Knowledge Specialization in PhD Student Groups

Abstract

It has been argued that specialization within groups yields productivity gains. We evaluate this statement with a focus on groups of PhD students. Using an established technique in computer science, the Latent Dirichlet Allocation (LDA), we construct a novel measure of the dispersion of PhD students' research interests based on their dissertation abstracts and relate it to PhD group publications and citations. We use a rich dataset on groups of PhD students who studied at the Swiss Federal Institute of Technology (EPFL) in Lausanne, Switzerland, during the 1993-2008 period. We find robust evidence that within-group knowledge specialization is associated with a larger number of publications. However, beyond a critical level, specialization hinders the group's publication output. We find similar results for the likelihood that the number of citations received by the group's publications falls in the last quartile. We interpret these results as an indication that gains, in terms of greater research output, can be achieved if PhD students specialize according to their different comparative advantages for the research areas that the group investigates. However, beyond a certain level, knowledge specialization has a detrimental impact on research output, due to increasing communication costs and to an increased likelihood of conflict insurgence.