Diagnostic Judgment as a Function of the Preprocessing of Evidence

Abstract

An experiment was conducted to determine how the quality of a human judgment (in this case, military threat diagnosis) is affected by various levels of preprocessing applied to the raw predictive events when such processing is carried out by the human and by a machine aid. Subjects were required to estimate the threat of attack on the friendly position (criterion) based on observed activity levels (cues) and designated threat potential weights of various enemy positions. Overall threat judgments were made under conditions in which a prior overt estimate of position activity levels was or was not required. Machine-aiding conditions were as follows: (1) no aiding, where subjects simply observed raw events in real time, (2) automatic tabulation of events, and (3) automatic computation of events. Finally, the rate of event occurrences was manipulated. When subjects made overall criterion judgments (threat evaluation) intuitively on the basis of events observed in real time, their performance improved markedly by interposing cue estimation, even if cue estimation was fairly inaccurate. If events were computed automatically, permitting a more analytic threat judgment, performance improved and the redundant estimation step was not helpful. If events were merely tabulated, estimation was helpful, but to an extent midway between the raw-observation and automatic computation conditions.