Paul Goldsmith-Pinkham

Research Economist




Best Figures


Beamer Tips

Applied Method PhD Course

Google Trends UI Forecast

I am an assistant professor at the Yale School of Management. If you need to reach me, please email me at paul.goldsmith-pinkham at yale dot edu.

Research Papers

My papers are organized by broad topic below:

Finance and Housing Papers:

Working Papers

Contagion Effects of the Silicon Valley Bank Run
with Dong Beom Choi and Tanju Yorulmazer

This paper analyzes the contagion effects associated with the failure of Silicon Valley Bank (SVB) and identifies bank-specific vulnerabilities contributing to the subsequent declines in bank stock returns. We find that uninsured deposits, unrealized losses in held-to-maturity securities, bank size, and cash holdings had a significant impact, while better-quality assets or holdings of liquid securities did not help mitigate the negative spillovers. Interestingly, banks whose stocks performed worse post SVB also had lower returns in the previous year following Federal Reserve interest rate hikes. The stock market partially anticipated risks associated with uninsured deposit reliance, but did not price in unrealized losses due to interest rate hikes nor risks linked to bank size. While mid-sized banks experienced particular stress immediately after the SVB failure, over time negative spillovers became widespread except for the largest banks.

Anonymity and Identity Online (slides)
with Florian Ederer and Kyle Jensen

Economics Job Market Rumors (EJMR) is an online forum and clearing house for information about the academic job market for economists. It also includes content that is abusive, defamatory, racist, misogynistic, or otherwise “toxic.” Almost all of this content is created anonymously by contributors who receive a four-character username when posting on EJMR. Using only publicly available data we show that the statistical properties of the scheme by which these usernames were generated allows the IP addresses from which most posts were made to be determined with high probability.1 We recover 47,630 distinct IP addresses of EJMR posters and attribute them to 66.1% of the roughly 7 million posts made over the past 12 years. We geolocate posts and describe aggregated cross-sectional variation—particularly regarding toxic speech—across sub-forums, geographies, institutions, and contributors.

Heterogeneous Real Estate Agents and the Housing Cycle
with Sonia Gilbukh (Revise and Resubmit at the Review of Financial Studies )

The real estate market is highly intermediated, with 90 percent of buyers and sellers hiring an agent to help them transact a house. However, low barriers to entry and fixed commission rates result in a market where inexperienced intermediaries have a large market share, especially following house price booms. Using rich micro-level data on 8.5 million listings and a novel instrumental variables research design, we first show that houses listed for sale by inexperienced real estate agents have a lower probability of selling, and this effect is strongest during the housing bust. We then study the aggregate implications of the distribution of agents’ experience on housing market liquidity by building a dynamic entry and exit model of real estate agents with aggregate shocks. We find that 3.7% more listings would have been sold in a flexible commission equilibrium. It would require a six-fold increase in entry costs for real estate agents to achieve this level of liquidity within the fixed commission framework. ∗

The Great Equalizer: Medicare and the Geography of Consumer Financial Strain
with Maxim Pinkovskiy and Jacob Wallace (Submitted)

We use a five percent sample of Americans’ credit bureau data, combined with a regression discontinuity approach, to estimate the effect of universal health insurance at age 65—when most Americans become eligible for Medicare—at the national, state, and local level. We find a 30 percent reduction in debt collections—and a two-thirds reduction in the geographic varia- tion in collections—with limited effects on other financial outcomes. The areas that experienced larger reductions in collections debt at age 65 were concentrated in the Southern United States, and had higher shares of black residents, people with disabilities, and for-profit hospitals.

Macroeconomic Effects of Debt Relief: Consumer Bankruptcy Protections in the Great Recession
with Adrien Auclert and Will Dobbie (Revise and Resubmit at the American Economic Review)

This paper argues that the debt forgiveness provided by the U.S. consumer bankruptcy system helped stabilize employment levels during the Great Recession. We document that over this period, states with more generous bankruptcy exemptions had significantly smaller declines in non-tradable employment and larger increases in unsecured debt write-downs compared to states with less generous exemptions. We interpret these reduced form estimates as the relative effect of debt relief across states, and develop a general equilibrium model to recover the aggregate employment effect. The model yields three key results. First, substantial nominal rigidities are required to rationalize our reduced form estimates. Second, with monetary policy at the zero lower bound, traded good demand spillovers across states boosted employment everywhere. Finally, the ex-post debt forgiveness provided by the consumer bankruptcy system during the Great Recession increased aggregate employment by almost two percent.

