Threshold setting and the cycling of a decision threshold

摘要

When policy-makers use a test result with a cut-off score in a decision, the cut-off threshold may change over time. An example is the threshold of "reasonable suspicion" used to justify a police search. Hammond (1996) postulated that a decision threshold will oscillate over time in response to competing pressures from affected constituencies, as unavoidable cases of false positives (e.g., innocent people searched) and false negatives (e.g., guilty people overlooked) emerge from the uncertainty of using an imperfect test (e.g., level of evidence) to predict the actual measure of interest (e.g., guilt). The structural underpinnings of a cycling threshold are analyzed in this theory-building article. First, we present a simplified converging model of Hammond's initial insight. Then, we present three alternative models: a model with delays in policy-maker responsiveness; one with stakeholders' shifting constituencies in response to recent errors; and one with integral control representing the historical dissatisfaction of competing constituencies. The "history" model meets the theoretical requirements set out by Hammond and fits historical data regarding FBI denials of access to information regarding a third party due to privacy concerns.