Scientific Networks and Co-authorship

Using a database of all published articles in economic journals over the last 30 years, we investigate the determinants of scientific co-authorship. We find that, controlling for pair-wise fixed effects and author productivity, a new collaboration emerges faster if the two authors are more closely connected, either directly or indirectly, through collaborations with others. This effect is strong and robust. It is consistent with a referral model of the formation of scientific collaborations. We also find that large differences in research output between authors favor co-authorship, suggesting that scientific collaboration is more likely between authors with dissimilar ability and experience. This effect disappears if we limit the data to highly productive authors.