Knowledge that Transforms

I am requesting withdrawal of

Siddique et al. (2024) report large effects of a mobile phone-based health awareness campaign in a field experiment conducted in rural Bangladesh and India during the COVID-19 pandemic. Treatments are stated as being randomized at the village level in the Bangladesh arm, but analyses of available replication materials from this and related field experiments show that this was not the case: participants retained their treatment group assignment from two earlier studies run by the same NGO. Because these earlier studies reported large and persistent treatment effects, the treatment groups in Siddique et al. (2024) could plausibly differ systematically, raising issues of confounding and interaction effects that invalidate the analyses and interpretations of this study, affecting both internal and external validity.

We test whether loan officers discriminate against female business owners in capital decisions affecting real businesses in an Ethiopian business plan competition. Randomizing owner gender across 3,696 evaluations, by 84 loan officers, for 916 businesses, we find no effect on prize scores or consideration for a loan. Incentivized beliefs about future performance did not differ by gender, ruling out statistical and taste-based discrimination. Machine learning predictions of future business performance show that loan officers beliefs were accurate. Gender does not meaningfully improve predictions, implying no trade-off between gender equity and effective capital allocation.

What are the consequences of losing political representation in a democracy? A reform of Germany’s mixed electoral system left more than 7 million citizens in 23 constituencies without a directly elected representative after the 2025 national election. Which and how many constituencies were affected was unpredictable and quasi-random. We show that losing geographical representation reduces citizens’ satisfaction with democracy. Moreover, discontent is independent of party preferences and spreads through social networks, reducing democratic satisfaction even among individuals who were unaffected or unaware of the reform’s outcome. We conclude that political representation is a cornerstone of democracy.

This paper provides new evidence on why men and women leaders make different choices. Exploiting Brazilian close elections, we show that female mayors responded differently to the COVID-19 crisis over the year 2020. Female mayors were less likely to close non-essential businesses early in the pandemic and female-led municipalities experienced higher mortality, while the reverse was true later on. We show that these findings can be rationalized by a simple political agency model in which politicians seek re-election and voters are gender biased. Consistent with this interpretation, the gender differences we find are driven exclusively by mayors facing re-election.

This paper investigates how parental connections to firms shape early-career earnings. I use employer-employee linked data to study one important type of connection: jobs obtained at a parent’s employer. Twenty-nine percent of individuals work for a parent’s employer at least once by age 30. Exploiting transitory, firm-specific fluctuations in hiring conditions at the parent’s employer, I estimate that working for a parent’s employer increases initial earnings by 24 percent, primarily by providing access to higher-paying firms. Individuals with higher-earning parents are more likely to work for a parent’s employer and experience larger earnings gains when they do. Consequently, the elasticity of initial earnings with respect to parental earnings would be 12 percent lower if no one found a job through these connections. The findings suggest that connections to firms through one’s social network may be an important determinant of intergenerational mobility.

We show that structural oil-market shocks identified in prominent SVAR models, as well as high-frequency oil price changes around OPEC announcements, are predictable, indicating informational deficiencies in standard oil-market VARs and contamination of commonly used instrumental variables. We resolve these identification problems for oil supply news shocks by augmenting the econometrician’s information set with financial indicators, yielding markedly different results: the shocks induce a sharper short-run output contraction, are less inflationary, contribute less to oil price variation, exhibit greater stability over time, and account for a substantial share of S&P 500 volatility.

We study the health impacts of job loss and unemployment insurance (UI) among Brazilian workers, using unique administrative data linking employment, hospitalizations, health insurance, and mortality records over 17 years. Leveraging mass layoffs for identification, we find that job loss raises hospitalization (+33%) and mortality (+23%) risks for men, with no significant effects for women. Impacts are pervasive across age, income, and education levels, with children of male workers also affected. External causes related to risky behavior and violence victimization explain half the effects. Using a regression discontinuity design, we show that UI partially mitigates health risks, suggesting that these are partially explained by income losses.

Debates over return-to-office mandates have intensified since the COVID-19 pandemic ended, though their economic implications remain poorly understood. Using large-scale resume data, we analyze how these policies affect employee tenure and seniority at three large U.S. tech companies: Microsoft, SpaceX, and Apple. Employing a distributional synthetic controls framework, we estimate a reduction in tenure and seniority at firms returning to the office. We extend the framework with a functional bootstrap procedure and prove its validity for constructing uniform confidence bands. Return-to-office mandates appear to drive senior employees away, posing a risk to firm productivity and innovation.