I am a labor economist who combines economics and econometrics methods with large administrative data to inform policy. My research focuses predominantly on two domains: education and housing.
My education work focuses on understanding individuals' decisions to invest in human capital, measuring the returns to education, and the role of non-cognitive skills. My housing work focuses on quantifying the prevalence and impact of evictions, and on evaluating policies designed to benefit low-income renters and prevent homelessness.
More than two million U.S. households have an eviction case filed against them each year. Policymakers at the federal, state, and local levels are increasingly pursuing policies to reduce the number of evictions, citing harm to tenants and high public expenditures related to homelessness. We study the consequences of eviction for tenants using newly linked administrative data from two major urban areas: Cook County (which includes Chicago) and New York City. We document that prior to housing court, tenants experience declines in earnings and employment and increases in financial distress and hospital visits. These pre-trends pose a challenge for disentangling correlation and causation. To address this problem, we use an instrumental variables approach based on cases randomly assigned to judges of varying leniency. We find that an eviction order increases homelessness and hospital visits and reduces earnings, durable goods consumption, and access to credit in the first two years. Effects on housing and labor market outcomes are driven by impacts for female and Black tenants. In the longer-run, eviction increases indebtedness and reduces credit scores.
We study the effects of conviction and incarceration on recidivism using quasi-random judge assignment. We extend the typical binary-treatment framework to a setting with multiple treatments, and outline a set of assumptions under which standard 2SLS regressions recover causal and margin-specific treatment effects. Under these assumptions, 2SLS regressions applied to data on felony cases in Virginia imply that conviction leads to a large and long-lasting increase in recidivism relative to dismissal, consistent with a criminogenic effect of a criminal record. In contrast, incarceration reduces recidivism, but only in the short run. The assumptions we outline could be considered restrictive in the random judge framework, ruling out some reasonable models of judge decision-making. Indeed, a key assumption is empirically rejected in our data. Nevertheless, after deriving an expression for the resulting asymptotic bias, we argue that the failure of this assumption is unlikely to overturn our qualitative conclusions. Finally, we propose and implement alternative identification strategies. Consistent with our characterization of the bias, these analyses yield estimates qualitatively similar to those based on the 2SLS estimates. Taken together, our results suggest that conviction is an important and potentially overlooked driver of recidivism, while incarceration mainly has shorter-term incapacitation effects.
The process of skill specialization starts before college, with different skills affecting students' choice of major and later labor market returns. This paper studies the role of multi-dimensional ability and high school track choices in college preparedness and labor market outcomes. We do so by estimating a sequential choice model of education using Swedish administrative data. Individuals sort at each stage based on prior choices and three dimensions of ability: cognitive, interpersonal, and grit. We find strong absolute and differential sorting on abilities in both high school and college choices. Both abilities and high school track choices are important determinants of college enrollment, college major choice, college graduation, and labor market outcomes. The labor market returns to abilities and high school track choices vary considerably by degree and major. Not accounting for multidimensional abilities and high school choices can overstate the role of preferences and understate selection on gains and the heterogeneous returns to different abilities across different college majors. While high school track choices tend to exacerbate inequality, we show that policies encouraging students to take more challenging high school tracks can help ameliorate it. (The project is funded by Riksbankens Jubileumsfond through the Stockholm School of Economics).
This paper investigates the determinants and consequences of entry into and exit from self-employment over the life cycle. It integrates traditional models of dynamic career choice that feature human capital investment with models of business start-up that feature costly capital investment. Applying machine learning methods to matched worker-firm data from Sweden, I isolate seven distinct patterns of participation in self-employment as part of broader life-cycle employment profiles. These patterns are rationalized using a dynamic Roy model with both human capital and physical capital. Using structural methods, I estimate the model and use it to evaluate policies designed to promote self-employment. Cognitive and non-cognitive skills, education, and past work experience are important determinants of which types of businesses individuals start, how much capital they employ, and how long they remain in self-employment. Subsidies that incentivize self-employment are generally ineffective, both in terms of promoting long-lasting firms and in terms of improving the welfare and earnings of those induced to enter self-employment.
