariance exceeds r = .30, the rate of ambiguous CCFIs increases, while the rate of inaccurate CCFI values (i.e., a CCFI value that indicates dimensional structure when the research data are taxonic and vice versa) remains low and relatively stable (Ruscio, Walters, Marcus, & Kaczetow, 2010). Figure 6 in Ruscio et al.’s (2010) simulation study shows the rate of correct, ambiguous, and incorrect CCFIs as the level of nuisance covariance in a dataset increases. Applying the rates found in Ruscio et al. (2010) to the level of nuisance covariance found in the present data, we would anticipate ~ 15–30% of the CCFIs to be ambiguous, ~ 2–18% to support dimensional structure, and ~ 62–90% to support taxonic structure, assuming taxonic structure is present in the data. In our first step of analysis, these are approximately the rates we found. In addition to this, the data conditions in the second step of analysis were generally improved and the CCFIs and curves were less ambiguous.

Some steps can be taken to limit the effects of skew and nuisance covariance on taxometric interpretation. A method to lessen the problems associated with skew (i.e., elevated Type I error rates) is to set a higher threshold for interpreting CCFIs (Ahmed, 2010; Ruscio, 2007; Ruscio et al., 2010). In the present study, we selected the most conservative threshold for interpreting CCFIs (i.e., CCFI > .600 indicate taxonic structure and CCFI < .400 indicated dimensional structure; relying on the majority rule and averaged CCFIs for interpretation) in order to strengthen the confidence in the findings under these data conditions.

Conclusions

Pedophilic interest was found to be trichotomous in structure. The implication of this finding is that a majority of men do not experience pedophilic interests, while a minority of men do. Within this group of men who experience pedophilic interest, our findings suggest some men are non-exclusively pedophilic and some men are exclusively pedophilic. The approach used in the present study is also instructive regarding interesting and important alternatives to consider when conducting taxometric analyses of human sexuality constructs. Considering trichotomous structure, or even more complex structures (see Borsboom et al., 2016), can provide a more accurate understanding of latent structure, and even has the potential to meaningfully alter the interpretation of taxometric results. Perhaps this lesson is summed up best by quoting Meehl (2004, p. 43), “No statistic is self-interpreting.” In the context of taxometric analysis, we suggest understanding latent structure goes beyond simple interpretation of CCFIs.

Notes

  1. 1.

    Using DSM-5 criteria, clinicians make a first categorical decision, whether a client has Pedophilic Disorder or does not have the disorder. Using the exclusivity specifier, clinicians make a second decision, whether a client diagnosed with Pedophilic Disorder is exclusively or non-exclusively attracted to children. This results in three classes of individuals, in terms of presence and intensity of pedophilic interest: teleiophilic, non-exclusively pedophilic, and exclusively pedophilic individuals.

  2. 2.

    Mackaronis, Strassberg, and Marcus (2011b) conducted a taxometric analysis using subscales from the Multiphasic Sex Inventory-2 (MSI-2; Nichols & Molinder, 1984), claiming to have assessed latent structure in pedophilic interest. However, the sexual obsessions and cognitive distortions subscales of the MSI-2 used in this study do not assess pedophilic interest. Given the choice of measure, we do not consider that study as having examined the latent structure of pedophilic interest.

  3. 3.

    Importantly, taxometric analysis appears to answer the simpler question, “Is pedophilic interest dimensional or taxonic?” Most taxometric analyses we are familiar with pose this form of question (e.g., is psychopathy dimensional or taxonic?) and not a more complex form of the question, such as, “Is the bipolar construct, sexual interest in children relative to sexual interest in adults, dimensional or taxonic?” To continue with a psychopathy analogy, this more complex question would be analogous to asking, “Is the bipolar construct, psychopathy relative to being a saint, dimensional or taxonic?” Further to this, theoretical work has suggested that the strength of pedophilic interest should be separated from the exclusivity of pedophilic interest in order to understand the construct (Finkelhor & Araji, 1986). Diagnostically, strength of interest in children is the main consideration (APA, 2013); however, there is an evidence base suggesting relative interest in children is important for specific validity purposes (Blanchard et al., 2009). Based on these considerations, we have intentionally chosen to define pedophilia as the sexual interest in prepubescent children, without considering concomitant interest in adults. This will have ramifications for our methodology and the interpretation of the results.

  4. 4.

