Distilling network effects from Steam

This paper develops a method to estimate the demand for network goods, using minimal network data, but leveraging within-consumer variation. I estimate demand for video games as a function of individuals’ social networks, prices, and qualities, using data from Steam, the largest video game digital distributor in the world. I separately identify price elasticities on individuals with and without friends with the same game, conditional on individual fixed effects and games’ characteristics. I then use the discrepancies between estimated price elasticities to identify the impact of social networks. I compare my method to “traditional-IV” strategies in the literature, which require detailed network data, and find similar results. A 1% increase in friends’ demands, increases demand by .13%. In counterfactual simulations, I find demand increases by about 5% from a promotional giveaway to “influencers,” those users in the top 1% of popularity in the network.