Cultic Studies Journal, Vol. 13, No. 1, 1996, page 43
Lindsay and Read (1994) used Bayesian statistical methods to predict the rate of false
positives and false negatives in the diagnosis of ICA. If a patient is suspected of a hidden
presentation of ICA (Gelinas, 1983), false positives occur when a history of ICA is
diagnosed, but, in fact, ICA never occurred. Deliberately using unrealistic assumptions
intended to minimize the prediction of false positives, Lindsay and Read (1994) estimated
that a minimum of 30% of the patients diagnosed as having repressed memories of ICA
would be false positives.
Extending the unrealistic assumptions of Lindsay and Read (1994) still further, Pezdek
(1994) was able to reduce the predicted false positive rate to 18%. This figure does not
assume incompetent diagnosticians--quite the contrary. Both the Lindsay and Read (1994)
prediction and the predictions made in the two tables below assume the ability to diagnose
the hidden presence of an etiologic factor (ICA) with as much accuracy as experts can
diagnose many overt conditions.
While the Lindsay and Read (1994) predictions provide a good starting place, their
assumptions were deliberately unrealistic, perhaps in part to avoid the appearance of
making overly pessimistic predictions about the proportion of false positives. In order to
establish a range for the incidence of incorrect diagnoses of hidden ICA, we will use two sets
of somewhat more realistic assumptions. One set of assumptions will be more realistic than
that of Lindsay and Read, but still strongly minimize false positives. The second set of
assumptions will be more middle of the road. Comparing these predictions may provide a
conservative range for the proportion of incorrect diagnoses of hidden ICA.
The authors‟ first set of assumptions--and computation of resulting false positive rate--
follows. Presuming that interested clinicians will be very careful not to miss a case of
repressed ICA memories, we assign a sensitivity of .95 to the diagnostic process: 1 patient
in 20 with a hidden presentation of ICA will be overlooked. Given the human tendency to
interpret data in light of our ideas (i.e., confirmation bias), let us assume a specificity of
.85, which is as good as or better than that for overt DSM-IV diagnoses. A specificity of .85
means that 85% of patients wrongly suspected of ICA will be correctly identified as
unabused. Assuming a specificity of .85 is unrealistically optimistic, but it provides a lower
limit on the proportion of false positives.
In this first computation, we will use a 33% prevalence estimate for ICA, similar to that
suggested by Bass and Davis (1988). However, we will make it more realistic by using the
33% figure as an estimate of prevalence in the population of women seeking
psychotherapy, and not the population as a whole. Given a sample of 1,000 women seeking
therapy, 330 would have suffered from ICA and 670 would not. Again, we use this figure
not because we think it correct, but because it establishes a lower limit on the proportion of
false positives.
The best study of the incidence of amnesia for ICA to date was done by Williams (1994),
who found that 12% of the abused women were entirely amnesic for ICA. If we assume
12% of the 330 women are amnesic for the ICA they really experienced, 40 women (.12 x
330) would have been abused and become totally amnesic for it, and there would be 290
women who were abused and remembered the abuse. These last 290 women raise no
diagnostic questions in the context of finding patients entirely amnesic for ICA and will be
excluded from further analysis. Given .95 sensitivity and .85 specificity, the following table
results:
Lindsay and Read (1994) used Bayesian statistical methods to predict the rate of false
positives and false negatives in the diagnosis of ICA. If a patient is suspected of a hidden
presentation of ICA (Gelinas, 1983), false positives occur when a history of ICA is
diagnosed, but, in fact, ICA never occurred. Deliberately using unrealistic assumptions
intended to minimize the prediction of false positives, Lindsay and Read (1994) estimated
that a minimum of 30% of the patients diagnosed as having repressed memories of ICA
would be false positives.
Extending the unrealistic assumptions of Lindsay and Read (1994) still further, Pezdek
(1994) was able to reduce the predicted false positive rate to 18%. This figure does not
assume incompetent diagnosticians--quite the contrary. Both the Lindsay and Read (1994)
prediction and the predictions made in the two tables below assume the ability to diagnose
the hidden presence of an etiologic factor (ICA) with as much accuracy as experts can
diagnose many overt conditions.
While the Lindsay and Read (1994) predictions provide a good starting place, their
assumptions were deliberately unrealistic, perhaps in part to avoid the appearance of
making overly pessimistic predictions about the proportion of false positives. In order to
establish a range for the incidence of incorrect diagnoses of hidden ICA, we will use two sets
of somewhat more realistic assumptions. One set of assumptions will be more realistic than
that of Lindsay and Read, but still strongly minimize false positives. The second set of
assumptions will be more middle of the road. Comparing these predictions may provide a
conservative range for the proportion of incorrect diagnoses of hidden ICA.
The authors‟ first set of assumptions--and computation of resulting false positive rate--
follows. Presuming that interested clinicians will be very careful not to miss a case of
repressed ICA memories, we assign a sensitivity of .95 to the diagnostic process: 1 patient
in 20 with a hidden presentation of ICA will be overlooked. Given the human tendency to
interpret data in light of our ideas (i.e., confirmation bias), let us assume a specificity of
.85, which is as good as or better than that for overt DSM-IV diagnoses. A specificity of .85
means that 85% of patients wrongly suspected of ICA will be correctly identified as
unabused. Assuming a specificity of .85 is unrealistically optimistic, but it provides a lower
limit on the proportion of false positives.
In this first computation, we will use a 33% prevalence estimate for ICA, similar to that
suggested by Bass and Davis (1988). However, we will make it more realistic by using the
33% figure as an estimate of prevalence in the population of women seeking
psychotherapy, and not the population as a whole. Given a sample of 1,000 women seeking
therapy, 330 would have suffered from ICA and 670 would not. Again, we use this figure
not because we think it correct, but because it establishes a lower limit on the proportion of
false positives.
The best study of the incidence of amnesia for ICA to date was done by Williams (1994),
who found that 12% of the abused women were entirely amnesic for ICA. If we assume
12% of the 330 women are amnesic for the ICA they really experienced, 40 women (.12 x
330) would have been abused and become totally amnesic for it, and there would be 290
women who were abused and remembered the abuse. These last 290 women raise no
diagnostic questions in the context of finding patients entirely amnesic for ICA and will be
excluded from further analysis. Given .95 sensitivity and .85 specificity, the following table
results:
