American Politics Workshop: Molly Offer-Westort, Data-Driven Hypotheses and Experimental Designs for Theory Testing and Theory Building

Please join the American Politics Workshop as they host Assistant Professor Molly Offer-Westort, University of Chicago. 

Abstract: Many theories in the social and behavioral sciences can be operationalized in more than one way: a single theoretical claim often corresponds to a class of interventions rather than a single treatment. This creates a mismatch between how we typically analyze experiments—testing a pre-specified arm (or a handful of arms with multiple-testing corrections)—and the theoretical question we often care about: e.g., whether any instantiation of a theory improves outcomes (an “existence” claim), whether all admissible instantiations improve outcomes (a “universal” claim), or whether one instantiation dominates the others by a meaningful margin. This project develops a framework for hypothesis testing and design when the object of interest is a treatment space (e.g., message variants in megastudies or crowdsourced intervention libraries). I formalize hypothesis families indexed by interventions, discuss classical and modern design options (uniform allocation, multiple testing, sample splitting/cross-fitting, two-stage “explore–confirm,” and adaptive allocation), and show how power can be understood as a product of (i) the probability of selecting a “good” intervention during exploration and (ii) standard confirmatory power conditional on that selection. I illustrate the implications with design comparisons and simulation evidence for existence-style hypotheses, including settings with adaptive assignment and design-based inference.

Molly Offer-Westort's work on statistical methodology for social science research integrates machine learning methods with experimental design to answer causal questions. She also has an ongoing substantive research program that examines online behavior to understand how people change their views and attitudes in response to the conversations they take part in and the information they engage with online. She combines these agendas in social media experiments, using approaches like adaptive assignment and policy learning, and incorporating natural language processing methods for flexible conversational interventions.

She has conducted social media experiments to identify the most effective interventions for curbing the spread of misinformation online, to optimally target informational messaging to people hesitant to adopt vaccines, and to measure the efficacy of online deep canvassing. Her work in statistical methodology develops and advances tools for experimental design and analysis, with a particular focus on adaptive experimentation.

Offer-Westort's PhD is joint in Political Science and Statistics & Data Science, conferred by Yale University in 2019; Offer-Westort also holds a Masters in Statistics, also from Yale, and a Masters in Public Affairs, from the Princeton School of Public and International Affairs.

Add Event To My Group

Please sign-in