Research

This short paper describes a simple estimator for compensating differentials in the presence of unobserved ability. First described by Bell (2020) and Bell et al. (2024), the method assumes selection into jobs is governed by a single latent index and uses a noisy proxy-an “anti-instrument”-to recover amenity prices without functional form assumptions. Assumptions are greatly relaxed relative to common approaches: conditional independence given a single index. Researchers are estimating the model using standard cross-sectional data and the companion aivreg package. This paper omits proofs and applications, which appear in a series of longer technical papers; the goal here is to provide a practical roadmap for implementation. I also demonstrate that this revealed-preference approach generates amenity price estimates in line with preferences inferred from the experimental data collected by Mas and Pallais (2017), further validating the strength of both the experimental and observational approaches.

A recent Handbook of Labor Economics chapter concludes that “[o]mitted measures of job and firm productivity are a primary source of bias in compensating differential estimates” (Mas, 2025). This paper shows that such bias can be resolved using a simple amenity pricing framework that leverages an imprecise proxy—or “anti-instrument”—for workers’ offer sets. The method identifies amenity prices while holding offer sets fixed and recovers intuitively signed estimates in widely used cross-sectional datasets. The approach relies only on a single vertical index of job quality and a conditional independence assumption standard in prior work; it requires no assumptions about functional forms, distributions, or correlations between ability and preferences. Relative to prior observational methods, the estimates suggest a substantial role for amenity substitution in explaining the gender pay gap—accounting for roughly two-thirds of the observed gap—and little role for amenities in explaining inequalities by race or parent background. These findings offer new insight into the structure of labor market inequality and the large scope for equity-focused policy.

How does exposure to adult mentors shape children’s long-run outcomes? This paper provides causal evidence that social exposure through mentoring can affect long-run trajectories. I analyze program administrative records and micro-data from a 1991 RCT that randomized disadvantaged children’s eligibility for a program that paired youth with stable adult mentors. Beyond short-run behavioral improvements during childhood, a linkage to later-life administrative records shows that treated youth were 10 percentage points more likely to attend college and also exhibited improvements in outcomes such as teen birth and marriage. While RCT estimates of earnings effects are imprecise, a supplementary research design in a larger dataset of program administrative records yields similar patterns and suggests positive differences in early-career earnings, though these estimates rely on stronger identifying assumptions. A back-of-envelope calculation based on the observational estimates suggests potential lifetime earnings gains, though such calculations should be interpreted cautiously given uncertainty in earnings effects. The program’s relatively low costs and potential benefits suggest that it may be a cost-effective intervention, and that increased tax revenue from higher earnings could offset program costs.

We investigate the impact of higher education financing on innovation, focusing on how student loan policies shape the innovative potential of individuals. Leveraging a novel dataset linking administrative patent records to consumer credit data, we first document a variety of stylized facts pertaining to household finance and innovation, including event studies of credit scores around the time of innovation and age-debt profiles of inventors in comparison to the general population. Then, we explore the causal relationship between access to student loans and the likelihood of subsequent patent filings. Our quasi-experimental design exploits changes in federal student loan limits to estimate causal effects, revealing that increased borrowing capacity causes higher rates of patenting among constrained borrowers. While preliminary, the findings suggest that alleviating credit constraints through student loan policy may generate significant social spillovers by enabling broader participation in innovation. This research contributes to our understanding of the role of education financing as a lever for sustaining economic growth and reducing innovation disparities across socioeconomic groups.