    We ran the taxometric analyses with and without the coerced stimuli removed from the audio stimuli datasets. The results did not change in a meaningful way and in order to retain a larger number of indicators, we report results including the coercive stimuli.

  5. 5.

    As a secondary check, MAXEIG Bayesian classification probabilities were also produced and visually inspected. A number of the Bayesian classification probabilities indicated a subset of the samples had a moderate chance of being classified to the taxon, which is consistent with trichotomous latent structure (McGrath, 2008).

  6. 6.

    We combined these datasets given the similarity of the phallometric stimuli used. The RTC 1: Audio and IPP datasets are based on a phallometric procedure using the same auditory stimuli (Quinsey & Chaplin, 1988); however, the IPP stimuli were translated to French (Barsetti et al., 1998). The RPC, RTC 1: Slide, and RTC 2 datasets all contain phallometric data using slide stimuli that are similar in terms of the age rages of persons depicted in the slides. We additionally ran the second step of analyses using only the RTC 2 dataset (n = 204). These analyses were conducted to protect against differences in the results that may have been caused by combining datasets. Conducting taxometric analyses using taxon members in the RTC 2 dataset resulted in CCFIs ranging from .669 to .879, which is consistent with the findings reported in Table 4. All validity estimates for the RTC 2 dataset were within expected limits.

  7. 7.

    Misclassifying complement members as taxon members in the second step of analysis inflates the risk of artificially identifying taxonic structure within the taxon identified in the first step of analysis (Ruscio & Ruscio, 2004). For this reason, in the second step of analysis, all members of the complement class should be excluded from analyses. To protect against artificially identifying taxonic structure within the taxon due to inclusion of complement class members, we reduced the taxon base rate because overestimating the taxon base rate risks complement members being falsely classified as taxon members. Running the second step of analysis with reduced taxon base rates did not change the CCFI results in a meaningful way and are not reported.

  8. 8.

    The Sexual Compulsivity item of the VRS-SO is a 4-point scale, with scores ranging from 0 to 3. We conducted group comparisons on the Sexual Compulsivity item in two ways. For the Mantel–Haenszel χ2 test, the Sexual Compulsivity item score of 0–3 was used. In Table 5, the Sexual Compulsivity item was coded as present (scores of 2 and 3) or absent (scores of 0 and 1) and odds ratios were computed. On the VRS-SO, items rated as 2 or 3 are indicative of problem areas and are used in clinical practice to identify treatment targets for sexual offenders (Olver et al., 2007).

References

  1. Ahmed, A. O. (2010). Differentiating classes from dimensions under unfavorable data conditions: Monte Carlo comparisons of taxometric and latent variable mixture models, Unpublished doctoral dissertation, University of Southern Mississippi.

  2. American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). Arlington, VA: American Psychiatric Press.

    Google Scholar 

  3. Babchishin, K. M., Nunes, K. L., & Hermann, C. A. (2013). The validity of Implicit Association Test (IAT) measures of sexual attraction to children: A meta-analysis. Archives of Sexual Behavior, 42, 487–499. https://doi.org/10.1007/s10508-012-0022-8.

    Article  PubMed  Google Scholar 

  4. Bailey, J. M., Bernhard, P. A., & Hsu, K. J. (2016a). An Internet study of men sexually attracted to children: Correlates of sexual offending against children. Journal of Abnormal Psychology, 125, 989–1000. https://doi.org/10.1037/abn0000213.

    Article  PubMed  Google Scholar 

  5. Bailey, J. M., Hsu, K. J., & Bernhard, P. A. (2016b). An Internet study of men sexually attracted to children: Sexual attraction patterns. Journal of Abnormal Psychology, 125, 976–988. https://doi.org/10.1037/abn0000213.

    Article  PubMed  Google Scholar 

  6. Banse, R., Schmidt, A. F., & Clarbour, J. (2010). Indirect measures of sexual interest in child sex offenders: A multimethod approach. Criminal Justice and Behavior, 37, 319–335. https://doi.org/10.1177/0093854809357598.

    Article  Google Scholar 

  7. Barsetti, I., Earls, C. M., Lalumière, M. L., & Bélanger, N. (1998). The differentiation of intrafamilial and extrafamilial heterosexual child molesters. Journal of Interpersonal Violence, 13, 275–286. https://doi.org/10.1177/088626098013002007.

