2. Previous Literature Reviews

Vorige Omhoog Volgende

[ Vorige = Back    -    Omhoog = Up    -    Volgende = Next ]

Go to >

          Qualitative Literature Reviews

23

Go to >

                   Causality

23

Go to >

                   Intensity

23

Go to >

                   Gender Equivalance 

24

Go to >

          Limitations of Qualitative Literature Reviews

24

Go to >

                  Sampling biases

24

Go to >

                  Subjectivity and imprecision

24

Go to >

          Quantitative Literature Reviews

24

Go to >

          Synthesis of Quantitative Reviews

25

Go to >

                    Causality

25

Go to >

                    Pervasiveness

25

Go to >

                   Intensity

26

Go to >

                   Gender Equivalence

26

 

Numerous literature reviews have appeared over the last 15 years that have attempted to synthesize the growing body of empirical investigations of CSA effects and correlates

(e.g., Bauserman & Rind, 1997 ; Beitchman, Zucker, Hood, DaCosta, & Akman, 1991 ; Beitchman et al., 1992 ; Black & DeBlassie, 1993 ; Briere & Elliot, 1994 ; Briere & Runtz, 1993 ; Browne & Finkelhor, 1986 ; Constantine, 1981 ; Glod, 1993 ; Jumper, 1995 ; Kendall-Tackett, Williams, & Finkelhor, 1993 ; Kilpatrick, 1987 ; Mendel, 1995 ; Neumann, Houskamp, Pollock, & Briere, 1996 ; Rind & Tromovitch, 1997 ; Urquiza & Capra, 1990 ; Watkins & Bentovim, 1992 ).
These reviews have not been unanimous in their conclusions. Below, we examine their conclusions regarding the four commonly assumed properties of CSA discussed previously. First we examine the qualitative literature reviews, then the fewer and more recent quantitative (i.e., meta-analytic) reviews.


Qualitative Literature Reviews

Causality.

Some qualitative reviewers have been cautious regarding the issue of causality

(e.g., Bauserman & Rind, 1997 ; Beitchman et al., 1991 ; Beitchman et al., 1992 ; Constantine, 1981 ; Kilpatrick, 1987 ),
arguing that the reliable confounding of family environment problems with CSA prevents definitive conclusions regarding the causal role of CSA in producing maladjustment.

Other reviewers, although recognizing limitations of correlational data, have nevertheless argued that causality is the likely state of affairs

(e.g., Briere & Runtz, 1993 ; Glod, 1993 ; Urquiza & Capra, 1990 ).
Some reviewers have strongly implied that CSA causes maladjustment by consistent use of phrases that imply causation (e.g., "effects of CSA," "impact of CSA") and by not addressing alternative explanations (e.g., third variables, such as family environment) that could account for the CSA-maladjustment link
(e.g., Black & DeBlassie, 1993 ; Briere & Elliot, 1994 ; Kendall-Tackett et al., 1993 ; Mendel, 1995 ; Watkins & Bentovim, 1992 ).


Pervasiveness.

Some reviewers have concluded that CSA outcomes are variable, rather than consistently negative

(e.g., Bauserman & Rind, 1997 ; Beitchman et al., 1991 ; Beitchman et al., 1992 ; Browne & Finkelhor, 1986 ; Constantine, 1981 ; Kilpatrick, 1987 ).
  • Constantine concluded that there is no inevitable outcome or set of reactions and that responses to CSA are mediated by nonsexual factors.
  • Beitchman et al. (1991) argued that the prevalence of negative outcomes may be overestimated because of overreliance on clinical samples.
  • Browne and Finkelhor noted that only a minority of both sexually abused (SA) children seen by clinicians and adults with a history of CSA show serious disturbance or psychopathology.
  • Other reviewers, however, have implied in several different ways that CSA effects or correlates are prevalent among persons with a history of CSA.

    First, some reviewers have claimed to have written "comprehensive" reviews of the literature or summaries of "what is currently known"

    (e.g., Briere & Elliott, 1994 ; Briere & Runtz, 1993 ; Glod, 1993 ; Urquiza & Capra, 1990 ; Watkins & Bentovim, 1992 );
    their conclusion that CSA is associated with numerous symptoms then implies that negative correlates are prevalent.

    Second, some reviewers have argued that studies showing a large percentage of asymptomatic persons with a history of CSA can be explained by factors such as insensitive measures or insufficient time for symptoms to have developed

    (e.g., Briere & Elliot, 1994 ; Kendall-Tackett et al., 1993 ).
    This argument implies that negative effects are prevalent, even if not yet observed in many cases.

