Preventing Child Sexual Abuse: Screening for Hidden Child Molesters Seeking Jobs in Organizations That Care for Children

Abstract

Approximately 10% of children and adolescents are sexually abused by adults caring for them outside the home. The current study tested the validity and reliability of a child protection screen to identify job applicants who pose a sexual risk to children. The screen uses three separate measures. In combination, they attempt to identify two types of sexually problematic job applicants: hidden abusers and people with cognitive distortions that encourage child sexual boundary violations by themselves or tolerate them by others. The high specificity (97.8% for males and 98.7% for females) favored the high number of job applicants and volunteers who have not crossed sexual boundaries with children. The study included over 19,000 participants, and the screen correctly identified 77% of the men and over 72% of the women who posed a sexual risk. The test–retest correlation was statistically significant at r(121) = .83, and the screening methodology is valid and reliable. By identifying most of the job applicants who are hiding their history of sexually abusing a child or hiding their belief that adult–child sex causes no harm from the organizations they are attempting to join, this new preemployment screen methodology can help child-centered organizations protect children and adolescents in their care.

Keywords

child abuse adolescent abuse child sexual abusers preemployment screening child sexual abuse prevention

The two current studies are intended to determine the viability (i.e., validity and reliability) of a preemployment screen to prevent hidden sexual abusers of children from obtaining jobs caring for children 17 years of age and younger. Study 1 demonstrates the screen’s validity. The screen uses three separate measures that utilize different methods to identify job applicants presenting a high sexual risk to children: (a) a classification model separating concealing sexual abusers of children from nonabusers, (b) a series of questions that allow job applicants to reveal their past sexual abuse of a child, and (c) a cognition section to identify the risk of job applicants committing sexual boundary violations. We present results for each of the three measures individually and then follow with the results for the screen as a whole when the three measures are combined. Study 2 analyzes the screen’s reliability. Such a screen might help youth-serving organizations to provide a safer environment for the children and teens in their care.

For survivors of child sexual abuse, both the abuse and the long-term health sequelae affect their lives. Such effects have been well documented in numerous studies. Victims of child sexual abuse experience loss of trust, depression, and identity confusion. They often become self-abusive and sometimes they act out sexually, risking pregnancy and increasing their chances of contracting sexually transmitted diseases, including those that are life threatening such as human immunodeficiency virus (HIV) and/or acquired immune deficiency syndrome (AIDS; Lalor & McElvaney, 2010). As a group, the children show a higher prevalence of major affective disorders, including mental, personality, and anxiety disorders, and they also exhibit physical and psychosomatic diseases, including enuresis, encopresis, chronic abdominal pain, headaches, anal or pelvic pain, vocal cord dysfunction, and fibromyalgia (Jenny, 2002; Spataro, Mullen, Burgess, Wells, & Moss, 2004). In a longitudinal study, Dube et al. (2005) found that victims of childhood sexual abuse were twice as likely to attempt suicide compared to individuals who were not sexually abused. Some child victims were successful in their attempts and died by suicide.

Literature Review

In 2014, the Centers for Disease Control and Prevention (CDC) reported a national rate for sexually abused children and adolescents in all settings to be 25% for girls and 16.6% for boys (20.8% of American children), with an estimated 15 million children being affected (CDC, 2014; U.S. Census Bureau, 2015). Based on additional studies, a substantial number of these children were sexually abused while in the care of youth-serving organizations.

A U.S. Department of Education study found that on average, nearly 10% of students have been victims of sexual abuse or misconduct by school employees or volunteers (Shakeshaft, 2004). The study also reported that at some high schools, as many as 50.3% of the students had been victimized at some point during their education. According to a study by the U.S. Department of Justice (2010), an average of 10.3% of the youth in juvenile detention across all custodial facilities reported having been sexually abused by the staff. The study revealed that many states had facilities with sexual abuse rates by staff higher than the national rate for sexually abused children and adolescents in all settings, including the familial setting, with as many as 36% of the youth having been sexually abused while in confinement.

Youth-serving organizations face loss of trust and legal consequences when they fail to protect children in their care. Sexual abuse of children by employees hired as caretakers is reported daily by the media, with stories citing abusers who are coaches, teachers, clergy, foster parents, daycare workers, children’s hospital workers, and workers in child residential facilities. In 2017, for example, The New York Times reported a sexual abuse cover-up by Choate Rosemary Hall, an elite Connecticut boarding school (whose alumni include John F. Kennedy and his brother Joseph P. Kennedy). According to the report, at least 12 former teachers from the 1960s onward had sexually molested or raped students. The newspaper article noted that “None of the teachers’ actions were reported to the police. In some cases, teachers were allowed to resign after being confronted with evidence of misconduct, and administrators wrote letters of recommendations for teachers who were fired” (Harris, 2017). In another example, following revelations of extensive cover-up, the Roman Catholic Church reported that up to 5% of its priests had sexually abused children, leading to US$2.1 billion in legal claims (Associated Press, 2007; Tomasi, 2009). The number of sexually abusive priests is thought to differ little from the number of clerical sex abusers in other denominations (Draper, 2010). Other youth organizations have also experienced legal claims. For instance, in 2010 a jury reached a US$1.4 million verdict against the Boy Scouts of America (Gibson, 2010). In another case, a woman filed a US$21 million lawsuit against the New York foster care system, alleging that she and her sisters were sexually abused for years (Burke, Woodby, & Bekiempis, 2016).

Given that child sexual abuse is a national health problem that often starts for children or adolescents when they are being cared for by workers in a youth service organization, what can such organizations do to protect the children in their care? The CDC recommends that all youth service organizations protect children by undertaking preventive screening of all job applicants (CDC, 2014). The Association for the Treatment of Sexual Abusers (2011) indicated that the primary prevention strategy to protect children and adolescents should be in place and implemented before sexual violence has occurred. The association suggests making “the healthy choice” of screening all employees and volunteers to be proactive in preventing child sexual abuse. However, at this time, we are unaware of any scientifically validated preemployment screens for the specific purpose of identifying prehires who are sexually problematic with regard to caring for children. Organizations that currently screen, and 12% of them do not (Webster & Whitman, 2008), commonly rely on criminal background checks to identify men and women who sexually abuse children. The greatest failing of these checks is that they identify so few adults (less than 0.2% of job applicants) as having sexual offense histories (Choice Point, 2008). The checks identify very few people who present a risk because they depend on the job applicant having already been convicted of child sexual abuse. In addition, most individuals who have sexually abused children are not convicted, as illustrated by two studies that found that only 6% of all child sexual abuse cases were reported to the police (Abel & Harlow, 2001; Kelly, Regan, & Burton, 1991).

