UNIDIMENSIONAL SCALING OF MULTIDIMENSIONAL FACIAL EXPRESSIONS1

It is often possible to construct unidimensional scales of multidimen-sional stimuli. Thurstonian scaling theory predicts that the standard deviations of differences in unidimensional paired-comparisons scaling of such stimuli should covary with the qualitative similarity between the stimuli being compared. In a test of this prediction, 78 5s judged 11 facial expressions from the Lightfoot series with respect to emotional intensity. Stimuli were selected so as to fall in 3 different clusters with large similarity within clusters and small similarity between clusters. Within-cluster comparisons gave smaller standard deviations than between-clusters comparisons for all 11 stimuli. The possibility of consistent judg-ments of complex stimuli has been rec-oganized for a long time (cf. Holling-worth, 1913). Unidimensional Thurs-tonian scales have been constructed from such continua as seriousness of criminal offenses and conservatism of attitude statements, to name just a few of many possible examples (Ekman & Sjoberg, 1965; Stevens, 1966). In addition, the newer so-called direct methods of scaling devised primarily by Stevens (Stevens, 1957; cf. Ekman & Sjoberg, 1965) have now been used successfully many times on complex, nonsensory continua such as those just mentioned (Stevens, 1966). The fact that unidimensional scales may be constructed from these continua constitutes somewhat of a paradox (Sjoberg, 1968) since stimuli are most often clearly multidimensional. Appar-ently, 5"s are able to collapse a multi-dimensional space into various unidi-mensional aspects or attributes