Adult Correlates of Child Sexual Abuse

A meta-analytic review of college student and national probability samples

[Start]     [Appendix] 

[Back to the Rind et al. Files] 

Society for Scientific Study of Sex, Eastern Regional Conference
SSSS-ER, April 18, 1997

Philip Tromovitch
Graduate School of Education, University of Pennsylvania (Philadelphia)

Bruce Rind
Department of Psychology, Temple University (Philadelphia, Pennsylvania)

Robert Bauserman
Department of Psychology, University of Michigan (Ann Arbor, Michigan)

Content

 Abstract  

[Note]

[Remark] 

[Introduction] 

The National Probability Studies 
Prevalence of CSA 
Meta-analyses of the national data 

Meta-analyses of the college data 
Estimating the impact of controlling for relevant confounds 

Summary 

Discussion 
Recommendations for future research 

Appendix 
Studies included in the meta-analyses of national probability samples 
Studies included in the meta-analyses of college student samples 
References 

Abstract

Child sexual abuse (CSA) is viewed by the lay public, and by many professionals, as one of the most psychologically damaging events that a child or adolescent can experience. Opinions expressed by many professionals imply that CSA possesses at least four fundamental qualities or properties:

(1) it causes harm,
(2) this harm occurs prevalently among people who have had experiences classifiable as CSA,
(3) this harm is typically intense, and
(4) CSA is at least as harmful an experience for males as it is for females.

The current paper examines these implied properties by reviewing the results of 54 college samples and 10 national probability samples (5 male and 5 female) which provide data relevant to psychological correlates of CSA. In order to minimize confirmation and sampling biases, 100% sampling of studies was attempted and quantitative analyses (i.e., meta-analyses) were conducted.

Meta-analyses of 18 symptom domains revealed that students with experiences classifiable as CSA are, on average, slightly less well adjusted than control subjects across all 18 symptom domains. Meta-analyses of a composite effect based on national probability samples showed an identical effect size to that found in the college data.

Further analyses, however, indicate that this poorer adjustment cannot be causatively attributed to the CSA experiences because of the reliable presence of confounding variables (in the general domain of family environment), which, when controlled for, rendered the majority of CSA-symptom relations nonsignificant in studies where statistical control could be applied.

Examination of the reported reactions to the CSA experiences also revealed significant gender differences, with males reporting significantly more positive experiences than females; it is further noted that the socio-legal definitions of CSA that are currently used in CSA research are so broad as to be capturing very different experiences under the same rubric.

It is concluded that:

(1) college student data on CSA-symptom relations is generalizable to the population at large,
(2) the assumptions of causality of harm, prevalence of harm, and intensity of effects are false (at least in college student and national samples), and
(3) reactions to experiences included under the wide scope of the currently used socio-legal definitions of CSA differ greatly between males and females.

The findings from this report contradict prevalently held assumptions about CSA — assumptions that may bias not only the lay public, but researchers studying and reporting on activities classifiable as CSA.

[Note]

This is the conference paper which accompanies the talk of the same name presented at the Society for the Scientific Study of Sexuality (SSSS) Eastern Region meeting, April 18, 1997. This talk and paper only include some of the findings of our research. Fuller descriptions, additional findings, and more extensive analyses, discussion, and background can be found in Rind & Tromovitch (in press), and in Rind, Tromovitch, & Bauserman (manuscript under submission). Copies of the slides for this presentation are attached as an appendix. [not true for email version]

[Remark

This text is sent electronically in ASCII-only format and without the slides. Thus, faults are possible where the statistics or data asked for more formatting. The same holds for the headers and sub(sub) headers.]

[Introduction]

Child sexual abuse (CSA) is an important contemporary issue. The past two decades have seen an extensive amount of media attention to this topic, as well as the completion of hundreds of scientific studies examining various aspects of CSA. Many of these studies have examined psychological correlates, and in the last decade, a number of researchers have attempted to bring this research together.

Most commonly they have used the qualitative literature review format, in which they gathered a collection of studies and attempted to summarize from them, in narrative fashion, the findings on the association between CSA and latter adjustment (e.g., sexual, psychological). Although this technique can yield many important findings, it is limited by subjectivity and imprecision (Jumper, 1995). A literature reviewer may be subject to confirmation bias, for example, noting findings consistent with his or her preexisting beliefs, while paying less attention to findings counter to what he or she expected to find (cf. Rosenthal, 1977). Furthermore, narrative summaries lack precision; a reviewer using this format may exaggerate or understate relations between variables.

Today, more powerful and precise methods are available for synthesizing a large number of studies when the data are quantifiable. Meta-analysis is the quantitative counterpart to the qualitative literature review. In meta-analysis, the findings of various studies are quantified into a common metric (e.g., Pearson’s r or Cohen’s d), referred to as the effect size, that can then be statistically combined, analyzed, and assessed with great objectivity.