Gender Representation in Economics Across Topics and Time: Evidence from the NBER Summer Institute
with Anusha Chari ( Reject and Resubmit at the Review of Economics and Statistics )

We document the representation of female economists on the conference programs at the NBER Summer Institute from 2001-2018. Over the period from 2016-2018, women made up slightly over 21 percent of all authors on scheduled papers. However, there was large dispersion in the female author share across programs. While the average share of women has slightly risen from 18 percent since 2001-2003, a persistent gap between finance, macroeconomics and microeconomics subfields remains. We examine several channels potentially affecting female representation including gender differences in acceptance and submissions rates, institutional rank, NBER affiliation, faculty seniority and the role of organizers.


Sea Level Rise and Municipal Bond Yields
with Matthew Gustafson, Ryan Lewis and Michael Schwert (May 2023, Review of Financial Studies )

Municipal bond markets began pricing sea-level rise (SLR) exposure risk in 2013, coinciding with upward revisions to worst-case SLR projections and accompanying uncertainty around these projections. The effect is larger for long-maturity bonds and not solely driven by near-term flood risk. We use a structural model of credit risk to quantify the implied economic impact and distinguish between the effects of underlying asset values and of uncertainty. The SLR exposure premium exhibits a trend different from house prices and is unaffected by house price controls. Together, our results highlight the importance of climate uncertainty in driving municipal bond prices.

The Gender Gap in Housing Returns
with Kelly Shue (March 2023, Journal of Finance)

Housing wealth represents the dominant form of savings for American households. Using detailed data on housing transactions across the United States since 1991, we find that single men earn 1.5 percentage points higher unlevered returns per year on housing relative to single women. The gender gap grows significantly larger after accounting for mortgage borrowing: men earn 7.9 percentage points higher levered returns per year relative to women. Approximately 45% of the gap in housing returns can be explained by gender differences in the location and timing of transactions. The remaining gap arises primarily from gender differences in execution prices: data on repeat sales reveal that women buy the same property for approximately 2% more and sell for 2% less. Women experience worse execution prices because of differences in the choice of initial list price and negotiated discount relative to the list price. Gender differences in upgrade and main- tenance rates, and preferences for housing characteristics and listing agents appear to be less important factors. Overall, the gender gap in housing returns is economically large and can explain 30% of the gender gap in wealth accumulation at retirement.

Predictably Unequal? The Effects of Machine Learning on Credit Markets
with Andreas Fuster, Tarun Ramadorai and Ansgar Walther ( Feburary 2022, Journal of Finance ))

Recent innovations in statistical technology, including in evaluating creditworthiness, have sparked concerns about impacts on the fairness of outcomes across categories such as race and gender. We build a simple equilibrium model of credit provision in which to evaluate such impacts. We find that as statistical technology changes, the effects on disparity depend on a combination of the changes in the functional form used to evaluate creditworthiness using underlying borrower characteristics and the cross-category distribution of these characteristics. Employing detailed data on US mortgages and applications, we predict default using a number of popular machine learning techniques, and embed these techniques in our equilibrium model to analyze both extensive margin (exclusion) and intensive margin (rates) impacts on disparity. We propose a basic measure of cross-category disparity, and find that the machine learning models perform worse on this measure than logit models, especially on the intensive margin. We discuss the implications of our findings for mortgage policy.

Changes in Racial and Ethnic Disparities in Access to Care and Health Among US Adults at Age 65 Years
with Jacob Wallace, Karen Jiang and Zirui Song ( July 2021, JAMA Internal Medicine ))

In this cross-sectional study that uses a regression discontinuity design, eligibility for Medicare at age 65 years was associated with marked reductions in racial and ethnic disparities in insurance coverage, access to care, and self-reported health.

