Ruling-In vs. Ruling-Out

Ultimately, all indicator lists of sexual abuse flounder because they neglect considerations of both “diagnostic sensitivity” and “diagnostic specificity” (Campbell, 1997, p. 4). Diagnostic sensitivity refers to how accurately an indicator identifies a population exhibiting some characteristic – such as sexual abuse. Diagnostic specificity refers to how accurately an indicator identifies the population that does not exhibit that characteristic.

Interestingly enough, the diagnostic sensitivity of an indicator might be quite good; but its specificity can be very poor.  For example, consider the following classification rule based on age and gender: Diagnose all males over 50 years of age as having prostate cancer. In fact, these are very sensitive indicators of prostate cancer. More than 99% of the population suffering from prostate cancer are males over 50 years of age. Thus, this decision making rule will miss less than one percent of the population with prostate cancer.

The obvious flaws of this classification rule only become evident when we consider that portion of the population that does not suffer prostate cancer. In fact, only 5% of the male population over 50 years of age ever develops prostate cancer. Therefore, this classification rule will be mistaken approximately 95% of the time. Nevertheless, this 95% rate of error does not reveal itself until we address considerations of diagnostic specificity.

Indicator lists of sexual abuse, however, preoccupy themselves excessively with considerations of diagnostic sensitivity while naively overlooking considerations of diagnostic specificity. In other words, the professionals who rely on these indicator lists are so determined to find evidence suggesting sexual abuse, they overlook evidence indicating no abuse. Because children who have not been sexually abused frequently exhibit the behaviors specified by various indicator lists, relying on these lists leads to an inordinate number of misclassifications. 

In particular, indicator lists result in many more false positive classifications (concluding a child has been sexually abused, when in fact no abuse occurred) than false negative classifications (concluding a child has not been sexually abused, when in fact the abuse did occur).

Overview of Effects

It is not uncommon for sexually abused children to exhibit an absence of clinically significant psychopathology. For example, only 27% of a sample of sexually abused children showed clinically significant psychopathology (Gomes-Schwartz, Horowitz, & Cardarelli, 1990). This varied according to the age of the child with 17% of the pre-school, 40% of the 7 to 13 year olds, and 8% of the adolescents designated as seriously disturbed. The effects ranged from the complete absence of symptoms to pervasive and serious problems.

Subsequent reviews have emphasized:

“In summary, the findings of the various studies reviewed indicated that molestations that included a close perpetrator; a high frequency of sexual contact; a long duration; the use of force; and sexual acts that included oral, anal, or vaginal penetration lead to a greater number of symptoms for victims. Similarly, as all the studies that included these variables indicated, the lack of maternal support at the time of disclosure and a victim’s negative outlook or coping style also led to increased symptoms.  The influence of age at the time assessment, age at onset, number of perpetrators, and time elapsed between the end of abuse and assessment is still somewhat unclear at the present time and should be examined in future studies on the impact of intervening variables” (Kendall-Tackett, Williams, & Finkelhor, 1993, p. 171).

Placing Effects of Child Sexual Abuse in Perspective

A 1989 study reported the effects of child sexual abuse (CSA) on a sample of female college students. This study concluded:

“The data do not support child sexual abuse as specific explanation of current emotional distress.  The data are best interpreted as supportive of other factors such as family violence … as having the greatest impact on current emotional adjustment” (Wisniewski, 1989, p. 258).

More recent work has emphasized that people with current adjustment problems – who attribute those problems to their CSA experiences – may make invalid causal inferences.  They may engage in “effort after meaning,” attempting to explain their current life difficulties vis-a-vis CSA.  Often they choose CSA as the causative factor because of the current salience of CSA in our culture as an explanation for maladjustment (Pope & Hudson, 1995).

Even more recent research, using a non-biased sample from the general population, emphasized:

“…clinical samples overestimate the adjustment variance accounted for by CSA in the general population by a factor of 15 and thereby substantially exaggerate the intensity of CSA correlates in the general population.  Researchers’ reliance on clinical samples in most previous literature reviews to estimate CSA-adjustment relations is therefore problematic” (Rind & Tromovitch, 1997, p. 250).

This same 1997 study concluded:

“Results from psychological adjustment measures imply that, although CSA is related to poorer adjustment in the general population, the magnitude of this relation is small.  Further, data on confounding variables imply that this small relation cannot safely be assumed to reflect causal effects of CSA” (Rind & Tromovitch, 1997, p. 253).

No Evidence of a Child Sexual Abuse Syndrome

The first and perhaps the most important implication is the apparent lack of evidence for a conspicuous syndrome in children who have been sexually abused (Kendall-Tackett, Williams, & Finkelhor, 1993). The evidence against such a syndrome includes the variety of symptoms children manifest and the absence of one particular symptom in a large majority of children. In other words, the vast majority of children who exhibit behaviors considered consistent with sexual abuse have not been sexually abused.

