Career Interests

Expertise: Bayesian Methods, Survey Data, Spatial Statistics

From a young age I have loved analyzing data (as evidenced by photos of me organizing my trick-or-treating spoils into candy bar graphs). Both the collection and subsequent analysis of data are powerful tools that historically have driven scientific inquiry and increasingly drive decision making in the private sector and public policy. I am passionate about continuing to narrowing the gap between the decision making process and well suited data/analysis for the question at hand.

Bayesian statistics are a natural and intuitive way to quantify uncertainty and risk in the face of decisions. As a Bayesian data scientist, I believe the Bayesian workflow (i.e. eliciting priors from subject matter experts, model specification, prior simulation, model fitting, posterior predictive checks, see Gelman et al 2020) is particularly well suited for understanding how current beliefs of stake holders affect down stream analyses/decisions. Often these decisions involve high-dimensional, multivariate outcomes which require increasingly complex statistical/machine learning models.

In my research I have been drawn to Statistical inference problems, where the primary goal is to draw conclusions or make judgements about a question/population of interest, especially in the social sciences. I am particularly interested in latent variable modeling, that is unobserved variables (i.e. individual preferences) that can be inferred from complex high dimensional data (i.e. shopping cart data, voting history, surveys).

There has been increased interest in explainable AI. In my opinion, when inference is the goal, the gold standard for interpretable models are identifiable models – models where unique sets of parameters lead to distinct probability models. Creating identifiable models which better represent the characteristics of the unobserved variables reduces uncertainty and improves confidence in the decision making process. (e.g. voting behavior of members of congress are explained better when political preferences are represented by a circle rather than a straight line (Yu and Rodriguez 2021)).

I am passionate about effective communication of research and statistical concepts to a larger community. The most interesting questions are always at the intersection of many domains of knowledge, and thus inherently require effective colloboration and communication. Scientific research and data science are useless unless the data-informed decisions can be clearly communicated to both broad and targeted audiences.