Understanding how markets price housing amenities is essential to addressing residential socioeconomic disparities. Traditional methods have struggled with unobserved quality, often producing wrong-signed estimates. This paper introduces a novel “anti-instrument” approach to estimating amenity prices amid unobserved quality. Our identification strategy relies on an “anti-instrumental” variable that is relevant to the unobserved quality, but conditionally independent from the amenity of interest, which are properties opposite of typical instruments for amenities. We apply this method to estimate the implicit price of flood risk using detailed housing transaction records and house-level flood risk measures. Household income, used as an anti-instrument, successfully recovers the negative price of flood risk. We find that being located in a FEMA-designated floodplain reduces home prices by 2.8% on average. Drawing on more granular measures of flood risk, we find increasing price discounts for homes with higher flood probabilities, with the most risky locations suffering a discount of nearly 5%. These price estimates are crucial for evaluating returns on investment in local public goods and environmental policies.

This paper introduces Geographic PageRank (GPR), an innovative measure of locational desirability that leverages migration decisions to assess housing quality at the neighborhood level. Conceptually, GPR employs a recursive algorithm that uses the full network of migration flows to derive a micro-founded measure of place quality. Our GPR measure can be particularly useful in settings where unobserved housing quality has led to biased, or even wrong-signed, estimates for standard hedonic regressions of housing prices on amenities. We also demonstrate how the GPR framework can be extended to finer geographic units and adapted to capture temporal dynamics, providing a versatile tool for housing analysis.

We investigate how academics trade off salary and job amenities, particularly the prestige of their departments, using a novel anti-instrumental variable approach (Bell 2022). While compensating differential theory predicts such trade-offs, standard regression techniques often fail to account for unobserved variation in faculty offer sets, leading to biased estimates. Our approach addresses this challenge by leveraging the relationship between a faculty member’s PhD-granting university quality, current institution salary, and current university prestige to better identify the salary-prestige tradeoff. Using de-identified data on over 180,000 faculty-field pairs in 325 U.S. universities across 107 fields, we find strong evidence that faculty are willing to forgo substantial earnings for positions at more prestigious institutions. The average faculty member must give up $2,000 in pay to move to the next-most-prestigious school. These findings contribute to a broader understanding of labor market dynamics in high-skilled professions, where non-monetary factors play a critical role in job selection. Our methodological approach offers new insights into the pricing of job amenities and how institutions compete for talent beyond academia.

This paper provides estimates of the effect of unemployment insurance benefits on labor supply outcomes over the business cycle using 20 years of administrative claims, earnings, and employer data from California. A regression kink design exploiting nonlinear benefit schedules provides experimental estimates of behavioral labor supply responses throughout the unemployment spell that are comparable over time. For a given unemployment duration, the behavioral effect of UI benefit levels on labor supply is unchanged over the business cycle from 2002 to 2019. However, due to increased coverage from extensions in benefit durations , the duration elasticity of UI benefits rises during recessions. The behavioral effect during the start of the COVID-19 pandemic is substantially lower at all weeks of the unemployment spell.

To what extent did jobless Americans benefit from unemployment insurance (UI) during the COVID-19 pandemic? This article documents geographic disparities in access to UI during 2020. We leverage aggregated and individual-level claims data to perform an integrated analysis across four measures of access to UI. In addition to the traditional UI recipiency rate, we construct rates of application among the unemployed, rates of first payment among applicants, and exhaustion rates among paid claimants. Through correlations across California counties and across states, we show that areas with more disadvantaged residents had less access to UI during the pandemic. Although these disparities are large in magnitude, cross-state analysis suggests that policy can play a salient role in mitigating them.

To better measure the full extent of the impact of the COVID-19 crisis on workers and the labor market, this paper estimates three measures of the cumulative impact of the pandemic on workers across intensive and extensive margins using longitudinal administrative unemployment insurance (UI) data from California. During the first year of the crisis, 30 percent of the labor force filed a UI claim, over 50 percent of recipients spent more than 6 months on the program, and the mean work time lost was 13 weeks. Less advantaged workers and counties saw much higher rates of claiming and long-term unemployment.

