A Monte Carlo study of cue salience measures obtained from the paired comparisons of multiple cue stimuli

Abstract

A policy capturing model was proposed and evaluated using Monte Carlo computer techniques. The model represents a particular adaptation of multiple regression techniques to the method of paired comparisons and can be used to develop cue salience measures based on the similarity judgments of multiple cue stimuli. Most policy capturing models depend on maintaining orthogonalality among cue values in order to provide measures of cue salience. The proposed model takes into account the difficulty of maintaining zero level correlations among cues when multiple cue stimuli are compared using a paired comparison format. As an alternative, the model allows the experimenter to select from various stimuli comparison strategies such that each comparison strategy represents a trade-off between the number of paired comparisons required and a level of uniform intercorrelation which must exist among all cues comprising the stimuli. The rationale for the model assumes that spurious effects due to cue intercorrelations which influence the magnitude of cue salience measures will be shared equally by all cues if intercorrelations are held constant and that these effects will not substantially influence the relative rank ordering of resulting cue salience measures. The model was evaluated under 189 different conditions defined by the number of cues comprising the stimuli, the number of these cues exhibiting valid relationships with criterion ratings, the level of valid cue-criterion relationships, and the stimuli comparison strategy employed. The adequacy of cue salience measures derived under these conditions were assessed using several criteria aimed at identifying the extent to which salience measures associated with valid cues were distinct from salience measures associated with random cues. The model was found to perform well under certain conditions and to perform poorly under others. The type of comparison strategy chosen, the number of valid cues, and the strength of cue-criterion relationships were all found to affect the level of adequacy associated with the cue salience measures. The model was found to work best for stimuli comparisons involving greater than four cues but fewer than four valid cues. A formula and a Table were developed which provide estimates of the expected adequacy of cue salience measures developed under all conditions for which the model was tested.