[Page 745 continued]
That correlation is not equivalent to causation is a fundamental tenet of research methodology. Using terms such as effect or impact is inappropriate in correlational research (especially when important confounds are present), as these terms imply causation (R. Rosenthal. 1994). Despite this, Dallam et al. (2001) used such causal language throughout their critique. They criticized our suggestion that the relation between CSA and adjustment might be spurious because of confounding, arguing that we made this suggestion "despite finding that students who reported a history of CSA were less well adjusted in 17 of 18" measures (Dallam et al., 2001, p. 725).
However, this argument is unconvincing because these correlations were all small in magnitude and very similar to
each other, as one might expect of correlated measures reflecting a single underlying construct: their consistency therefore does not add weight to the argument for causation.
Ondersma et al. (2001) questioned whether efforts to control for FE are appropriate at all, claiming this is problematic when "environment and CSA events are so thoroughly intertwined" (p. 709). However, the overall relation between FE and CSA was .13, which, although representing confounding, does not constitute being thoroughly intertwined. We now respond to arguments that our approach to statistical control was invalid, and present new data and analyses that further question the proposition that the CSA-adjustment association can safely be assumed to be causal.
Dallam et al. (2001) argued that the "deck was stacked against" (p. 725) CSA when compared with FE in accounting for adjustment because of the way CSA was treated statistically. CSA was often a dichotomous measure, whereas FE was often continuous, and they argued that our discussion of continuous CSA measures was unconvincing because these measures had no reliability or validity data and were retrospective.
These latter criticisms apply to all CSA research, including that cited by our critics to bolster their arguments, and can hardly be selectively applied to our studies.
We cited Wisniewski (1990) as a study that used a continuous measure for CSA yet found FE substantially more important in predicting current adjustment. Wisniewski constructed the continuous measure from degree of felt victimization and negative reactions, a face-valid measure of actual severity that would also have face validity for adult-adult sex, as opposed to severity based on a hierarchy of types of sex from exhibitionism to intercourse. Moreover, she used a large sample (N = 3,187) of female college students drawn from 32 colleges and universities fairly representative of those across the United States, increasing the generalizability of her inferences to the college population. In a path analysis, she found that family violence, not CSA, was behind current adjustment problems. Further, her zero-order effect size for CSA (r = .11) was consistent with most other college studies.
In their national study, Laumann et al. (1994) created a more conventional severity index based on a hierarchy. They found that this continuous measure did not improve prediction over their dichotomous measure. Like the Wisniewski study, the Laumann et al. study is important because of its representativeness (the U.S. adult population) and large sample size. These findings, cited in our original review, are an important counterweight to the "stacked deck" argument.
Dallam et al. (2001) went on to argue that nonclinical samples do not have many severe cases. If severity means intercourse, we have already shown that this claim is incorrect. They next argued that our statistical control came predominantly from analyses of covariance (ANCOVAs) and hierarchical regression, where the covariate (FE) should be unaffected by the independent variable (CSA) for valid control. From this, they claimed that CSA and FE would have to be uncorrelated.
This last claim is incorrect. Partialing a third variable from two primary variables, both of which are correlated with it, is valid if one assumes the third variable is either a mediator of the two primary variables (i.e., caused by one and causing the other in turn) or a common cause of both of them (Pedhazur, 1997). Partialing is problematic when, for example, the third variable is affected (caused) by the independent variable, which affects the dependent variable, and the dependent variable is also affected by the third variable. Then, partialing may remove too much of the relation between the primary variables - for example, if CSA causes FE, which causes symptoms, and CSA directly causes symptoms, then controlling for FE removes part of the CSA effect.
The latter is the critical case and is what Briere (1988) and others have argued happens in the real world: CSA causes harm and disrupts families, which results in more harm. We addressed this issue in our original article (Rind et al., 1998), presenting evidence to argue that, although Briere's concern may be valid in incest cases, it was unlikely to apply in the bulk of the cases we examined in the college samples, where close family CSA was rare and CSA involving a parent even rarer.