    Article  Google Scholar 

  8. Baxter, D. J., Marshall, W. L., Barbaree, H. E., Davidson, P. R., & Malcolm, P. B. (1984). Deviant sexual behavior: Differentiating sex offenders by criminal and personal history, psychometric measures, and sexual responses. Criminal Justice and Behavior, 11, 477–501. https://doi.org/10.1177/0093854884011004007.

    Article  Google Scholar 

  9. Beauchaine, T. P. (2003). Taxometrics and developmental psychopathology. Development and Psychopathology, 15, 501–527. https://doi.org/10.1017/S0954579403000270.

    Article  PubMed  Google Scholar 

  10. Becker, J. V., Kaplan, M. S., & Kavoussi, R. (1988). Measuring the effectiveness of treatment for the aggressive adolescent sexual offender. Annals of the New York Academy of Sciences, 528, 215–222. https://doi.org/10.1111/j.1749-6632.1988.tb50865.x.

    Article  PubMed  Google Scholar 

  11. Becker, J. V., Stein, R. M., Kaplan, M. S., & Cunningham-Rathner, J. (1992). Erection response characteristics of adolescent sex offenders. Annals of Sex Research, 5, 81–86. https://doi.org/10.1177/107906329200500202.

    Article  Google Scholar 

  12. Beier, K. M. (1998). Differential typology and prognosis for dissexual behavior—A follow-up study of previously expert-appraised child molesters. International Journal of Legal Medicine, 111, 133–141. https://doi.org/10.1007/s004140050133.

    Article  PubMed  Google Scholar 

  13. Blanchard, R., Klassen, P., Dickey, R., Kuban, M. E., & Blak, T. (2001). Sensitivity and specificity of the phallometric test for pedophilia in nonadmitting sex offenders. Psychological Assessment, 13, 118–126. https://doi.org/10.1037/1040-3590.13.1.118.

    Article  PubMed  Google Scholar 

  14. Blanchard, R., Kuban, M. E., Blak, T., Cantor, J. M., Klassen, P. E., & Dickey, R. (2009). Absolute versus relative ascertainment of pedophilia in men. Sexual Abuse: A Journal of Research and Treatment, 21, 431–441. https://doi.org/10.1177/1079063209347906.

    Article  Google Scholar 

  15. Blanchard, R., Kuban, M. E., Blak, T., Klassen, P. E., Dickey, R., & Cantor, J. M. (2012). Sexual attraction to others: A comparison of two models of alloerotic responding in men. Archives of Sexual Behavior, 41, 13–29. https://doi.org/10.1007/s10508-010-9675-3.

    Article  PubMed  Google Scholar 

  16. Borsboom, D., Rhemtulla, M., Cramer, A. O. J., van der Maas, H. L. J., Scheffer, M., & Dolan, C. V. (2016). Kinds versus continua: A review of psychometric approaches to uncover the structure of psychiatric constructs. Psychological Medicine, 46, 1567–1579. https://doi.org/10.1017/S0033291715001944.

    Article  PubMed  Google Scholar 

  17. Bouchard, K. N., Dawson, S. J., & Lalumière, M. L. (2017). The effects of sex drive and paraphilic interests on paraphilic behaviours in a nonclinical sample of men and women. Canadian Journal of Human Sexuality, 26, 97–111. https://doi.org/10.3138/cjhs.262-a8.

    Article  Google Scholar 

  18. Bradford, J. M., & Pawlak, A. (1993). Effects of cyproterone acetate on sexual arousal patterns of pedophiles. Archives of Sexual Behavior, 22, 629–641. https://doi.org/10.1007/BF01543305.

    Article  PubMed  Google Scholar 

  19. Canales, D. D., Olver, M. E., & Wong, S. C. (2009). Construct validity of the Violence Risk Scale—Sexual Offender Version for measuring sexual deviance. Sexual Abuse: A Journal of Research and Treatment, 21, 474–492. https://doi.org/10.1177/1079063209344990.

    Article  Google Scholar 

  20. Cantor, J. M., & Blanchard, R. (2012). White matter volumes in pedophiles, hebephiles, and teleiophiles. Archives of Sexual Behavior, 41, 749–752. https://doi.org/10.1007/s10508-012-9954-2.

    Article  PubMed  Google Scholar 

  21. Cantor, J. M., Kabani, N., Christensen, B. K., Zipursky, R. B., Barbaree, H. E., Dickey, R., … Richards, B. A. (2008). Cerebral white matter deficiencies in pedophilic men. Journal of Psychiatric Research, 42, 167–183. https://doi.org/10.1016/j.psychires.2007.10.013.

