Multidimensional Diffusion Processes in Dynamic Online Networks

(Job market paper - Coauthored with David Easley and Eleonora Patacchini). We develop a matched sample estimation framework to estimate peer influence effects on item adoption in multidimensional diffusion processes by first inferring preferences using a machine learning algorithm applied to previous adoption behaviors. We show that ignoring previous behaviors leads to significantly overestimating the role of peer influence in the diffusion of information, mistakenly confounding influence-based contagion with diffusion driven by common preferences. Our matching-on-preferences algorithm with machine learning reduces the estimate of peer influence effects significantly more than matching on earlier adoption behaviors, as well other observable characteristics, using directly observable covariates in the data. We also show significant and intuitive heterogeneity in the relative amount of peer influence.

Influence and Homophily in Directed Triadic Closure

Previous work has demonstrated a correlation between directed links created by an agent, and those created by the agents whom she follows. This finding has been used to support claims of a peer influence effect on link copying, a process known as directed triadic closure. However, just like with the diffusion of other behaviors, homophily could explain directed triadic closure in an information-social hybrid network. I provide the first evidence of peer influence causing directed triadic closure in an information-social hybrid network, after controlling for homophily, by adapting the matched sampling estimation framework from (Easley, Patacchini & Rojas 2018) to estimate peer influence effects on link copying. I also show significant heterogeneity in the relative size of the peer influence effect depending on the characteristics of the agents involved in the potential triangle.

Group-Size Effects on Incentives to Contribute to an Online Public Good

Social media platforms depend on the voluntary contributions of individual users, which can be subject to free-riding and social benefit incentives. Using data from an online social network embedded in an open-source platform, I analyze how changes in the number of an agent's followers and followees affect her contributions to open-source software. To control for potential homophily which could bias the estimation of the peer influence effects, I adapt the methodology from (Easley, Patacchini & Rojas 2018) for a 2x2 panel regression setting. The results show that an increase in the number of an agent's followers causes her to contribute more, whereas an increase in the number of an agent's followees cause her to contribute less.