Results 2

Statistical Analysis: Sources of Attenuation Choice of Effect Size

Content References Tables

Vorige Start Volgende

Choice of Effect Size 
Other sources of attenuation 
Base-rate differences for men and women 
The cost of dichotomization 
Interpretation of effect sizes 

Examination of Moderators 
Contact versus non-contact CSA 

Family Environment 

Qualitative Analysis 
Length of time to recover 
Effects on current life 
Initial emotional reactions 
Differences between subjective and objective measures 

[Page 720 continued]

Choice of Effect Size 

Throughout their article Rind et ai. (1998) used the Pearson correlation coefficient (r) as the effect size statistic rather than Cohen's measure of effect size (d), the standardized mean difference, between the CSA and control groups. 

Although there is no problem with the conversion of d to r per se, one must be careful with its interpretation because unlike d, r is affected by the base rates of the independent variable when it is dichotomous. 

As the prevalence of the independent variable approaches either 0 or 1, the value of r, for any fixed value of d, approaches 0. For example, an effect size of d = 0.8 is one that Cohen (1977) described as "large," and indeed with this effect size, comparing a randomly selected CSA and non-CSA sample, 71% of the time the CSA group is "more distressed" than the non-CSA group.

However, as Figure 1  [→ ] [*↓ ]  demonstrates, this "large" effect size (based on interpretation of d) translates to an r anywhere from 0 (when the prevalence of CSA approaches 0% or 100% ) to a maximal value of .37 (slightly above medium on the r scale) when the prevalence is 50%. Thus, Rind et al.'s decision to use r as the effect size statistic for studies that mainly used group comparison designs with very unequal proportions in the respective groups created a situation in which clinically large effects could be represented by what appear to be small r values.

[* ↑ ] Figure 1. The relationship between r and d, based on the prevalence of the independent variable in the population studied. In this case, P represents the proportion of those in the sample with reporting child sexual abuse. Assuming that the two samples have the same variance, the exact relationship between r and d is: 
= d / (d2 + 1 / P.P')˝, where P is the probability that X= 1, and P' is the probability that X= 0 in the population sampled. As P approaches either 0 or 1, the value of r, for any fixed value of d, approaches 0.

Other sources of attenuation

Hunter and Schmidt (1994) listed 10 sources of attenuation in meta-analyses (see Table 4), and the current study had problems with attenuation from most of these sources. For example, there was attenuation due to dichotomization for the obvious reason that the studies were using dichotomies created from an underlying continuum. The attenuation due to error of measurement was large because of the definitional problems described above, and the use of biased samples led to attenuation due to range restriction. 

Some of these limitations were also mentioned by Rind et al. (1998). For example, they noted that correlations attenuate with small base rates (p. 41) and dichotomized or unreliable variables (p. 41) and that semi-partial correlations will be conservative (p. 33). Although Rind et at. could not correct for all sources of attenuation, many sources, like attenuation due to base-rate differences and dichotomized variables, are easily correctable.

Base-rate differences for men and women

Although Rind et al. ( 1998) correctly noted that correlations attenuate with small base rates and that this problem was larger for male samples than female ones (p. 41 ), they chose not to correct for this problem, suggesting the results would be too small to bother with. They stated,

"the attenuation is small in absolute magnitude for small effect sizes. For the small CSA-symptom effect size estimates obtained in the current review, adjusted effect size estimates, based on a 50-50 split increase at most [italics added] by .03 (based on formulas provided by Rosenthal & Rosnow, 1991)" (p. 41).
At the same time, Rind et al. concluded that "the relation between CSA and adjustment problems was generally stronger for women than men" (p. 42).

Although we did not find such formulas in Rosenthal and Rosnow (1991), an appropriate formula can be found in Becker (1986). [*4]

[*4] Becker (1986) noted that it may or may not be appropriate to use such adjustments, depending on the construct being investigated. However, if one is attempting to compare the size of the r values for men versus women, it would seem imperative to base such comparisons on r values unaffected by base rates. Furthermore, given that the research question examines whether CSA is harmful, we take the position that it is also appropriate to correct the point-bi-serial r for unequal sample sizes in the CSA and non-CSA groups. If the question of interest were to know, from a public health perspective, the effect of CSA on the entire population, r attenuations due to low base rates of CSA would accurately reflect this type of effect. However, the question of interest in this meta-analysis is a quasi-experimental one, and thus correction for unequal cells in the control (CSA) and experimental (non-CSA) group is warranted.