Bad Credit, No Problem? Credit and Labor Market Consequences of Bad Credit Reports
with Will Dobbie, Neale Mahoney and Jae Song ( 2020, Journal of Finance )

Credit reports are used in nearly all consumer lending decisions and, increasingly, in hiring decisions in the labor market, but the impact of a bad credit report is largely unknown. We study the effects of credit reports on financial and labor market outcomes using a difference-in-differences research design that compares changes in outcomes over time for Chapter 13 filers, whose personal bankruptcy flags are removed from credit reports after 7 years, to changes for Chapter 7 filers, whose personal bankruptcy flags are removed from credit reports after 10 years. Using credit bureau data, we show that the removal of a Chapter 13 bankruptcy flag leads to a large increase in credit scores and economically significant increases in credit card and mortgage borrowing. Using administrative tax records linked to personal bankruptcy records, we estimate a precise zero effect of flag removal on employment and earnings outcomes. We rationalize these contrasting results by showing that, conditional on basic observables, “hidden” bankruptcy flags are strongly correlated with adverse credit market outcomes but have no predictive power for labor market outcomes.

Consumer Bankruptcy and Financial Health
with Will Dobbie and Crystal Yang (Review of Economics & Statistics, 2017)

This paper estimates the effect of Chapter 13 bankruptcy protection on post-filing financial outcomes using a new dataset linking bankruptcy filings to credit bureau records. Our empirical strategy uses the leniency of randomly-assigned judges as an instrument for Chapter 13 protection. Over the first five post-filing years, we find that Chapter 13 protection decreases an index measuring adverse financial events such as civil judgments and repossessions by 0.316 standard deviations, increases the probability of being a homeowner by 13.2 percentage points, and increases credit scores by 14.9 points. Chapter 13 protection has little impact on open unsecured debt, but decreases the amount of debt in collections by $1,315. We find evidence that both debt forgiveness and protection from debt collectors are important drivers of our results.

Opting Out of Good Governance
with Fritz Foley, Jonathon Greenstein and Eric Zwick (Journal of Empirical Finance, 2018)

Cross-listing on a US exchange does not force foreign firms to follow the exchange’s corporate governance rules. Hand-collected data show that 80% of cross-listed firms opt out of at least one exchange governance rule and those that opt out have a smaller share of independent directors. Cross-listed firms opt out more when coming from countries with weak corporate governance rules, but if these firms are growing and need external financing, they are more likely to comply. For firms in such countries, opting out also lowers firm valuations, decreases the value of cash holdings, and reduces investment sensitivity to market valuations.

Credit Ratings and Security Prices in the Subprime MBS Market
with Adam Ashcraft, Peter Hull and James Vickery ( AER P&P , 2011)

We present and discuss preliminary evidence suggesting that credit ratings significantly influenced prices for subprime mortgage-backed securities issued in the period leading up to the recent financial crisis. Ratings are closely correlated with prices even controlling for a rich set of security- and loan-level controls. This incremental variation in ratings has much less predictive power for security defaults, however, based on findings to date from our ongoing research, suggesting prices were excessively sensitive to ratings relative to their informational content.

Liquidity, Bank Runs, and Bailouts: Spillover Effects During the Northern Rock Episode
with Tanju Yorulmazer (Journal of Financial Services Research, 2009)

In September 2007, Northern Rock—the fifth largest mortgage lender in the United Kingdom—experienced an old-fashioned bank run, the first bank run in the U.K. since the collapse of City of Glasgow Bank in 1878. The run had been contained by the government’s announcement that it would guarantee all deposits in Northern Rock. This paper analyzes spillover effects during the Northern Rock episode and shows that both the bank run and the subsequent bailout announcement had significant effects on the rest of the U.K. banking system, as measured by abnormal returns on the stock prices of banks. The paper also shows that the effects were a rational response by investors to market news about the liability side of banks’ balance sheets. In particular, banks that rely on funding from wholesale markets were significantly affected, a result consistent with the drying up of liquidity in wholesale markets and the record-high levels of the London Interbank Offered Rate (LIBOR) during the crisis.

