The Visible Hand of Social Networks in Electronic Markets

Abstract:     
The sustained increase in different forms of electronic interaction over the last decade has led to the emergence of a number of electronic and visible networks that connect consumers, products and businesses. In this paper, we conjecture that the visibility of these networks influences a wide variety of choices and outcomes in electronic markets, and analyze the nature and extent of influence that their visibility induces. We do so by developing an extended model of social or "peer" effects that separates network-induced social influence from demand correlations that are caused by product complementarity, by product characteristics, and by self-selected groupings. Our main analytical result provides a simple set of conditions under which the influence that visible social networks have on demand can be econometrically identified, and we show that our conditions place very minimal empirical restrictions on the structure of the networks that define the pattern of social influence. We estimate this model using data about the demand and co-purchase networks for over 250,000 books offered on Amazon.com over the period of a year. Our empirical results show that the explicit visibility of a co-purchase relationship more than triples the average influence that complementary products have on each others' demand. Furthermore, the magnitude of this social influence is higher for more popular books, for more recently published books, and varies in counter-intuitive ways with changes in pricing, secondary market activity and assortative mixing across product categories. Our paper presents new evidence quantifying the role of network "position" and the influence of visible social networks in electronic markets, highlighting the power of basing virtual shelf position or slotting on consumer preferences that are revealed through shared purchasing patterns. It also offers new results for the identification of peer effects which reverse the impossibility issue associated with the reflection problem of Manski (1993), establishing in the process that robust identification is simplified considerably when there are multiple overlapping networks mediating peer influence.

Keywords: networks, social networks, peer effects, electronic commerce, ecommerce, recommender systems, identification, selection, influence

JEL Classifications: D85, L14, Z13

Working Paper Series