We characterize the factors that determine who becomes an inventor in the United States, focusing on the role of inventive ability (“nature”) versus environment (“nurture”). Using deidentified data on 1.2 million inventors from patent records linked to tax records, we first show that children’s chances of becoming inventors vary sharply with characteristics at birth, such as their race, gender, and parents’ socioeconomic class. For example, children from high-income (top 1%) families are 10 times as likely to become inventors as those from below-median income families. These gaps persist even among children with similar math test scores in early childhood—which are highly predictive of innovation rates—suggesting that the gaps may be driven by differences in environment rather than abilities to innovate. We directly establish the importance of environment by showing that exposure to innovation during childhood has significant causal effects on children’s propensities to invent. Children whose families move to a high-innovation area when they are young are more likely to become inventors. These exposure effects are technology class and gender specific. Children who grow up in a neighborhood or family with a high innovation rate in a specific technology class are more likely to patent in exactly the same class. Girls are more likely to invent in a particular class if they grow up in an area with more women (but not men) who invent in that class. These gender- and technology class–specific exposure effects are more likely to be driven by narrow mechanisms, such as role-model or network effects, than factors that only affect general human capital accumulation, such as the quality of schools. Consistent with the importance of exposure effects in career selection, women and disadvantaged youth are as underrepresented among high-impact inventors as they are among inventors as a whole. These findings suggest that there are many “lost Einsteins”—individuals who would have had highly impactful inventions had they been exposed to innovation in childhood—especially among women, minorities, and children from low-income families.

Many countries provide financial incentives to spur innovation, ranging from tax incentives to research and development grants. In this paper, we study how such financial incentives affect individuals’ decisions to pursue careers in innovation. We first present empirical evidence on inventors’ career trajectories and income distributions using deidentified data on 1.2 million inventors from patent records linked to tax records in the United States. We find that the private returns to innovation are extremely skewed—with the top 1% of inventors collecting more than 22% of total inventors’ income—and are highly correlated with their social impact, as measured by citations. Inventors tend to have their most impactful innovations around age 40 and their incomes rise rapidly just before they have high-impact patents. We then build a stylized model of inventor career choice that matches these facts as well as recent evidence that childhood exposure to innovation plays a critical role in determining whether individuals become inventors. The model predicts that financial incentives, such as top income tax reductions, have limited potential to increase aggregate innovation because they only affect individuals who are exposed to innovation and have essentially no impact on the decisions of star inventors, who matter most for aggregate innovation. Importantly, these results hold regardless of whether the private returns to innovation are fully known at the time of career choice or are fully stochastic. In contrast, increasing exposure to innovation (e.g., through mentorship programs) could have substantial impacts on innovation by drawing individuals who produce high-impact inventions into the innovation pipeline. Although we do not present direct evidence supporting these model-based predictions, our results call for a more careful assessment of the impacts of financial incentives and a greater focus on alternative policies to increase the supply of inventors.

We establish the importance of team-specific capital in the typical inventor's career. Using administrative tax and patent data for the population of US patent inventors from 1996 to 2012, we find that an inventor's premature death causes a large and long-lasting decline in their co-inventor's earnings and citation-weighted patents (–4 percent and –15 percent after 8 years, respectively). After ruling out firm disruption, network effects, and top-down spillovers as main channels, we show that the effect is driven by close-knit teams and that team-specific capital largely results from an "experience" component increasing collaboration value over time.

This note clarifies the identifying assumptions underlying the approach in Bell (2020) and addresses a specific misinterpretation appearing in Mas (2025). That secondary source stated that the method requires independence between workers’ productivity and preferences over amenities. This characterization is not implied by the method, nor any previous iterations of it. Identification instead relies on a conditional independence assumption that pertains to a third observed variable, such as education, which the method casts as a shifter of individuals’ offer sets. The assumption stipulates that the shifter must be relevant to the quality of the offer set, but irrelevant to how workers split their compensation into pay versus amenities conditional on the quality of the offer set. Contrary to the Mas (2025) critique, the model in fact allows for arbitrary correlations between productivity and preferences. This note’s purpose is to clarify the record regarding these identifying assumptions, which in turn shape how empirical evidence using this approach is evaluated in practice.

