Research

My work sits at the intersection of political science, statistics, and applied methodology. I’m motivated by a single question: when the conditions for a clean experiment aren’t available, how can we still learn something honest from the data we have?

That question breaks into three methodological threads and several application areas.

Methodological threads

Identification strategies

How do we translate substantive assumptions into claims the data can speak to? And what do our conclusions become when those assumptions fail? Work here includes dedicated treatments of instrumental-variables-like designs under imperfect conditions, difference-in-differences extensions, and synthetic-control-style estimators tailored to rare, high-stakes outcomes.

Sensitivity analysis

Even well-identified analyses rest on assumptions that can’t be fully tested. Sensitivity analysis makes the size of that uncertainty explicit — how strong would unobserved confounding or violations of exchangeability have to be to overturn a conclusion? I’ve contributed tools for sensitivity to omitted variable bias in regression, in instrumental-variable settings, in weighting- and matching-based analyses, and in synthetic-control designs. The goal is to give readers a principled vocabulary for debating results rather than arguing past one another.

Estimation and modeling

Many causal estimators embed strong parametric choices. I work on methods that relax those choices while keeping interpretability: kernel-based regression (KRLS), kernel balancing (KBAL), covariate balancing propensity scores for continuous and multi-valued treatments, and prognostic-score augmented estimators. The emphasis is on making flexible methods safe — well-behaved in small samples, transparent about what they’re doing, and accompanied by honest standard errors.

Application areas

Political violence and mass atrocity prevention

With the US Holocaust Memorial Museum I co-lead the Early Warning Project’s statistical risk assessment — annual forecasts of which countries are at heightened risk of mass killing. Related work studies how terrorist attacks reshape domestic coalitions and support for security forces, and how conflict and its aftermath interact with institutional trust.

Environmental and health policy

Questions about wildfire exposure and climate attitudes, air pollution and mortality, and evaluating state-level policies (e.g., firearm regulations, treatment-access laws) where randomized experiments are impossible.

Medicine

Collaborations with medical colleagues on causal questions in clinical data — from immunotherapy response to treatment sequencing — where small samples and rich covariates demand flexible, interpretable methods.

Get in touch

If you’re working on a methodological or applied problem that sounds like one of these, I’m always glad to hear about it — see the home page for contact info.