Fishbein and Middlestadt ( 1995) rather recently initiated a discussion about the role of noncognitive factors in the formation of attitudes, in the Journal of Consumer Psychology. They argued that when attitudes, beliefs, and values are measured by "standard procedures" (i.e., Fishbein-type procedures), regression analyses of attitude and other analyses show that beliefs and values absorb all the predictive power. Noncognitive factors allegedly are only statistically potent when the models are inadequately specified. This argument met with a number of opponents ( Haugtvedt et al., 1997; Miniard & Barone, 1997; Priester & Fleming, 1997; Schwarz, 1997) who pointed to an extensive literature in experimental social psychology demonstrating what seems to be effects of noncognitive factors, such as the effect of mere exposure. Fishbein and Middlestadt ( 1997) countered, however, that cognitive factors have rarely, if ever, been adequately investigated in such work and concluded that cognitive factors still must be seen as the determining factors of attitudes.
In my view, beliefs and values may well appear to be determiners of attitudes, when analysis is confined to simple regression analysis. But the correlations that are the basis of the regression analyses may well reflect a very different psychological dynamics, as I have demonstrated in the present chapter. Causal relations are not proven by correlations and regression analyses. Expectancy-value models are basically very misleading because they appear to fit, while not capturing the psychological processes in an adequate manner ( Sjöberg & Montgomery, in press).
The author is grateful to Martin Fishbein, Henry Montgomery, Norman Anderson, and Amos Tversky for discussions of some of the problems treated in this paper.