should) be set differently, depending on policy needs. (Considerations for setting thresholds are discussed below, “Selection of Decision Thresholds.”)
The concept of a decision threshold (and other concepts from signal detection theory) have been little used in the U.S. government-supported polygraph research, though they have been used for decades in other studies (see, e.g., Ben-Shakhar, Lieblich, and Kugelmass, 1970; Ben-Shakar, Lieblich, and Bar-Hillel, 1982; Hammond, Harvey, Jr., and Hastie, 1992; Swets, 1992, 1996:Chapter 5; Szucko and Kleinmuntz, 1981). The committee’s discussions with representatives of government agencies reveal little awareness of the concept in polygraph practice. There may indeed be some resistance to the idea that polygraph examiners can set various thresholds, perhaps because the idea makes the polygraph sound less scientific or objective. However, the need to set thresholds with diagnostic tests does not make them any less accurate or objective. Different thresholds simply reflect different tradeoffs between false positives and false negatives: for a test of any given level of accuracy, setting a threshold to decrease false negatives means accepting more false positives, and vice versa.
We have some concern that in practice, polygraph programs and examiners may ostensibly adhere to a given threshold—reflected by a mandated point on a scoring scale—while accomplishing the equivalent of varying the threshold in other ways, for instance, by altering the test conditions to affect the strength of the examinee’s autonomic response. That examiners can do so is reflected in their own claims to the committee about their ability to influence examinees’ physiological reactions and by the small worth typically assigned to a polygraph chart collected under circumstances friendly to an examinee. Test conditions may vary systematically according to such factors as expectancies of guilt about individuals and expected base rates of guilt in a population of examinees. If they do, and if different test conditions yield different physiological responses, the effect would be similar to varying the threshold—but less transparent and more difficult to control. The effect would be to undermine claims that the quality of polygraph examinations is sufficiently controlled that a polygraph test result has the same meaning across test formats, settings, and agencies.
As shown in the second panel of Figure 2-1, any given decision threshold will produce a certain proportion of true-positive decisions (equal to the shaded proportion of total area under the curve in the upper part of the panel, which represents examinees with the target condition present) and a certain proportion of false-positive decisions (similarly represented in the lower part of the panel). These two proportions vary together from 0 to 1 as the threshold is moved continuously