When the point-bi-serial correlation is given directly, the formula for the corrected point-bi-serial r, rc, is rc = r / sqrt (r2 + 4pq(1-r2), where r is the uncorrected bi-serial correlation and p and q are the proportions in each of the two groups.

By applying this formula to the overall ruS (listed in Rind et al., 1998, Table 4) [*5] of .07 for men and .10 for women, and using the weighed prevalence rates of CSA that Rind et al. reported for men and women (14% and 27%, respectively), one can calculate rcs of .11 for women and .10 for men.

[*5] These adjustments should in practice be made at the level of individual study effect sizes. We report them here on overall effect sizes as an illustration of how they change, and do so differentially for men and women.

Thus, the overall effect size for men increases by .03, making the effect sizes for men and women, based on corrected point-bi-serial correlations, nearly identical.

The authors admitted this fact in a subsequent article in which they defended their use of correlations and their decision not to correct for attenuation, stating,

"we directly addressed this issue in our review. ... We noted that effect size attenuation small in magnitude for small effect sizes - i.e., an r = .07 based on a 14-86 split [CSA prevalence for male samples] increases at most by .03 (to r = .10) in a 50-50 split" (Rind, Tromovitch, & Bauserman, 2000a, p. 29).

By applying the same correction, for the four attenuated ru values of .13 (for anxiety, depression, paranoia, and psychotic symptoms) in their Table 3 (Rind et al., 1998, p. 32), one would obtain a corrected point-bi-serial rc of .16 for both men and women.

However, in practice, the base-rate problem should be attended to at the sample level, where, contrary to Rind et al.' s claim, effect sizes in some cases would have increased by much more than .03.

For example, we examined Bendixen, Muus, and Schei (1994), whose male sample reported a CSA prevalence of only 3%. Our unbiased effect size estimate for the male sample (ru = .09) was similar to the one reported by Rind et al. (ru = .08). However, after correction for attenuation due to the low base rate of CSA, the effect size estimate was .26 (an increase of .17).

These results show that using correlations without correcting for unequal base rates of the independent variable minimized effect sizes and created the appearance of gender-related differences in CSA adjustment, when effect sizes for men and women were actually equivalent.

The cost of dichotomization

Effect sizes were also attenuated because the original studies used dichotomies created from an underlying continuum. Although it is clear that CSA is a problem with varying degrees of severity (Bouvier et al., 1999; Rind et al., 1998; Wyatt & Peters, 1986), most studies dichotomized CSA. Point-bi-serial rs should be used only when there is a true dichotomy (e.g., men vs. women). When a dichotomy is created on the basis of an underlying continuum (as we would argue is the case with CSA), [*6] power to detect differences is lost and the appropriate r is the bi-serial correlation, or the tetrachoric correlation if the dependent variable is also dichotomized (see Hunter & Schmidt, 1990).

[*6] Throughout their article, Rind et al. ( 1998) also conceptualized CSA as a continuous variable by noting that there are different types and levels of abuse. For example, in comparisons to national samples, the authors discussed CSA in terms of four levels of severity (i.e., exhibitionism. fondling, oral sex, and intercourse).

Cohen (1983) noted that even with a 50-50 distribution (which minimizes the attenuation) there is a 20% reduction in r (or a multiplicative factor of .80) if one variable is dichotomized, and a 40% reduction (or multiplicative factor of .64) if both are. With one variable dichotomized and a 90-10 split (as was frequently the case with the male samples), the attenuation is 41%.

The corrected point-bi-serial rcs of .11 for women and .10 for men can be converted to the bi-serial r by dividing the correlation by the multiplicative factor of .80. This yields an overall effect of .14 for women and .13 for men (see Table 5). Biserial rs for the individual symptoms reported by Rind et al. (1998) in their Table 3 (e.g., anxiety, depression, paranoia. and psychotic symptoms) could be as high as .20 for both men and women.

Interpretation of effect sizes

In addition to failing to correct for effect size attenuation, Rind et al. (1998) interpreted the magnitude of the overall CSA-adjustment effect using the concept of variance accounted for, or r2. They wrote, "according to Cohen's (1988) guidelines; in terms of variance accounted for, CSA ac- counted for less than 1% of the adjustment variance" (p. 31).

