Reply to Eysenck*

In the interest of brevity, we might limit our reply to Eysenck (1992) to an invitation to the reader to re-read our original article (Costa & McCrae, 1992); most of the points he makes are anticipated and answered there. For example, he discusses the Zuckerman, Kuhlman, Thornquist and Kiers (1991) article and comments that their five factors do not match ours very well. We acknowledge this, and showed how a re-rotation could substantially improve the match. He points out that cross-observer correlations for Conscientiousness (C) are “derisorily small,” noting a 0.30 between peer raters. But Table 2, from which he extracts this figure, also shows peer/self correlations of 0.46 for C-higher than those found for the uncontested dimensions of Neuroticism (N) (0.33) and Extraversion (E) (0.43). There are some points, however, we would like to clarify here. In our original article, we argued that the dimensions of the five-factor model are real, pervasive, universal, and biologically based. In his reply, Eysenck grants these points and acknowledges that they are necessary features of basic dimensions of personality, but argues that,; they are not sufficient: they do not distinguish the five-factor model from its competitors-at least from Eysenck’s three-factor model. He is of course correct in one sense. If we assume that the three-factor model is a subset of the five-factor model [Psychoticism (P) being viewed as a combination of low Agreeableness (A) and low C], then the dimensions of the three-factor model must also be real, pervasive, universal, and biologically based. We are happy to grant this. The three-factor model, however, is not comprehensive, most obviously because it does not account for traits related to Openness (0). By listing a catalog of related traits, we intended to show that this dimension has recurred in too many conceptualizations to be ignored. By reviewing its minimal correlation with IQ, we have shown that 0 cannot be reduced to intellectual ability. At one time, Eysenck suggested that 0 “represents possibly the opposite end of a continuum to psychoticism” (Eysenck & Eysenck, 1985, p. 138), but in fact the correlation between 0 and P is 0.05, N = 586, NS (McCrae & Costa, 1985). It would seem that, at a minimum, Eysenck should recognize the need for a four-factor model: N, E, P and 0. We concur with Eysenck’s view that purely psychometric criteria cannot resolve the issue of which factors are basic (although Everett’s 1983 formalization of the criterion of factor replicability is a marked advance over eigenvalue rules). Ultimately, the decision of what factor structure is best requires scientific judgment on the recurrence of regular patterns that make sense of the broadest range of data. Eysenck claims that the five-factor model has not been recovered by researchers “outside the narrow circle of five-factor enthusiasts,” and cites a 1983 review by Royce and Powell as evidence of the wider acceptability of the three-factor model. It is possible that, as of 1983, there was more evidence in favor of a three-factor model than a five-factor model, but much has changed since then. Perhaps we should set aside the voluminous research on trait adjectives (e.g. Goldberg, 1989) and adjective-based scales (e.g. Trapnell & Wiggins, 1990; Brand & Egan, 1989), because the five factors were first discovered in analyses of adjectives, and these studies could be considered