Recent decades have seen great advances in the methods we use to understand cause and effect in the world of work. Building on that tradition, this dissertation explores two broad topics in econometrics as tools to address specific questions in labor economics. The main econometric contributions are to extend identification results for research designs based on bunching (Chapter 1) and those that make use of instrumental variables (Chapters 2 and 3). The empirical questions that compel them are described below.
Chapter 1 examines the effect of overtime regulation on hours of work in the United States, extending a recently popularized technique that uses bunching observed at kinks in agents' choice sets for identification. In the U.S., most workers are required to be paid one-and-a-half times their typical rate of pay for any hours in excess of forty within a week. While prominent and long-standing, this policy has not been meaningfully reformed since it was first established at the federal level in 1938. As a result, few studies have been able to leverage causal research designs to assess its labor market impacts. I use bunching in the distribution of weekly hours at forty--where the policy introduces a convex "kink" in firms' costs--to estimate this effect. To do so, I develop a framework in which bunching at a choice-set kink is informative about causal effects under substantially weaker assumptions than those maintained in existing work. This allows the effect of the overtime policy to be partially identified without making parametric assumptions about firms' objective functions, or about the distribution of hours they would set in the absence of the policy. Using an administrative dataset of weekly hours derived from payroll records, I find that the bounds are informative and that covered hourly workers in the U.S. work an average of at least half an hour less as a result, in affected weeks.
Chapter 2 turns to a still-more popular strategy in applied microeconomics: the instrumental variables research design. I propose a new method for estimating causal effects when a researcher has more than one such instrument, and apply it to reassess the labor market returns to college education. The method is motivated by the following issue. When treatment effects are heterogeneous, it is known that instruments can be used to identify local average treatment effects under an assumption known as "monotonicity”. However, when a researcher wishes to use multiple instruments together, this assumption can become quite restrictive, and empirical conclusions may be misleading if it is violated. I propose an alternative assumption that I call "vector monotonicity", which is quite natural in typical settings with multiple instruments. I show that vector monotonicity leads to identification of a useful class of treatment effect parameters, but the two-stage-least-squares estimator popular in applied work does not consistently estimate them. I propose an alternative estimator, and apply it to the classic question of the returns to schooling. I find that the approach based upon vector monotonicity reveals new patterns of heterogeneity in the earnings effect of college education.
Chapter 3, with coauthors Ashna Arora and Jonas Hjort, considers the effects of a worker's first job on outcomes later in their career. This is typically a difficult question to answer empirically, as workers entering the labor force are not randomly assigned to employers. We make use of a unique opportunity to study this question in the context of medical residencies in Norway. For decades, medical school graduates in Norway were matched to residencies based on a random serial dictatorship mechanism, in which doctors could choose--in an order determined by lottery--among available positions in the country. We develop an econometric framework in which the random choice set a doctor is presented with provides a collection of instruments for their choice of residency hospital, and hence first job as a doctor. Because we only observe choices and not a doctor’s full preferences, this requires new methods--related to those of Chapter 2. We find persistent effects of a doctor’s first job on earnings, specializations, and mid-career moves. We use the estimates to assess the replacement of the serial-dictatorship by a decentralized labor market in 2013, which we find led to a small increase in resident welfare.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/d8-vt2z-4g72 |
| Date | January 2021 |
| Creators | Goff, Leonard |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0044 seconds