The Role of Gender in Academic Seminar Interactions: A Machine Learning Approach

Recent research has highlighted an unequal treatment of women in academic economics. This paper focuses on economic seminars and the effect of being a female presenter. For that I collected a database of more than 2,000 audio recordings of seminars and presentations in academic settings. Making use of different machine learning algorithms I construct a "diarisation" of each seminar identifying, among others, the interruptions that a presenter has and the gender of all the speakers at that seminar. I find evidence supporting this unequal treatment received by women. On average they receive between 1 and 2 more interruptions per presentation. In addition the interruption arrives earlier on the presentation and they take a longer time than the ones received by males presenters. Interestingly, I also find evidence supporting the fact that an important part of those extra interruptions received by female presenters are driven by females in the audience. The results are robust when I control by affiliation and seniority of the presenter, ranking of the department to which the speaker belongs and topic of the presentation.