Fig. 10

ROC-Curve for algorithmic gender detection. The intersection point of the country-specific gender detection ratio (sorted in descending order, blue line) and the corresponding cumulative author count (red line, countries with a large amount of authors are indicated by an additional label) defines a adaptive threshold detection criterion θ for the inclusion of a country in the analysis. By definition, θ also defines the fraction of considered authors. In this example countries with a detection rate of at least theta = 0.776 (i.e. 77.6 % of all authors from N = 67 countries) are included in the analysis