However, texts on meta-analysis, including the one most relied on by Rind et al. (Rosenthal, 1984), consistently discourage the use of the r2 statistic as an effect size estimate. For example, Rosenthal wrote,

[Page 722]

"because of the probability of seriously misinterpreting its practical significance ... we shall not use r2 as an effect size estimate" (p. 24).

It is interesting to note that in an earlier article, Rind and Tromovitch ( 1997) recognized this fact when they wrote,

"r2 is not currently considered to be a good measure of effect size by leading methodologists because it underestimates this metric" (p. 240).

To more accurately reflect the real world significance of an effect size, Rosenthal and Rubin (1982) developed the binomial effect size display (BESD). Although the BESD is most appropriate for situations with approximately equal size groups with similar variances, some examples presented in Rosenthal and Rubin ( 1982; see also Rosenthal, 1984) make it quite clear that effect sizes associated with CSA may be much more meaningful than argued by Rind et al. (1998).

For example, López et ai. (1995) reported the lifetime prev- alence of diagnosed depression to be 19% in students who reported CSA and 12% in students who did not. After correcting for the low base rate of CSA, L6pez et al.'s results correspond to an r = .10, or an r2 of .01.

One percent of explained variance might not be very impressive in theory, but in practice it represents a 7% increase of diagnosed depression in the Spanish population as a whole and could mean up to a 10% absolute increase in the prevalence of depression for an at-risk population (e.g., fraternal twins of individuals diagnosed with depression; see Bertelsen, Harvald, & Hauge, 1977).

Because Rind et al.'s findings show CSA to be highly prevalent and associated with a wide range of problems, the relatively small r values they reported could represent enormous costs to society in terms of human suffering and lost productivity (see Dallam, 2001, for a review).

Examination of Moderators

Rind et al. (1998) performed multiple regression analyses on three variables (contact, gender, and consent) to determine whether they accounted for some of the variability found in sample-level effect sizes.

Contact versus non-contact CSA

Rind et al. (1998) reported that they found no significant differences in outcome for contact versus non-contact forms of abuse. This is a surprising finding which bears closer examination, as it runs contrary to the results of numerous nonclinical studies, which have reported higher rates of adjustment problems for individuals who acknowledged more severe forms of abuse

( e.g., Bennett, Hughes, & Luke, 2000; Bifulco, Brown, & Alder, 1991; Cheasty, Clare, & Collins, 1998; Fergus- son, Horwood, & Lynskey, 1996; Fleming, Mullen, Sibthorpe, & Bammer, 1999; Mullen, Martin, Anderson, Romans, & Herbison, 1993).

After carefully reviewing the data, we found reason to believe that significant differences in outcome may have been present, but not demonstrable because of the inability of moderator analysis to compensate for attenuated and overlapping variables.

In all, 73% of the studies used by Rind et al. (1998) included non-contact forms of abuse in their definitions of CSA, and the vast majority of these did not differentiate outcomes for contact and non-contact forms of abuse in their data. Thus, rather than making direct comparisons, the moderator analysis compared studies whose definition was restricted to contact CSA with studies that included varying mixtures of both contact and non-contact forms of abuse. This, of course, diminished the precision by which Rind et al. could measure differences and, as they correctly noted (p. 33), likely underestimated the relationship.

The possibility that such differences would have been demonstrated if direct comparisons had been possible is supported by the results of those studies in the primary data set that did directly compare outcomes for contact versus non-contact forms of abuse (e.g., Bergdahl, 1983; Collings, 1995; Haugaard & Emery, 1989; Peters & Range, 1995; Rew, Esparza, & Sands, 1991).

The results of these studies were remarkably consistent: The adjustment of respondents reporting only non-contact forms of abuse usually did not differ significantly from that of their controls. However, when investigators restricted their definition of CSA to experiences involving physical contact, significant differences between abused students and non-abused controls emerged. The impact of definitional differences was most clearly demonstrated by Haugaard and Emery (1989), who systematically examined the effect of four levels of CSA definition on psychological and social adjustment. A dose - response relationship was reported, with increasing levels of severity associated with poorer adjustment.

It is also important to remember that the effect Rind et al. (1998) reported for contact versus non-contact CSA, sr(41) = .15, was a semi-partial r, which means that it was the effect after accounting for every other variable in the equation (see p. 33 for a description of this analysis).