Resting Papers

Parsing the Content of Bank Supervision
with Beverly Hirtle and David Lucca

We measure bank supervision using the database of supervisory issues, known as matters requiring attention or immediate attention, raised by Federal Reserve examiners to banking organizations. The volume of supervisory issues increases with banks’ asset size, especially for the largest and most complex banks, and decreases with profitability and the quality of the loan portfolio. Stressed banks are faster at resolving issues, but all else equal, resolving new issues takes longer the more issues a bank faces, which may suggest capacity constraints in addressing multiple supervisory issues. Using computational linguistic methods on the text of the issue description, we define five categorical issue topics. The subset of issues related to capital levels and loan portfolio are the most consequential in terms of regulatory rating downgrades and are directly related to changes in banks’ balance sheet characteristics and profitability. Other issues appear to reflect soft information and are less correlated with bank observables. By categorizing questions asked by analysts at banks’ quarterly earnings calls using the same linguistic approach, we find that market monitors raise issues similar to those of supervisors when the issues are related to hard information (such as loan quality or capital) and public supervisory assessment programs.

MBS Ratings and the mortgage credit boom
with Adam Ashcraft and James Vickery (In Revision)

We study credit ratings on subprime and Alt-A mortgage-backed-securities (MBS) deals issued between 2001 and 2007, the period leading up to the subprime crisis. The fraction of highly rated securities in each deal is decreasing in mortgage credit risk (measured either ex ante or ex post), suggesting that ratings contain useful information for investors. However, we also find evidence of significant time variation in risk-adjusted credit ratings, including a progressive decline in standards around the MBS market peak between the start of 2005 and mid-2007. Conditional on initial ratings, we observe underperformance (high mortgage defaults and losses and large rating downgrades) among deals with observably higher risk mortgages based on a simple ex ante model and deals with a high fraction of opaque low-documentation loans. These findings hold over the entire sample period, not just for deal cohorts most affected by the crisis.

Econometrics & Networks Papers:

Working Papers

Contamination Bias in Linear Regression
with Peter Hull and Michal Kolesár ( Revise and Resubmit, American Economic Review

We study the interpretation of regressions with multiple treatments and flexible controls. Such regressions are often used to analyze stratified randomized control trials with multiple intervention arms, to estimate value-added (for, e.g., teachers) with observational data, and to leverage the quasi-random assignment of decision-makers (e.g. bail judges). We show that these regressions generally fail to estimate convex averages of heterogeneous treatment effects, even when the treatments are conditionally randomly assigned and the controls are sufficiently flexible to avoid omitted variables bias. Instead, estimates of each treatment's effects are generally contaminated by a non-convex average of the effects of other treatments. Thus, recent concerns about heterogeneity-induced bias in regressions leveraging potential outcome restrictions (e.g. parallel trends assumptions) also arise with "design-based" identification strategies. We discuss solutions to the contamination bias and propose a new class of efficient estimators of weighted average effects that avoid bias. In a re-analysis of the Project STAR trial, we find minimal bias because treatment effect heterogeneity is largely idiosyncratic. But sizeable contamination bias arises when effect heterogeneity becomes correlated with treatment propensity scores.


Measuring Changes in Disparity Gaps: An Application to Health Insurance
with Karen Jiang, Zirui Song and Jacob Wallace ( May 2022, AER P&P ))

We propose a method for reporting how program evaluations reduce gaps between groups, such as the gender or Black-white gap. We first show that the reduction in disparities between groups can be written as the difference in conditional average treatment effects (CATE) for each group. Then, using a Kitagawa-Oaxaca-Blinder-style decomposition, we highlight how these CATE can be decomposed into unexplained differences in CATE in other observables versus differences in composition across other observables (e.g. the “endowment”). Finally, we apply this approach to study the impact of Medicare on American's access to health insurance.

