Fig. 2: The gender gap in self-promotion is not fully explained by women’s under-representation on Twitter.

A, Predicted probability of self-promotion by gender based on a mixed-effects logistic regression model (Table 1) fitted to all 11,396,752 (paper, author) pairs where the dependent variable is coded as 1 if the author has self-promoted the paper. B Same as A, but for the predicted probability of active presence on Twitter by gender, based on a regression model with the same set of controls (Table 2). The dependent variable is coded as 1 if the author is active on Twitter at the paper’s publication date. C Predicted probability of self-promotion for the subset of 618,742 (paper, author) pairs where the author is active on Twitter (Table 3). All three regression models have the same set of controls, including publication year, journal impact factor, authorship position, number of authors, affiliation rank and ___location, author productivity and number of citations, research topics, and paper random effects. Control variables have been set to their median values to create these plots. Error bars indicate 95% bootstrapped confidence intervals.