Effect sizes indicate the strength or magnitude of a relationship between two variables, which is often of greater interest than whether the variables are statistically significantly related (that is, rather than addressing the question of whether or not there is a difference, effect size gets at the magnitude of the difference). In a meta-analysis, the researcher combines the effect sizes from all of the included studies to obtain a weighted mean effect size that estimates the magnitude of the relationship between the variables in the population. When the effect sizes are weighted by their degrees of freedom (i.e., N - 3), the resulting weighted mean effect size is referred to as the unbiased effect size estimate (ru).

Meta-analysts frequently also assess the homogeneity of the individual effect sizes from the studies included in the analysis. Homogeneity asks the question: "Are all of the studies telling the same story?" If the effect sizes are homogeneous, additional confidence can be placed in the unbiased effect size estimate. If the effect sizes are heterogeneous, then meta-analysts often attempt to account for this heterogeneity by determining what factors moderate this variability in effect sizes.

Most statistics that measure the association between two groups (e.g., Student’s t, Pearson’s r, an F from a 2-group ANOVA, 2, z, are mathematically related. Just as the length of a piece of paper is the same whether it is measured in inches or in centimeters, the size of the association between two groups (i.e., the effect size) is the same regardless of the metric used to report it.

In this research we used Pearson’s r (a correlation coefficient) as our measure of effect size. We chose r since it has a number of desirable features:

first, most researchers are familiar with r and have a sense of what various r values mean;
second, once a mean effect size has been computed for a set of studies, the mean r can be squared to form the coefficient of determination (r2), indicating the proportion of variance in one variable that is accounted for by the other.

In order to get a rough idea of what constitutes a small, medium, and large effect size, Cohen (1988) has suggested the following guidelines:

r =.1 is small,
r =.3 is medium, and
r = .5 is large.

The National Probability Studies

Most researchers who have reviewed literature on CSA have focused on clinical and legal samples. These samples are, however, biased in that their results cannot be assumed to generalize to the population of persons who have experienced CSA (Pope & Hudson, 1995). In order to avoid sample bias, we focused first on studies designed to be nationally representative.

To reduce the chance of sampling bias we attempted to collect and analyze every study, whether published or unpublished, that met the following four criteria:

(1) the study had to be designed to be nationally representative,
(2) the study had to separately report data on males and females,
(3) the data had to examine the association between CSA and some form of functioning (typically psychological or sexual) via a comparison with a control group (so that the data could be quantified as r), and
(4) the study had to be obtainable (obtainable means that it was available in one of the author’s university libraries, through inter-library loan, at the Library of Congress in Washington, DC, or was sent to the authors in response to our requests).

Our efforts yielded five studies.

The study by Bigler (1992) was a mail survey that obtained a 33% response rate with a total N = 14 [?]. This was the smallest study both in terms of response rate and number of subjects. The definition of CSA was typical of those used in scientific research over the past decade, including any sexual experiences that occurred prior to age 18, whether contact or noncontact (e.g., exhibitionism), in which the other person was a family member or five or more years older or where there was coercion. Sexual functioning was assessed.

The study by Boney-McCoy and Finkelhor (1995) was a telephone survey that obtained an effective response rate of 72% with a total N = 98 [?]. This study differed from the other national studies in that the subjects were aged 10 to 16 at the time of the survey rather than being ‘adults’. In this study telephone contact was made with a subject’s caretaker, and if permission was granted, the phone was passed to the subject. As a consequence of this methodology, under-reporting of CSA incidents may have occurred due to the lack of privacy in many or most interviews, thus making the findings of this study difficult to generalize. This study also differed from the other national studies in that CSA experiences were limited to those of an unwanted nature. Post-traumatic-stress-disorder symptomatology, sadness, depression, and trouble with teachers were assessed.

The study reported by Finkelhor et al. (1989) was an LA Times telephone poll which obtained a response rate of 76% and a total N = 18 [?]. CSA was defined as any childhood sexual experience now seen as abuse. Marital disruption, satisfaction with intimate relationships with the other gender, and attitudes potentially influenced by CSA experiences were assessed.

The study by Laumann et al. (1994) utilized face-to-face interviews and obtained a response rate of 79% with a total N = 19 [?]. This study was the only national study which defined CSA using a biological definition of child, as well as requiring physical contact; they defined CSA as physical contact experiences occurring prior to puberty with someone past puberty. Psychological and sexual functioning were assessed.

The study by López et al. (1995) utilized face-to-face interviews (to gather data on the occurrence of CSA) and self-administered-questionnaires (to gather psychological adjustment data) in Spain; the other national samples were based in the U.S.A. The questionnaires were returned by mail, creating two levels of responding; this study obtained an effective response rate of 40% with a total N = [??]. The definition used was typical for CSA research; it included all types of sexual experiences prior to age 17 where there was coercion or an age difference of five or more years. Psychological adjustment was assessed.

