Research Opportunities for Undergraduates

Summer 2026

"Divergence in beliefs"

with Marina Agranov

A large literature in economics and political science studies how individuals form beliefs and update them upon receiving new information. Prior work has documented systematic deviations from Bayesian updating and shown that these deviations vary with factors such as signal precision, the contentiousness of the topic (political vs. neutral), and individual characteristics of the respondent. In this project, I propose a different perspective. Rather than focusing on the gap between observed and Bayesian beliefs, we will examine the dispersion of beliefs—that is, the extent of belief divergence—and how this dispersion changes in response to new information. I have the dataset from 600 responders (general population of U.S. stratified by gender). The student research will analyze this data and should have a basic knowledge of probability and statistics and be comfortable working with empirical data. Experience with basic statistical tests and, regression analysis is desirable. Prior experience with statistical software (e.g., Stata, R, or Python) is a plus but not strictly required.

"A model of generative AI and research output"

with Michael Gibilisco and Alex Hirsh

How does generative AI affect the modes of research and the long-run production of knowledge? In this project, we aim to model researchers whose goal is to produce applications of knowledge. In the model, researchers have two tools with which to do so: they can either innovate or mine the current stock of knowledge. Innovation potentially adds to the stock of knowledge for future researchers but mining does not. Generative AI increases the productivity of mining current knowledge, which increases the ability to solve applications but does not add to the stock of knowledge for future use.

The Political Economy of Municipal Austerity in California during the Great Depression

with Pawel Janas

This project examines how the identities of elected city councilmembers, their party/reform affiliation, occupational background, and demographics shaped the austerity measures adopted by California cities during the Great Depression. Inspired by the regression-discontinuity (RD) designs of Ferreira & Gyourko (2009, 2014), which use close elections to isolate the policy effects of political leaders, we apply the same logic to the Depression, when cities faced severe fiscal shocks yet responded with sharply different mixes of layoffs, wage cuts, service reductions, and bond defaults.

Choices vs. Valuations

with Kirby Nielsen

Empirical researchers in economics typically observe one of two types of data: choices (i.e., a decision-maker's selection of one or more items from a menu of options) or valuations (i.e., a decision-maker's subjective price or value for a single option). Prior research has documented inconsistencies between choices and valuations: Individuals often choose an item to which they ascribe lower valuation. For example, in isolation, an individual might say that they like apples a lot and think that oranges are just okay, but in a choice between an apple and an orange they might choose the orange. This presents a paradox---which item is truly preferred when one item is valued higher but the other item is chosen? Knowing how to answer this is critical to be able to understand preferences, predict choices, and model decision-making behavior.

Mathematical Models of Consumer Choice

with Kota Saito

This project asks: what are the observable implications of a random utility model (RUM) when the available data are aggregated or partially unobservable? In many applications, researchers only observe market shares for a subset of alternatives, or they see aggregated shares such as the total market share of all Toyota cars. Classical characterization results for RUM apply only to complete datasets in which all choice frequencies for all underlying (non-aggregated) alternatives are observed, and extending these characterizations to incomplete datasets has been an open question since the 1980s. Building on new characterizations of RUM, we study the limitations of this outside-option approach by clarifying which implications of RUM are lost upon aggregation, and how much information about preferences and substitution patterns is discarded by aggregation and partial observability. The SURF project will extend these characterizations, explore applications to empirical IO, study systematic deviations from RUM, and apply the same methodology to deterministic individual choice to propose a new measure of irrationality.