Interacting Regional Policies in Containing a Disease
with Arun G. Chandrasekhar, Matthew Jackson and Samuel Thau ( April 2021, Proceedings of the National Academy of Sciences )

Regional quarantine policies, in which a portion of a population surrounding infections are locked down, are an important tool to contain disease. However, jurisdictional governments -- such as cities, counties, states, and countries -- act with minimal coordination across borders. We show that a regional quarantine policy's effectiveness depends upon whether (i) the network of interactions satisfies a balanced-growth condition, (ii) infections have a short delay in detection, and (iii) the government has control over and knowledge of the necessary parts of the network (no leakage of behaviors). As these conditions generally fail to be satisfied, especially when interactions cross borders, we show that substantial improvements are possible if governments are outward-looking and proactive: triggering quarantines in reaction to neighbors' infection rates, in some cases even before infections are detected internally. We also show that even a few lax governments -- those that wait for nontrivial internal infection rates before quarantining -- impose substantial costs on the whole system. Our results illustrate the importance of understanding contagion across policy borders and offer a starting point in designing proactive policies for decentralized jurisdictions.

Bartik Instruments: What, When, Why and How
with Isaac Sorkin and Henry Swift ( 2020, American Economic Review )

The Bartik instrument is formed by interacting local industry shares and national industry growth rates. We show that the Bartik instrument is numerically equivalent to using local industry shares as instruments. Hence, the identifying assumption is best stated in terms of these shares, with the national industry growth rates only affecting instrument relevance. We then show how to decompose the Bartik instrument into the weighted sum of the just-identified instrumental variables estimators, where the weights sum to one, can be negative and are easy to compute. These weights measure how sensitive the parameter estimate is to each instrument. We illustrate our results through three applications: estimating the inverse elasticity of labor supply, estimating local labor market effects of Chinese imports, and using simulated instruments to study the effects of Medicaid expansions.

Social Networks and the Identification of Peer Effects
with Guido Imbens (Journal of Business & Economic Statistics, 2013)

There is a large and growing literature on peer effects in economics. In the current article, we focus on a Manski-type linear-in-means model that has proved to be popular in empirical work. We critically examine some aspects of the statistical model that may be restrictive in empirical analyses. Specifically, we focus on three aspects. First, we examine the endogeneity of the network or peer groups. Second, we investigate simultaneously alternative definitions of links and the possibility of peer effects arising through multiple networks. Third, we highlight the representation of the traditional linear-in-means model as an autoregressive model, and contrast it with an alternative moving-average model, where the correlation between unconnected individuals who are indirectly connected is limited. Using data on friendship networks from the Add Health dataset, we illustrate the empirical relevance of these ideas.

Composite Poisson Models for Goal Scoring
with Phil Everson (Journal of Quantitative Analysis in Sports, 2008)

Goal scoring in sports such as hockey and soccer is often modeled as a Poisson process. We work with a Poisson model where the mean goals scored by the home team is the sum of parameters for the home team's offense, the road team's defense, and a home advantage. The mean goals for the road team is the sum of parameters for the road team's offense and for the home team's defense. The best teams have a large offensive parameter value and a small defensive parameter value. A level-2 model connects the offensive and defensive parameters for the k teams. Parameter inference is made by imagining that goals can be classified as being strictly due to offense, to (lack of) defense, or to home-field advantage. Though not a realistic description, such a breakdown is consistent with our model assumptions and the literature, and we can work out the conditional distributions and generate random partitions to facilitate inference about the team parameters. We use the conditional Binomial distribution, given the Poisson totals and the current parameter values, to partition each observed goal total at each iteration in an MCMC algorithm.

Covid-19 Papers:

Working Papers

Can a Trusted Messenger Change Behavior when Information is Plentiful? Evidence from the First Months of the COVID-19 Pandemic in West Bengal
with Abhijit Banerjee, Marcella Alsan, Emily Breza, Arun G. Chandrasekhar, Abhijit Chowdhury, Esther Duflo, and Benjamin A. Olken (Revise and Resubmit at the Review of Economics and Statistics )

Can information from a credible messenger shift behavior in an information-saturated environment? In a randomized controlled trial involving twenty-eight million individuals in West Bengal, we find that SMS-delivered video messages containing information about COVID-19 symptoms and health-preserving behaviors recorded by a credible messenger increased adherence to targeted and nontargeted preventive behaviors, measured by two objective measures (symptoms reported to a health worker, and phone usage at home),as well as self-reported behaviors. We find large spillovers onto nontargeted recipients. Credible light-touch messaging can play an important role in crisis response, even when similar information is widely available.