Leveraging data from Sweden and Chicago, we study the educational pipeline for science, technology, engineering, and mathematics (STEM) and economics majors to better understand the determinants of the gender gap and when these determinants arise. We present three findings. First, females are less likely to select STEM courses in high school despite equal or better preparation. Second, there are important gender differences in preferences and beliefs, even conditional on ability. Third, early differences in preferences and beliefs explain more of the gaps in high school sorting than other candidate variables. High school sorting then explains a large portion of the gender difference in college majors.
This paper studies the effects of the COVID-19 pandemic on small businesses between March and November 2020 using new survey data on 35,000 small businesses in eight Latin American countries. We document that the pandemic had large negative impacts on employment and beliefs regarding the future, which in turn predict meaningful economic outcomes in the medium-term. Despite the unprecedented amount of aid, policies had limited impact for small and informal firms. These firms were less aware of programs, applied less, and received less assistance. This may have lasting consequences, as businesses that received aid reported better outcomes and expectations about the future.
The This paper uses a college-by-graduate degree fixed effects estimator to evaluate the returns to 19 different graduate degrees for men and women. We find substantial variation across degrees, and evidence that OLS overestimates the returns to degrees with high average earnings and underestimates the returns to degrees with low average earnings. Second, we decompose the impacts on earnings into effects on wage rates and effects on hours. For most degrees, the earnings gains come from increased wage rates, though hours play an important role in some degrees, such as medicine, especially for women. Third, we estimate the net present value and internal rate of return for each degree, which account for the time and monetary costs of degrees. We show annual earnings and hours worked while enrolled in graduate school vary a lot by gender and degree. Finally, we provide descriptive evidence that gains in overall job satisfaction and satisfaction with contribution to society vary substantially across degrees.
The Paycheck Protection Program (PPP) extended 669 billion dollars of forgivable loans in an unprecedented effort to support small businesses affected by the COVID-19 crisis. This paper provides evidence that information frictions and the “first-come, first-served” design of the PPP program skewed its resources towards larger firms and may have permanently reduced its effectiveness. Using new daily survey data on small businesses in the U.S., we show that the smallest businesses were less aware of the PPP and less likely to apply. If they did apply, the smallest businesses applied later, faced longer processing times, and were less likely to have their application approved. These frictions may have mattered, as businesses that received aid report fewer layoffs, higher employment, and improved expectations about the future.
This paper estimates returns to education using a dynamic model of educational choice that synthesizes approaches in the structural dynamic discrete choice literature with approaches used in the reduced form treatment effect literature. It is an empirically robust middle ground between the two approaches which estimates economically interpretable and policy-relevant dynamic treatment effects that account for heterogeneity in cognitive and non-cognitive skills and the continuation values of educational choices. Graduating college is not a wise choice for all. Ability bias is a major component of observed educational differentials. For some, there are substantial causal effects of education at all stages of schooling.
This paper analyzes the non-market benefits of education and ability. Using a dynamic model of educational choice we estimate returns to education that account for selection bias and sorting on gains. We investigate a range of non-market outcomes including incarceration, mental health, voter participation, trust, and participation in welfare. Unlike previous evidence on the monetary benefits of education, the benefits to education for many non-market outcomes appear to be larger for low-ability individuals. College graduation decreases welfare use, lowers depression, and raises self-esteem more for less-able individuals. Accounting for the non-market benefits of education is an important component of any analysis of educational policy.
Across academic sub-fields such as labor, education, and behavioral economics, the measurement and interpretation of non-cognitive skills varies widely. As a result, it is difficult to compare results on the importance of non-cognitive skills across literatures. Drawing from these literatures, this paper systematically relates various prototypical non-cognitive measures within one data set. Specifically, we estimate and compare several different strategies for measuring non-cognitive skills. For each, we compare their relative effectiveness at predicting educational success and decompose what is being measured into underlying personality traits and economic preferences. We demonstrate that the construction of the non-cognitive factor greatly influences what is actually measured and what conclusions are reached about the role of non-cognitive skills in life outcomes such as educational attainment. Furthermore, we demonstrate that, while sometimes difficult to interpret, factors extracted from self-reported behaviors can have predictive power similar to well established taxonomies, such as the Big Five.