Finkelhor et al.'s ( 1990) national sample findings are also relevant to this issue. They conducted a series of analyses that indicated that "unhappy family life is a true risk factor [for CSA] and not simply a distorted perception that a victim develops as a result of having been abused" (p. 24). In other words, contrary to Briere's concern, their findings favored causality going from FE to CSA and not the reverse. This is consistent with Pedhazur's criteria for valid partialing.
Dallam et al. (2001) next claimed that our findings are at variance with other large-scale nonclinical studies "whose methods are more appropriate for disentangling the effects of CSA from those of family dysfunction" (p. 725, italics added) and presented these studies in their Table 8.
We find it curious that the methods of ANCOVA and hierarchical regression were labeled inappropriate in the studies we used but were considered more appropriate when used in 5 of 13 samples cited by Dallam et al. (see our Table 6).
In any case, the studies cited are quite mixed in terms of degree of including relevant covariates: In half the samples, these control variables were weak (see Table 6), not including factors such as physical abuse and emotional neglect. Effects of statistical control were mixed and were compromised in the studies with weak covariates.
On a further note, we computed sample-level effect sizes for each sample, combining emotional and behavioral measures.
As with the results of the high school and junior high samples discussed previously, these results are consistent with those from the college and national samples.
Clearly, meta-analyses of nonclinical samples show that both men and women with a history of CSA are slightly less well adjusted than controls. However, in our society, minors in general who have precocious sex - for example, willing peer intercourse at a young age - are also less well adjusted
Early sex is non-conventional in our society, but not cross-culturally (Ford & Beach, 1951), and reflects a package of
emotional, behavioral, familial, and social correlates that are maladaptive according to our society's standards.
Jessor et al. (1983) described this package in terms of behavior problem theory, which is composed of three systems that promote non-normative behavior.
Early sex is seen as originating in these systems rather than causing them. This point is relevant to cause and effect regarding CSA. .
Using the four studies just cited, we meta-analyzed various correlates of precocious consensual sex that did not involve CSA (consensual is the authors' term). Table 7 provides the results for each study.
Precocious sex was associated with
Jessor et al. (1983) found that precocious sex was strongly associated with general problem behavior (r = .45).
Resnick and Blum (1994) found that precocious sex, defined as consensual intercourse before age 10 (the authors' term), was associated with mental health problems (r = .10). In short, this research shows that early sex is associated with a wide range of problems, emotional and behavioral, and/or non-conventional behavior. Of importance is that there is no indication or argument by these researchers that precocious sex causes other problems.
Instead, it is a correlate of other problems with the common cause of personality, home, and social environmental factors.
Behavioral and emotional correlates of CSA look very similar to those of peer precocious sex. Therefore, causal inferences about CSA effects need to take into account the findings from research on precocious sex with peers. Controlling for FE alone in CSA research may in fact insufficiently control for other key influences on the CSA-adjustment relationship. Problem behavior theory questions even further the degree to which CSA, particularly a single event occurring many years previously and not perceived as especially important at the time, can be labeled the sole or even primary cause of current adjustment difficulties.
Finally, it is important to add that correlational research has serious limitations in drawing causal inferences. This is much less of a problem, however, when zero-order correlations are non-significant or valid statistical control eliminates significant relations. In these situations, inferences can be made, especially when results are replicable, because the absence of correlation implies a lack of causation.
When correlations remain significant after statistical control, however, causation cannot be firmly assumed, given the problem of unexamined third variables. Future research, to more adequately address causality in the case of CSA and adjustment, needs to focus more on alternative research strategies, such as
Regarding the last of these, the Dinwiddie et al. (2000) and Kendler et al. (2000) studies using twin controls are a start, one finding an elimination of significant CSA-symptom relations and the other not. Such research would be improved if it included cases in which the twins perceived themselves as willing or assenting in the CSA, as the two studies just cited included unwanted CSA only yet assumed to be addressing the more general issue of socio-legal CSA. Also, such studies could examine precocious sex as discussed above to further clarify the issue of CSA and causality.