Figures and data
Change in women’s overall faculty representation for 111 academic fields between 2011–2020, decomposed into change due to hiring (horizontal axis) and change due to attrition (vertical axis, see Supplementary Sec. S1), showing that hiring increased women’s representation for a large majority (87.4%) of fields, while it decreased women’s representation for five fields. Point size represents the relative size of each field by number of faculty in 2020, and points are colored by STEM (black) or non-STEM (gray).
Gendered faculty attrition has caused a differential loss of women faculty in both STEM and non-STEM fields. (A) Gendered attrition in psychology has caused a loss of − 1.83 pp (p < 0.01) of women’s representation between 2011-2020, relative to a counterfactual model with gender-neutral attrition (see Methods Sec. C). In contrast, (B) gendered attrition in Ecology has not caused a statistically significant loss (+1.42 pp, p = 0.24). Relative to their field-specific counterfactual simulations, 15 academic fields and the STEM and non-STEM aggregations exhibit significant losses of women faculty due to gendered attrition (circles on figure; two-sided test for significance relative to the gender-neutral null distribution derived from simulation, α = 0.1). The differences in the remaining 95 fields were not statistically significant (crosses on figure), but we note that their lack of significance is likely partly attributable to their smaller sample sizes at the field-level compared to the all STEM and all non-STEM aggregations, which exhibited large and significant differences. Error bars for the non-STEM and STEM aggregations contain 95% of n = 500 stochastic simulations. No bars are included for field-level points to preserve readability.
(A) Observed (dotted line, 2011-2020) and projected (solid lines, 2021-2060) faculty gender diversity for Natural Sciences over time and (B) projections for 11 academic domains over 40 years under five policy scenarios. Line widths span the middle 95% of N = 500 simulations and 
The change in gender diversity between 2011 and 2020 can be approximately decomposed into parts due to hiring and attrition for each academic field, but there is a leftover residual term. In practice, we find that the residual term tends to be very small, such that the decomposition is nearly ideal. The dotted line represents an ideal decomposition, where the change in women’s representation among faculty due to hiring and attrition perfectly matches the total observed change.
Model validation: Differences between observed gender diversity outcomes and model-based outcomes. (A) The mean outcomes of model-based simulations in psychology differ from the observed outcomes by − 0.04pp, and (B) in Ecology by +0.11pp, but these differences are not statistically significant. (C) Gender diversity outcomes from model-based simulations of hiring and attrition are statistically indistinguishable from observed gender diversity outcomes for all 111 fields, and for STEM and non-STEM aggregations, based on a two-sided test for significance relative to the model-based null distribution derived from simulation, α = 0.1). Error bars for the non-STEM and STEM aggregations contain 95% of stochastic simulations. No bars are included for field-level points to preserve readability.
Trends in women’s representation among new hires from 2012 to 2020 for 11 academic domains, along with academia overall. We use linear regression to measure the expected change in women’s concentration among new hires each year, and find that women’s representation has been increasing in 6 of the 11 domains over time, at rates ranging from 0.58 pp to 1.30 pp per year. The remaining 5 domains have not exhibited significant linear trends. Overall, the fraction of women among hires has been increasing in academia over time (Fig. S7). These findings are qualitatively replicated using logistic regression, so we present the linear regression results here for enhanced interpretability.
Model validation: Projected 2060 faculty career age distributions for Natural Sciences from Fig. 3 are similar to the observed career age distribution for Natural Sciences in 2020, for each projection scenario. Line widths for the simulated scenarios span the middle 95% of simulations. OA = observed attrition, GNA = gender-neutral attrition, IR = increasing representation of women among hires (+0.5 pp each year), ER = equal representation of women and men among hires.
Model selection. (A) Four logistic regression models fit to observed faculty attrition data. Each model includes career age up to a different power, e.g., the model labeled “Career age order 3” includes career age up to its third power: logit(p) = β0 + β1a + β2a2 + β3a3 + β4t where a represents career age and t represents year (see Methods Sec. D for details). The pattern in observed attrition risk becomes more noisy at higher career ages, because (B) there are relatively low numbers of faculty at the highest observed career ages.
Model selection. (A) Four logistic regression models fit to observed faculty hiring data, where the outcome variable is the gender of the faculty hire (1 = woman, 0 = man). Each model includes career age up to a different power, e.g., the model labeled “Career age order 3” includes career age up to its third power: logit(p) = β0 + β1a + β2a2 + β3a3 + β6t where a represents career age and t represents year (see Methods Sec. D for details). The pattern in the gender representation among new faculty hires becomes more noisy at higher career ages, because (B) there are relatively low numbers of faculty hired at higher career ages.
Sensitivity analysis: Replicating the counterfactual analysis from results Sec. B using career age up to its third power in the associated logistic regressions model, instead of the fifth power (see Supplementary Sec. S1 A for details). Findings under this parameterization are qualitatively very similar to those presented in Fig. 2, indicating that the results are robust to modest changes to model parameterization.
Fraction of women among tenure-track faculty hires over time at U.S. PhD granting institutions. Women’s share of new hires is observed to increase at around 0.91 pp annually (t-test, p < 0.001), measured by an ordinary least squares regression fit (shown in purple).
Career age distribution of women (red) and men (blue) tenured and tenure-track faculty across all academic fields. Career age is measured as the number of years since earning a PhD. There are substantially more men faculty with high career ages than women faculty.
Changes in Women’s Representation through Hiring, Attrition, and Gendered Attrition in Academic Fields (2011-2020).
Observed changes in women’s representation resulting from hiring and attrition, expressed in percentage points (pp), based on data from Fig. 1, and the estimated average change in women’s representation due to gendered attrition as depicted in Figure 2, accompanied by the 2.5 percentile and 97.5 percentiles of simulations in parentheses. The analysis covers 111 academic fields.
Number of faculty by field and gender, 2020.
Estimated counts of women and men faculty based on 2020 faculty rosters and name-based gender inference [25].
