Research

The United States House of Representatives is intended to be the locally representative body of federal government.  While only district residents are eligible to vote, residents of the entire country can donate to any candidate.  This paper explores how much the pursuit of money from ideologically extreme donors explains candidates' deviations from most closely representing their constituents.  Using geographically-targeted Facebook ads for all 2022 House candidates, I first present evidence that candidate positioning responds to fundraising incentives: Candidates portray themselves as more ideologically extreme in messaging to donors than in communication to voters.  Second, I estimate a structural model of a candidate's ideology choice in which candidates care about winning, which tends to pull them to the ideological center, and fundraising, which tends to pull them to the ideological extremes.  A counterfactual prohibiting out-of-district giving results in the average congressperson choosing an ideology that is ten percent closer to their district's median voter, reducing polarization by about one fifth. 

A growing share of U.S. citizens live in electorally lopsided congressional districts, which may depress their political participation.  While citizens can only vote in their districts' elections, they may donate to candidates anywhere in the country, raising the question of whether individuals disengaged by their electorally lopsided districts find voice through greater non-local giving.  I use the post-2010 congressional redistricting that exogenously reassigned individuals to more, or less, competitive districts to explore this directly.  When an individual's district becomes less competitive, she  donates less to her district's House candidates and more to out-of-district candidates.  Hence, givers regard local and non-local giving as substitutes: a dollar reduction in local giving increases non-local giving by $0.48.  The substitution is strongest for competitive out-of-district races, suggesting that individuals give with the intention of helping their party win nationally.

Platform participation is effective advertising if it delivers net new customers.  I study the efficacy of platform participation as advertising using restaurants' decisions to join OpenTable.  Using a cross-section of 8,000 Texas restaurants, I document that restaurants less likely to have repeat customers have higher probabilities of participation, suggesting that participation would deliver new customers.  Using monthly revenue data of Texas restaurants, 423 of which join the platform, I find revenue effects that are indistinguishable from zero, however.  Moreover, restaurants participate for years, suggesting their operators overestimate the platform's benefits, providing another example of the challenges of measuring advertising effectiveness.

Econometric testing of models of firm conduct when true markups are unobserved is based on a falsifiable restriction (Berry and Haile, 2014). We reinterpret this restriction to shed light on the economic determinants of model falsifiability. We show that whether a model of conduct can be falsified largely depends on an interplay between the variation induced by the instruments, and the differences in pass-through matrices between that model and the truth. Through a set of examples that include the leading models used in empirical work, we illustrate why falsification succeeds or fails. 

This paper explores the contrast between how candidates appeal to ideologically-distinct local voters and national donors.  Using a dataset containing all 2022 general election House candidate Facebook ads, I document that candidates highlight different campaign priorities to local versus national audiences.  Candidates are more likely to mention rising prices and health care in locally-targeted versus nationally-targeted ads.   In electorally balanced districts, candidates are especially likely to highlight these issues locally, generating more issue overlap between district rival candidates.  These results suggest the rise in gerrymandered districts, and separately, candidate incentives to appeal to ideologically-extreme non-voters have contributed to more polarized candidate messaging.

In a differentiated products setting when costs are unobserved, the Cournot model of quantity-setting competition attributes a greater share of prices to markups than does the Bertrand–Nash model of price-setting, leading to lower estimates of marginal costs.

Platforms are growing increasingly powerful, raising questions about whether their power might be exercised with bias. While bias is inherently difficult to measure, we identify a context within the music industry that is amenable to bias testing. Our approach requires ex ante platform assessments of commercial promise – such as the rank order in which products are presented – along with information on eventual product success. A platform is biased against a product type if the type attains greater success, conditional on ex ante assessment. Theoretical considerations and voiced industry concerns suggest the possibility of platform biases in favor of major record labels, and industry participants also point to bias against women. Using data on Spotify curators’ rank of songs on New Music Friday playlists in 2017, we find that Spotify’s New Music Friday rankings favor independent-label music, along with some evidence of bias in favor of music by women. Despite challenges that independent-label artists and women face in the music industry, Spotify’s New Music curation appears to favor them.