Prevalence of CSA

Due to the potential sampling biases in the Bigler study (i.e., low N & low response rate) and the Boney-McCoy and Finkelhor study (i.e., potential lack of privacy during interview), we regard the remaining three studies as probably the best indications of the prevalence of CSA in the population at large. [...] the prevalence rate for males runs from 12% to 15% and the rate for females runs from 17% to 28%. The differences between studies may be accounted for by the differing ages used in the definitions of CSA. Using ‘typical’ definitions, 15% for males and 25% for females are probably close to the true prevalence rates.

Meta-analyses of the national data

For each study, an effect size was computed for each gender measuring the magnitude of the association between CSA and adjustment. These effect sizes were then meta-analyzed by gender to obtain an unbiased effect size estimate of the association between CSA and adjustment in the male and female populations at large [...].

The unbiased effect size estimate for males (combined N = [??]) was ru=.07 with a 95% confidence interval running from r = .04 to r =.10; the effect sizes were homogeneous. The unbiased effect size estimate for females (combined N = [??]) was ru=.10 with a 95% confidence interval running from r = .07 to r = .12; the effect sizes were heterogeneous.

The female effect size from the Boney-McCoy and Finkelhor study was a statistical outlier; its removal left the remaining effect sizes homogeneous.

[The slightly larger effect size found in the Boney-McCoy and Finkelhor study may be due to factors such as:

(a) it used subjects aged 10-16,
(b) the definition of CSA was restricted to unwanted experiences, or there may have been sampling biases due to a lack of privacy.]

Meta-analyses of the college data

Having established a good estimate of the typical association between CSA and adjustment in the population at large, we now turn to data from studies that used college students as their subjects.

As with the national studies, we tried to obtain all studies, both published and unpublished, that met our inclusion criteria.

We were able to obtain CSA prevalence data on 72 samples; the average prevalence rates were 14% for males and 27% for females, very similar to the 15% and 25% figures found in the national studies.

We were also able to obtain effect sizes for the association between CSA and adjustment from 54 college samples. When meta-analyzed, the unbiased effect size estimate between CSA and adjustment was ru = .09; separated by gender, the effect size for males was ru = .07 and for females was ru = .10. When rounded to two digits as reported here, these effect sizes are identical to those found in the meta-analyses of the national data [...].

Given the identical effect sizes and nearly identical prevalence rates, it appears that findings from college student samples are relevant to the population at large.

To determine if the association between CSA and adjustment varied as a function of type of adjustment, we conducted 18 meta-analyses, each on a different symptom domain [...]. The 18 domains assessed were: trouble with alcohol, anxiety, depression, dissociation, eating disorders, hostility, interpersonal sensitivity, locus of control, obsessive compulsiveness, paranoia, phobias, psychoticism, self-esteem, sexual adjustment, social adjustment, somatization, suicidal ideation, and measures of wide psychological adjustment (typically the ‘total’ or ‘global’ score on a multidimensional instrument).

For each of these 18 meta-analyses, data from a minimum of 5 studies with a combined N  = 24 [?} were included (the largest had 22 studies with a combined N = 78 [?]). The unbiased effect size estimates ran from a high of ru = .13 (anxiety) to a low of ru = .04 (self-esteem, locus of control). After the removal of some statistical outliers, the effect sizes for 15 of the 18 symptom domains were homogeneous.

Estimating the impact of controlling for relevant confounds

Having established a small but statistically significant association between CSA and adult adjustment, an important question was whether this association was causal in nature; that is, did CSA cause poorer adjustment?

To begin to address this question we gathered data from the college samples on the association between CSA and six domains that loosely fall under the rubric of family environment. The six domains were: nonsexual abuse & neglect, adaptability, conflict & pathology, family structure, support & bonding, and traditionalism.

The mean association between CSA and family environment was r = .13, indicating a confound. We then examined the mean association between family environment and symptoms, and found the mean association to be r = .29.

Because family environment tends to be a more stable and long-lived phenomenon than CSA, and given that family environment was approximately 10 times better at predicting adult adjustment than the presence of CSA (CSA - symptom r2 = 0.8%, family environment-symptom r2 = 8.4%), and given that CSA and family environment were confounded, it is reasonable to posit that certain attributes of the family environment predispose young people to have experiences classifiable as CSA; consequently, some of the association between CSA and symptoms found in the previously presented analyses was likely to have been spurious, due to family environment rather than experiencing CSA.

To estimate the effect of controlling for one or two family environment variables, we conducted an analysis using the mean effect sizes reported earlier (CSA - symptoms r = .09, family environment - CSA r = .13, family environment - symptoms r = .29). In doing so we were conducting a conservative test of the extent to which the CSA - symptom association would decrease while controlling for one or two family environment variables (if the data were sufficient to control for all six of the environment variables, the effects of statistical control would be even more pronounced, hence the value about to be presented may be considered an upper-bound on any causative association between CSA and symptoms). The results of this analysis reduced the association between CSA and symptoms from r = .09 to r = .05.

