Global method for gender profile estimation from distribution of first names

Current approaches to infer the gender profile of a group exploit empirical correlations observed in the population at large to guess, one by one, the likely gender of every name in the group. We show that such "individual-based gender estimation methods" (iGEMs) are logically inconsistent due to implicit reliance on a fair gender sampling assumption. Moreover, their gender estimates are intrinsically biased, systematically overestimating the participation of the minority gender. We introduce an inference strategy based on a global and self-consistent analysis of the target list of names that, by eliminating the fair gender sampling assumption and taking into account contextual information, aims at a more accurate gender classification. Our "global gender estimation method" (gGEM) relies on a leaky pipeline model of a social dynamic and is devoid of the intrinsic methodological errors of iGEMs. We demonstrate the effectiveness of gGEM against iGEMs using real-world examples. We also employ artificially generated populations of varying gender compositions to determine the quantitative scale of the problem. Our method outperforms iGEM approaches, particularly for populations showing a high degree of gender bias or large numbers of unisex names.