Measuring the importance of cues in policy capturing

Regression analysis has often been used to "capture" the decision policies of raters or judges. The usefulness of policy capturing generally has been thought to be limited to the case in which the evaluative cues are uncorrelated; otherwise, the various indices of cue importance often do not agree. We argue that the raw-score regression weight is the most appropriate index of importance in that it represents the expected change in overall rating per unit change on a given cue and is invariant across changes in the cue intercorrelations. To test the proposition that raw-score regression weights do not change as a function of cue structure, the decision policies of 14 subjects were captured under three cue intercorrelation structures. The raw-score regression weights did not change significantly as a function of cue structure, whereas the simple correlations, the semipartial correlations, and the standardized regression weights did.