Summary

Analyses of the data gathered to be nationally representative imply that the association between having an experience classifiable as CSA and having poorer adult functioning is small. The size of this association is identical to that found when using college samples, suggesting that data based on college samples may be generalizable to the population at large. The college data imply that this small association applies regardless of the symptom under examination. Finally, the reliable confounding of CSA and family environment imply that this small association cannot be assumed to represent a causal association.

Discussion

McMillen et al. (1995, p. 1037) asserted that the "experience of child sexual abuse is a traumatic event for which there may be few peers". Results from the current meta-analyses show that this commonly expressed view is highly exaggerated. 

This exaggeration can be accounted for by several factors. 

First, as Money (1979) has pointed out, we as a society tend to associate harmfulness with wrongfulness in sexual matters. Consequently, since we view CSA as morally and legally wrong, we assume that it is harmful, regardless of its actual impact. 
.
Second, as is often the case in sexuality research on taboo behaviors, most findings in this area have been based on clinical and legal samples—samples that are highly biased and not generalizable. 
.
Third, overstatement of the potential impact of CSA has been fueled by ideological and political motives (Okami, 1990, 1994); purporting to act to ‘save the children’ has long been an effective route to election, re-election, and the implementation of legal policy. 
.
Fourth, in part due to the large number of studies based on clinical and legal samples available today, many literature reviews on the assumed effects of CSA have been based exclusively on data from these biased samples (e.g., Beitchman et al., 1991; Black & DeBlassie, 1993; Glod, 1993; Kendall-Tackett et al., 1993; 

Watkins & Bentovim, 1992) and many others have been based primarily on these biased samples

 (e.g., Beitchman et al., 1992; Briere & Elliot, 1994; Briere & Runtz, 1993; Browne & Finkelhor, 1986; Constantine, 1981; Kilpatrick, 1987; Mendel, 1995; Urquiza & Capra, 1990), 

further promoting a belief that CSA is a "special destroyer of adult mental health" (Seligman, 1994, p. 232). 

Finally, perhaps the common belief in the harmfulness of these acts leads to a selective focus on experiences reported or perceived as negative.

In college sample studies where the subjects were asked to retrospectively report their reactions at the time of the experience, 

11% of women and 37% of men reported that the experience was a positive one, 
16% of women and 29% of men reported that it was neutral, and 72% of women and 
33% of men reported that it was negative. 

Clearly, experiences defined as CSA by researchers are characterized by a broad range of subjective perceptions. Although many experiences are perceived as negative, and many lead to harm, it is unrealistic to expect all such experiences to produce harm regardless of the child’s or adolescent’s perception of the experience.

Recommendations for future research

Future research into the possible effects of experiences classifiable as CSA should control for relevant family environment variables. At a minimum the presence of physical and emotional abuse and neglect should be controlled for. 

Future research will make a significant contribution if it additionally controls for a variety of stable attributes of the family that may be causatively associated with adult adjustment.

Future research should also report clear quantifiable data both before and after statistical control. Today, many researchers who conduct studies with statistical control clearly report their findings before control, but fail to precisely report them after control, often including only a brief comment in the text such as: ‘after statistical control, none of the relations remained significant.’ If the data before control are worth reporting precisely (e.g., as t values, rs, z, etc.), then the data after control are worth reporting precisely. Only with precise reporting will future meta-analysts be able to precisely quantify the findings of multiple studies.

Future research should also separately analyze positive, neutral, and negative experiences. Since CSA typically is defined to include all of these experiences, it is important to break it down to find out what the differing associations are, because it cannot be assumed that positive and negative experiences have the same consequences.

Finally, if the goal is to determine what is harmful, rather than to find out the associations between CSA and adjustment when using socio-legal definitions, then we must explore the use of definitions that are likely to measure harmful events. We suggest that based on the current data and the significant homogeneity among studies using differing (socio-legal) definitions, the following should be explored:

(1) Define ‘child’ based on biological childhood (e.g., less than puberty, or less than age 12); studies can then see if there is a difference between the experiences of biological children and those of young adults (i.e., adolescents).
(2) Physical contact should be required or physical contact abuse should be separately analyzed from noncontact abuse (rather than lumping these experiences together, potentially obscuring a difference).
(3) The presence of force or significant coercion should be separately analyzed from experiences where these factors are not present; a lack of willing participation may be the key element.

Appendix

Studies included in the meta-analyses of national probability samples 
Studies included in the meta-analyses of college student samples 
References 

 

[Back to the Rind et al. Files]

[Start]     [Appendix]