Another problem for organizations that care for youth is that sexual abusers of children are within all strata of our society, equally distributed among educational and economic groups, various racial groups, and religious groups (Abel & Harlow, 2001). The abusers are virtually invisible. Men and women who sexually abuse children resemble all other job applicants, and they keep their sexual interest in children secret, leading organizations, children, and parents to trust them (Elliott, Browne, & Kilcoyne, 1995).

Study 1: Validity

Portions of the screen tested in this study incorporated questions based on information gathered by the National Center for Missing and Exploited Children (Wolff, 1986) and the Federal Bureau of Investigation (Lanning, 1992) about the characteristics of sexual abusers of children who had successfully molested in large child care organizations, as well as items suggested by a screening questionnaire used by Big Brothers Big Sisters (McCormack & Selvaggio, 1989). The development of such a child protection screen presented five practical challenges. We had to (a) design two methodologies, one applicable to males and one applicable to females, (b) design a screen free of prejudice against race and age, (c) determine an estimated base rate of child molesters in the general population, (d) account for job applicants who were hidden child molesters who would attempt to conceal that behavior (i.e., by faking or manipulating the screen beyond simply denying their sexual abuse of children), and (e) collect a large sample of sexual abusers of children concealing their behavior.

The most difficult challenge was the last one. Collecting a sample of concealing sexual abusers of children took a great deal of time because the sample had to contain both males and females and also had to be sufficiently large to build and cross-validate a classification model to distinguish job applicants who were hidden abusers from those who were not (most applicants). Our biggest concern was the risk of falsely identifying individuals as having a high probability of posing a sexual risk simply because they coincidentally had some demographic or behavioral variables in common with molesters. To combat this potential problem, we focused on specificity. We set the screen’s specificity (the percentage of applicants who had a low risk of sexually abusing a child) at 97.8% for the male screen and 98.7% for the female screen. Another challenge centered on choosing a sufficient number of variables that were complex enough to be difficult to fake. Most important, the screening methodology needed to obfuscate which items (and how they were scored and combined) were being used to identify applicants who posed a sexual risk. Describing the development and empirical validation of the screen here, even without publishing an answer key, is difficult, given the possibility of enabling a savvy job applicant to dodge the screening methodology.

Few studies examine the prevalence of adults in the general population who are sexual abusers of children. In addition, they make determinations using dissimilar questions and differing populations, and they fail to include the prevalence of abusers of children beyond the ages of 12 or 13. Two studies of college men found that 3% of them reported having sexually abused a child (Ahlers et al., 2011; Fromuth, Burkhart, & Jones, 1991). However, noting that study participants may be reluctant to admit to an illegal act, even anonymously, researchers have also inquired about the participants’ sexual interest in children and their sexual fantasies about children. When asked about their sexual interest in young children, 5% to 21% of male students and 10.4% of male adults reported some sexual interest in children (Ahlers et al., 2011; Briere & Runtz, 1989; Templeman & Stinnett, 1991). Perhaps even more revealing were answers to the hypothetical question, “Would you have sex with a child if you could avoid detection or punishment?” In one study of male college students, 7% reported some likelihood that they would have sex with a child if they were sure they wouldn’t get caught (Briere & Runtz, 1989). Asked the same question in another study, 19% of adult men said they also had some interest in having sex with a child (Hayashino, Wurtele, & Klebe, 1995).

With regard to the prevalence of women in the general population who sexually abuse children, studies of self-reports by survivors of child sexual abuse found that 42% to 78% of male survivors and 6% to 10% of female survivors reported having been abused by a female (Denov, 2003). However, these results do not provide an estimate of the prevalence of female sexual abusers in the general population. In one study that asked 546 college women about sexual contact with children, Fromuth and Conn (1997) found that 4% of the students surveyed reported at least one sexual experience that met the criterion for sexually abusing a child (i.e., sexual experiences with a child at least 5 years younger than themselves).

The published prevalence studies focus only on prepubescent children and so must necessarily have lower prevalence numbers than would account for the approximately 20.8% of children up to 17 years of age who have been sexually abused. Although choosing the conservative prevalence of 7% for men (Briere & Runtz, 1989) over 19% for men (Hayashino et al., 1995) and 4% for women (Fromuth & Conn, 1997), we present various values for our study’s statistics predicated on a range of abuser prevalence from 1% to 10%.

The purpose of the current study was to test both the validity and reliability of a prehire sexual risk screen—more specifically to identify job applicants with a high probability of sexually abusing children in their care. To demonstrate the screen’s validity and reliability, we conducted two studies. The first study, addressing validity, analyzes each of the screen’s three measures separately as they apply to males and to females. It then combines the three measures to show results of the full screen for each sex. The second study, which focuses on reliability, shows the analysis and results for test–retest.

Method: Study 1

The preemployment screen uses three measures to identify applicants who pose a sexual risk to children and should be precluded from working with them. Each measure uses a separate methodology and has a separate group of participants. Measure 1 is a classification measure that matches job applicants either to a group of sexual abusers of children who are concealing their abusive behavior or to a group of adults who have not sexually abused a child. Measure 2 allows job applicants to admit to having already sexually abused a child. Measure 3 allows job applicants to say “yes” to extreme beliefs or cognitions about the rights of adults to have sexual interactions with children. Separate results are presented for each of the three measures, and final results for the full screen are then presented by combining all three measures. Statistics are presented for each of the three measures separately, and results are then presented for the full screen combing all three measures.

Sample and participant selection for Study 1

Participants were drawn from two national populations: (a) men and women who had been evaluated by therapists who specialize in treating patients with paraphilias and sexually problematic behaviors (sex-specific therapists) across the United States because they had been accused or convicted of sexually abusing a child through 17 years of age and (b) child care job applicants. Job applicants were gathered from men and women who were given the screen by various organizations in 15 states. They were screened as candidates for ordination, foster care parents, mentors, and workers in daycare, summer camps and afterschool programs, community service centers for families, and a range of jobs in schools and in residential treatment and juvenile detention facilities. To be eligible for participation, the men and women had to be at least 18 years of age. After they provided voluntary consent on forms approved by the Georgia State University Institutional Review Board, individuals from the sample of participants accused or convicted of child sexual abuse (concealing abusers) answered the screen’s 120 questions as the first section of The Abel Questionnaire for Men^© or The Abel Questionnaire for Women^©, part of the Abel Assessment for sexual interest™ (AASI or AASI-II). Male and female job applicants who were not sexual abusers of children (nonabusers) also completed the 120 questions on the screen. The female version of the screen’s 120 questions only differed from the male version in pronouns; the female version used the words “she” and “her” to replace “he” and “his.” Men and women answered the screen’s questions differently; because of this, although we have used the same screen, we designed separate methodologies for males and for females for Measure 1, the classification model.

