Correlational bias in observer ratings


A variety of evidence suggests that implicit assumptions about traits and behaviors may seriously bias the correlations derived from observer ratings. Up to now, however, there has been little experimental examination of the process. In the present study with 50 Ss, pairs of traits that are commonly assumed to be either positively correlated, uncorrelated, or negatively correlated were assigned to fictitious stimulus persons in such a way that there actually was either a positive, zero, or negative correlation between the 2 traits in each pair. Ss made trait ratings of the stimulus persons, and correlations were derived from these ratings. Results indicate that implicit assumptions concerning the co-occurrence of the traits had a systematic effect on immediate as well as delayed ratings made by the Ss. Moreover, the degree of this correlational bias appeared to be constant not only across different levels of actual correlation but also across ratings made at different points in time. These findings have important implications for the use of human observers and raters in the social sciences. (41 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved). © 1976 American Psychological Association.

Publication Title

Journal of Personality and Social Psychology