ABSTRACT
We analyse academic success using a genealogical approach to the careers of over 95,000 scientists in mathematics and associated fields in physics and chemistry. We look at the effect of Ph.D. supervisors (one's mentors) on the number of Ph.D. students that one supervises later on (one's mentees) as a measure of academic success. Supervisors generally provide important inputs in Ph.D. projects, which can have long-lasting effects on academic careers. Moreover, having multiple supervisors exposes one to a diversity of inputs. We show that Ph.D. students benefit from having multiple supervisors instead of a single one. The cognitive diversity of mentors has a subtler effect in that it increases both the likelihood of success (having many mentees later on) and failure (having no mentees at all later on). We understand the effect of diverse mentorship as a high-risk, high-gain strategy: the recombination of unrelated expertise often fails, but sometimes leads to true novelty.