    Article  Google Scholar 

  22. Cantor, J. M., Klein, C., Lykins, A., Rullo, J. E., Thaler, L., & Walling, B. R. (2013). A treatment-oriented typology of self-identified hypersexuality referrals. Archives of Sexual Behavior, 42, 883–893. https://doi.org/10.1007/s10508-013-0085-1.

    Article  PubMed  PubMed Central  Google Scholar 

  23. Cantor, J. M., Lafaille, S., Soh, D. W., Moayendi, M., Mikulus, D. J., & Girard, T. A. (2015). Diffusion tensor imaging of pedophilia. Archives of Sexual Behavior, 44, 2161–2172. https://doi.org/10.1007/s10508-015-0629-7.

    Article  PubMed  Google Scholar 

  24. Cantor, J. M., & McPhail, I. V. (2015). Sensitivity and specificity of the phallometric test for hebephilia. Journal of Sexual Medicine, 12, 1940–1950. https://doi.org/10.1111/jsm.12970.

    Article  PubMed  Google Scholar 

  25. Carvalho, J., Stulhofer, A., Vieira, A. L., & Jurin, T. (2015). Hypersexuality and high sexual desire: Exploring the structure of problematic sexuality. Journal of Sexual Medicine, 12, 1356–1367. https://doi.org/10.1111/jsm.12865.

    Article  PubMed  Google Scholar 

  26. Cohen, L. J., & Galynker, I. I. (2002). Clinical features of pedophilia and implications for treatment. Journal of Psychiatric Practice, 8, 276–289. https://doi.org/10.1097/00131746-200209000-00004.

    Article  PubMed  Google Scholar 

  27. Davis, K. (2017). The relation of childhood abuse experiences to problematic sexual behaviors in male youths who have sexually offended, Unpublished doctoral dissertation, Brandeis University.

  28. Dyer, T. J., & Olver, M. E. (2016). Self-reported psychopathy and its association with deviant sexual fantasy and sexual compulsivity in a nonclinical sample. Sexual Offender Treatment, 11, 1–18.

    Google Scholar 

  29. Dyshniku, F., Murray, M. E., Fazio, R. L., Lykins, A. D., & Cantor, J. M. (2015). Minor physical anomalies as a window into the prenatal origins of pedophilia. Archives of Sexual Behavior, 44, 2151–2159. https://doi.org/10.1007/s10508-015-0564-7.

    Article  Google Scholar 

  30. Eher, R., Olver, M. E., Heurix, I., Schilling, F., & Rettenberger, M. (2015). Predicting reoffense in pedophilic child molesters by clinical diagnoses and risk assessment. Law and Human Behavior, 39, 571–580. https://doi.org/10.1037/lhb0000144.

    Article  PubMed  Google Scholar 

  31. Eher, R., Rettenberger, M., Matthes, A., & Schilling, F. (2010). Stable dynamic risk factors in child sexual abusers: The incremental predictive power of narcissistic personality traits beyond the Static-99/Stable-2007 priority categories on sexual reoffense. Sexual Offender Treatment, 5, 1–12.

    Google Scholar 

  32. Fazio, R. L., Dyshniku, F., Lykins, A. D., & Cantor, J. M. (2017). Leg length versus torso length in pedophilia further evidence of atypical physical development early in life. Sexual Abuse: A Journal of Research and Treatment, 29, 500–514. https://doi.org/10.1177/1079063215609936.

    Article  Google Scholar 

  33. Finkelhor, D., & Araji, S. (1986). Explanations of pedophilia: A four factor model. Journal of Sex Research, 22, 145–161. https://doi.org/10.1080/00224498609551297.

    Article  Google Scholar 

  34. Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual differences in implicit cognition: The Implicit Association Test. Journal of Personality and Social Psychology, 74, 1464–1480. https://doi.org/10.1037/0022-3514.74.6.1464.

    Article  PubMed  Google Scholar 

  35. Hall, R. C., & Hall, R. C. (2007). A profile of pedophilia: Definition, characteristics of offenders, recidivism, treatment outcomes, and forensic issues. Mayo Clinic Proceedings, 82, 457–471. https://doi.org/10.4065/82.4.457.

    Article  PubMed  Google Scholar 

  36. Hanson, R. K. (2010). Dimensional measurement of sexual deviance. Archives of Sexual Behavior, 39, 401–404. https://doi.org/10.1007/s10508-009-9575-6.

