Research

Publications

R tutorial

 

Home

There is a growing literature on the relationship between reactions to emotionally charged stimuli and political preferences, and I am working with Mike Alvarez and Ralph Adolphs to improve that literature through more reliable measurements of reactions and finer-grained models of ideology.

I am also working with Mike Alvarez and Laura Loesch on a study relating self-reported fear to biases in item nonresponse in political surveys.

Empirical research that involves mass-level ideology generally uses one of three models: a single liberal-conservative or party ID dimension, completely independent question-specific dimensions, or a priori groupings of questions (economic, social, foreign affairs, etc.).

Ordinal item response theory models are a flexible intermediate between question-by-question analyses or single-dimensional spaces, and they let the data determine how questions group. Unfortunately, such models are difficult to interpret without a hierarchical structure since there is so little information about each individual respondent, but hierarchical multidimensional models are difficult to fit in practice.

I have extended the R package MCMCpack to fit Bayesian ordinal IRT models with hierarchical predictors of ideal points, and I am currently working on further extensions to allow for heteroskedasticity and elliptical salience weights over the ideal point space. Both of these new extensions will also allow for hierarchical predictors.

Geocoding Addresses to Census Blocks

I patched the open-source geocommons Ruby geocoder to output census block information directly from the Census Bureau's TIGER Shapefiles. Email me if you would like the code.