How does remote work affect productivity and how productive are workers who choose remote jobs? We decompose these effects in a Fortune 500 firm. Before Covid-19, remote workers answered 12% fewer calls per hour than on-site workers. After the offices closed, the productivity gap narrowed by 4%, and formerly on-site workers’ call quality and promotion rates declined. Even with everyone remote, an 8% productivity gap persisted, indicating negative selection into remote jobs. A cost-benefit analysis indicates that savings in reduced turnover and office rents could outweigh remote work's negative productivity impact but not the costs of attracting less productive workers.

In an increasingly digital world, how much does sitting near coworkers matter for on-the-job training? And who is most affected by proximity? We study software engineers at a Fortune 500 firm. When offices were open, engineers working in the same building as all their teammates received 21 percent more online feedback on their computer code than engineers with distant teammates. After offices closed for COVID-19, this advantage shrank by 15 percentage points. Sitting near coworkers increases how much junior engineers can learn from their senior colleagues --- not only in-person but also online. However, sitting together reduces senior engineers' programming output, suggesting a tradeoff between short-term productivity and long-run human-capital development. Proximity particularly increases feedback to female engineers from both male and female colleagues and reduces female engineers' quit-rates. Even pre-COVID, gaining one distant teammate reduced online feedback among coworkers sitting together: thus, remote-work policies may impact even workers who choose to go into the office. 

In criminal cases, prosecutors can adjust police officers’ charges between arrest and sentencing — and therefore have the discretion to check any racial bias in arrests. Yet previous research suggests that prosecutors simply compound earlier racial disparities with their own biases. We investigate prosecutors’ impacts on racial disparities using discontinuities in North Carolina’s sentencing laws. For defendants with criminal histories above certain thresholds, the law mandates prison sentences. However, prosecutors can sidestep mandatory prison for these defendants by reducing their charge. We ask who benefits from prosecutors’ charging responses to the mandatory-prison discontinuities. From 1995 to 2019, we find that Black defendants were initially less likely — but ultimately became more likely — to benefit from charge reductions to avoid mandatory prison. The reversal is concentrated in arrests typically initiated by police stops and absent from arrests typically initiated by victim reports, suggesting that prosecutors have increasingly questioned disparities introduced by the police.

In criminal courts, decision-makers often aim to selectively incarcerate people with a high risk of future violent crime. If decision-makers have more accurate beliefs about this risk, can they reduce violent crime without simply incarcerating more people? We survey 162 prosecutors about how violent re-arrest varies across defendants of different ages and criminal records. We link prosecutors’ beliefs to their 104,039 cases, which are assigned to prosecutors quasi-randomly. Prosecutors’ beliefs vary widely and predict their incarceration rates for different age-groups and criminal-record groups. Prosecutors with more accurate beliefs (by 1 sd) reduce violent crime (by 6%) without incarcerating more people.

Can firms' efficiency-wage incentives counterbalance their monopsony power in wage-setting decisions? We study a Fortune 500 firm's voluntary firm-wide $15/hour minimum wage, which affected some of its warehouses more than others. In a continuous difference-in-differences design, we find that a $1/hour (5.5 percent) increase in pay halves worker departures and increases the number of boxes moved per hour by 5.9 percent. These productivity gains fully defrayed increased labor costs, thereby offsetting the firm's incentive to mark down wages. We develop a simple model that connects efficiency-wage incentives and monopsony power, showing how the two pull in opposite directions.

Standard accounts of discrimination in hiring, lending, and sentencing focus on decision-makers' introduction of disparities --- and abstract away their mental models of the discretionary decisions of past employers, lenders, and police. We extend the canonical statistical-discrimination model to allow signals to come from earlier actors who may make mistakes and introduce biases. We use prosecutors’ responses to police as our test case. We survey North Carolina prosecutors to elicit their beliefs about defendant conduct and police reliability and link these responses to prosecutors’ real-world cases. We first show that prosecutors vary widely in their impact on incarceration disparities (SD = 3.6 pp) using rotational case assignment. We find that prosecutors who believe that (a) police are less reliable and (b) criminal conduct differs less by race reduce incarceration disparities.  Each belief explains over the twice the variation in incarceration disparities as prosecutor race or politics.

One in five Americans bears the mark of a criminal record as they navigate the job-search process. How do the resulting search frictions affect equilibrium wages in low-wage labor markets? We use data from a Fortune 100 staffing agency to calibrate and test a wage-posting model, in which workers with and without records face different search frictions but earn commonly posted wages. We find that workers with records have 17 percent less elastic labor supply to each firm, suggesting that firms can post lower wages in labor markets where more workers have records. When the same job is offered by a firm in multiple locations, we find that the firm's posted wage tends to be lower in places where the share of workers with criminal records in that specific occupation is higher. This pattern holds equally for workers with and without records. These equilibrium consequences suggest that the increase in records in the past few decades has reduced earnings of all low-wage workers by 1.2-1.5 percent or about $400 per year.