How Large Are Cognitive Gender Differences?

Abstract

Maccoby and Jacklin concluded that the following cognitive gender differences are well-established: verbal ability, quantitative ability, and visualspatial ability. The present study applied meta-analysis techniques to the studies cited by Maccoby and Jacklin, assessing the magnitude of gender differences using both o> and A statistics. The results indicated that gender differences in all of these abilities are very small: For verbal ability, the median u? was .01 and the median d was .24; for quantitative ability, the median values of (a and d were .01 and .43, respectively; for visualspatial ability, they were .043 and .45, respectively; and for field articulation, to was .025 and d was .51. Discussion focused on the practical implications of the finding that these "well-established" differences were in fact very small. Concerns about sampling were raised. The problem was also discussed in the context of a larger issue in psychological research: the limitations of the hypothesis-testing approach and the need to estimate the magnitude of effects. Maccoby and Jacklin (1974) reviewed the enormous literature on psychological gender differences. In particular, they concluded that three cognitive gender differences were "well-established": Girls have greater verbal ability than boys, and boys have better visual-spatial ability and better mathematical ability than girls. Sherman (1978) re-reviewed the evidence on cognitive gender differences and pointed out that even for these supposedly well-established differences, the magnitude of the gender difference was very small. For example, Maccoby and Jacklin (1974, Table 3.4) computed the magnitude of gender differences in verbal ability for a subset of studies providing sufficient data. Typically, the magnitude of the difference was only about .25 standard deviations. Sherman (1978, p. 43) noted that the proportion of variance (w) accounted for by gender differences in verbal ability for the 1955 892 • AUGUST 1981 • AMERICAN PSYCHOLOGIST standardization of the WAIS (Matarazzo, 1972) was less than .01: Meta-analysis is a technique for analyzing a body of research on a particular topic by statistical analysis of the analyses of the individual studies (Glass, 1976). It is becoming increasingly popular as a technique for evaluating a given area of research (for examples, see Hall, 1978; Kulik, Kulik, & Cohen, 1979; Smith, 1980; Smith & Glass, 1977). Typically, these studies use the measure d = MI M2 ' —as a measure of the magnitude of difoU ferences between two groups, that is, as a measure of effect size. The purpose of the present article is to reanalyze the studies on cognitive gender differences considered to be well-established by Maccoby and Jacklin (1974) and to determine the magnitude of these gender differences. Two measures of effect size were used: co (Hays, 1963) and d. Such an analysis is particularly important because the term "well-established" is often taken to mean "large." Maccoby and Jacklin's review has had a widespread influence, and their conclusions are cited in many introductory-level psychology texts. Thus, large numbers of people may have the impression that there are large gender differences in cognitive abilities. This, in turn, may affect practices such as vocational counseling. For example, a girl might be discouraged from a career in mathematics or science because of the "well-established" and "large" superiority of males in quantitative ability (despite repeated caveats in many texts about the great overlap in male and female distributions). Thus, it seems important to determine whether Requests for reprints should be sent to Janet Shibley Hyde, Department of Psychology, Denison University, Granville, Ohio 43023. Vol. 36, No. 8, 892-901 Copyright 1981 by the American Psychological Association, Inc. (K)03-066X/81/3608-0892$00.75 these "well-established" differences are in fact "large." This study can be viewed as one example of a larger issue in psychological research: criticism of the hypothesis-testing approach and a suggested alternative, estimating the magnitude of effects. Cognitive gender differences are good examples because they are widely believed to be well-established, yet the size of the gender difference has rarely been estimated.