Job applicants who took the screen were assured of the confidentiality of their answers in two ways: (a) the hiring agency would have no access to their individual answers, and (b) the staff processing the results would not have their names; each screen when sent for processing would be identified by a number only.

Measure 1, classification model

Measure 1 matched applicants on a series of variables to a group of child molesters who were hiding or concealing their sexual behavior.

Method for Measure 1, classification model

From over a possible 100 variables on the screen, 45 variables from the screen that showed the greatest variance between the two groups were included in the predictive variable pool: 23 hobby and interest items, two cognition questions, eight identification with children items, 11 adolescent experience items, and the number of home relocations the job applicant had over the past 10 years. An example of one of the hobby and interest items was, “Science hobbies like astronomy and nature study.” For this item, the instructions were “Please rate you current interest in each of the following activities. For each of the statements, you will answer one of the following ranges of answers: 1. Very High, 2. High, 3. Some, 4. None.” An example of one of the adolescent experiences items was “One of my best friends was at least five years younger than me.” For this item, the instructions were “Were these statements true about you when you were a teenager? For each of the statements, you will answer either: Yes or No.”

Sample and participant selection for Measure 1, classification model

The male model building sample included concealing sexual abusers (n = 2,000) and nonabusers (n = 1,803), while the validation sample included abusers (n = 2,798) and nonabusers (n = 3,283). The female model building sample included concealing sexual abusers (n = 394) and nonabusers (n = 2,938), while the validation sample also included abusers (n = 124) and nonabusers (n = 5,805).

The statistics listed below are calculated from the total sample which combined the model building and validation samples. In the male study, the 4,798 participants from the concealing child sexual abuser group ranged from 18 to 91 years of age, with a mean age of 40.8(SD=13.7), and those 5,086 participants from the nonabuser group ranged from 18 to 83 years of age, with a mean age of 35.4(SD=12.2). In the female study, the 518 participants from the concealing abuser group ranged from 18 to 73 years of age, with a mean age of 34.3(SD=9.7), and those 8,743 participants from the nonabuser group ranged from 18 to 85 years of age, with a mean age of 34.7(SD=11.9). See Table 1 for a breakdown of participants by gender, sex abuse status, and race.

Table 1.

Measure 1: Male and Female Participants by Child Sexual Abuser Status and Race.

	Child sexual abusers	Nonabusers	Total
Males	4,798	5,086	9,884
	n (%)	n (%)	n (%)
African American	611 (13)	1,848 (36)	2,459 (25)
Asian American	65 (1)	50 (1)	115 (1)
Caucasian	3,083 (64)	2,583(51)	5,666 (57)
Hispanic/Latino	705 (15)	365 (7)	1,070 (11)
Native American	148 (3)	13 (0)	161 (2)
Other ethnicity	111 (2)	35 (1)	146 (1)
Multiracial	75 (2)	138 (3)	213 (2)
Race not reported	0 (0)	54 (1)	54 (1)
Females	518	8,743	9,261
	n (%)	n (%)	n (%)
African American	41 (8)	3,303 (38)	3,344 (36)
Asian American	6 (1)	135 (2)	141 (2)
Caucasian	386 (75)	4,232 (48)	4,618 (50)
Hispanic/Latina	44 (8)	592 (7)	636 (7)
Native American	19 (4)	39 (0)	58 (1)
Other ethnicity	13 (3)	60 (1)	73 (1)
Multiracial	9 (2)	282 (3)	291 (3)
Race not reported	0 (0)	100 (1)	100 (1)
Total	5,316	13,829	19,145

Participants had to meet two criteria for inclusion in the concealing child sexual abuser group. Both males and females had to deny current accusations or charges of having sexually abused a child. In addition, participants had to meet one of two further criteria: (a) having been accused of sexual abuse by children from more than one family or (b) having been convicted of sexual abuse against a child. Participants who met these criteria were included in the concealing child sexual abuser group.

Participants in the nonabuser group were job applicants (or candidates for ordination as clergy). We removed job applicants who were at a higher probability of being sexual abusers of children because either they had admitted to having sexually abused a child or they had answered positively to one or more of five cognitions asserting their beliefs in an adult’s right to have sex with a minor (see Figures 1 and 2).

Figure 1.

Flow chart showing the sequence for determining the participants’ inclusion in the concealing child sexual abuser group.

Figure 2.

Flow chart showing the sequence for determining the participants’ inclusion in the nonabuser group.

Data analysis for Measure 1, classification model

The 120 screen questions were included based on prior research by the lead author (Abel, Huffman, Warberg, & Holland, 1998; Abel, Jordan, Hand, Holland, & Phipps, 2001; Abel, Lawry, Karlstrom, Osborn, & Gillespie, 1994). Some items were excluded from the variable pool because they were legally problematic in certain states (e.g., marital status and number of times married), were seen to favor certain groups (e.g., questions about religious participation were seen to favor religious applicants), or were items unique to an organization (such as numbers of prior attempts to match with a child for Big Brothers Big Sisters volunteers). Each of the 45 variables selected as producing the greatest separation between the two groups had a coefficient with a different weight, either positive or negative. As males and females answered the questions differently, the coefficients for the male and female models were estimated separately.

Misclassification effect for Measure 1, classification model

Sensitivity and specificity were measured separately (Altman & Bland, 1994). Sensitivity was the percentage of the group of concealing sexual abusers of children who failed the screen. Specificity was the percentage of nonabusers who passed the screen. For the specificity of the nonabuser group, although we eliminated both job applicants who admitted to sexually abusing a child and those who failed cognitions, concern remained that the nonabuser group still contained concealing sexual abusers.

