Analysis of Relational Data

Broadly speaking, relational data are observations and outcomes as measured between two individual units: people, schools, countries, and so forth. I focus on methods for evaluating and predicting relations based on individual and relational characteristics. In particular, I use hierarchical/multilevel modelling and tools provided by Bayesian computational statistics to shed new light on old methods and models. Unless stated otherwise, all work is coauthored with Joe Blitzstein.

Projects and Applications:

• Dissertation.
• The ElectroGraph package for R, including several supporting documents.
• Probit Models for Binary Relational Data (contact for working paper). Traditional social network models use log-odds connections to study whether a "tie" exists or not between two individuals. We argue that a model based on the normal distribution (further generalized to the t distribution) has several advantages over the log-odds interpretation of ties.
• Data Augmentation and Gibbs Sampling in Relational Data Models (contact for working paper). Methods for computing binary and non-binary relational data algorithms using ideas popular in Bayesian simulation.
• "Shot Noise" in Count-Based Relational Data (contact for working paper). Poisson and negative-binomial models may undercount large deviations from expected values in cases where unobserved connections exist. We hypothesize using a standard Erdos-Renyi random graph as a latent covariate to generate a "hidden network", whose connective structure can best act as a tool to hypothesize relevant relational types that were previously not considered.
  • An example with Senate Press Releases with Justin Grimmer.