Document Type


Publication Date



This Article has two major purposes. First, it provides a mathematical
description of an ideal procedure for making clinical decisions about patients'
future violence, a description that provides a context for evaluating clinicians'
"dangerousness decisions." For purposes of illustration, the Article uses a specific clinical situation-deciding whether to hospitaize involuntarily a patient
based on his risk of harming another. The Article argues that the decision
involves balancing potential risks to third parties (often the patient's family
members) with the "massive deprivation of liberty and other potential
harms to the patient that could result from confinement. The mathematical
description of the decision procedure consists of a comprehensive method for
describing the accuracy of predictions or prediction instruments, a method
for assigning values to correct and incorrect predictions, a method for
adjusting predictions based on those values, and-most importantly-an explicit means for expressing uncertainty in those values.

Second, the Article evaluates the actual impact of uncertainty on an ideal
decision procedure. When we combine our uncertainty about moral valuations
of right and wrong decisions, our uncertainty about base rates, and our
uncertainty about the relevant time periods over which predictions should
apply, what results is an uncertainty about the correctness of prediction-based
decisions that makes most criticism of those decisions untenable. This
Article shows that our uncertainty about the factors intrinsic to a hypothetical,
best-case prediction procedure usually would preclude valid post hoc criticism
of wrong decisions about dangerousness; a fortiori, most real-life prediction
errors also should be beyond criticism.