    Third, some reviewers have not discussed limitations on generalizability from their sample of (usually clinical) studies to other CSA populations

    (e.g., Black & DeBlassie, 1993 ; Kendall-Tackett et al., 1993 ; Mendel, 1995 ),
    again implying that findings of negative correlates apply to the entire population of persons with CSA experiences.

    Intensity.

    Some reviewers have concluded that the intensity of CSA outcomes varies, rather than usually being intensely negative

    (e.g., Bauserman & Rind, 1997 ; Beitchman et al., 1991 ; Beitchman et al., 1992 ; Browne & Finkelhor, 1986 ; Constantine, 1981 ; Kilpatrick, 1987 ).

  • Browne and Finkelhor noted that SA persons in community samples tend to be either normal or only slightly impaired on psychological measures.
  • Constantine and Kilpatrick found that negative outcomes were often absent in SA persons in nonclinical samples.
  • Other reviewers, however, have implied that negative psychological effects are frequently intense by describing the "extreme psychic pain" ( Briere & Runtz, 1993 , p. 320) or the "pronounced deleterious effects" ( Mendel, 1995 , p. 101) that CSA is assumed to produce.

    Some reviewers have further implied the intensity of CSA effects or correlates by presenting long lists of severe disorders (e.g., posttraumatic stress, self-mutilation) associated with CSA

    [Page 24]

    (e.g., Black & DeBlassie, 1993 ; Briere & Elliot, 1994 ; Briere & Runtz, 1993 ; Glod, 1993 ; Kendall-Tackett et al., 1993 ; Mendel, 1995 ; Urquiza & Capra, 1990 ; Watkins & Bentovim, 1992 ).


    Gender equivalence.

    Several reviewers have argued that the data are insufficient to address the issue of gender differences in outcomes

    (e.g., Beitchman et al., 1991 ; Beitchman et al., 1992 ; Browne & Finkelhor, 1986 ).
  • Constantine (1981) concluded that girls react more negatively than boys, attributing this difference to differences between girls' and boys' CSA experiences. 
  • Bauserman and Rind (1997) , on the basis of a review of college, national, and convenience samples, concluded that reactions and outcomes for boys are more likely to be neutral or positive than for girls.
  • Many reviewers, however, have concluded or implied that CSA is an equivalent experience for boys and girls in terms of its negative impact

    (e.g., Black & DeBlassie, 1993 ; Briere & Runtz, 1993 ; Mendel, 1995 ; Urquiza & Capra, 1990 ; Watkins & Bentovim, 1992 ).
  • Black and DeBlassie stated that CSA "has, at the very least, an equivalent impact on males and females" (p. 128).
  • Watkins and Bentovim claimed that one prevalent myth about CSA is that boys are less psychologically affected than girls.
  • Mendel dismissed as an "exercise in futility" efforts to determine whether boys or girls are more adversely affected by CSA, and concluded that CSA "has pronounced deleterious effects on its victims, regardless of their gender" (p. 101).

  • Limitations of Qualitative Literature Reviews

    The qualitative literature reviews present a mixed view of causality, pervasiveness, intensity, and gender equivalence. This inconsistency suggests the need for additional work in synthesizing the literature.

    Two other considerations also indicate such a need:

  • sampling biases in many of the qualitative reviews, and
  • the vulnerability of qualitative reviews to subjectivity and imprecision.
  • Sampling biases.

    Qualitative literature reviews have been primarily based on clinical or legal samples, which cannot be assumed to be representative of the population of persons with a history of CSA

    (Bauserman & Rind, 1997 ; Okami, 1991 ; Rind, 1995 ).
    Some reviews were based exclusively or almost exclusively on clinical and legal samples
    (e.g., Beitchman et al., 1991 ; Black & DeBlassie, 1993 ; Glod, 1993 ; Kendall-Tackett et al., 1993 ; Mendel, 1995 ; Watkins & Bentovim, 1992 ).
    Others were based on a majority of clinical and legal samples but included a sizable minority of nonclinical and nonlegal samples
    (e.g., Beitchman et al., 1992 ; Briere & Elliott, 1994 ; Briere & Runtz, 1993 ; Browne & Finkelhor, 1986 ; Constantine, 1981 ; Kilpatrick, 1987 ; Urquiza & Capra, 1990 ).
    Only one of the qualitative reviews cited previously (Bauserman & Rind, 1997 ) included a majority of nonclinical and nonlegal samples.

    Drawing conclusions from clinical and legal samples is problematic not only because these samples cannot be assumed to be representative of the general population, but also because data coming from these samples are vulnerable to several biases that threaten their validity ( Pope & Hudson, 1995 Rind & Tromovitch, 1997 ).

    Okami (1991) studied adults who had experienced CSA as negative, neutral, or positive. Negative responders included both clinical and nonclinical subjects. Clinical negative responders showed substantially more pronounced adjustment problems than nonclinical negative responders.
    Okami argued that clinical participants with negative CSA experiences constitute the negative extreme of CSA outcomes.

    Pope and Hudson argued that reliance on clinical samples is problematic for several reasons.

  • One problem is information bias, in which clinical patients, in a search for the causes of their problems (termed effort after meaning ), are more likely than nonclinical participants to recall events that can be classified as CSA, thus inflating the CSA-maladjustment relationship.
  • Another potential bias is investigator expectancies (cf. Rosenthal, 1977 ), in which clinical researchers who believe that CSA is a likely cause of their patients' difficulties may transmit this expectancy to patients, thereby increasing confirming responses.
  • Finally, Pope and Hudson argued that causality cannot be inferred from clinical samples because CSA and family disruption are highly confounded in this population ( Beitchman et al., 1991 ; Ney, Fung, & Wickett, 1994 ).
  • Legal samples are also likely to contain the more serious cases, limiting their generalizability.


    Subjectivity and imprecision.

    Qualitative reviews are entirely narrative and therefore susceptible to reviewers' own subjective interpretations ( Jumper, 1995 ). Reviewers who are convinced that CSA is a major cause of adult psychopathology may fall prey to confirmation bias by noting and describing study findings indicating harmful effects but ignoring or paying less attention to findings indicating nonnegative outcomes.

    For example, Mendel (1995) focused on results from Fromuth and Burkhart's (1989) midwestern sample of males to argue that boys are harmed by their CSA experiences but paid little attention to the southeastern sample of males reported in the same article, for whom all CSA-adjustment correlates were nonsignificant.

    In a quantitative review, the latter sample would typically have received more weight because it had 30% more participants than the former.

    Even when study results generally indicate statistically significant differences in adjustment between CSA and control participants, summarizing this information alone is inadequate ( Rosenthal & Rosnow, 1991 ).

    The sizes of these differences (i.e., effect sizes) are also important; these effect sizes can be used to assess the intensity of CSA effects or correlates (Rind & Tromovitch, 1997 ). Only quantitative (i.e., meta-analytic) reviews can provide this important information.


    Quantitative Literature Reviews

    Three recent quantitative literature reviews ( Jumper, 1995 ; Neumann et al., 1996 ; Rind & Tromovitch, 1997 ) represent a significant advance in assessing CSA-adjustment relations because they all

  • (a) included a sizable proportion of nonclinical and nonlegal samples and
  • (b) avoided subjectivity and imprecision by using meta-analysis.
  • Meta-analysis is a statistical technique in which statistics from a set of studies are converted to a common metric (e.g., standard normal deviate z s, Cohen's d s, Pearson's r s), which are then combined into one overall statistic that can be used to

  • (a) infer whether one variable (e.g., CSA) is significantly associated with another (e.g., adjustment) and
  • (b) estimate the strength of this association (Rind & Tromovitch, 1997 ).
  • Common metrics such as d and r are referred to as effect sizes and can be interpreted as assessing the size of the difference of some attribute between two groups or the magnitude of association between two variables.

    As a guideline,

    [Page 25]

    Cohen (1988) has suggested that small, medium, and large effect sizes correspond, respectively, to

  • d s of .20 [small], .50 [medium], and .80 [large], and to
  • r s of .10 [small], .30 [medium], and .50 [large].
  • Thus, these reviews were well suited to examining not only whether control and SA respondents differ in adjustment, but also to what extent they differ.

    Two of the reviews ( Jumper, 1995 ; Rind & Tromovitch, 1997 ) were also able to precisely compare the genders in terms of CSA outcomes.

    Jumper (1995) examined CSA-adjustment relations from 26 published studies with 30 samples. Of 23 samples with identified sources,

  • 30% were clinical,
  • 26% community,
  • 22% student, and
  • 22% mixed.
  • Thus, at least 48% of the identified samples were nonclinical and nonlegal. Most samples (83%) consisted of female participants.

    Using a weighted means approach ( Shadish & Haddock, 1994 ), Jumper meta-analyzed effect sizes ( r s) across samples for

  • depression,
  • self-esteem, and
  • symptomatology (i.e., psychological difficulties other than depression and self-esteem problems).
  • The overall magnitude of the relation between CSA and symptomatology was of medium size, r = .27.