    Article  PubMed  Google Scholar 

  37. Hanson, R. K., & Morton-Bourgon, K. E. (2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73, 1154–1163. https://doi.org/10.1037/0022-006X.73.6.1154.

    Article  PubMed  Google Scholar 

  38. Haslam, N., Holland, E., & Kuppens, P. (2012). Categories versus dimensions in personality and psychopathology: A quantitative review of taxometric research. Psychological Medicine, 42, 903–920. https://doi.org/10.1017/S0033291711001966.

    Article  PubMed  Google Scholar 

  39. Howes, R. J. (1995). A survey of plethysmographic assessment in North America. Sexual Abuse: A Journal of Research and Treatment, 7, 9–24. https://doi.org/10.1007/BF02254871.

    Article  Google Scholar 

  40. Hunter, J. A., & Goodwin, D. W. (1992). The clinical utility of satiation therapy with juvenile sexual offenders: Variations and efficacy. Annals of Sex Research, 5, 71–80. https://doi.org/10.1007/BF00849732.

    Article  Google Scholar 

  41. Kafka, M. P. (2010). Hypersexual disorder: A proposed diagnosis for DSM-V. Archives of Sexual Behavior, 39, 377–400. https://doi.org/10.1007/s10508-009-9574-7.

    Article  PubMed  Google Scholar 

  42. Kafka, M. P., & Hennen, J. (2003). Hypersexual desire in males: Are males with paraphilias different from males with paraphilia-related disorders? Sexual Abuse: A Journal of Research and Treatment, 15, 307–321. https://doi.org/10.1023/A:1025000227956.

    Article  Google Scholar 

  43. Kingston, D. A., Firestone, P., Moulden, H. M., & Bradford, J. M. (2007). The utility of the diagnosis of pedophilia: A comparison of various classification procedures. Archives of Sexual Behavior, 36, 423–436. https://doi.org/10.1007/s10508-006-9091-x.

    Article  PubMed  Google Scholar 

  44. Klein, V., Schmidt, A. F., Turner, D., & Briken, P. (2015). Are sex drive and hypersexuality associated with pedophilic interest and child sexual abuse in a male community sample? PLoS ONE, 10, e0129730. https://doi.org/10.1371/journal.pone.0129730.

    Article  PubMed  PubMed Central  Google Scholar 

  45. Knight, R. A., & Graham, F. J. (2017). Hypersexuality: Equifinal, cohesive, clinical presentation or symptom cluster with multiple underlying mechanisms? Archives of Sexual Behavior, 46, 2261–2264. https://doi.org/10.1007/s10508-017-1089-z.

    Article  PubMed  Google Scholar 

  46. Långström, N., & Hanson, R. K. (2006). High rates of sexual behavior in the general population: Correlates and predictors. Archives of Sexual Behavior, 35, 37–52. https://doi.org/10.1007/s10508-006-8993-y.

    Article  PubMed  Google Scholar 

  47. Långström, N., & Seto, M. C. (2006). Exhibitionistic and voyeuristic behavior in a Swedish national population survey. Archives of Sexual Behavior, 35, 427–435. https://doi.org/10.1007/s10508-006-9042-6.

    Article  PubMed  Google Scholar 

  48. Långström, N., & Zucker, K. J. (2005). Transvestic fetishism in the general population. Journal of Sex and Marital Therapy, 31, 87–95. https://doi.org/10.1080/00926230590477934.

    Article  PubMed  Google Scholar 

  49. Laws, D. R. (2009). Penile plethysmography: Strengths, limitations, innovations. In D. Thornton & D. R. Laws (Eds.), Cognitive approaches to the assessment of sexual interest in sexual offenders (pp. 7–29). London: Wiley-Blackwell.

    Google Scholar 

  50. Looman, J., & Marshall, W. L. (2001). Phallometric assessments designed to detect arousal to children: The responses of rapists and child molesters. Sexual Abuse: A Journal of Research and Treatment, 13, 3–13. https://doi.org/10.1177/107906320101300102.

    Article  Google Scholar 

  51. Mackaronis, J. E., Byrne, P., Strassberg, D., Marcus, D., & Solari, J. (2011a). What does phallometry assess: Categories or continua? Presentation given at the Conference of the Association for the Treatment of Sexual Abusers, Toronto.

