Research

Working Papers

What do firms gain from raising pay for low-wage workers? Focusing on a Fortune 500 retailer, we estimate the impact of higher wages on employee productivity, turnover, and recruitment among warehouse and call-center workers, using the quasi-randomness induced by sticky wage-setting policies. We document finite wage elasticities of turnover (between −3.0 and −4.5) and recruitment (between 3.2 and 4.2), which suggest the firm has some wage-setting power. Yet, on the margin, raising wages by $1 increases productivity by more than $1, giving the firm an incentive to pay more, even if they could pay lower wages. These responses to pay emerge both in a setting where the firm discretely raised wages and in a setting where its wages remained constant while other firms raised pay. These effects reflect both changes in worker selection and changes in behavior of existing workers. We estimate that over half of the turnover reductions and productivity increases arise from changes in workers’ behavior. Finally, our estimates suggest considerable gender heterogeneity: Men’s turnover is more responsive to higher wages than women’s. But turnover effects are swamped by women’s stronger productivity response to higher pay. Together, the gender-specific elasticities suggest firms have an implicit incentive to set female wages above male wages and thus firm profits cannot explain the gender pay gap.

Marginal Revolution


Why Do Women Earn Less Than Men? Evidence from Bus and Train Operators

with Valentin Bolotnyy

Female workers earn $0.89 for each male-worker dollar even in a unionized workplace where tasks, wages, and promotion schedules are identical for men and women by design. We use administrative timecard data on bus and train operators to show that the earnings gap can be explained by female operators taking, on average, 1.5 fewer hours of overtime and 1.3 more hours of unpaid time-off per week than male operators. Female operators, especially those who have dependents, pursue schedule conventionality, predictability, and controllability more than male operators. Analyzing two policy changes, we demonstrate that while reducing schedule controllability can reduce the earnings gap, it can also make workers—particularly female workers—worse off.

Wall Street JournalMarginal RevolutionSociety of Human ResourcesQuartz


"Working" Remotely? Selection, Treatment, and Market Provision of Remote Work

with Emma Harrington

Why was remote work rare prior to Covid-19? We ask whether remote work reduces worker productivity or attracts less productive workers. Using data on call-center workers at a Fortune 500 online retailer, we find that remote work improves worker productivity but worsens worker selection. Prior to the pandemic, remote workers were less productive than on-site workers, answering 10% fewer calls. In contrast, we estimate positive causal effects of working remotely. When on-site workers opted into remote work in 2018-2019, their hourly call volumes rose by 7%. Similarly, when Covid-19 forced all on-site workers to work remotely, their call volume rose by 8-10% relative to their already remote peers, without compromising customer satisfaction. Given the positive treatment effect of remote work, remote workers' lower productivity implies negative selection --- an inference that is only magnified when we condition on worker observables. Consistent with their lower productivity, remote workers received lower wages. Yet, worker surplus from remote work averages $1.48/hr, based on our estimates of workers' demand for remote work. In the absence of adverse selection, the retailer would hire 57% more remote workers and worker surplus from remote work would rise by 32%. Our paper concludes that Covid-19's effect on remote work will persist if the pandemic changes worker selection: if more productive workers disproportionately become more willing to work remotely because of the pandemic, this shock will attenuate adverse selection and permanently increase the prevalence of remote work.

Marginal Revolution The Economist The Financial Times


Behavioral Biases and Legal Compliance: Evidence from an RCT on Failure to Appear

with Helen Ho

Many defendants fail to appear (FTA) for court despite the prospect of legal consequences. In a field experiment, we compare the effectiveness of text message reminders to an intervention that combines reminders with personalized assistance. The treatments are equally effective, reducing FTA by 8 percentage points from a 21 percent baseline rate. However, personalized assistance facilitates greater take-up of court accommodations such as rescheduling and payment plans. For more serious cases, the treatments reduce arrests by two percentage points, implying FTAs have a large effect on arrests. For the least serious cases, an FTA has small effects on fines.