  • Community ( r = .29) and
  • clinical samples ( r = .27) were similar in magnitude,
  • but student samples were substantially lower ( r = .09).
  • For self-esteem,

  • community ( r = .34) and
  • clinical samples ( r = .36) were also similar,
  • whereas student samples were much lower ( r = -.02).
  • For depression,

  • the community samples ( r = .17) were closer to
  • student ( r = .09) than
  • clinical samples ( r = .34).
  • Jumper concluded that the student samples were anomalous, possibly because symptoms had not yet manifested at college age.

  • The CSA-symptomatology relation was the same for men ( r = .29) and women ( r = ..26);
  • the CSA-self-esteem relation, however, was lower for men ( r = -.02) than women ( r = .24).
  • On the basis of the symptomatology results, which were derived from nearly twice as many samples as the self-esteem results, Jumper concluded that SA men and women do not differ in terms of subsequent psychological adjustment.

    Neumann et al. (1996) examined CSA-adjustment relations using 38 published studies consisting exclusively of female participants, half of which were based on nonclinical samples.

    These researchers computed an overall effect size ( d ) for each study (i.e., a study-level effect size) and then meta-analyzed them, obtaining a small to medium weighted mean effect size ( d = .37).

    Using Rosenthal's (1984) formula, and assuming a 19% CSA prevalence rate for women in the general population based on Rind and Tromovitch's (1997) estimate, we converted this d to an r . The obtained result ( r = .14) was considerably smaller than Jumper's estimate of r = .27.

    Neumann et al. also found that the magnitude of the effect sizes differed between nonclinical ( d = .32) and clinical ( d = .50) samples. Converting these values to r with the procedure described previously yielded r = .12 and .19, respectively.

    Thus, whereas Jumper found that community and clinical samples were similar in terms of mean effect sizes, Neumann et al. found that nonclinical samples had a lower mean effect size than clinical samples.

    This difference might be due to the fact that Neumann et al.'s nonclinical samples included student samples (but see below).

    Finally, Neumann et al. found virtually identical effect sizes for samples with a mean age of 30 or below ( d = .39) and above 30 ( d = .40), casting doubt on Jumper's speculation that her student results might be attributable to a lack of time for symptoms to manifest.

    Rind and Tromovitch (1997) examined CSA-outcome relations from 7 male and 7 female national probability samples from the United States, Canada, Great Britain, and Spain. These results are especially important for estimating population parameters because these samples were all chosen to be representative of their national populations.

    Rind and Tromovitch meta-analyzed mean effect sizes from each sample (i.e., sample-level effect sizes) separately by gender and found that the magnitude of CSA-adjustment relations was small for both men ( r = .07) and women ( r = .10). These mean effect sizes were not statistically different.

    For self-reports of CSA effects, significantly more women (68%) reported the presence of some type of negative effect at some point after their CSA experience than did men (42%); the size of this difference was small to medium ( r = .23).

    Self-reports in Baker and Duncan's (1985) national study in Great Britain suggested that lasting negative effects for SA persons are rare: 13% for women and 4% for men. [*]

    [* Inserted by Ipce:] In the presentation of Rind c.s. in Rotterdam, The Netherlands, December 18, 1998, they have presented the next figures from Baker & Duncan (1985):
     
    Baker & Dunca's (1985) questions  Males (n-79)  Females (n-119) 
    Permanent damage          4%        13%
    Harmful at the time,
    but no lasting effects 
          33%      51%
    No effect         57%      34%
    Improved quality of life         6%      2%

    Several of the national studies also examined third variables that might account for CSA-adjustment relations.

    In one study, greater sexual activity in adulthood was confounded with CSA ( Laumann, Gagnon, Michael, & Michaels, 1994 ).

    In two others ( Boney-McCoy & Finkelhor, 1995 ; Finkelhor, Hotaling, Lewis, & Smith, 1989 ), most CSA-adjustment relations remained statistically significant after controlling for several possible confounds.

    However, nonsexual abuse and neglect variables were not held constant in these analyses, weakening any causal interpretations because CSA often occurs along with physical abuse or emotional neglect ( Ney et al., 1994 ) and because CSA-adjustment relations have been shown to disappear when these factors are held constant (e.g., Eckenrode, Laird, & Doris, 1993 ; Ney et al., 1994).

    Finally, Rind and Tromovitch reviewed the results of another national study that found that SA girls tended to have disruption in their family, school, and social environments both before and after their CSA experience ( Ageton, 1988 ), weakening causal interpretations regarding CSA effects in the general population.


    Synthesis of the Quantitative Reviews

    Causality.

    All three reviews expressed caution regarding causal inferences about CSA-adjustment relations.

    Jumper (1995) noted that researchers need to differentiate between effects related to CSA and those related to other traumatic events, and to control for family variables.