  52. Mackaronis, J. E., Strassberg, D. S., & Marcus, D. K. (2011b). The latent structure of Multiphasic Sex Inventory—Assessed Pedophilic Interest. Psychological Assessment, 23, 1017–1022. https://doi.org/10.1037/a0024625.

    Article  PubMed  Google Scholar 

  53. Marques, J., Nelson, C., West, M. A., & Day, D. M. (1994). The relationship between treatment goals and recidivism among child molesters. Behaviour Research and Therapy, 32, 577–588. https://doi.org/10.1016/0005-7967(94)90148-1.

    Article  PubMed  Google Scholar 

  54. Marshall, W. L. (1997). The relationship between self-esteem and deviant sexual arousal in nonfamilial child molesters. Behavior Modification, 21, 86–96. https://doi.org/10.1177/01454455970211005.

    Article  PubMed  Google Scholar 

  55. Marshall, W. L., O’Brien, M. D., & Marshall, L. E. (2009). Modifying sexual preferences. In A. R. Beech, L. A. Craig, & K. D. Browne (Eds.), Assessment and treatment of sex offenders. A handbook (pp. 311–327). Chichester: Wiley.

    Google Scholar 

  56. McGrath, R. E. (2008). Inferential errors in taxometric analyses of ordered three-class constructs. Journal of Personality Assessment, 90, 11–25. https://doi.org/10.1080/00223890701356755.

    Article  PubMed  Google Scholar 

  57. McPhail, I. V. (2018). Age of onset in pedohebephilic interests [Letter to the Editor]. Archives of Sexual Behavior, 47, 1313–1317. https://doi.org/10.1007/s10508-018-1198-3.

    Article  PubMed  Google Scholar 

  58. McPhail, I. V., & Cantor, J. M. (2015). Pedophilia, height, and the magnitude of the association: A research note. Deviant Behavior, 36, 288–292. https://doi.org/10.1080/01639625.2014.935644.

    Article  Google Scholar 

  59. McPhail, I. V., Hermann, C. A., Fernane, S., Fernandez, Y. M., Nunes, K. L., & Cantor, J. M. (2017). Validity in phallometric testing for sexual interests in children: A meta-analytic review. Assessment. https://doi.org/10.1177/1073191117706139.

    Article  PubMed  Google Scholar 

  60. McPhail, I. V., & Olver, M. E. (2018). Effectiveness of treatments for pedohebephilic interest in sexual offenders against children: A meta-analytic review. Manuscript in preparation.

  61. Meehl, P. E. (1992). Factors and taxa, traits and types, differences of degree and differences in kind. Journal of Personality, 60, 117–174. https://doi.org/10.1111/j.1467-6494.1992.tb00269.x.

    Article  Google Scholar 

  62. Meehl, P. E. (1995a). Bootstraps taxometrics: Solving the classification problem in psychopathology. American Psychologist, 50, 266–275. https://doi.org/10.1037/0003-066X.50.4.266.

    Article  PubMed  Google Scholar 

  63. Meehl, P. E. (1995b). Extension of the MAXCOV–HITMAX taxometric procedure to situations of sizeable nuisance covariance. In D. Lubinski & R. V. Dawis (Eds.), Assessing individual differences in human behavior: New concepts, methods, and findings (pp. 81–92). Palo Alto, CA: Davies-Black.

    Google Scholar 

  64. Meehl, P. E. (2004). What’s in a taxon? Journal of Abnormal Psychology, 113, 39–43. https://doi.org/10.1037/0021-843X.113.1.39.

    Article  PubMed  Google Scholar 

  65. Meehl, P. E., & Yonce, L. J. (1994). Taxometric analysis: I. Detecting taxonicity with two quantitative indicators using means above and below a sliding cut (MAMBAC procedure). Psychological Reports, 74, 1059–1274.

    Google Scholar 

  66. Meehl, P. E., & Yonce, L. J. (1996). Taxometric analysis: II. Detecting taxonicity using covariance of two quantitative indicators in successive intervals of a third indicator (MAXCOV procedure). Psychological Reports, 78, 1091–1227. https://doi.org/10.2466/pr0.1996.78.3c.1091.

    Article  Google Scholar 

  67. Moulden, H. M., Firestone, P., Kingston, D., & Bradford, J. (2009). Recidivism in pedophiles: An investigation using different diagnostic methods. Journal of Forensic Psychiatry & Psychology, 20, 680–701. https://doi.org/10.1080/14789940903174055.