Considering that the model included gender, contact, consent, and two interactions (Contact x Gender and Consent x Gender} - variables for which there was no doubt a problem of multi-collinearity - and that the correlation suffered from many of the sources of attenuation listed in our Table 4, finding a semi-partial r of .15 suggests a robust effect. However, because the correlation was not significant, Rind et al. concluded that "contact and Contact x Gender were not related [to effect sizes]" (p. 33} - implying that the null hypothesis had been proven. In our opinion, a more appropriate wording would have been to simply say that the effect sizes in question were not statistically significant, while noting the appropriate limitations of the data and their methods.

In summary, Rind et al.'s (1998) finding that physical contact during sexual abuse was not related to adjustment should be viewed cautiously. Problems with multi-collinearity and attenuation, along with the lack of differentiation in the original studies and the limited ability of moderator analysis to compensate for this lack, make it very questionable whether Rind et al.'s findings apply to more serious forms of CSA.


In their abstract, Rind et al. (1998) stated, "Self-reported reactions to and effects from [italics added] CSA indicated that ... men reacted much less negatively than women" (p. 22), suggesting that gender moderated adjustment. This finding, however, is not supported by their moderator analysis of gender. The global unbiased effect sizes Rind et al. reponrtd for males and females were .07 and .10, respectively, which moderator analysis revealed "was non-significant" (p. 33). Moreover, as we previously demonstrated and as Rind et al. (2000a, p. 29) have admitted, correcting for base rates would have increased effect sizes for male samples from .07 to at least .10.


It appears that some of the evidence Rind et al. (1998) used to support their claim of significant gender differences in CSA adjustment was based on their finding that consent was an important moderator of adjustment in males. The authors later summarized their findings, stating, "We showed that for boys in nonclinical populations, willing relations are generally experienced positively or neutrally and are not associated with maladjustment" (Rind, Bauserman, & Tromovitch, 1999, p. 2185). To

[Page 723]

evaluate the validity of this claim, it is important to carefully examine how Rind et al. conceptualized and performed their analysis of consent.

It should be noted that whereas some studies asked respondents to report only unwanted experiences, few of the other studies included in the authors' primary data set asked participants to specify whether the sexual experiences reported were wanted. Thus, Rind et al. (1998) attempted to examine a variable that was, for the most part, never directly measured. To get around this problem, the authors assumed that detectable portions of "willing" sex were included in any study that did not explicitly state that the student should report only unwanted experiences.

Rind et al. then divided studies into two groups: unwanted and all levels of consent. Studies that required all CSA to be unwanted were placed in the unwanted group. All other studies, most of which used standard definitions of CSA based predominantly on the age difference between the child and the perpetrator, relationship to the perpetrator, or the use of force, were placed in the all-levels-of-consent group. Rind et al. then attributed all differences between the two groups to the presence or absence of willing experiences, or what they termed consent.

Rind, Tromovitch, and Bauserman (2000b) argued that their use of consent in this manner was "scientifically justified because ... the same construct appeared in many of the primary studies" (p. 108).

However, we were able to find only one study that directly asked students whether they had participated in the CSA experience willingly. Bergdahl (1983) studied 430 college women, 154 of whom reported some form of CSA, which was broadly defined as any type of sexual experience (including non-contact) with an adult while they were under the age of 18. Thirty-five respondents (23% of abused women) placed a check mark next to an item that indicated a willingness to participate in at least one of the sexual experiences they had reported. It is important to note that the mean age at first occurrence of willing participation was 15.8 years, which was higher than the mean age of any other group. Because of how Bergdahl defined CSA, a consensual peer relationship between a 17-year-old girl and an 18-year-old boy was categorized as CSA. Most other studies in the all-levels-of-consent group required a 5-10-year age difference before a relationship was categorized as CSA. As a result, Bergdahl provides little information on whether the other groups in the all-levels-of- consent group could be expected to contain detectable amounts of willing experiences.

When Rind et al. (1998) compared 12 samples in which CSA was defined so as to only include unwanted CSA with 35 samples of all levels of consent, the results were not significant (see Rind et aI., 1998, Table 4). The authors then looked at the interaction of gender and consent (see Rind et aI., 1998, Table 5). When they compared female samples, women in the group that putatively included wanted sex appeared to be less well adjusted. However, when four samples of males reporting only unwanted CSA were compared with 10 male samples categorized as including all levels of consent, the latter group had a smaller combined effect size (suggesting better adjustment) and the lower limit of the 95% confidence level was zero. Despite the fact that effect sizes for male samples were attenuated, Rind et al. took this result to mean that "consenting" males were as well adjusted as their control counterparts.

