Figure 1: An illustration of Berkson’s effect when hiring faculty in different institutions.

(a) When applying for faculty positions, a candidate’s success generally depends on their skills at both teaching and research. We assume that these abilities are statistically independent in the overall field of applicants. (b) At comprehensive institutions, the hiring process considers both skills and eliminates candidates who are both bad teachers and bad researchers. Consequently, the two abilities become negatively correlated among successful candidates. (c) A set of specialized institutions, each one dedicated either purely to teaching or purely to research, select faculty based solely on the relevant ability in each case—a probabilistic mixture of both causal mechanisms as shown in the inset. Knowing that a candidate was successful in this scenario only reveals information about one of their abilities, and consequently induces weaker negative correlations than in b, due to the larger fraction of faculty members who are skilled at both.