2016

Dean Knox (MIT)

Essays on Modeling and Causal Inference in Network Data

Selection committee: Justin Grimmer (Chicago, chair), Matt Blackwell (Harvard) and Teppi Yamamoto (MIT)

Citation:

We are thrilled to select Dean Knox's dissertation proposal "Essays on Modeling and Causal Inference in Network Data" for the Williams Prize. Knox's dissertation proposal provides new tools for incorporating networks into several methodological traditions including Bayesian statistics, causal inference, and machine learning. Along the way, Knox makes several contributions to political methodology. Notably, he proposes a novel statistical model for "path data," which is an entirely new type of dependent variable in political science research. This model has the potential to answer a large set of important questions: Does distributive politics influence the paths of highway networks? How does information flow over social networks? Knox also develops a sampling algorithm to estimate the parameters of this model, removing the largest obstacle to inference in this setting. Knox also applies network reasoning to the estimation of causal effects. The problem of spillover is well-known and particularly vexing for researchers interested in estimating causal effects. Scholars of networks and computer science have provided numerous tools for analyzing the structure of networks, but those tools have not regularly been applied to estimate causal effects. Knox's dissertation proposal provides a major attempt at bridging the two disparate literatures. Knox breaks ground in introducing new ways of modeling causal effects along a network, new ways to make minimal assumptions to detect network effects, and provides excellent applications of the techniques to problems of interest. In our view the dissertation will represent a major step forward in incorporating networks into political science research.

John T. Williams Dissertation Prize