Assumptions regarding child sexual abuse syndromes mistakenly rely on relatively high base rate behaviors (anxiety, depression, withdrawal, etc.) attempting to identify a relatively low base rate event (sexual abuse). Relying on comparatively high base rate behaviors attempting to identify a low base rate event inevitably leads to false positive classifications (mistakenly concluding that a child has been sexual abused, when in fact, she has not). Despite the lack of a single symptom that occurs in the majority of victims, both sexualized behavior and symptoms of PTSD occurred with relatively high frequency. These also appeared to be “the only two symptoms more common in sexually abused children than in other clinical groups” (Kendall-Tackett, Williams, & Finkelhor, 1993, p. 173).

“Sexualized Behaviors”

The frequency of sexualized behavior in sexually abused children (including frequent and overt self-stimulation; inappropriate sexual overtures toward other children and adults; and compulsive talk, play, and fantasy with sexual content) is somewhat difficult to determine. Although it is the most regularly studied symptom, its occurrence varies enormously. Across six studies of preschoolers (the children most likely to manifest such symptoms) an average of 35% exhibited sexualized behavior. Besides sample and methodological differences, other variations may well arise because the concept itself can be vague (sometimes it is called inappropriate sexual acting out, and other times it is called sexual acting out) (Kendall-Tackett, Williams, & Finkelhor, 1993).

Although sexualization is relatively specific to sexual abuse (more so than symptoms such as depression), nonsexually abused children may also be sexualized. For example, Deblinger et al. (1989) found that 17% of physically (but not sexually) abused children exhibited sexually inappropriate behavior. From a clinical point of view, this symptom may indicate sexual abuse but is not completely diagnostic because children can appear to be sexualized for other reasons.

Base Rate Issues

Relying on behavioral indicators to identify children who have been sexually abused neglects to consider what is known as base rates. Children who have been sexually abused often exhibit a broad range of non-specific symptoms from depressive withdrawal to aggressive acting out. In fact, however, the vast majority of children who exhibit this range of non-specific behavioral symptoms have not been sexually abused. In the terms of mathematical psychology, relying on behavioral indicators to identify sexually abused children involves using high base-rate behaviors to identify a comparatively low base rate event.

Non-specific behavioral symptoms ranging from depressive withdrawal to aggressive acting out is a high base-rate behavior. Most children exhibit some symptoms within this range at some point in their development. Fortunately, sexual abuse is a comparatively low base-rate event.  Depending on the definition of sexual abuse, only 10-20% of children are ever sexually abused. Therefore, relying on high base-rate behaviors (behavioral symptoms) to identify a low base rate event inevitably leads to a substantial frequency of false-positive classifications — Mistakenly concluding that a child has been sexually abused when she has not.

Example of mistakenly relying on high base-rate behaviors to identify a low base-rate event:

(a) Sexually abuse children regularly walk, talk, and drink water.

(b) This child regularly walks, talks, and drinks water.

(c) Therefore this child has been sexually abused.

To belabor the obvious, these assumptions result in an inordinate number of false positive classifications.

Logical Error of Affirming the Consequent

Example of logical error of affirming the consequent: (a) All humans who are pregnant are female. (b) This human is a female. (c) This human is pregnant. Correspondingly, then, it also mistaken to assume: (a) All sexually abused children exhibit behavioral symptoms. (b) This child exhibits behavioral symptoms. (c) This child has been sexually abused.

In fact, (a) If this person is pregnant, she must be female. (b) But if she is female, she is not necessarily pregnant. Relatedly, (a) if this child has been sexually abused, she likely exhibits behavioral symptoms. (b) But if a child exhibits behavioral symptoms, she has not necessarily been sexually abused.

Sensitivity-Specificity Issues

The sensitivity of any indicator (behavioral or otherwise) refers to how accurately the indicator rules-in some condition. The specificity of any indicator (behavioral or otherwise) refers to how accurately the indicator rules-out some condition. For example, consider the following indicator for identifying prostate cancer in males: Diagnose all males 50 years and older as having prostate cancer. Because 99% of the population with prostate cancer are males 50 years and older, the sensitivity of this indicator would be approximately 99%.

The obvious flaws of this diagnostic procedure only become evident when consider the male population without prostate cancer. Because only about 10% of males ever develop prostate cancer, this classification procedure will misclassify the 90% of males who do not have prostate cancer. The sensitivity of this procedure is therefore quite good (99%); but its sensitivity is unacceptably low (0%).  In other words, this procedure can only rule in prostate cancer. It cannot accurately rule-out prostate cancer.

Relatedly, behavioral indicators of sexual abuse can only rule-in sexual abuse. These indicators cannot rule-out sexual abuse. As a result, behavioral indicators of sexual are systematically biased because of their rule-in emphasis.