Excess Death Rates for Republican and Democratic Registered Voters in Florida and Ohio During the COVID-19 Pandemic
with Jacob Wallace and Jason Schwartz ( JAMA IM July 2023) )

Between January 1, 2018, and December 31, 2021, there were 538 159 individuals in Ohio and Florida who died at age 25 years or older in the study sample. The median age at death was 78 years (IQR, 71-89 years). Overall, the excess death rate for Republican voters was 2.8 percentage points, or 15%, higher than the excess death rate for Democratic voters (95% prediction interval [PI], 1.6-3.7 percentage points). After May 1, 2021, when vaccines were available to all adults, the excess death rate gap between Republican and Democratic voters widened from −0.9 percentage point (95% PI, −2.5 to 0.3 percentage points) to 7.7 percentage points (95% PI, 6.0-9.3 percentage points) in the adjusted analysis; the excess death rate among Republican voters was 43% higher than the excess death rate among Democratic voters. The gap in excess death rates between Republican and Democratic voters was larger in counties with lower vaccination rates and was primarily noted in voters residing in Ohio.

The impact of large-scale social media advertising campaigns on COVID-19 vaccination: Evidence from two randomized controlled trials
with Lisa Y. Ho, Emily Breza, Marcella Alsan, Abhijit Banerjee, Arun G. Chandrasekhar, Fatima Cody Stanford, Renato Fior, Kelly Holland, Emily Hoppe, Louis-Maël Jean, Lucy Ogbu-Nwobodo, Benjamin A. Olken, Carlos Torres, Pierre-Luc Vautrey, Erica Warner, and Esther Duflo ( AEA P&P 2023) )

COVID-19 vaccines are widely available in wealthy countries, yet many people remain unvaccinated. Understanding the effectiveness -- or lack thereof -- of popular vaccination campaign strategies is therefore critical. In this paper, we report results from two studies that tested strategies central to current vaccination outreach: (1) direct communication by health professionals addressing questions about vaccination and (2) efforts to motivate individuals to promote vaccination within their social networks. Near the peak of the Omicron wave, doctor- and nurse-produced videos were disseminated to 17.8 million Facebook users in the US and 11.5 million in France. In both countries, we cannot reject the null of no effect of any of the interventions on any of the outcome variables (first doses - US and France, second doses and boosters - US). We can reject very small effects on first doses during the interventions in both countries (0.16pp - US, 0.021pp - France). In contrast with similar campaigns earlier in the pandemic to encourage health-preserving behaviors, messaging at this stage of the pandemic -- whether aimed at the unvaccinated or those tasked with encouraging others -- did not change vaccination decisions.

End COVID-19 in low- and middle-income countries
with Ahmed Mushfiq Mobarak, Edward Miguel, Jason Abaluck, Amrita Ahuja, Marcella Alsan, Abhijit Banerjee, Emily Breza, Arun G. Chandrasekhar, Esther Duflo, James Dzansi, Denise Garrett, Gregg S. Gonsalves, Muhammad Maqsud Hossain, Aleksandra Jakubowski, Gagandeep Kang, Arjun Kharel, Michael Kremer, Niccolo Meriggi, Carol Nekesa, Benjamin A. Olken, Saad B. Omer, Firdausi Qadri, Helen Rees, Babatunde Salako, Maarten Voors, Shana Warren and Witold Więcek ( Science (Policy Forum, 10 March 2022) )

Vaccines are changing the course of the COVID-19 pandemic, but in grossly uneven ways. Low- and middle-income countries (LMICs) face considerable obstacles in both receiving and distributing doses. To limit virus transmission, its devastating impacts, and opportunities for further mutations, this must change. Until it does, nonpharmaceutical interventions such as masking must remain a priority. Science invited global experts to highlight research and innovations aimed at quickening the end of COVID-19 in LMICs.

