Janet Hyde and Marcia Linn on the Psychological Similarity between Men and Women

In my previous post arguing for the relatively large psychological similarity between men and women -- in great contrast to the public conception -- I drew heavily on the work of Janet Hyde, a professor of Psychology at Berkeley.

Now Janet Hyde and Marcia Linn have published an editorial and review in Science summarizing their work. Money quote:

A review of meta-analyses of research on psychological gender differences identified 46 reports, addressing a variety of psychological characteristics, including mathematical, verbal, and spatial abilities; aggression; leadership effectiveness; self-esteem; and computer use (5). These research syntheses summarize more than 5000 individual studies, based on the testing of approximately 7 million people. The findings are represented on a common scale based on the standard deviation. The magnitude of each gender difference was measured using the d statistic (6), d = (MM - MF)/sw,where MM is the mean score for males, MF is the mean score for females, and sw is the pooled within-sex standard deviation. The d statistic measures the distance between male and female means, in standard deviation units. In each individual meta-analysis, the values of d from multiple investigations of the same outcome were weighted by sample size and combined.

A total of 124 synthesized effect sizes resulting from meta-analysis were extracted from the reports. Following convention, d values in the range 0.11 to 0.35 were classified as small, 0.36 to 0.65 as moderate, and 0.66 to 1.00 as large (6). Values greater than 1.0 were categorized as very large and values between 0 and 0.10 were considered trivial.

Of the effects for gender differences, 30% were trivial and an additional 48% were small. That is, 78% of the effects for psychological gender differences were small or near zero. For example, for mathematics problem-solving, d = 0.08 (7); for leadership effectiveness, d = -0.02 (8); and for negotiator competitiveness, d = 0.07 (9).

An essential implication of these findings is that the overlap of distributions for males and females is substantial for most outcomes. For example, the chart below shows the distribution of male and female performance for a small effect size of 0.20. Assuming that performance fits a normal distribution (supporting online material text), for means that are 0.20 standard deviations apart (d = 0.20), the populations show 85.3% overlap. Only 54% of members of one gender exceed the 50th percentile for the other gender. For d = 0.10, there is 92.3% overlap in the distributions of the two groups, and 52% of those of one gender exceed the 50th percentile for the other. Even for a moderate effect size (d = 0.50), 60% of members of one gender exceed the 50th percentile for the other gender.

Figure 1 Effect size of 0.20. When the effect size between two groups is 0.20, 85.3% of the distributions overlap.

Most relevant here is the meta-analysis of research on gender differences in mathematics performance, which was based on 100 studies and the testing of more than 3 million people (7). Patterns emerged as a function of the age of test takers and the cognitive level of the test. Girls outperformed boys on computation in elementary school and middle school (d = -0.20). There was no gender difference in high school. There was no gender difference in deeper understanding of mathematical concepts at any age. For complex problem solving, a skill that is highly relevant for science, technology, engineering, and mathematics careers, there was no gender difference in elementary or middle school; a small difference favoring boys emerged in high school (d = 0.29). Consistent with these findings of gender similarities in mathematics performance, in 2001 women earned 48% of the bachelor's degrees in mathematics in the United States (10), demonstrating that substantial numbers of women do have the ability to engage mathematics successfully at the advanced levels required of a mathematics major.

A few exceptions to the pattern of gender similarities were identified in the review of meta-analyses. Most relevant to educational settings are the gender differences in activity level and physical aggression, with effect sizes for aggression ranging between 0.40 and 0.60 (males more aggressive) across several meta-analyses (11-14). Males display a higher activity level than females, d = 0.49 (15). It is widely believed that there are substantial gender differences in interests in fields such as psychology compared with physics, although we found no meta-analysis of this research literature.

With respect to mathematical ability there are some documented differences between men and women; however, these differences are not pervasive -- present across all mathematical aptitudes -- nor do they consistently favor men, nor are they consistent over time. This suggests that the use of gender as a proxy for mathematical ability is a very poor one.

Drs. Hyde and Linn's core argument -- and mine -- is not that there are no documented psychological differences between men and women. Rather, the argument is that the effect sizes for those differences which do exist are not sufficient to explain the disparity in participation in science, math, and engineering -- suggesting some other social factor influencing the outcome.

Women earn 46% of the Ph.D.'s in biology but, despite evidence for gender similarities, they earn only 25% of the Ph.D.'s in physical science and 15% in engineering. Women comprise 30% of the assistant professors in biology but only 16% in physical science and 17% in engineering (20). Too often, small differences in performance in the NAEP and other studies receive extensive publicity, reinforcing subtle, persistent, biases (20, 21). Indeed, the magnitude of the attitudinal association between science and males is large, d = 0.72 (22, 23). These biases can have an impact on decisions about admissions, hiring, and promotion (24, 25). These biases may contribute to popular beliefs about same-sex education and learning styles, and dissuade some individuals from persisting in science (26).

For example, advocates of same-sex education claim advantages for both boys and girls (2). Some argue that boys' great activity level and aggressiveness make it difficult for girls to learn and participate actively, and at the same time, boys need a classroom that tolerates their active style. Activity level and physical aggression are two exceptions to the gender similarities rule, with effect sizes around 0.50 for each. Yet even a gender difference of that magnitude means that 40% of one group (in this case, girls) score higher than the average for the other group (boys). If the idea is to separate children into classrooms for the more active and aggressive and the less active and aggressive, gender is not the best indicator. A teacher's rating of activity level would be far more accurate.

The phenomenon of gender similarities has implications for schooling. Emphasis on gender differences in the popular literature reinforces stereotypes that girls lack mathematical and scientific aptitude. However, gender is a poor indicator of whether one will major in mathematics or the biological sciences as an undergraduate. A better predictor would be actual mathematics achievement scores in middle school or high school (27). A cultural overemphasis on gender differences may mask critical predictive variables and lead to decision-making that is empirically unsupported. To help teachers succeed, we may need to address variability in aggression and activity level for all learners. To neutralize traditional stereotypes about girls' lack of ability and interest in mathematics and science, we need to increase awareness of gender similarities. Such awareness will help mentors and advisers avoid discouraging girls from entering these fields. Continued monitoring of the relative progress of boys and girls is essential so that neither group falls behind.

Rather than focusing on gender differences, mathematics and science educators and researchers could more profitably examine ways to increase awareness of the similarities in performance and in ability to succeed.

This is the real issue. Our stereotypes about men and women are not serving us well. What we would really like to do is put children in the best possible environment for them to follow their interests and succeed. This decision requires evaluation of their capacities. If indeed a child -- male or female -- has trouble with math, then they need remediation. If they possess an aptitude for it, then they should be encouraged and pushed forward.

Gender is neither a morally acceptable nor a particularly effective indicator in this evaluation.