depends on the degree of evidence favoring one or the other alternative. With a single diagnostic test, the raw score on the test is typically interpreted as indicating strength of evidence—for example, stronger differential responses to relevant questions on the polygraph are taken as stronger evidence of deception. A diagnostic decision is determined by how much positive evidence the diagnostician requires to make a positive diagnosis or how much negative evidence to make a negative diagnosis. This reasoning is the basis for the most common polygraph scoring systems, which base diagnostic decisions on numerical representations of the strength and consistency of physiological responses.
Degree of evidence can be represented along a decision axis as shown in the left panel of Figure 2-1. In general, greater amounts of positive evidence (higher eye pressure test scores, in this example) are associated with the presence of the underlying condition (the right-hand distribution, for glaucoma cases) than with its absence (the left-hand distribution, for healthy eyes). However, the two distributions overlap, and intermediate degrees of evidence are often interpreted as inconclusive. A diagnostician may use two cutoff points, as in the left panel of the figure (such as 10 and 40), and call the intermediate values inconclusive, or he or she may choose to make only a positive or negative decision, as based on a single cutoff point (e.g., 20, in the second panel of the figure). The choice of this particular cutoff point represents the judgment, common in medical diagnosis, that it is more important to avoid false negatives than to avoid false positives.
Signal detection theory distinguishes two independent features of a test that contribute to its diagnostic performance: (1) the accuracy of the test for the application being studied, which depends on the amount of overlap of the test score distributions when the target condition is present and absent (more accurate tests have less overlap), and (2) a measure of the decision threshold(s)—the cutoff point(s) along the decision or evidence axis—used by the diagnostician.
This distinction—and particularly the concept of decision threshold— deserves further explanation in relation to polygraph testing. The familiar scoring of each question comparison and each physiological response on a polygraph chart on a scale of +3 to –3 (Backster, 1963, 1973; Swinford, 1999) sets thresholds in the form of numerical scores (for example, sums of item scores) that must be attained for a chart to be considered conclusively indicating deception or nondeception. It is not always appreciated, however, that these thresholds are policy choices made by polygraph researchers or polygraph program managers. Thresholds could (and