    Neumann et al. (1996) argued that third variables such as other forms of maltreatment may be responsible for the CSA-adjustment relation, and that most studies in their review did not consider the possible role of family dynamics.

    About 72% of the studies in Jumper's review were also reviewed by Neumann et al., suggesting that most of Jumper's studies also did not consider the role of family environment. 

    Rind and Tromovitch (1997) found that the studies in their review usually did not use statistical control, and when they did, it was inadequate.

    Thus, a quantitative review of studies using statistical control of important potential confounds (e.g., family environment) has yet to be done and is needed to address the issue of causality.


    Pervasiveness.

    Only Rind and Tromovitch's (1997) review

    [Page 26]

    presented data relevant to how widespread negative outcomes are in the population of persons with a history of CSA. Their findings suggest that lasting negative effects are rare, but these results are based on only one study ( Baker & Duncan, 1985 ). These considerations point to the need for further attention to this issue.

    Intensity.

    The meta-analytic reviews were especially useful for assessing the intensity of CSA correlates or effects, indicated by weighted mean effect sizes.

    Neumann et al. (1996) and Rind and Tromovitch (1997) found that the magnitude of the relation between CSA and adjustment in the general population is small.

    In contrast, Jumper's (1995) meta-analysis of community samples suggests that the magnitude of the CSA-adjustment relation in the general population is medium in size and equivalent to that in the clinical population.

    To investigate this discrepancy, we examined the community samples used by Jumper.

    For symptomatology, Jumper reported the following effect sizes:

  • Bagley and Ramsay (1986) , r = .13;
  • Mullen, Romans-Clarkson, Walton, and Herbison (1988) , r = .16;
  • Murphy et al. (1988) , r = .13;
  • Peters (1988) , r = ..30;
  • Stein, Golding, Siegel, Burnam, and Sorenson (1988) , r = .31 for the female sample and r = .37 for the male sample.
  • We calculated the effect sizes for these samples and obtained, respectively, r s = .21, .16, .16, .14, .15, and .12. Because we obtained substantially lower effect sizes in the last three samples, we asked an expert meta-analyst to calculate these values independently; his calculations confirmed ours. [*1] 

    [*1] Ralph Rosnow served as the expert meta-analyst. In an attempt to resolve our discrepancies with Jumper, we contacted her. She informed us that her meta-analysis came from her master's thesis and that all her data and calculations were in storage in a different part of the country. She therefore advised us that she would be unable to help but nevertheless suggested that we proceed with our report, mentioning that we were unable to resolve the discrepancies with her.

    We meta-analyzed the recomputed effect sizes, obtaining a small weighted mean effect size ( r = .15), which is consistent with the results of the other two meta-analytic reviews.

    We next examined the four community samples in Jumper's meta-analysis of depression and the three in her meta-analysis of self-esteem. Although we obtained similar effect sizes, two of the samples used in each meta-analysis (from Hunter, 1991 ) were not valid community samples.

    Hunter recruited participants through newspaper advertisements and community notices asking for volunteers who were "sexually molested as children" (p. 207). The recruitment method suggests a convenience sample rather than a community sample; further, the notice wording was likely to attract volunteers who had more negative experiences.

    Thus, the results of Jumper's meta-analyses of depression and self-esteem for community samples have limited generalizability.

    In sum, the quantitative reviews indicate that in the entire population of persons with a history of CSA, the magnitude of the CSA-adjustment relation is small, implying that CSA does not typically have intensely negative psychological effects or correlates.

    The results from the Neumann et al. (1996) and Rind and Tromovitch (1997) meta-analyses, as well as results from the recomputed meta-analysis of Jumper's (1995) community samples, suggest that the student population is not anomalous with respect to CSA-adjustment relations. Instead, it appears that the clinical population is anomalous.


    Gender equivalence.

    Using the recomputed effect sizes for Jumper's (1995) community samples, we recalculated the weighted mean effect sizes for male and female participants for symptomatology and found r s = .11 and .22, respectively, compared with reported values of r = .29 and r = .26, respectively.

    These revised results suggest a sex difference. Rind and Tromovitch's (1997) meta-analysis did not reveal a sex difference in CSA-adjustment relations

    (although the direction of the mean effect sizes was consistent with greater problems for SA women),
    although it did show a sex difference in self-reported effects.

    Each meta-analysis was based on only a small number of male samples

    (Jumper used four; Rind and Tromovitch used five for CSA-adjustment relations and three for self-reported effects).
    Neumann et al. (1996) examined only female samples. The mixed results regarding CSA-adjustment relations, along with the small number of samples used, suggest the need for a more extensive meta-analytic examination of sex differences.

     

    Vorige Omhoog Volgende