Predicting scientific output from interactions at conferences

Across scientific fields, it is common to have annual meetings, conferences, or symposia, where scientists can meet, present their work and hear about their peers' progress. These meetings often serve as incubators for new collaborations. But how good are they at generating successful collaborations that actually lead to scientific output? Here we analyze a unique dataset that enables us to draw connections between initial meetings of scientists and eventual peer-reviewed publications years after. We analyze data from thousands of pairs of scientists spanning multiple conferences in different fields. We propose a simple mathematical model to predict which pairs of authors will end up publishing, and show that it outperforms competing models. This provides new insight into the drivers of collaboration, but also enables us to make predictions as to whether a pair of scientists will be productive together or not — perhaps enabling us to design future conferences more effectively.