September 30, 2015

Tree-based models for political science data

In this paper, we introduce a family of tree-based nonparametric techniques that are often more appropriate than traditional methods for confronting these data challenges. In particular, tree models are extremely effective for detecting nonlinearities and interactions in datasets with many (potentially irrelevant) covariates.

Forthcoming at American Journal of Political Science.  With Santiago Olivella
Local copy (pdf) | Supplemental Information (pdf) 

September 30, 2015

The Effects of Congressional Staff Networks in the U.S. House of Representatives

In this paper, we use a novel dataset of comprehensive longitudinal employment records from the U.S. House of Representatives to show that Congressional staff -- whose careers often cross multiple offices -- help disseminate legislative expertise within parties and develop and reinforce the voting patterns of legislators.

Forthcoming at the Journal of PoliticsWith Brendan Nyhan.
Local Copy (pdf)

September 29, 2015

An Informed Forensics Approach to Detecting Vote Irregularities

We deploy a Bayesian additive regression trees (BART) model -- a machine-learning technique -- on a large cross-national dataset to explore the dense network of potential relationships between various forensic indicators of anomalies and electoral fraud risk factors, on the one hand, and the likelihood of fraud, on the other. 

2015. Political Analysis 23 (4): 488-505 With Santiago Olivella, Joshua D. Potter, and Brian F. Crisp.