The formula that takes this problem into account is based on the formula for the observed pass rate for the nonabuser group of 94.6% for males and 96.7% for females:

P a s s R a t e = p r e v a l e n c e \times (1 - s e n s i t i v i t y) + (1 - p r e v a l e n c e) \times s p e c i f i c i t y

Solving for specificity, we get:

S p e c i f i c i t y = \frac{S e n s i t i v i t y \times P r e v a l e n c e + P a s s R a t e - P r e v a l e n c e}{1 - P r e v a l e n c e}

Assessment of fit for Measure 1, classification model

In addition to sensitivity and specificity, a measure of fit for classification models is the area under the receiver operating characteristic (ROC) curve. The ROC curve ranges between 0 and 1, where 1 is perfect prediction and 0.5 is prediction no better than chance (Hosmer & Lemeshow, 2000).

Results for Measure 1, classification model

The predicted values from the model were rescaled as z scores based on the nonabusers in the sample. This action meant that a job applicant with a score of 0 would have scored as an average nonabuser and a job applicant with a score of 2 would have scored two standard deviations above what an average nonabuser would have scored.

For the male model, we chose a cut-point score that was 1.67 above the mean for a nonchild molester. The estimated sensitivity for males using only Measure 1 of the three measures was 51.6%, with a pass rate of 94.6%. Specificity depended on sensitivity and prevalence of sexual abusers of children. Using a prevalence estimate of 7% (Briere & Runtz, 1989), the specificity for Measure 1 was 98.1%, which resulted in a 1.9% false-positive rate. A pass rate of 94.6% meant that approximately 94.6% of male applicants would be classified in the lower probability category as being a nonabuser.

For the female model, we chose a cut-point score of 1.91. The estimated sensitivity for Measure 1 was 47% with a pass rate of 96.6%. With the use of a prevalence rate of 4% (Fromuth & Conn, 1997), the specificity was 98.4% which resulted in a 1.6% false positive rate. The percentage correctly classified was 96.4%. A pass rate of 96.6% meant that approximately 96.6% of female applicants would be classified in the lower probability category as being a nonabuser.

The area under the ROC curve was calculated using the continuous predicted value. For males, the area under the ROC curve was 0.89 when applied to the model building sample, while the area under the ROC curve was 0.86, 95% confidence interval (CI) [0.85, 0.87] when applied to the validation sample. For females, the area under the ROC curve was 0.89 when applied to the model building sample, while the area under the ROC curve was 0.84, 95% CI [0.80, 0.88] when applied to the validation sample. Both ROC curves 0.86 and 0.84, when applied to the validation samples demonstrated a moderately strong positive relationship between the model prediction and the dependent variable measuring child sexual abuse committed.

Race and age prejudice testing for Measure 1, classification model

If a model to classify applicants who have a higher probability of sexually abusing children is valid and unprejudiced, then the predicted value from that model should be significantly related to child sexual abuser status and not significantly related to age and race. The present study used the same general linear model that was constructed in the Abel et al. (2012) study to test for prejudice.

A joint test of the interaction of race and age, controlling for concealing child sexual abuser status, was not statistically significant F(13, 3,788) = 0.7, p = .78 in the male model or in the female model F(13, 3,317) = 0.5, p = .93. A test of the main effect of race, controlling for age and child sexual abuser status, was also not statistically signiﬁcant in the male model F(6, 3,794) = 0.2, p = .98 or in the female model F(6, 3,332) = 0.1, p = .99. In addition, a test of the main effect of age, controlling for race and child sexual abuser status, was not statistically signiﬁcant in the male model F(1, 3,794) = .1, p = .80 or the female model F(1, 3,323) = 0.01, p = .98. These hypothesis tests failed to find prejudice with nearly 4,000 males and over 3,000 females, providing evidence that the models were free of prejudice based on age or race.

ROC by race and age

Table 2 shows area under the curve by race and age groups. Because of the sparsity of cases for many of the races, all non-White races were folded into one category. Ages were divided into four groups based on the 25th percentile, median and 75th percentile for each gender.

Table 2.

Area Under the ROC Curve by Gender, Race, and Age.

	N	ROC	Pass rate (%)	Sensitivity (%)	Specificity (%)
Male
White	3,404	0.86 [0.85, 0.87]	95	51	98
Non-White	2,677	0.87 [0.85, 0.88]	94	53	98
18–27	1,671	0.89 [0.87, 0.91]	94	58	98
28–35	1,377	0.87 [0.85, 0.89]	95	54	99
36–47	1,529	0.87 [0.85, 0.88]	96	48	99
Over 47	1,504	0.83 [0.81, 0.85]	93	49	97
Female
White	2,792	0.84 [0.80, 0.89]	97	48	99
Non-White	3,137	0.85 [0.76, 0.93]	96	41	98
18–25	1,652	0.93 [0.87, 0.98]	95	67	98
26–31	1,325	0.81 [0.72, 0.91]	96	46	98
32–43	1,532	0.82 [0.76, 0.88]	98	41	99
Over 43	1,420	0.82 [0.72, 0.91]	97	36	98

Note. ROC = receiver operating characteristic.

Measure 2, admissions

Totally, 12 questions allow job applicants to admit that they have sexually abused a child in the past. We believed that Measure 2 had strong face validity. Based on long-term studies, the reconviction rate of untreated males who are admitted sexual abusers of children was 34.1% and the reconviction rate for untreated females was 15.7%, for an average of 25.8% (Alexander, 1999).

Method for Measure 2, admissions

The participants answered 12 questions that allowed them to admit to sexually abusing a child in the past.

Participants and results for Measure 2, admissions

For this study, the number of participants who admitted to having sexually abused a child was extremely small: only 23 male applicants (0.4%) of 5,260 men and 29 female applicants (0.3%) of 9,115 women.

As the screen collected no information on the victims of the abuse, the hiring agency had no access to individual answers and the screen processing staff had no access to the applicants’ names, neither the hiring agency nor the processing staff had enough information to make a report.

Measure 3, cognitions tolerant of adult–child sex

Measure 3, cognitions tolerant of adult–child sex reveals job applicants’ failure to understand sexual boundaries with children.

Method for Measure 3, cognitions tolerant of adult–child sex

Totally, 52 therapists who were experienced in treating sexual abuse victims and 51 therapists who were experienced in treating sex abusers evaluated 22 cognitions. Conceptually, these cognitions were outlined in two previous studies by the lead author (Abel, Becker, & Cunningham-Rathner, 1984; Abel et al., 1989). The first study demonstrated that most child molesters had cognitions tolerant of adult–child sex. The second study showed that one could separate child molesters from nonmolesters based on the abusive group’s failure to understand sexual boundaries with children.

