99% of Fortune 500 companies use algorithms in hiring, yet identical scores can imply different expected productivity levels across groups. I introduce Cross-Group Mapping Bias, the bias that arises when decision-makers treat scores as comparable despite group-specific signal–outcome mappings. Transformations using group-specific marginals and dependence, estimated via advanced empirical copulas, enable cross-group comparability. In two incentivized MTurk hiring experiments in India, participants learn within-group mappings but do not learn cross-group differences, leading to persistent bias. In mixed-gender pairwise choices, participants select women 42% of the time, even though the optimal female hiring share is 57% due to differential selection in the experimental environment. Participants select higher-productivity candidates 53% of the time, compared to 67% under Bayesian updating. Neither information provision nor gender masking improves decisions, but active debiasing reduces gender bias by 10-12 percentage points and increases accuracy by 3 percentage points. However, treatment reduces reliance on the signal, limiting productivity gains. These group-specific signal–outcome mappings are commonplace in decision and policy evaluation contexts that rely on proxies such as test scores, education, and earnings residuals.
Reject with possibility of resubmission at the American Economic Journal: Economic Policy.
This paper bridges the gap between models of discrimination and policing. Using a series of conditional likelihood hit-rate tests on arrest, conviction, and exoneration data, I find that the American judicial system is prejudiced against Black Americans, and American police officers are more prejudiced than the judicial system. Innocent Black people are more likely to be arrested and convicted than White people. Further, prejudice reduces the quality of police reporting and increases both type one and type two errors in judicial decision-making. This prejudice by police officers and in criminal courts costs the US over $94.4-124.4 billion annually.
Men exhibit the classic self-serving bias, with nearly 80% attributing high pay to high performance and low income to chance. Women display the opposite behaviour, attributing high pay to chance and low pay to their performance, even controlling for actual performance. In contrast to existing theories, men’s self-serving bias leads them to be more generous than their counterfactual redistributive behaviour. Women exhibit a type of certainty aversion and redistribute less when they know with certainty that their payment was due to chance.
Despite abundant solar resources and heavily subsidized financing, solar panel adoption in Yemen remains puzzlingly low, with minimal loan take-up even when panels offer over 100% annual returns. This study investigates the behavioral barriers preventing efficient energy technology adoption in fragile, low-income settings with limited financial markets.
We develop a theoretical framework decomposing technology adoption decisions into distinct behavioral frictions—risk aversion, loss aversion, debt aversion, and present bias—examining how these interact with credit constraints. Using a survey of 1,422 Yemeni adults across three governorates, we employ hypothetical vignettes and discrete choice experiments to separately identify and quantify each behavioral parameter.
By quantifying the relative importance of each friction and their interactions, this research provides actionable guidance for designing financial products and development interventions that can overcome behavioral barriers to technology adoption in contexts where poverty, conflict, and limited financial infrastructure compound the challenges of sustainable development.
Funding from STICERD
Launching January 2026.
The U.S. Defense Department budget for the 2023 financial year was $797 billion. Total U.S. Military Budget Resources for the 2025 Fiscal Year exceed $1.5 trillion (15.5% of GDP). Foreign military aid exceeds $12 billion a year. The United States comprises over 40% of global military spending. Votes on military intervention, military expenditures, and veteran benefits are some of the most impactful mechanisms through which politicians shape foreign and domestic policy. Policy analysts, economists, and politicians have assorted views on the efficacy of this spending and what costs it imposes.
Similarly, attitudes towards military spending are heterogeneous within and across parties. Given the extensive global impacts of military expenditures, understanding the drivers of attitudes toward war is essential. Politicians’ own military experience and the war experiences of others in their cohorts are plausibly one of these determinants.
Do people whose peers (or themselves) are drawn into conflict have differences in their support for military interventions, military expenditures, or veteran benefit policies? What mechanism drives this? We will explore whether cohort effects shape the policy preferences of politicians by changing who becomes a politician or whether cohort effects influence otherwise similar politicians to have different attitudes regarding war. This project primarily focuses on having Americans classify historical government decisions to evaluate political behaviour.
Does cohort exposure drive politicians of different characteristics to contest elections or get elected? In addition, we will examine the role of a transformation in preferences arising from cohort-level exposure to war and whether this varies according to the public’s view of the war.