    Article  Google Scholar 

  68. Müller, K., Curry, S., Ranger, R., Briken, P., Bradford, J., & Fedoroff, J. P. (2014). Changes in sexual arousal as measured by penile plethysmography in men with pedophilic sexual interest. Journal of Sexual Medicine, 11, 1221–1229. https://doi.org/10.1111/jsm.12488.

    Article  Google Scholar 

  69. Nichols, H. R., & Molinder, I. (1984). Manual for the Multiphasic Sex Inventory. Tacoma, WA: Crime and Victim Psychology Specialists.

    Google Scholar 

  70. Ó Ciardha, C., Attard-Johnson, J., & Bindemann, M. (2018). Latency-based and psychophysiological measures of sexual interest show convergent and concurrent validity. Archives of Sexual Behavior, 47, 637–649. https://doi.org/10.1007/s10508-017-1133-z.

    Article  PubMed  Google Scholar 

  71. Olver, M. E., Wong, S. C., Nicholaichuk, T., & Gordon, A. (2007). The validity and reliability of the Violence Risk Scale-Sexual Offender version: Assessing sex offender risk and evaluating therapeutic change. Psychological Assessment, 19, 318–329. https://doi.org/10.1037/1040-3590.19.3.318.

    Article  PubMed  Google Scholar 

  72. Quinsey, V. L., & Chaplin, T. C. (1988). Penile responses of child molesters and normals to descriptions of encounters with children involving sex and violence. Journal of Interpersonal Violence, 3, 259–274. https://doi.org/10.1177/088626088003003001.

    Article  Google Scholar 

  73. Ricci, R. J., Clayton, C. A., & Shapiro, F. (2006). Some effects of EMDR on previously abused child molesters: Theoretical reviews and preliminary findings. The Journal of Forensic Psychiatry & Psychology, 17, 538–562. https://doi.org/10.1080/14789940601070431.

    Article  Google Scholar 

  74. Ruscio, J. (2007). Taxometric analysis: An empirically-grounded approach to implementing the method. Criminal Justice and Behavior, 34(12), 1588–1622. https://doi.org/10.1177/0093854807307027.

    Article  Google Scholar 

  75. Ruscio, J. (2014). Taxometric programs for the R computing environment: User’s manual. Retrieved from ruscio.pages.tcnj.edu.

  76. Ruscio, J., Haslam, N., & Ruscio, A. M. (2006). Introduction to the taxometric method: A practical guide. Mahwah, NJ: Lawrence Erlbaum Associates.

    Google Scholar 

  77. Ruscio, J., & Kaczetow, W. (2009). Differentiating categories and dimensions: Evaluating the robustness of taxometric analysis. Multivariate Behavioral Research, 44, 259–280. https://doi.org/10.1080/00273170902794248.

    Article  PubMed  Google Scholar 

  78. Ruscio, J., & Ruscio, A. M. (2004). A nontechnical introduction to the taxometric method. Understanding Statistics, 3, 151–194. https://doi.org/10.1207/s15328031us0303_2.

    Article  Google Scholar 

  79. Ruscio, J., Ruscio, A. M., & Meron, M. (2007). Applying the bootstrap to taxometric analysis: Generating empirical sampling distributions to help interpret results. Multivariate Behavioral Research, 42, 349–386. https://doi.org/10.1080/00273170701360795.

    Article  PubMed  Google Scholar 

  80. Ruscio, J., Walters, G. D., Marcus, D. K., & Kaczetow, W. (2010). Comparing the relative fit of categorical and dimensional latent variable models using consistency tests. Psychological Assessment, 22, 5–21. https://doi.org/10.1037/a0018259.

    Article  PubMed  Google Scholar 

  81. Schmidt, A. F., Babchishin, K. M., & Lehmann, R. J. (2017). A meta-analysis of viewing time measures of sexual interest in children. Archives of Sexual Behavior, 46, 287–300. https://doi.org/10.1007/s10508-016-0806-3.

    Article  PubMed  Google Scholar 

  82. Schmidt, A. F., Mokros, A., & Banse, R. (2013). Is pedophilic sexual preference continuous? A taxometric analysis based on direct and indirect measures. Psychological Assessment, 25, 1146–1153. https://doi.org/10.1037/a0033326.

    Article  PubMed  Google Scholar 

  83. Seto, M. C. (2008). Pedophilia and sexual offending against children: Theory, assessment, and intervention. Washington, DC: American Psychological Association.