Each of the therapists read the 22 cognitions. The statements embodied extreme beliefs similar to “A four-year-old girl has the right to have sex with any adult she wants” or “No child has ever been hurt by having sex with an adult.” The therapists were asked to answer the following question to evaluate the sexual risk to children of job applicants who held these cognitions: “If a qualified person applied to work with children and said “yes” to these statements, would you recommend they be hired to work with children?” The therapists had a choice of three answers, “no,” “maybe,” or “yes.” Rather than picking an arbitrary cutoff line for consensus on which of the 22 statements would preclude a job applicant from being hired, we included a benchmark statement: “I have been convicted of child sexual abuse.” The therapists agreed by a consensus of 87.4% that being convicted would disqualify an applicant from being hired. Therefore, their consensus on each of the other statements had to be higher than 87.4% for it to be chosen as a disqualification cognition.

The therapists identified nine cognitive beliefs (by a consensus of 89.3% to 99%) that could be used to discriminate applicants who presented such a great sexual risk to children and teenagers than they should not be hired. To fail Measure 3 on the screen, a job applicant had to endorse two or more of these nine extreme beliefs on the right of adults to have sexual interactions with children.

Sample and participant selection for Measure 3, cognitions tolerant of adult–child sex

The therapists who treated sexual abuse victims were sent surveys through children’s advocacy centers and residential centers for children and teens with behavioral problems; the therapists who treated adults accused of sexually abusing children were sent surveys through the Abel Screening network of therapists as well as through additional sex specific treatment centers. The 103 participating therapists were located in 19 states: California, Colorado, Georgia, Idaho, Illinois, Kentucky, Louisiana, Maryland, Massachusetts, Missouri, Montana, Nevada, New York, North Carolina, Oklahoma, Oregon, Pennsylvania, Rhode Island, and Virginia. Their education level had a broad range: doctorate, 32.0%; masters, 59.2%; bachelors, 5.8%; and no degree, 1.0%. The 103 therapists had a mean of 15.3 years of experience (median = 15, min = 3, and max = 40). The therapists who treated victims had a mean of 12.3 years working with patients who had been sexually abused (SD = 7.5), and the therapists who treated sexual abusers of children had an average of 18.0 years working with that population (SD = 8.9). The therapists for victims included 17 males and 35 females, and the therapists for sexual abusers of children included 34 males and 17 females.

Results for Measure 3, cognitions tolerant of adult–child sex

The reliability of cognitions pertaining to tolerance of adult–child sex was estimated by using Cronbach’s alpha (Cronbach, 1951). Cronbach’s alpha is a measure of internal consistency that generally ranges between 0 and 1. Based on the validation sample of 3,283 male job applicants, the Cronbach’s alpha for males was 0.75. Based on the Measure 1 validation sample of 5,805 female job applicants, the Cronbach’s alpha for females was 0.74.

Results for Study 1: Full Screen Combining Three Measures

We combined the results from Measure 1, the classification model, separating concealing sexual abusers from nonabusers; Measure 2, the admission of having sexually abused a child; and Measure 3, the possession of cognitions tolerant of adult–child sex. Sensitivity and specificity were estimated from the model sample.

Because the control group for Measure 1, the nonabusers, contained an unknown number of concealing sexual abusers of children, specificity had to be derived from the failure rate as described in the misclassification effect on summary statistics for the classification models. The issue was more complicated for the full screen. In addition to accounting for the concealed and admitted sexual abusers, prevalence must factor in individuals with cognitions that indicate they fail to understand sexual boundaries with children.

For males, the prevalence rate for hidden child molesters or admitters was assumed to be 7%. The prevalence rate for failure on the cognitions scale was 6.2%. Deriving the prevalence rate for the three measures combined, however, was more complicated because two different, but overlapping groups were being combined. Because the groups were positively related, simply adding the two prevalences would have overstated the prevalence.

The formula for the combined prevalence of hidden child sexual abusers/admitters and applicants who failed cognitions was:

\begin{array}{l} Prevalence = P (Cog = F) + P (CSA = 1) - P (ADM = 1) \times P (Cog = F | ADM = 1) - P (DIS = 1) \times \\ P (Cog = F | DIS = 1) \end{array}

Where: P(Cog=F) is the probability of failing Measure 3, cognitions tolerant of adult–child sex, P(CSA=1) is the probability of being a child sexual abuser, P(ADM=1) is the probability of being an admitting child sexual abuser, P(Cog=F|ADM=1) is the probability of failing Measure 3 and being an admitting child sexual abuser, P(DIS=1) is the probability of being a concealing child sexual abuser, and P(Cog=F|DIS=1) is the probability of failing Measure 3 and being a concealing child sexual abuser.

We estimated a prevalence for the two groups combined of 12.1%. For a range of prevalences of child sexual abusers alongside a range of combined prevalences, see Table 3.

Table 3.

Male Full Screen: Classification Statistics Dependent on Prevalence.

Child sex abuser prevalence (%)	Combined prevalence (%)	Specificity (%)	Sensitivity (%)	Correctly classified (%)	Positive predictive value (%)	Negative predictive value (%)
1	7.0	95.2	97.6	95.4	60.4	99.8
2	7.8	95.6	92.3	95.3	64.0	99.3
3	8.7	96.0	88.1	95.3	67.7	98.8
4	9.5	96.4	84.6	95.3	71.4	98.3
5	10.4	96.9	81.7	95.3	75.1	97.9
6	11.2	97.3	79.2	95.3	78.8	97.4
7	12.1	97.8	77.1	95.3	82.4	96.9
8	12.9	98.2	75.2	95.2	86.1	96.4
9	13.8	98.7	73.6	95.2	89.8	95.9
10	14.6	99.1	72.2	95.2	93.5	95.4

For females, the prevalence rate for hidden child molesters and admitters was assessed to be 4% (Fromuth & Conn, 1997). The prevalence rate for failure on the cognitions scale was 3.4%. Combining the two groups: hidden molesters/admitters and those failing cognitions, the prevalence rate, accounting for overlap between the groups was estimated at 7%. For a range of prevalences of child sexual abusers alongside a range of combined prevalences, see Table 4.

Table 4.

Female Full Screen: Classification Statistics Conditional on Prevalence.

