In this presentation, leaving academic science is explored: how attrition differs across genders, disciplines, and over time. Traditional narratives of women leaving science earlier than men (and leaving it in larger proportions) are revisited, using currently available structured Big Data on academic careers. In this cohort-based, longitudinal approach, the details of careers of STEMM scientists who started publishing in different points in time (year 2000, N=142,776; and year 2010, N=232,843) are examined. Survival analysis shows that attrition is amazingly high – one third of scientists disappears after 5 years and a half after 10 years. However, probabilities of leaving science are powerfully differentiated. In regression models, predictors of staying in science are sought, with special interest in the quantity and quality of publications. Overall, probabilities of attrition are high – but over time, they are ever less gendered. Global bibliometric datasets are tested, opening new opportunities to explore careers nationally and globally. Limitations and trade-offs are discussed and wider implications are shown.