We obtained the 14 studies of male college students used to form this conclusion and replicated the moderator analysis. Our results did not support Rind et al. ' s ( 1998) conclusions. In Table 6 we have listed the prevalence rates, definitional criteria, dependent measures, and effect sizes we obtained for the male samples analyzed.

On close inspection we could find no evidence to suggest studies included in the all-levels-of-consent group included much in the way of willing experiences.  For example, male students in the all-levels-of-consent group did not report higher prevalence rates, which we would expect to find if in addition to unwanted CSA, the original investigators had captured significant portions of wanted experiences.

In addition, many studies in the all-levels-of-consent group defined experiences after age 12 or 13 as CSA only if they were unwanted (Finkelhor, 1979; Fishman, 1991; Greenwald, 1994) or defined experiences as CSA only if they were accomplished through the use of force (Greenwald, 1994; Predieri, 1992; Preuss, 1988). These studies appear to be unlikely sources for finding significant amounts of wanted experiences. Moreover, the moderator analysis was confounded, as effect sizes reflected not only differences in definitional criteria but also differences in the instruments being used and the outcomes being studied. [*7]

[*7] The problem with confounding can best be illustrated by examining Greenwald (1994), a study from the male all-levels-of-consent group with a negative effect size (Rind et al., 1998, reported an effect size of -.09), suggesting better adjustment for abused men.
However, this low effect size cannot be attributed to wanted experiences, as Greenwald reported that over two thirds of the abuse reported by men was accomplished by "more than a moderate degree of force" (p. 114).
In this case, the most obvious reason for the lower effect size was the outcome being studied. Greenwald studied a new and un-validated construct called romantic and sexual self-esteem (Reed, 1988) and found that for males, scores correlated positively with increased sexual activity and negatively with a need for and sense of mutuality in a relationship.
Thus, rather than measuring psychological adjustment in males, this measure appeared to serve as a proxy for increased sexual activity among abused males.

When we calculated effect sizes for the objective measures (see Table 6) we found effects sizes that were slightly higher than reported by Rind et al. (1998, Appendix), even before correcting for base-rate differences. [*8]

[*8] For Fishman (1991), Rind et al. (1998) reported a negative effect size of -.04 (suggesting better adjustment for abused males). It should be noted that Fishman compared abused and non-abused students on only one objective measure, sexual self-esteem and dysfunction, and found abused males to be less well adjusted on every item.
For instance, in his abstract, Fishman stated that males who reported CSA "were more likely to ... identify higher levels of sexual dysfunction, and lower levels of sexual self-esteem" (p. viii). On the basis of the results he reported for this measure (see Fishman, 1991. p. 150), we calculated an uncorrected effect size of .07.

Table 7 compares the global effect sizes we computed for both groups before and after correcting for base rates (to be conservative, we did not adjust for attenuation due to dichotomization of the independent variable).

 It is worth noting that even before correcting for attenuation due to low base rates, the confidence interval (CI) for the male all-levels-of-consent group (95% CI = .03-.19) did not include zero, suggesting that contrary to Rind et al.'s conclusion, men in this category did differ from their controls. In addition, the confidence levels for the male unwanted and all-levels-of-consent groups overlapped, suggesting that the two groups were not significantly different. It should be noted that Rind et al.'s own data (see Rind et al., 1998, Table 5) also showed overlapping CIs between effect sizes for these groups, although they disregarded this result in forming their conclusions.

[Page 724]

We also failed to find support for Rind et al.'s (1998) conclusion that "adjustment was associated with level of consent for men, but not for women" (p. 34), [*9] Because the effect sizes Rind et al. estimated for men were more attenuated than those for women, we would expect the range of effect sizes for men to be lower. After correcting for base-rate differences, effect sizes estimates for male (rc = .11) and female (rc = .12 [*10] samples in the all-levels-of-consent group were nearly identical.

[*9] Rind et al.'s (1998) data also showed overlapping CIs between effect sizes for males and females in both groups (see Rind et al., 1998, Table 5), undermining their own argument that there is a gender difference.