Child sex abuser prevalence (%)	Combined prevalence (%)	Specificity (%)	Sensitivity (%)	Correctly classified (%)	Positive predictive value (%)	Negative predictive value (%)
1	4.3	97.5	90.80	97.2	61.5	99.6
2	5.2	97.9	82.60	97.1	67.8	99.0
3	6.1	98.3	76.80	96.9	74.0	98.5
4	7.0	98.7	72.60	96.8	80.3	98.0
5	7.9	99.1	69.30	96.7	86.5	97.4
6	8.8	99.5	66.70	96.6	92.8	96.6

Assessment of Fit for the Full Screen Combining All Three Measures

For males, the area under the ROC curve was 0.93, 95% CI [0.92, 0.93] when applied to the validation sample. For females, the area under the ROC curve was 0.95, 95% CI [0.93, 0.97] when applied to the validation sample. The area under the ROC curve of 0.93 for males demonstrated a moderately strong positive relationship between the model prediction and the dependent variable measuring child sexual abuse committed. The area under the ROC curve of .95 for females demonstrated an even stronger relationship.

Sensitivity and Specificity for the Full Screen Combining All Three Measures

For males, the estimated sensitivity (identifying abusers) was 77.1%. The estimated specificity (identifying nonabusers) was 97.8%, with a resulting false-positive rate of 2.2%. The estimated sensitivity was based on a cut-point score of 1.67, which was 1.67 standard deviations above the mean for nonabusers. The estimated specificity depended on sensitivity and the prevalence for all three measures combined. Male job applicants passed the full screen at a rate of 88.7%. For females, the estimated sensitivity was 72.6%. The estimated specificity was 98.7%, which gives a 1.3% false-positive rate. The sensitivity was based on a cut-point score of 1.91, which was 1.91 standard deviations above the mean for nonabusers. This resulted in 93.7% of female applicants passing the screen.

Study 2: Reliability

Method for Study 2: Reliability

The participants completed the screen’s 120 questions as the first section of the Abel Questionnaire for Men^© part of the Abel Assessment for sexual interest™ (AASI-II or AASI-III).

Sample and participant selection for study 2, reliability

All participants provided voluntary consent on forms approved by the Georgia State University Institutional Review Board. These participants were not included in any of the previous samples for the model building or validation. The sample includes only males because we were not able to recruit enough females (n = 10) for the study. The participants consisted of 123 males receiving evaluation or treatment from the practice of the lead author between 2008 and 2014. Each participant was assessed twice. Both assessments were completed within a 2-week period. The participants ranged from 20 to 82 years of age, with a mean age of 46. The participants were 79% Caucasian, 8% African American, 7% Asian American, 2% Latino/Hispanic, and 4% listed their race as either “other” or “more than one race.” Totally, 73% of the sample were college educated.

Data analysis for study 2, reliability

Test–retest reliability of the screen was assessed by measuring the Pearson’s correlation between the first and second assessment scores.

Results for Study 2: Reliability

The test–retest correlation was statistically significant r(121) = .83, p < .0001, 95% CI [0.77–0.88]. This correlation indicates that about 83% of the variance was systematic as opposed to random.

Discussion

Discussion of Study 1: Validity

Sexual abuse of children and adolescents in youth service organizations is a problem that society is only beginning to address. National studies on the prevalence of child care workers crossing sexual boundaries with children and adolescents indicate a range from 3.7% to 50.3% (Shakeshaft, 2004; U.S. Department of Justice, 2010). Regardless of the actual prevalence rate, children and adolescents in the care of organizations are clearly at risk of being sexually targeted by their caretakers.

Although criminal background checks are the most frequently used means of keeping child molesters from getting child care jobs, these checks are inadequate, resulting in less than 0.2% of abusers being identified (Choice Point, 2008).

No definitive numbers are attached to the prevalence of sexual abusers in child care organizations or in the general population, as suggested by diverse studies. We chose to base the screen’s Measure 1, the classification model, on prevalence estimates for the general population because the screen is designed for a wide and diverse number of prehire applicants or volunteers who come into contact with children, from clergy to foster parents to workers in amusement parks to volunteers in hospitals, summer camps, child and adolescent sport teams and individual sports, and more. The number of sexual abusers is believed to have declined based on recent studies of reduced conviction rates (Finkelhor & Jones, 2006; Sandler & Freeman, 2009). However, to our knowledge, no studies have suggested a reduction in the number of victims. Our choice of a 7% prevalence of male sexual abusers of children and 4% of female abusers is believed to be low because any prevalence has to coincide in some manner to the 20.8% of sexually abused child victims (CDC, 2014). Furthermore, existing studies have focused on children up to 12 or 13 years of age and have failed to include adolescents through 17 years of age. Still, the lack of a definitive answer led us to provide the classification analysis based on a broad prevalence range (1% to 10%); see Tables 3 and 4.

An advantage of this screening methodology is its basis on questions suggested by the work of the Federal Bureau of Investigation, National Center for Missing and Exploited Children, and Big Brothers and Big Sisters. To our knowledge, this methodology is the first child protection screen to be based on the characteristics and behavior of known sexual abusers of children.

This screen does not assume that hidden child molesters in youth service organizations are pedophiles or hebephiliacs purposely seeking access to children. It also does not rely on measuring the individuals’ sexual interest pattern, but rather tests for attitudes, interests, and cognitive distortions typically held by individuals who have molested children, regardless of the “reason” they engaged in the activity. The sample of concealing sexual abusers of children includes hebephiliacs, as well as individuals who molest children or adolescents as a result of opportunity, proximity, and misdirected “affection.” Thus it is applicable to the pool of applicants seeking to volunteer or work in youth service organizations.

We felt it was important to minimize the false positives based on the assumption that the greatest number of job applicants would be unlikely to cross sexual boundaries with children. A specificity of 97.8% for males and 98.7% for females lowered the false-positive rate to 2.2% and 1.3%, respectively. Using a penalized least squares methodology, the classification models have been demonstrated to be nearly free of race and age bias. However, despite sample sizes of over 3,000 and nearly 4,000, a model could be prejudiced against a nearly infinite set of demographic subgroups. Even if the data were available, only a limited number of demographic variables could be feasibly handled by this method.

Committing child sexual abuse was a direct risk factor. If applicants had a high probability of having sexually abused a child in the past either because they failed the classification models (Measure 1) or admitted to the act (Measure 2), then it was reasonable to assume that they were at greater risk to do so in the future because recidivism rates for sexual abusers of children range from 22% to 32%. These rates based on a sexual abuser of children being convicted and then being reconvicted are likely well below the recidivism rates of such abusers who have never been convicted.

Job applicants who held cognitions that revealed that they failed to understand sexual boundaries with children (Measure 3) could be seen as an indirect risk factor. These applicants may pose a greater risk of crossing sexual boundaries with a child, and they may also be more likely to overlook or condone the sexually abusive behavior of their coworkers.

