My primary academic interests are the development and application of new methodological approaches to questions regarding American political institutions. At present, my research is focused on identifying challenges associated with the use of ideal point estimation across various institutions and improving on inadequacies.
In particular, my recent work has focused on the quality of different sorts of bridge observations, modeling non-observations, and unifying interesting extensions to the IRT ideal point approach with the original spatial voting model.
My dissertation is compose of three essays on the extension of political measurement to both “big-data” and new kinds of data. A brief description of this collection of essays is below:
Scaling at Scale: Ideal Point Estimation with ‘Big-Data’
Here, I derive and demonstrate approximate inference for ideal point estimation. Existing methods have worked well for the modestly sized data typical of common applications. However, as researchers look to incorporate new kinds of data and bridge across time and actors, the amount of data grows exponentially. The scale of the problems renders current methods computationally intractable for ‘big-data’. Approximate inference, on the other hand, produces equivalent results with just a fraction of the computational resources.
- early draft (about 3MB)
Missing the Point: The President, Abstention, and Ideal Point Estimation
Members of Congress in the United States face considerable pressure to cast a vote one way or another on important legislation. So, on roll calls where votes are recorded, there are very few “abstentions” observed. In other legislative settings, though, where abstentions are frequent, simplistic assumptions about these missing data are shown to bias the recovered estimates. Unlike actual members of Congress, the president’s “voting record” is rife with “abstentions”. Instead of assuming the president abstains at random, I explicitly model the decision to refrain from making a public announcement about his preference. The standard “missing at random” model (and its assumptions) is a special case of this more general approach and can be directly tested. Not only do the data strongly reject a “missing at random” assumption, the bias induced by this assumption on the president’s estimation ideal point is considerable. Estimates from the more general approach reduce the “artificial extremeness” in ideal point estimates for presidents been observed by substantive scholars.
The President’s ‘Votes’ on Roll Calls
The US president announces support or opposition to many of the bills proposed in Congress. These preferences are typically used to recover ideal point estimates for the president along members of Congress. More recently, the decision to sign or veto legislation has also been included. In addition to these sources of expressed preference, I incorporate the president’s issuance of signing statements and the nature of these signing statements. But, in contrast to previous use of heterogeneous data sources for the president, I allow each decision to follow from a slightly different utility calculation. Because of these, one can test hypotheses about whether support, a signature, and a signature with a signing statement are all equivalent. And, in fact, they are not. This provides both an improved understanding of the president’s unobservable ideology and new insights into the nature of these very different decision-making processes.
Any questions regarding the research listed can be sent to my email address.