[*10] The correction for females was made by using the unbiased effect size estimate provided by Rind et al. (1998, Table 5) for females in the all-types-of-consent group (ru = .11) using the overall prevalence rate that Rind et al. reported for female samples of 27%.

In summary, the lack of any direct measurement of "willing" experiences and the presence of confounding variables raise doubts as to what Rind et al.'s (1998) analysis of "consent" was actually measuring. Despite their claim of significant findings, the differences that Rind et al. reported for effects between male unwanted and all-levels-of-consent groups were marginal at best.

Our own analysis of the data failed to reveal any significant effects for gender or what Rind et al. called "consent." Our findings also failed to support Rind et al.'s conclusion that willing sex with adults is not harmful to children.

Family Environment

Despite finding that students who reported a history of CSA were less well adjusted in 17 of the 18 types of psychological adjustment examined, Rind et al. ( 1998) suggested that the

[Page 725]

relationship may be spurious because of the confounding of CSA with family dysfunction. Although there has been debate in the literature concerning whether some effects associated with CSA are artifacts of family dysfunction (see Briere & Elliot, 1993; M. R. Nash, Hulsey, Sexton, Harralson, & Lambert, 1993), we would argue that Rind et al.'s conclusions in this regard are stronger than their data warrant.

First, the stronger relationship between environment and CSA symptoms may merely be an artifact of the way CSA was treated statistically. In the vast majority of the original research Rind et al. ( 1998) analyzed, CSA was treated as a dichotomous variable. At the same time, family environment was usually measured by well-validated measures such as the Family Environment Scale, which has multiple continuous scales. By comparing the predictive ability of one dichotomous variable (CSA) with multiple continuous variables, the deck was stacked against sexual abuse before any statistical analysis occurred. [*11]

[*11] Rind et al. (1998) attempted to counter this criticism (regarding artificial dichotomization) by noting that a few studies have used continuous scales for CSA and found that CSA did not explain adjustment variance above and beyond that explained by various family environment factors in a hierarchical regression.
We find this argument unconvincing given that no reliability or validity data were presented for these scales and that the studies mentioned were retrospective in nature and thus subject to the partialing effect previously noted.

Second, the strong relationship between family dysfunction and distress in college samples using broad definitions of CSA is not surprising. As Briere (1988) noted,

"nonclinical samples (especially involving students) may not include many severe cases of sexual or physical victimization ... and thus any 'abuse effects' may be trivial in comparison to the variance accounted for by the 'control' variables" (pp. 83-84).

This contention is supported by the results of several studies included in the current meta-analysis. For instance, Collings (1995) reported that outcomes associated with non-contact forms of CSA were not significant after controlling for the influence of dysfunctional parenting behaviors. However, CSA involving physical contact was associated with elevated scores on all subscales of the Brief Symptom Inventory even after controlling for family dysfunction. Similar results were reported by Edwards and Alexander (1992), who found that

"psychological distress was significantly predicted by sexual abuse severity and marginally predicted by parental conflict" (p. 315).

Finally, the fact that both family dysfunction and abuse exposure are related to subsequent distress does not allow the inference that the shared variance is due to family dysfunction rather than abuse. The data examined by Rind et al. (1998) came predominantly from the partialization of family environment based on either analysis of covariance (ANCOVA) or hierarchical regression.

For a covariate to be effective in controlling for an extraneous variable, the covariate must be linearly related to the dependent variable and be unaffected by the treatment or independent variabies (Hinkle, Wiersma, & Jurs, 1988; Porter & Raudenbush, 1987).

Given that the relevant independent variable in the studies reviewed by Rind et al. was CSA, to meet this condition one would have to assume that sexual abuse was uncorrelated with family dysfunction. This is clearly not the case, as families that abuse their children are by by definition dysfunctional, and children are often targeted by sex offenders because of parental non-involvement (Budin & Johnson, 1989; Christiansen & Blake, 1990).

When a covariate is significantly correlated with both the dependent variable(s) and the independent variables, the results of an ANCOVA may be misleading (Pedhazur, 1982), as the analysis will partial out not just variance attributable to the covariate but also variance shared by the covariate and the independent variable. That is, while partialing out the effects of family environment one is also removing the effects of CSA (see Briere, 1988; Briere, 1992; Briere & Elliot, 1993). [*12] Furthermore, when there is a high degree of multi-collinearity, standard errors will be larger, parameter estimates will be unstable, and p values will be increased (e.g.. Wampold & Freund. 1987).

