Systematic comparison of gender inequality in scientific rankings across disciplines

TL;DR

It is highlighted that while the participation of women has risen in some fields, they remain under-represented in top-ranking positions, and in most fields, women and men with comparable productivity levels and career age tend to attain different levels of citations.

Abstract

The participation of women in academia has increased in the last few decades across many fields (e.g., Computer Science, History, Medicine). However, this increase in the participation of women has not been the same at all career stages. Here, we study how gender participation within different fields is related to gender representation in top-ranking positions in productivity (number of papers), research impact (number of citations), and co-authorship networks (degree of connectivity). We analyzed over 80 million papers published from 1975 to 2020 in 19 academic fields. Our findings reveal that women remain a minority in all 19 fields, with physics, geology, and mathematics having the lowest percentage of papers authored by women at 14% and psychology having the largest percentage at 39%. Women are significantly underrepresented in top-ranking positions (top 10% or higher) across all fields and metrics (productivity, citations, and degree), indicating that it remains challenging for early researchers (especially women) to reach top-ranking positions, as our results reveal the rankings to be rigid over time. Finally, we show that in most fields, women and men with comparable productivity levels and career age tend to attain different levels of citations, where women tend to benefit more from co-authorships, while men tend to benefit more from productivity, especially in pSTEMs. Our findings highlight that while the participation of women has risen in some fields, they remain under-represented in top-ranking positions. Greater gender participation at entry levels often helps representation, but stronger interventions are still needed to achieve long-lasting careers for women and their participation in top-ranking positions.

Figures (4)

• Figure 1: Participation of women and men researchers per field. Each panel shows the proportion (number) and number (bar chart) of papers in A (authors in B) written by at least one (classified as) women and men per field in the Semantic Scholar dataset. All orange (diagonal lined) bars refer to women, and green (dots design) bars refer to men. In A, the order of the fields goes from top (smallest) to bottom (largest) proportion of papers written by women in the field. In B, the order corresponds to the proportion of authors in the field with arrows comparing both panels: $(\uparrow)$ when the field decreased its position, $(\downarrow)$ increased, and $(\updownarrow)$ maintained the same place, and the number of changed positions when necessary. The right panel in B refers to the growth over time of the percentage of women authors per field, with each line located in the respective % value and its color corresponding to the year referred to in the color bar. The value in parenthesis corresponds to the total growth in the percentage of women from 1975 to 2020. In this panel, we show value proportions solely for women, as the values for men are complementary. In the A and B-left, the x-axis is in logarithmic scale to show the smallest values for women in some fields with lines in 2.5 and 5 million to guide the reader.
• Figure 2: Representation of women and men in top-ranked positions. The left panel corresponds to the position of the top 1,000 most cited women and men per field. We ranked all researchers by citations, and the box plots correspond to the variance of their position in the ranking. For the panels in the right: in the left subpanels, the $y$-axis refers to the representation of each gender in the $x\%$ top-ranked position, orange for women and green for men. $y\approx1$ is a fair representation, $y>1$ is over-representation (in most cases for men), and $y<1$ is under-representation (in most cases for women)--lines in 1.5, 1, and 0.5 are to guide the reader. The representation is the division between the proportion of women (men) in the top-ranked $x\%$ of citations over the participation of women (men). The line in 1 finishes when the minimum value to be included in the $x\%$ consists of the entire field corresponding to the number at the end of each plot. The right subpanels correspond to the minimum number of citations necessary to be in the top-ranked $x\%$, and the $G_{metric}$ term represents the Gini coefficient for the distribution of citations. Both subpanels present the results of fields with the lowest (1), medium (11), and largest (19) participation of women (further details in Section S4).
• Figure 3: Change of the rankings over time. On the left side of the plots, we display the flux values of the top 1,000 most cited researchers in each field across years. Each line starts at the flux value in 1976 represented with a circle (first year, $\mdsmblkcircle$). The final flux value in 2020 is represented with a triangle to the left (decreased flux, $\smallblacktriangleleft$) and the right (increased flux, $\smallblacktriangleright$), and the line in the middle is the average value of the 45 years (average, |). Fields with decreasing flux values are shown in red, while those with increasing flux values are in dark blue. As in previous plots, fields are ordered by the participation of women from lowest (top) to highest (bottom), and a vertical dashed line in 0.06 can guide a baseline for the reader. On the right side, there is a detailed visualization of the number of researchers women/men in the ranking for different career stages per year for the field with the lowest (physics), medium (history), and highest (psychology) participation of women. These curves are derived from polynomial regressions, with further details provided in Section S5.2.
• Figure 4: Effect of productivity and the number of co-authorships on attaining citations.A Toy example of our first approach using the Ordinary Least Squares (OLS) regression to analyze top-ranked women and men researchers. The independent variables were normalized using Min-Max scaling to ensure comparability across fields and to represent the full range of ranking positions, from the beginning to the end. The dependent variable is indicated after the $\sim$ symbol. After the dashed line, we exemplify the log of the number of citations as a function of a researcher’s position in the ranking (inverse, starting from the bottom of the ranking to the top), to account for the exponential growth of citations relative to the ranking position. B Results of OLS regressions analyzing gender differences in the impact of productivity and co-authorships on achieving top-ranked citation status. The left panel shows the $R^{2}$ values for regressions in each field ( S2). The middle panel presents the $F_P$ values, representing the odds ratio for productivity between women and men, while the right panel displays the $F_C$ values, representing the odds ratio for co-authorships between women and men ( S3). If $F_P$ (or $F_C$) is positive, the effect is greater for women than for men; if it is negative, the effect is stronger for men than for women. The marker sizes in the middle and right panels are proportional to the respective $F_P$ and $F_C$ values. A detailed explanation of the regression methods and results can be found in Section S6.