ABSTRACT
Clinicians routinely draw inferences from test results to the test taker’s latent condition. After all, “a large part of medicine is practiced on people who do not have obvious illnesses, but rather have signs, symptoms, or findings that may or may not represent an illness that should be treated” (Eddy, 1984, p. 75). In the simplest case, both test results and latent conditions are dichotomous variables. Tests turn out either positive or negative, and test takers either do or do not have the disease in question. The inference of interest is a predictive judgment of whether a client with a positive test result has the disease. This judgment depends on several cues. Some of these cues are external, such as the information contained in the client’s file (including test scores), whereas other cues are internal, such as the prevalence of the disease in the population or the clinician’s experience and memory of related cases. In this chapter, we address the integration of external and internal cues in simple Bayesian inference and in more extended belief networks.