[*12] Rind et al. (1998) attempted to counter this type of criticism by pointing out (p. 41) that in one study physical abuse predicted adjustment above and beyond family environment even though CSA did not.

Although Rind et al. (1998) briefly addressed some of these issues, they suggested that these concerns "do not appear to be problematic in the current review". (p. 43). We disagree. Their statistical methods could not compensate for the limitations of retrospective, quasi-experimental data, and their conclusions are at variance with those of numerous large scale nonclinical studies whose methods are more appropriate for disentangling the effects of CSA from those of family dysfunction (see Table 8).

Moreover, some investigators have found that family dysfunction and abuse each exert a negative influence independent of the other and that these effects appear to be additive (e.g., Felitti et al., 1998; Yama, Tovey, Fogas. & Teegarden, 1992). Finally. we would submit that much of the etiological picture of the complexities of human functioning is missed when researchers attempt to reduce something as broad as psychological maladjustment to uni-variate causation.

Qualitative Analysis

In the qualitative section of their article, Rind et al. (1998) examined college students' retrospective reports of their initial emotional reactions to CSA, how long it took them to recover from the experience, and whether they continued to be affected.

Length of time to recover

On the basis of the data summarized in their Table 8, Rind et al. (1998) concluded that "when negative effects occur, they are often temporary" (p. 37). This conclusion, however, is severely limited by the fact that only three studies examined this variable, and these studies included very little in the way of serious abuse. In fact, the largest of these studies, Landis (1956), did not even study CSA. Instead, Landis examined college students' encounters with "a sexual deviate" (p. 93). Because no age limits were placed on when the experience occurred, a significant

[Page 726]

percentage of the experiences occurred after age 17. Consensual experiences were also included, and the vast majority (83%) of the experiences reported by male participants were homosexual advances during older adolescence or adulthood, which usually occurred in public places and were easily rebuffed.

Most of the experiences reported by the female participants involved exhibition. The majority of these women reported they immediately left the scene. None of the men reported being raped, and less than 2% of the women reported attempted or committed rape. Although the students had an overwhelming negative perception of the experience, as might be expected, few reported that the episode had a long-terrn negative effect on their life.

Similarly, West and Woodhouse (1993) noted that most of the incidents recalled by their male sample were either relatively trivial homo-sexual approaches that were soon rejected or minor indecencies, that boys looked on as unimportant. The most relevant of the three studies, C. Nash and West (1985), used a broad definition of CSA and reported a prevalence rate of 54%. Of note, 22% of students

[Page 727]

who acknowledged abuse reported that they were still being adversely affected by the experience.

It is obvious that these data are unsuitable to answer the question of how long it takes a child to recover from CSA. The most scientifically valid method of determining the length of time to recover from CSA is to follow abused children prospectively. Longitudinal studies of sexually abused children contradict Rind et al. ' s ( 1998) findings, as they show that the pattern of recovery is different for different symptoms and often unpredictable (Calam, Home, Glasgow, & Cox, 1998; McLeer et al., 1992; Tebbutt, Swanston, Oates, & O'Toole, 1997).

For instance, Tebbutt et al. (1997) reassessed 68 children 5 years after their abuse was discovered. There were no significant changes in depression, self-esteem, or behavior over 5 years. At follow-up, 43% of the children were sad or depressed, 43% had low self-esteem, and 46% had behavioral dysfunction. Whereas some children improved, a nearly equal number deteriorated, with no clear pattern of change. Moreover, family variables did not predict outcome.

Effects on current life

On the basis of self-reported effects from seven studies, Rind et al. (1998) concluded that "lasting negative effects are not prevalent" (p. 37) among college students who report CSA.

Our review of the original studies revealed that this conclusion could be considered valid only for those studies looking at milder forms of abuse or, as in the case of Landis (1956), experiences that do not fulfill normal criteria for CSA. Very different results were reported by studies that included more serious CSA.

For instance, Fishman (1991), reported that "47% [of male respondents] stated that this experience has had a negative effect on their life, and 23% reported that the CSE [child sexual experience] had a negative effect on their current sexual life" (p. 162; Rind et al. misreported these results as 27% and 13%, respectively). Fischer (1991) reported that the majority of students reporting CSA (93% of females and 79% of males), reported that the experience had caused or was continuing to cause them stress.

Initial emotional reactions

Rind et al. (1998) reported that "two-thirds of SA men and more than one-fourth of SA women reported neutral or positive reactions," which they suggested is "inconsistent with the assumption of pervasive and intense harm" (p. 44).

On examination, we found there were numerous problems with how Rind et al. reported and interpreted the results of the original studies. For example, in three instances we found that Rind et al. misreported the participants' reactions to CSA (Fishman, 1991[*13]; Brubaker, 1991, 1994 [*14]). In each case, the misreporting inflated positive reactions and simultaneously diminished negative ones.

[*13] For Fishrnan (1991), Rind et al. (1998, Table 7) reported that 27% reported positive immediate reactions, 43% neutral, and 30% negative. However, Fishman reported "53% of these men felt the CSE [child sexual experience; was negative at the time] ... 12% positive and 35% neutral" (p. 162).

[*14] In Brubaker (1991 and 1994), 83% and 92% of the participants reported feeling initially upset, with the largest category reporting having felt extremely upset (35% and 40%, respectively). These figures contrast greatly with those reported by Rind et al. (1998, Table 7): that 40% and 27% of the participants felt neutral or positive at the time of the abuse.

Although males were more likely than females to ascribe neutral or positive initial feelings to the experience, our review of the original data revealed a number of mediating variables that should be considered when interpreting these findings.

For example, Rind et al. (1998) reported only males' initial reactions to the abuse. Studies that compared initial reactions with current ones found that males' initial reactions tended to become more negative over time. Therefore, whereas almost 38% of males surveyed by Urquiza (1989) reported a positive or mostly positive initial reaction to CSA, only 15% indicated that they still felt positively about the experience as an adult.

Another factor that mediated reactions was the type of abuse experienced. As Rind et al. noted, "college men and women ... tended to have different experiences; SA women experienced close family CSA more than twice as often as SA men and experienced force about twice as often" (p. 43).

In general, males reacted most negatively to encounters involving other males, force, or close relatives (Finkelhor, 1979; Fishman, 1991; Fromuth & Burkhart, 1989; O'Neill, 1991; Risin & Koss, 1987; Urquiza, 1989). Positive reactions were most apparent when the experience was with an unrelated female, someone who was closer in age to the boy, and when the boy felt that he had initiated the contact (Condy, Templer, Brown, & Veaco, 1987). In other words, the experiences most likely to be perceived positively by males were those that most closely resemble the types of sexual contacts that a boy might have with a peer.

Differences between subjective and objective measures

 Studies that compared qualitative perceptions with the results of objective measures demonstrated that self-reported ratings of adjustment may not provide the best criterion of harm in abused men.

For example, three studies, Fishman (1991), Fromuth and Burkhart (1989), and Urquiza (1989), compared men's overall subjective ratings of current adjustment with scores on objective measures and found that abused men tended to present themselves as doing much better than their objective scores indicated.

For instance, Fishman (1991) reported that, compared with their non-abused peers, men acknowledging a history of CSA rated themselves higher on a subjective assessment of overall sexual adjustment and current adjustment. However, an objective measure of sexual dysfunction showed these same men to be less well adjusted in every area assessed. Moreover, nearly three times as many of the abused men reported having been treated for emotional problems as their non-abused counterparts (29% vs. 10%, respectively), Urquiza ( 1989) noted similar findings. Most men in his study reported neutral reactions to their abuse; however, compared with their non-abused peers, over two times as many abused men reported using illegal drugs, three times as many had sought therapy for emotional problems, and five times as many had attempted suicide.

In summary , although men tended to report less negative perceptions of CSA and attributed less harm to the experience, objective measures did not support their appraisals. A more correct statement of Rind et al. (1998) findings is that men often claimed to be unaffected by CSA but simultaneously demonstrated negative effects similar to those displayed by their female counterparts. In this respect, the findings of the current meta-analysis are consistent with those of numerous other studies and reviews, which have reported that men frequently deny that their abuse had any serious effect on them yet at the same time display poorer adjustment than their non-abused peers

(e.g., Briggs & Hawkins, 1996; Fondacaro, Holt, & Powell, 1999; Holmes & Slap, 1998; Myers, 1989; Varia, Abidin, & Dass, 1996; Watkins & Bentovim, 1992)


Content References Tables

Vorige Start Volgende