Data & Analysis

I build datasets from sources that arrive scattered, unstructured, or only on paper.

• Web data at scale. I scraped more than 1.6 million posts from the campaign, personal, and official Twitter/X accounts of sitting U.S. senators, from 2008 to the present.
• Administrative records. I assembled more than 64,000 franking advisory-opinion submissions from the House Clerk’s records, 2018 through January 2025, the first systematic dataset of its kind.
• Paper archives. Records filed before 2018 exist only on paper. With a team of undergraduates, I copied more than 10,000 documents in one week at the Legislative Resource Center and am converting them into digital data.

Most of my questions require joining data that were never designed to fit together, at the level of the individual legislator over time.

• I linked senators’ tweets to monthly campaign-finance records to test whether local messaging is associated with donor support.
• I combined monthly counts of franked communications with the House’s quarterly Statements of Disbursements to track how members allocate spending on print and digital mail.
• I compared inflation-adjusted communication spending from 1991–1995 with spending from 2018–2024, putting historical and modern records on a common scale.

I translate abstract concepts into measures that capture more than one dimension, and I test whether those measures hold up.

• Localness. Existing measures treat any mention of a state as “local.” I am developing a continuous measure that places communication on a range from national to local, distinguishes passing references from sustained attention, and records how the state is invoked, for example as a hashtag or as the beneficiary of constituency service. I will validate it against survey respondents’ perceptions.
• Electoral pressure. I operationalize pressure several ways: reelection timing, first terms, prior margins, primary challenges, and departures from the chamber through retirement, runs for other office, or temporary appointment.
• Institutional power and rhetoric. I measure agenda control through committee and subcommittee chairs and majority-party seniority, and I classify messages as problem-oriented or solution-oriented.

Communication and spending data are counts that are overdispersed, clustered by legislator, and often full of zeros. I choose models that fit those features rather than forcing the data into a default.

• Count models, including negative binomial and zero-inflated negative binomial models. I have taught graduate students when and why to use zero-inflated models.
• Analysis of timing, such as how outreach changes in the months before a primary or the quarter leading up to it.
• Regression, hypothesis testing, and sampling, which I also teach in quantitative methods courses.

Numbers explain patterns, and qualitative evidence explains why they occur. My training in public law and political history shapes how I read primary sources.

• Archival research in congressional records and founding-era documents.
• Interviews with communication directors for sitting members of the House, conducted with my undergraduate research team.
• Doctrinal analysis of case law, including a study of Mississippi case law on race and abortion before and after Dobbs.