Computing Gender Difference Effects in Tails of Distributions: The Consequences of Differences in Tail Size, Effect Size, and Variance Ratio.

TLDR: It is thought that the "sex difference in the percentages of low scorers" meant the difference obtained by putting all the low Scorers of both genders together and computing the gender difference in this group is the more conventional and policy-relevant conception of "effect size in the tails."

Abstract: Quantitative methods in empirical research are not as objective as their label suggests: Decisions about what to calculate and how to interpret those calculations are necessarily highly subjective. The only step in the process of producing quantitative measures that can be called fully explicit, objective, and reproducible is the process of calculation itself. We (Hedges & Friedman, 1993a) were interested in Feingold's (1992a) work because we felt we could make calculation of the joint effects of differences in means and variability more explicit and objective. From Feingold's reply (Feingold, 1993), it is clear that his steps in characterizing tail effect sizes in his Table 5 (1992a) are different from the calculations we had imagined. We had supposed that the "sex difference in the percentages of low scorers" (p. 75) meant the difference obtained by putting all the low scorers of both genders together and computing the gender difference in this group. We think this is the more conventional and policy-relevant conception of "effect size in the tails."