Knowledge that Transforms

We propose a Cramér–von Mises–type test for testing whether the mean potential outcome given a specific treatment level has a weakly monotonic relationship with the continuous treatment under unconfoundedness. To flexibly control for a possibly high-dimensional set of covariates, our test is based on a double debiased machine learning method. We show that our test controls asymptotic size and is consistent against any fixed alternative. We apply our test to evaluate the Job Corps program and reject a weakly negative relationship between the treatment (hours in academic and vocational training) and labor market performance among relatively low treatment values.

We study the influence of workplace peers on a high-stakes, irreversible retirement plan choice. Midcareer U.S. military personnel choose between higher future pension payouts or an immediate bonus plus lower future payouts. With peers defined as those who have locked in their choices and personnel assignment rules ensuring that peer groups are exogenously formed, we capture the causal impact of peers. Greater peer take-up of the bonus, which is difficult to compare to the alternative plan but often extremely costly over one’s lifetime, discourages choosers from selecting the bonus. Peers have special impact within professional, race, and gender groups.

We provide the first estimates of intergenerational income mobility using tax data for a large developing country, namely Brazil. We measure formal income from tax and payroll data, and we train machine learning models on census and survey data to predict informal income. We quantify the estimation bias resulting from income imputation and other sources of measurement error, and show that such bias remains negligible in our context. A 10 percentile increase in parental income rank is associated on average with a 5.5 percentile increase in child income rank, with considerable variation across sociodemographic groups and geographical areas.

Predictive algorithms inform consequential decisions in settings with selective labels: outcomes are observed only for units selected by past decision makers. This creates an identification problem under unobserved confounding — when selected and unselected units differ in unobserved ways that affect outcomes. We propose a framework for robust design and evaluation of predictive algorithms that bounds how much outcomes may differ between selected and unselected units with the same observed characteristics. These bounds formalize common empirical strategies including proxy outcomes and instrumental variables. Our estimators work across bounding strategies and performance measures such as conditional likelihoods, mean squared error, and true/false positive rates. Using administrative data from a large Australian financial institution, we show that varying confounding assumptions substantially affects credit risk predictions and fairness evaluations across income groups.

This paper estimates the size of the media multiplier, an easily generalizable model-based measure of how far media coverage magnifies the economic response to shocks. We combine monthly aggregated and anonymized credit card activity data from 114 card-issuing countries in 5 destination countries with a large corpus of news coverage in issuing countries reporting on violent events in the destinations. To define and quantify the media multiplier, we estimate a model in which latent beliefs, shaped by either events or news coverage, drive card activity. According to the model, media coverage can more than triple the economic impact of an event. We document, through our model, that this effect is highly heterogeneous and depends on the broader media representation of countries in each other’s news. We speculate about the role of the media in driving international travel patterns.

We examine how the 1993 Family and Medical Leave Act (FMLA) impacts the gender composition at U.S. firms experiencing a negative demand shock. Combining changes in Chinese imports across industries between 2000 and 2003 and a sharp regression discontinuity to identify FMLA status, we find that an increase in import competition decreases the share of female employment, earnings, and promotions at FMLA relative to non-FMLA firms. This effect is driven by women in prime child-bearing ages and without college degrees and is pronounced at firms with all male managers. These results suggest that job-protected leave mandates may exacerbate gender inequalities in response to adverse shocks.

This paper revisits the relationship between federal debt and interest rates in the U.S. A common approach is to regress long-term forward interest rates on long-term projections of federal debt. We show that issues regarding nonstationarity have become more pronounced over the last 20 years, significantly biasing recent estimates. Estimating the model in first differences rather than in levels addresses these concerns. We find that a 1 percentage point increase in the debt-to-GDP ratio raises the 5-year-ahead, 5-year Treasury rate by about 3 basis points. Roughly half of the interest rate response is driven by a higher nominal term premium.

We study how economic development affects female labor force participation, focusing on the role of gender norms. Analyzing quasi-random variation in provincial exports in reunified Vietnam from 2002 to 2018, we find that positive economic shocks reduced women’s labor market engagement, particularly among married women from wealthier households and those with husbands in more skilled occupations. This trend is more pronounced in the South (formerly capitalist) than in the North (always socialist), and among native Southerners compared to Northerners relocated to the South after the war. Our findings highlight how gender role attitudes shape women’s responses to rising incomes.

I examine the long-run effects of the timing of railroad construction on city sizes. I first present a stylized model that predicts that towns that are railroad end points for a longer period of time become persistently larger. I then show that, in a sample of Brazilian railroad towns, time as end point strongly predicts town size: each additional year that a town was a railroad end point in the past is associated with a town population 0.107 log points larger in 2010. Additional testable implications of the model and an instrumental variable approach suggest that such an association reflects a causal effect.

Aid fragmentation—the simultaneous operation of multiple development agencies in one setting—has long raised concerns about coordination challenges and opportunities for corruption. Leveraging unique data on project delivery in Afghanistan, we present the first microlevel empirical analysis of aid fragmentation. We find that aid delivered by a single donor can significantly reduce corruption. Projects delivered under conditions of aid fragmentation, by contrast, can facilitate corruption. We find evidence for a theoretical mechanism linking infrastructure and physical goods with waste and leakage. Our results clarify the policy losses tied to fragmentation, yielding insights for combating misappropriation of aid.