Now to discuss the significance of the table. The issue in false memories is to review those persons in the False Alarm box who previously were in the Normal box. If one can objectively trace when and under what conditions a person made the transition from Normal to False Alarm, they should attempt to do so. The underlying issue is how this transition occurred.

If it occurred in a person who was anxious, who was suggestible, and was in a mode of being influenced by "new age" material such as self-help books or by a psychotherapeutic clinician who used treatment methods associated with recovered memory therapy or the ability to develop a trance state,¹ there is great concern. The client may have been led by an "expert" interrogator and/or counselor to assume that their primary adult caretakers abused them as a child. Thus, the person may come to believe that traumatic images that have developed either while awake or asleep are recreations of historical events, especially if the "expert" convinces them that this is the case.

The examples shown below have occurred in regard to contested custody battles in which a spouse (usually mother) accuses the other of sexually abusing one or more children. As agencies and courts get involved in making custody decisions, various judgments are applied to assess the likelihood of knowing who is telling the truth (the accuracy dilemma), and to assess an appropriate set of conditions for custody, visitation, etc. (the risk dilemma).

Apart from other data which may influence an expert reviewer's judgment of character of either spouse, the discussion focuses on an examination of the yes/no choices an expert may make when choosing to accept or reject the allegation of abuse, and coming to an average value of estimating that abuse occurred. While the model itself is void of case-unique features, the net effect of an expert being biased toward a given level of percentage estimates that abuse occurred can be examined in terms of societal effects as well as for overall accuracy over a large number of cases.

A 2 x 2 table is used to examine the choices made by the expert. These choices are made in the context of having some data which assumes that, within the population group represented, there is an average percentage of molestations. The expert may have additional biases that assume it is in society's interest to favor a high level of choices for judging that abuse occurred, or a low level.

Estimated Rate of Actual Molestation
with analytical symbols shown

For a given levels of p (and q) and a (and b ), values for a, b, c, and d may be entered into the corresponding cells by multiplying the chosen values along the boundary of the table. The individual values of cells a, b, c, and d represent the decision outcomes associated with the differently applied prediction and actual estimate values. Note that cell a represents "true positive", cell b "false positive", cell c "true negative", and cell d "false negative". The best that can be done is to maximize the sum of a and d, which minimizes the sum of b and c.

An analysis shows that the numerically best results occur when a = 0, if p is less than 0.5. That is, if the expert says no every time, the sum of b and c is least when the reviewer says no every time when the stipulated "positive" average is less than 50 percent.³

Friends of FMSF are concerned with excessive values of false positives (large values of b).

One of the examples used in the Horner & Guyer article is based on estimated data on the problem of Sexual Abuse allegations In Divorce cases (SAID cases). The estimated values are based on 1986 data supplied by Underwager, et. al. Their estimated rate for this population is 20 percent "founded" and 80 percent "unfounded". Thus, p = 0.2 and q = 0.8. Figure 2 shows an example of using the table for this condition and for the condition that the expert predicts "yes" 20 percent of the time.

Estimated Rate of Actual Molestation
with sample estimates shown

This result shows that false positive and false negative outcomes would each occur one time in six for the stated conditions.

The above example assumes that the expert knows a valid assessment of the percentage of actual in the population (base rate) and comes to an average assessment value that matches the base rate for the larger population.

The Horner & Guyer article cites rationale that an expert may use when choosing various values of average yes choices. Sample a values used in the article for this 20 percent base rate example are 0.00 (Pollyanna), 0.20 (base rate), 0.80 (expert), and 1.00 (cynical).

There may be societal or personal biases that influence an expert to tend toward one of the above values of a . One basis of English jurisprudence is that it is better to acquit a person if it cannot be shown that he (or she) is guilty. It's a different ball game when it comes to accusations of child abuse.

A significant topic to examine is the basis by which experts claim their expertise in regard to making a discerned choice on a when a specific case is reviewed, and a tailored judgment (specific to a given case) is applied.

The Horner & Guyer article gives examples of information that would suggest that there is a tendency by experts to believe that their training provides them with an ability to make correct choices more often than the untrained. Further, that the untrained may have a greater ability to make correct choices than the experts if they follow a general set of guidelines and are not influenced by beliefs that they have an ability to exercise better judgment.

Let us make an estimate of actual molestation that may apply to a set of women who enter therapy for some general problem which typically is not related to knowledge of being sexually abused as a child. For this purpose, let us take Diana Russell's 1983 survey of a large number of women which found that 5 percent reported actual or attempted contact child sexual abuse (CSA). Note that the operative word contact sets a narrower definition than used by some other surveys.

Note from the Lindsay reference⁴ that only a small percentage has reported repeated incestuous contact CSA. Further, few victims of repeated contact CSA that occurred beyond the first few years of life are likely to forget that it occurred.

Other survey data indicates that actual contact CSA is perpetrated more often by a stepfather than by a biological father.

Considering this data, a high estimate of base rate prevalence of contact CSA by a natural father for the assumed population could reasonably be set near one percent.

Conclusions

More realistic estimates of base rates should be considered by clinicians. With an estimated base rate of one percent and no other factual data on which to base a conclusion, an expert evaluator would do well to tend toward the "Pollyanna" approach when estimating a reasonable value for a (i.e., a fairly low presumed level of contact CSA occurrence, which would be representative within a large sample). The Pollyanna approach (a limiting case used as an example) sets a = 0, with the evaluator always estimating that molestation did not occur. Of course, each case should be judged on its merits.

______________________________________

1. This may include the following methods: age regression therapy, recovered memory therapy, free-association, relaxation techniques, guided imagery, and holotropic breathing.

2. A three-part article by Horner and Guyer titled "Prediction, Prevention, and Clinical Expertise in Child Custody Cases in Which Allegations of Child Sexual Abuse Have Been Made" begins in the Summer 1991 issue of Family Law Quarterly. Part I of the article contains a discussion of means for estimating the net accuracy of an evaluator's ability to make the highest percentage of correct estimates of who's telling the truth when no information other than conflicting narratives exist.

3. In the analytical examples, it should be noted that judgment is not made on the basis of any individual case. That is, the expert could line up 1000 cases, and make a priori judgments (in advance) on how to apply the yes/no judgment for each case, based solely on the order of appearance of each case within the set.