Salience and the cognitive mediation of attribution.

Two studies of the impact of salience and informational factors on attribution and memory yielded findings on two general issues. First, manipulations of the amount of thought subjects gave to their attributions and of a delay before responding to attribution questions did not diminish the effect of salience on attribution; in fact, the delay increased the effect. Second, recall of the stim-ulus material was shown to be influenced by salience and by covariation in-formation (consensus, distinctiveness, and consistency) and to be related to attributions. These findings, together with theory and data from the literature on comprehension and representation of linguistic material in memory, are used to argue that salience is not simply a process by which people make attribu-tions without giving much thought to them. Instead, salience effects reflect the close relationships among the processes of comprehension, remembering, and attribution, and the fact that attributional processing can take place at the time of the encoding and storage of information, as well as at the time of its re-trieval from memory. The literature on the perception of causa-tion currently contains support for two dis-tinct and even opposed theories of the attribu-tion process. Kelley (1967, 1972) has pro-posed that the perceiver makes attributions after logically weighing information about the covariation of an effect with various possible causes. This covariation information may be observed or assumed (Kelleys, 1972, causal schema), but in either case it is processed more or less rationally, in a fashion resem-bling the analysis of variance (ANOVA), Kel-leys model of attribution has come to be known as the ANOVA model for this reason. On the other hand, several recent writers (Pryor Kriss, 1977; Taylor Fiske, 1975, 1978) have argued that perceivers use a much simpler information-processing strategy: at-The authors are grateful to Susan T. Fiske, Reid Hastie, and Shelley E. Taylor for their comments on an earlier version of this paper. The research re-ported here was supported by the University of