The screen was designed to be difficult to fake. The male and female classification models each involve 45 different variables, from a total pool of over 100 variables in each of the questionnaires. To fake the models, a test-taker would need to know which questions are included as variables and how to answer the items in a manner that approximates the direction and magnitude of the regression coefficients for each of these variables. These models are also free of age and race bias. It is important to remember that no methodology, regardless of the degree of empirical validation, can definitively discern whether a person has sexually abused a child. As with other psychological or medical screens, individuals identified as having a higher probability require further and more detailed testing and evaluation. Such further “testing” could involve currently used methods, such as a thorough structured clinical interview, background/criminal history checks, and the thorough checking of references, as well as psychological and personality testing, a polygraph examination, or both. However, these models, when used as part of a screening tool for people applying for positions of trust with children, substantially increase an organization’s ability to protect children by identifying those men and women who have the greatest likelihood of having sexually abused children in the past.

Discussion of Study 2: Reliability

Reliability of the classification model was assessed by the use of test–retest correlation. The correlation coefficient of 0.83 between assessments indicated temporal stability and reliability. One caveat about assessing reliability through the test–retest correlation over a short period of time is that test-takers may remember their answers, thus inflating the measure of reliability. However, the full screens, both for males and females, contain 120 questions, and it is unlikely the applicants would be able to remember their answers to all of the questions.

Limitation of the Present Studies

As with any research, this study has limitations. The comparison groups, concealing sexual abusers and nonabusers, are unavoidably imperfect samples. For the sample of concealing sexual abusers, we were unable to accommodate the way new technologies (e.g., the Internet) change the manner in which child sexual abuse occurs. Similarly, it was very possible that some individuals in the nonabuser group concealed their prior history of being sexual abusers of children or that they may go on to commit sexual abuse. To the extent that this possibility may be the case, the accuracy of the statistical classification models would be reduced.

A potential criticism is that the sampling was not “pure” because we were unable to assemble a group of concealing sexual abusers and a group of nonabusers, all of whom had worked in youth service organizations. As the literature suggests that adults who sexually abuse children come equally from every socioeconomic group, as well as coming in exactly the same percentages as the three major races in the United States (Abel & Harlow, 2001), and that they may more frequently target organizations that care for groups of children, one would expect that their presence in youth service organizations would be the same or higher than in the general population.

While the purpose of this study was to identify job applicants who have an increased risk of sexually abusing children or violating sexual boundaries with children, this issue could only be resolved in a prospective study. Researchers would have to follow job applicants for several years and then compare their placement in the group of a “high probability of a sexual risk or a low probability of sexual risk” with those who had sexually abused a child and those who had violated sexual boundaries. This prospective design was beyond our capabilities.

A further limitation is that the screening measure is vulnerable to faking if the items and scoring algorithms were to become public knowledge thereby increasing the ability of motivated applicants to intentionally falsify their application. A nondisclosure agreement was purposely added to protect the data and scoring algorithms.

Conclusions and Future Study

Future research might focus on helping to determine the true prevalence of adults who sexually abuse minors by designing studies that ask about the adults’ current or probable sexual abuse of children through their 17th year. Other studies might explore the dynamics between sexual abusers discovered in schools, foster care, religious organizations, juvenile custody, and other youth service organizations to discover both the admittance procedures that allowed the job applicant to be hired and the ongoing supervisory policies that allowed them to continue sexually abusing a minor.

The major contribution of this study is its development of a methodology using a combination of three measures to identify job applicants from the general population who have the highest probability of crossing sexual boundaries with children. The methodology is an improvement over criminal background checks that identify a tiny fraction of 1% of child sexual abusers. The screen in this study identified 77% of male job applicants and 73% of female job applicants who posed a sexual risk to children—job applicants who might then be prevented from being hired for positions as caretakers of children. However, child service organizations utilizing this methodology need to be reminded that the findings on this screen should not be used alone, but should be one of several criteria used to approve job applicants. No test is perfect, and no screen, regardless of the degree of empirical validation, can discern whether an applicant will definitely sexually abuse a child. Nevertheless, the predictive ability of this screen presents an advance in the field and a promising screening methodology that youth service organizations might use to keep the job applicants who pose a sexual risk to children and adolescents from gaining access to them.

Footnotes

Authors’ Note

Details about the items and data analysis were not reported in this paper to protect the proprietary nature of the measure that was used in this study; two of the authors have a financial interest in this measure. These details were also held back because the validity of the measure would be adversely affected if the items or data analytic approach were widely known. At the same time, we recognize that scientific reporting requires transparency and accountability, so that independent, qualified professionals can verify the results reported in this paper. Therefore, the items and data used in this study can be made available for the purpose of verifying the results upon vetting and signing of a nondisclosure agreement, available from the first author. This nondisclosure prevents the signee from revealing the items or the scoring algorithm, however, it does not prevent the signee from discussing his or her reanalysis.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

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

References

Abel

G. G.

Becker

J. V.

Cunningham-Rathner

(1984). Complications, consent, and cognitions in sex between children and adults. International Journal of Law and Psychiatry, 7, 89-103.

Abel

G. G.

Gore

D. K.

Holland

C. L.

Camp

N. L.

Becker

J. V.

Rathner

(1989). The measurement of cognitive distortions of child molesters. Annals of Sex Research, 2, 135-153.

Abel

G. G.

Harlow

(2001). The stop child molestation book: What ordinary people can do in their everyday lives to save three million children. Bloomington, IN: Xlibris Corporation.

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, 81-94.

Abel

G. G.

Jordan

A. D.

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

E. M.

Osborn

C. A.

Gillespie

C. F.

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

Abel

G. G.

Wiegel

Jordan

Harlow

Hsu

Martinez

(2012). Development and validation of classification models to identify hidden child molesters applying to child service organizations. Children and Youth Services Review, 34, 1378-1389.

Ahlers

C. J.

Schaefer

G. A.

Mundt

I. A.

Roll

Englert

Willich

S. N.

Beier

K. M.

(2011). How unusual are the contents of paraphilias? Paraphilia-associated sexual arousal patterns in a community-based sample of men. The Journal of Sexual Medicine, 8, 1362-1370.

Alexander

M. A.

(1999). Sexual offender treatment efficacy revisited. Sexual Abuse: A Journal of Research and Treatment, 10(2), 81-94. doi:10.1177/107906329901100202

10.

Altman

D. G.

Bland

J. M.

(1994). Diagnostic tests. 1: Sensitivity and specificity. British Medical Journal, 308, Article 1552.