This paper develops robust models for estimating and interpreting treatment effects arising from both ordered and unordered multi-stage decision problems. Identification is secured through instrumental variables and/or conditional independence (matching) assumptions. We decompose treatment effects into direct effects and continuation values associated with moving to the next stage of a decision problem. Using our framework, we decompose the IV estimator, showing that IV generally does not estimate economically interpretable or policy-relevant parameters in prototypical dynamic discrete choice models, unless policy variables are instruments. Continuation values are an empirically important component of estimated total treatment effects of education. We use our analysis to estimate the components of what LATE estimates in a dynamic discrete choice model.
The option to obtain a General Educational Development (GED) certificate changes the incentives facing high school students. This article evaluates the effect of three different GED policy innovations on high school graduation rates. A 6-point decrease in the GED pass rate produced a 1.3-point decline in high school dropout rates. The introduction of a GED certification program in high schools in Oregon produced a 4% decrease in high school graduation rates. Introduction of GED certificates for civilians in California increased the dropout rate by 3 points. The GED program induces students to drop out of high school.
Intelligence quotient (IQ), grades, and scores on achievement tests are widely used as measures of cognition, but the correlations among them are far from perfect. This paper uses a variety of datasets to show that personality and IQ predict grades and scores on achievement tests. Personality is relatively more important in predicting grades than scores on achievement tests. IQ is relatively more important in predicting scores on achievement tests. Personality is generally more predictive than IQ on a variety of important life outcomes. Both grades and achievement tests are substantially better predictors of important life outcomes than IQ. The reason is that both capture personality traits that have independent predictive power beyond that of IQ.
This paper discusses and illustrates identification problems in personality psychology. The measures used by psychologists to infer traits are based on behaviors, broadly defined. These behaviors are produced from multiple traits interacting with incentives in situations. In general, measures are determined by these multiple traits and do not identify any particular trait unless incentives and other traits are controlled for. Using two data sets, we show, as an example, that substantial portions of the variance in achievement test scores and grades, which are often used as measures of cognition, are explained by personality variables.
The General Educational Development (GED) credential is issued on the basis of an eight hour subject-based test. The test claims to establish equivalence between dropouts and traditional high school graduates, opening the door to college and positions in the labor market. In 2008 alone, almost 500,000 dropouts passed the test, amounting to 12% of all high school credentials issued in that year. This chapter reviews the academic literature on the GED, which finds minimal value of the certificate in terms of labor market outcomes and that only a few individuals successfully use it as a path to obtain post-secondary credentials. Although the GED establishes cognitive equivalence on one measure of scholastic aptitude, recipients still face limited opportunity due to deficits in noncognitive skills such as persistence, motivation and reliability. The literature finds that the GED testing program distorts social statistics on high school completion rates, minority graduation gaps, and sources of wage growth. Recent work demonstrates that, through its availability and low cost, the GED also induces some students to drop out of school. The GED program is unique to the United States and Canada, but provides policy insight relevant to any nation's educational context.
Objective: To design and implement a tool that creates a secure, privacy preserving linkage of electronic health record (EHR) data across multiple sites in a large metropolitan area in the United States (Chicago, IL), for use in clinical research.
Methods: The authors developed and distributed a software application that performs standardized data cleaning, preprocessing, and hashing of patient identifiers to remove all protected health information. The application creates seeded hash code combinations of patient identifiers using a Health Insurance Portability and Accountability Act compliant SHA-512 algorithm that minimizes re-identification risk. The authors subsequently linked individual records using a central honest broker with an algorithm that assigns weights to hash combinations in order to generate high specificity matches.
Results: The software application successfully linked and de-duplicated 7 million records across 6 institutions, resulting in a cohort of 5 million unique records. Using a manually reconciled set of 11 292 patients as a gold standard, the software achieved a sensitivity of 96% and a specificity of 100%, with a majority of the missed matches accounted for by patients with both a missing social security number and last name change. Using 3 disease examples, it is demonstrated that the software can reduce duplication of patient records across sites by as much as 28%.
Conclusions: Software that standardizes the assignment of a unique seeded hash identifier merged through an agreed upon third-party honest broker can enable large-scale secure linkage of EHR data for epidemiologic and public health research. The software algorithm can improve future epidemiologic research by providing more comprehensive data given that patients may make use of multiple healthcare systems.