    Google Scholar 

  84. Seto, M. C. (2012). Is pedophilia a sexual orientation? Archives of Sexual Behavior, 41, 231–236. https://doi.org/10.1007/s10508-011-9882-6.

    Article  PubMed  Google Scholar 

  85. Seto, M. C. (2017). The puzzle of male chronophilias. Archives of Sexual Behavior, 46, 3–22. https://doi.org/10.1007/s10508-016-0799-y.

    Article  PubMed  PubMed Central  Google Scholar 

  86. Stephens, S., Leroux, E., Skilling, T., Cantor, J. M., & Seto, M. C. (2017). Taxometric analyses of pedophilia utilizing self-report, behavioral, and sexual arousal indicators. Journal of Abnormal Psychology, 126, 1114–1119. https://doi.org/10.1037/abn0000291.

    Article  PubMed  Google Scholar 

  87. Stulhofer, A., Jurin, T., & Briken, P. (2016). Is high sexual desire a facet of male hypersexuality? Results from an online study. Journal of Sex and Marital Therapy, 42, 665–680. https://doi.org/10.1080/0092623X.2015.1113585.

    Article  PubMed  Google Scholar 

  88. Sutton, K. S., Stratton, N., Pytyck, J., Kolla, N. J., & Cantor, J. M. (2015). Patient characteristics by type of hypersexuality referral: A quantitative chart review of 115 consecutive male cases. Journal of Sex and Marital Therapy, 41, 563–580. https://doi.org/10.1080/0092623X.2014.93553.

    Article  PubMed  Google Scholar 

  89. Tabachnick, B. G., & Fidell, L. S. (2013). Using multivariate statistics (6th ed.). Boston: Pearson.

    Google Scholar 

  90. Tozdan, S., & Briken, P. (2015). The earlier, the worse? Age of onset of sexual interest in children. Journal of Sexual Medicine, 12, 1602–1608. https://doi.org/10.1111/jsm.12927.

    Article  PubMed  Google Scholar 

  91. Waller, N. G., & Meehl, P. E. (1998). Multivariate taxometric procedures: Distinguishing types from continua. Thousand Oaks, CA: Sage.

    Google Scholar 

  92. Walters, G. D., McGrath, R. E., & Knight, R. A. (2010). Taxometrics, polytomous constructs, and the comparison curve fit index: A Monte Carlo analysis. Psychological Assessment, 22, 149–156. https://doi.org/10.1037/a0017819.

    Article  PubMed  Google Scholar 

  93. Walton, M. T., Cantor, J. M., Bhullar, N., & Lykins, A. D. (2017). Hypersexuality: A critical review and introduction to the “sexhavior cycle”. Archives of Sexual Behavior, 46, 2231–2251. https://doi.org/10.1007/s10508-017-0991-8.

    Article  PubMed  Google Scholar 

  94. Wilcox, R. R. (2005). New methods for comparing groups: Strategies for increasing the probability of detecting true differences. Current Directions in Psychological Science, 14, 272–275. https://doi.org/10.1111/j.0963-7214.2005.00379.x.

    Article  Google Scholar 

  95. Wilson, R. J., Abracen, J., Looman, J., Picheca, J. E., & Ferguson, M. (2011). Pedophilia: An evaluation of diagnostic and risk prediction methods. Sexual Abuse: A Journal of Research and Treatment, 23, 260–274. https://doi.org/10.1177/1079063210384277.

    Article  Google Scholar 

Download references

Acknowledgements

This research was supported by the Social Sciences and Humanities Research Council of Canada.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Ian V. McPhail.

Ethics declarations

Conflict of interest

The author declares that they have conflicts of interest.

Ethics Approval

Ethics approval for this research was received from University of Saskatchewan Behavioural Research Ethics Board to conduct all analyses conducted in the manuscript and to combine datasets in the manner outlined in the manuscript. This article does not contain any studies with human participants performed by any of the authors. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. For this type of study, formal consent is not required.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (DOCX 2084 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

McPhail, I.V., Olver, M.E., Brouillette-Alarie, S. et al. Taxometric Analysis of the Latent Structure of Pedophilic Interest. Arch Sex Behav 47, 2223–2240 (2018). https://doi.org/10.1007/s10508-018-1225-4

Download citation

Keywords

  • Pedophilia
  • Sexual interest in children
  • Taxometric analysis
  • Human sexuality
  • Phallometric testing
  • DSM-5