11.

Associated Press. (2007, July 14). L.A. archdiocese to settle suits for $660 million. Retrieved from http://www.nbcnews.com/id/19762878/ns/us_news-life/t/la-archdiocese-settle-suits-million/#.V2l4JrgrLbI/

12.

Association for the Treatment of Sexual Abusers. (2011, March 26). Sexual abuse as a public health problem. Retrieved from http://www.atsa.com/sexual-abuse-public-health-problem

13.

Briere

Runtz

(1989). University males’ sexual interest in children: Predicting potential indices of “pedophilia” in a nonforensic sample. Child Abuse & Neglect, 13, 65-75.

14.

Burke

Woodby

Bekiempis

(2016, March 30). Ex-foster child claims years of sexual abuse in $21M suit. Daily News. Retrieved from http://www.nydailynews.com/new-york/manhattan/ex-foster-child-claims-years-sexual-abuse-21m-suit-article-1.2582118

15.

Centers for Disease Control and Prevention. (2014). Child maltreatment: Risk and protective factors. Retrieved from http://www.acf.hhs.gov/sites/default/files/cb/cm2014.pdf

16.

Choice Point. (2008). The importance of background screening for nonprofits: An updated briefing. Alpharetta, GA: Author.

17.

Cronbach

L. J.

(1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297-334. doi:10.1007/bf02310555

18.

Denov

M. S.

(2003). The myth of innocence: Sexual scripts and the recognition of child sexual abuse by female perpetrators. Journal of Sex Research, 40, 303-314.

19.

Draper

(2010, May 25). Scandal creates contempt for Catholic clergy. The Denver Post. Retrieved from http://blogs.denverpost.com/hark/2010/05/25/scandal-creates-contempt-for-catholic-clergy/39/

20.

Dube

S. R.

Anda

R. F.

Whitfield

C. L.

Brown

D. W.

Felitti

V. J.

Dong

Giles

W. H.

(2005). Long-term consequences of childhood sexual abuse by gender of victim. American Journal of Preventive Medicine, 28, 430-438.

21.

Elliott

Browne

Kilcoyne

(1995). Child sexual abuse prevention: What offenders tell us. Child Abuse & Neglect, 19, 579-594.

22.

Finkelhor

Jones

(2006). Why have child maltreatment and child victimization declined? Journal of Social Issues, 62, 685-716.

23.

Fromuth

M. E.

Burkhart

B. R.

Jones

C. W.

(1991). Hidden child molestation an investigation of adolescent perpetrators in a nonclinical sample. Journal of Interpersonal Violence, 6, 376-384.

24.

Fromuth

M. E.

Conn

(1997). Hidden perpetrators: Sexual molestation in a nonclinical sample of college women. Journal of Interpersonal Violence, 12, 456-465.

25.

Gibson

(2010, April 18). Five myths about the Catholic sexual abuse scandal. The Washington Post. Retrieved from http://www.washingtonpost.com/wp-dyn/content/article/2010/04/16/AR2010041602026.html

26.

Harris

E. A.

(2017, April 13). Sexual abuse at Choate went on for decades, school acknowledges. The New York Times. Retrieved from https://www.nytimes.com/2017/04/13/nyregion/sexual-abuse-choate-connecticut-school.html?smprod=nytcore-iphone&smid=nytcore-iphone-share

27.

Hayashino

D. S.

Wurtele

S. K.

Klebe

K. J.

(1995). Child molesters: An examination of cognitive factors. Journal of Interpersonal Violence, 10, 106-116.

28.

Hosmer

D. W.

Lemeshow

(2000). Applied logistic regression (2nd ed.). New York, NY: John Wiley.

29.

Jenny

(2002). Medical issues in sexual abuse. In Myers

J. E.

Berliner

Briere

Hendrix

C. T.

Jenny

Reid

T.A.

(Eds.). The APSAC handbook on child maltreatment (2nd ed., pp. 235-247). Thousand Oaks, CA: Sage.

30.

Kelly

Regan

Burton

(1991). An exploratory study of the prevalence of sexual abuse in a sample of 16-21-year-olds. London, England: Polytechnic of North London.

31.

Lalor

McElvaney

(2010). Child sexual abuse, links to later sexual exploitation/high-risk sexual behavior, and prevention/treatment programs. Trauma, Violence, & Abuse, 11, 159-177. doi:10.1177/1524838010378299

32.

Lanning

(1992). Child molesters: A behavioral analysis for law enforcement officers investigating cases of child sexual exploitation. Washington, DC: Center for Missing and Exploited Children.

33.

McCormack

Selvaggio

(1989). Screening for pedophiles in youth-oriented community agencies. Social Casework: Journal of Contemporary Social Work, 70, 37-42.

34.

Sandler

J. C.

Freeman

N. J.

(2009). Female sex offender recidivism: A large-scale empirical analysis. Sexual Abuse: A Journal of Research and Treatment, 21, 455-473.

35.

Shakeshaft

(2004). Educator sexual misconduct: A synthesis of existing literature (PPSS 2004-09). Washington, DC: U.S. Department of Education.

36.

Spataro

Mullen

P. E.

Burgess

P. M.

Wells

D. L.

Moss

S. A.

(2004). Impact of child sexual abuse on mental health. The British Journal of Psychiatry, 184, 416-421.

37.

Templeman

T. L.

Stinnett

R. D.

(1991). Patterns of sexual arousal and history in a “normal” sample of young men. Archives of Sexual Behavior, 20, 137-150.

38.

Tomasi

S. M.

(2009). Vatican sets record straight on sexual abuse. Retrieved from http://www.catholiceducation.org/en/controversy/common-misconceptions/vatican-sets-record-straight-on-sexual-abuse.html/

39.

U.S. Census Bureau. (2015). Population estimates, July 1, 2015 (V2015). Retrieved from https://www.census.gov/quickfacts/table/PST045215/13

40.

U.S. Department of Justice. (2010). National strategy for child exploitation prevention and interdiction. Retrieved from https://www.justice.gov/psc/docs/natstrategyreport.pdf

41.

Webster

M. W.

Whitman

(2008, April). Who’s lending a hand? A national survey of nonprofit volunteer screening practices. Washington, DC: The National Center for Victims of Crime.

42.

Wolff

D. L.

(1986, June). Child sexual abuse: What we have learned. Presented at the Child Sexual Abuse Symposium, 1986. BB/BSA National Conference, Clearwater Beach, FL.