|
Abstract
|
Tasks used to study human judgment and decision making have generally employed "cues" that have already undergone some processing. For example, a subject may be asked to diagnose a hypothetical patient based on a set of test results, or to rate a hypothetical job applicant based on a set of predictor scores. Many decision situations, however, require dealing with relatively unprocessed data: the decision maker must estimate "cue" states from raw observations before making an overall judgment. Tactical military commanders, for example, may have to estimate the level of activity or strength or readiness of multiple enemy units in order to arrive at a judgment of overall threat. It is not known how such initial processing (e.g., cue estimation) affects final decisions (e.g., cue utilization), although we have some evidence that prior estimation can influence subsequent choices (Howell & Kerkar, 1982).
Two studies were designed to explore this issue in the context of military threat diagnosis. Subjects made overall threat judgments of several kinds based upon "real-time observations" of events occurring at four hypothetical enemy positions. The positions differed in threat potential or weight to be accorded the observations.
In the first study, the main independent variables, both manipulated in a between-subjects design, were presence vs. absence of a prior cue-estimation requirement (evaluation of readiness for each position) and manual vs. automatic tabulation of observed events. Subjects' cue-weighting strategies were computed by regressing their overall threat assessment on the event frequencies at the four positions. Also, correlations of actual with optimal threat assessments were computed. Results showed that prior readiness assessment (cue estimates) improved use of frequency information in assessing threat. The readiness variable significantly affected the subjects' weighting strategies. While all subjects overestimated the importance of the most "diagnostic" position, those who first estimated readiness produced b-weights for the three other positions that more closely approximated the optimal b-weights than did the no-estimation subjects. Furthermore, subjects who estimated readiness more closely approximated optimal overall threat assessment than did no-estimation subjects. The tabulation variable, however, had no effect on performance. Therefore, in the second study, this variable was eliminated.
The second study replicated the cue estimation (readiness assessment) manipulation together with a second variable, the extent to which the cues had undergone prior processing. Keeping in mind that the cue values were in all cases levels of readiness indexed by frequency of enemy activity, the three levels of the latter variable were (1) a raw data condition which required subjects to deal with observed events in real time, (2) a stored event condition (as in Exp. 1) in which the events were preserved but not computed, and (3) an automatically pre-processed condition in which event frequencies were computed and displayed as cue values in the typical "policy capturing" paradigm. Theoretically, these conditions were intended to represent three levels along Hammond's cognitive processing continuum (1980): a relatively intuitive level (raw data), a relatively analytic level (automatically preprocessed), and an intermediate level (stored event). Overall threat judgments based on raw data observations (the intuitive end of the continuum) without an estimation step correlated with optimal values r = .48, while those preceded by estimation of activity level correlated r = .79. By contrast, under the automatically preprocessed condition (the analytical end of the continuum), estimation had no beneficial effect--in fact, it produced a non-significant decrement in performance (r = .83 vs. r = .90). The higher overall performance associated with the automatic preprocessing is, of course, to be expected given the perfect accuracy of the cue values. The intermediate stored event condition produced a significant estimation effect which was properly located between the two extremes (r = .77 without estimation vs. r = .86 with estimation). The cue-weighting results also confirmed the previous study with regard to the estimation effect.
In sum, it appears that when decision makers are forced to make overall criterion judgments (such as threat assessments) intuitively on the basis of events observed in "real time", their performance can be improved markedly by interposing a processing step (cue estimation). However, if this processing is done automatically, permitting a more "analytic" threat judgment, performance improves and the redundant estimation step is not helpful. If the event occurrences are merely preserved but not processed, estimation is again helpful, but to an extent midway between the raw-observation and the automatic-processing conditions.
Language: en |