Probable Causation Podcast

In Progress

Snowed In: School Closures and Family Violence

I use data on police reports of domestic violence and cases of confirmed child maltreatment to identify proximate events that coincide with reports of intra-household violence. After three-day weekends and snow-days, reports of family violence increase. I further show that these effects are concentrated in counties with low median income. These findings suggest that when unsure, policy-makers may want to keep school going or open snow-day locations so individuals may have a physical refuge to avoid intra-household violence.


Smudges: Criminal Records' Employment Signal

with Emma Harrington

We use data from a national staffing agency to compare the predictive content of criminal background checks to that of other signals that an employer observes about prospective employees. Even when managers are blind to criminal records that do not disqualify the candidate from the job, such a record does signal that the worker is less likely to complete the temporary assignment and receive a favorable review from the manager. However, the negative signal of a criminal record is more than outweighed by having completed a temporary assignment in the past. Our evidence suggests that prior work experience should outweigh criminal records when managers make hiring decisions. In our ongoing work, we highlight particular types of convictions that do not contain insight about workers' productivity and what types of prior convictions may have particularly racially disparate impacts in the labor market.


Determinants of Racial Disparities in Prosecutors’ Charging Decisions

with Emma Harrington, William Murdock III, and Hannah Shaffer

We explore to what extent institutional discrimination contributes to observed racial disparities in prosecutors' cases. To assess this, we link prosecutor survey responses to administrative records from North Carolina's Superior Court. Our survey presents prosecutors with hypothetical cases to test how prosecutors (consciously or unconsciously) respond to past actions of police and statutes that have disparate impact on black men, while holding fixed true defendant conduct. By linking our survey results to administrative data, we test whether prosecutors' responses predict their historical charging decisions. This allows us to quantify how much racial gaps in prosecutors' charging decisions are influenced by these up-stream disparities.


Biased Beliefs in the Criminal Justice System

with Emma Harrington, William Murdock III, and Hannah Shaffer

Prosecutors have considerable discretion over defendant outcomes in criminal courts and aim to apply their discretion to reserve incarceration for those who are more likely to re-offend. However, it is unclear whether they have accurate beliefs about the correlates of re-offense. In a survey of over 100 prosecutors in North Carolina, we find that prosecutors have systematically biased beliefs about how age and prior criminal history relate to re-offense. Prosecutors underestimate the likelihood of criminal re-offense among young offenders and overestimate the likelihood of criminal re-offense among those with long criminal histories. In future work, we will link these beliefs and the extent of each prosecutor’s inaccuracies to court records to investigate how prosecutors’ beliefs relate to the outcomes in their cases.




Older Papers

The Prevalence of Confirmed Maltreatment Among US Children, 2004 to 2011 • JAMA Pediatrics (2014)

with Christopher Wildeman, John Leventhal, Emily Putnam-Hornstein, Jane Waldfogel, and Hedwig Lee.

National Public RadioUS News and World ReportReuters


Cumulative Risks of Foster Care Placement by Age 18 for U.S. Children, 2000-2011 PLoS One (2014)

with Christopher Wildeman

Cited in Evicted


Cumulative Risks of Foster Care Placement for Danish Children • PLoS One (2014)

with Peter Fallesen and Christopher Wildeman


Mortality among White, Black, and Hispanic Male and Female State Prisoners, 2001–2009SSM Population Health (2016)

with Christopher Wildeman, E. Ann Carson, Daniela Golinelli, Margaret Noonan


State-Level Variation in the Imprisonment-Mortality Relationship, 2001-2010Demographic Research (2016)

with Christopher Wildeman, Margaret E. Noonan, Daniela Golinelli, and E. Ann Carson.


State-Level Variation in the Risk of Child Welfare System Contact among U.S. Children, 2008-2018 Working Paper

with Youngmin Yi, Frank Edwards, Hedwig Lee, John Leventhal, Jane Waldfogel, and Christopher Wildeman