This chapter focuses on preliminary research surrounding the three largest income support programs totaling $1.6 trillion that were paid directly to U.S. individuals, workers, and families amid the first two years of the COVID-19 pandemic: economic impact payments, Unemployment Insurance, and the expanded Child Tax Credit.

When workers choose jobs with predictable schedules, safer conditions, or greater autonomy, they are often paying a price: lower wages. In some cases, these hidden trade-offs are considerable. Indeed, my research shows that these “amenity trade-offs” can account for as much as two-thirds of the gender pay gap. In this column, I first introduce a new approach built upon my research to measuring these nonwage amenities trade-offs and then explain what the findings mean for both economists and policymakers seeking to understand and address persistent income inequality. I close with suggestions for how to foster a more equitable U.S. labor market that reframes how our nation values work, wages, and amenities.

The California Policy Lab partnered with the Employment Development Department (EDD) early in the crisis to analyze unemployment claims data. On April 29th, 2020, CPL released its first analysis, and it was clear, even then, that the crisis would disproportionately harm the state’s most vulnerable workers. The report is a snapshot of ten key trends from the unemployment crisis, based on 16 reports the research institute has released during the crisis, and suggests steps for how to translate this evidence into policy to reform the unemployment system.

Unemployment Insurance (UI) benefits provided a lifeline to workers who lost their jobs during the pandemic. However, access to these benefits has been uneven across communities and states (Edwards, 2020). Identifying and documenting these disparities is an important step to addressing them and to rendering the UI system more equitable. Utilizing a  conceptual framework of unemployment claims, we developed three metrics to measure access to UI benefits across the claim lifecycle. We then analyzed these measures to provide insight into differential access to UI benefits across U.S. states and across counties within California.

The Extended Benefits (EB) program automatically extends how long a person can claim Unemployment Insurance (UI) benefits when the share of workers claiming regular UI benefits reaches a certain level. However, a key measure of unemployment through which EB triggers on and off does not count individuals receiving benefits through extension programs, such as Pandemic Emergency Unemployment Compensation (PEUC) or EB. As a result, when a large share of unemployed workers transition from regular UI to extension programs, EB can mechanically trigger off, even if the total number of UI claimants remains unchanged or is increasing.

Historically, the share of unemployed workers receiving regular UI benefits (recipiency rate) in California has been relatively low (as has also been the case in other states). This Data Point combines administrative data from California’s Employment Development Department (EDD) with monthly Current Population Survey (CPS) data to construct an improved recipiency rate to measure the extent to which unemployed and underemployed Californians are receiving regular UI benefits.

Under the CARES Act, Congress created two federal unemployment programs, Pandemic Unemployment Assistance (PUA) and Pandemic Emergency Unemployment Compensation (PEUC). However, these two programs are slated to expire on December 26th, 2020 unless Congress passes new legislation to extend them. If Congress is unable to pass such legislation, our latest projections find that over one million Californians will stop receiving unemployment insurance benefits by the end of 2020, despite remaining unemployed. This Data Point analyzes a bi-partisan proposal, the Emergency COVID Relief Act of 2020 that would extend the PUA and PEUC programs by 16 weeks (through April 17th) while also extending the maximum length a person could receive benefits under these programs. We show this proposal would be extremely effective in reducing the incidence of benefit exhaustion, reducing the number of claimants who stop receiving benefits before April 10th by 95%. This new proposal would provide nearly $10 billion in direct payments to unemployed Californians by mid-April, with indirect benefits spilling over to business owners and the wider labor market.

The “Lost Wages Assistance” (LWA) program provides grants to participating states in order to fund a $300 per week supplementary payment to individuals receiving unemployment insurance (UI) benefits, so long as they meet certain eligibility requirements. This Data Point provides an overview of the impact of the LWA program in California as well as policy considerations for any additional federal unemployment supports.