Effects of a large-scale social media advertising campaign on holiday travel and COVID-19 infections: a cluster randomized controlled trial
with Emily Breza, Fatima Cody Stanford, Marcela Alsan, M.D. Ph.D., Burak Alsan, Abhijit Banerjee, Arun G. Chandrasekhar, Sarah Eichmeyer, Traci Glushko, Kelly Holland, Emily Hoppe, Mohit Karnani, Sarah Liegl, Tristan Loisel, Lucy Ogbu-Nwobodo, Benjamin A. Olken Carlos Torres, Pierre-Luc Vautrey, Erica Warner, Susan Wootton, and Esther Duflo ( Nature Medicine , August 2021 )

During the Coronavirus Disease 2019 (COVID-19) epidemic, many health professionals used social media to promote preventative health behaviors. We conducted a randomized controlled trial of the effect of a Facebook advertising campaign consisting of short videos recorded by doctors and nurses to encourage users to stay at home for the Thanksgiving and Christmas holidays (NCT04644328 and AEARCTR-0006821). We randomly assigned counties to high intensity (n = 410 (386) at Thanksgiving (Christmas)) or low intensity (n = 410 (381)). The intervention was delivered to a large fraction of Facebook subscribers in 75% and 25% of randomly assigned zip codes in high- and low-intensity counties, respectively. In total, 6,998 (6,716) zip codes were included, and 11,954,109 (23,302,290) users were reached at Thanksgiving (Christmas). The first two primary outcomes were holiday travel and fraction leaving home, both measured using mobile phone location data of Facebook users. Average distance traveled in high-intensity counties decreased by −0.993 percentage points (95% confidence interval (CI): –1.616, −0.371; P = 0.002) for the 3 days before each holiday compared to low-intensity counties. The fraction of people who left home on the holiday was not significantly affected (adjusted difference: 0.030; 95% CI: −0.361, 0.420; P = 0.881). The third primary outcome was COVID-19 infections recorded at the zip code level in the 2-week period starting 5 days after the holiday. Infections declined by 3.5% (adjusted 95% CI: −6.2%, −0.7%; P = 0.013) in intervention compared to control zip codes. Social media messages recorded by health professionals before the winter holidays in the United States led to a significant reduction in holiday travel and subsequent COVID-19 infections.

Effect of Physician-Delivered COVID-19 Public Health Messages and Messages Acknowledging Racial Inequity on Black and White Adults’ Knowledge, Beliefs, and Practices Related to COVID-19
with Carlos Torres, Lucy Ogbu-Nwobodo, Marcella Alsan, Fatima Cody Stanford, Abhijit Banerjee, Emily Breza, Arun Chandrasekhar, Sarah Eichmeyer, Mohit Karnani, Tristan Loisel, Benjamin Olken, Pierre-Luc Vautrey, Erica Warner and Esther Duflo ( JAMA Network Open, July 2021 ))

In this randomized clinical trial of 18,223 White and Black adults, a message delivered by a physician increased COVID-19 knowledge and shifted information-seeking and self-protective behaviors. Effects did not differ by race, and tailoring messages to specific communities did not exhibit a differential effect on knowledge or individual behavior.

Comparison of Knowledge and Information-Seeking Behavior After General COVID-19 Public Health Messages and Messages Tailored for Black and Latinx Communities
with Marcella Alsan, Fatima Cody Stanford, Abhijit Banarjee, Emily Breza, Arun Chandrasekhar, Sarah EichMeyer, Lucy Ogbu-Nowbodo, Ben Olken, Carlos Torres, Anirudh Sankar, Pierre-Luc Vautrey and Esther Duflo ( Annals of Internal Medicine, Dec 2020 ))

The paucity of public health messages that directly address communities of color might contribute to racial and ethnic disparities in COVID-19–related knowledge and behavior. This randomized trial examined whether physician-delivered prevention messages affect knowledge and information-seeking behavior of Black and Latinx persons and if this differs according to the race/ethnicity of the physician and the tailored content.