Research

I explore the causal effects of single-sex high schools on long-term labor market outcomes using a unique feature of educational policy in South Korea—the random assignment of students to single-sex versus coeducational high schools. I find that the effect of single-sex schooling on earnings is positive for women but not for men. The effect varies with job experience, increasing by 1.9 percentage points with each additional year of experience. In addition, single-sex schooling affects other labor market outcomes, including women's work hours and job satisfaction. Finally, I use measurements of completed education and job training as human capital accumulation to study potential mechanisms for gender differences in the effect of single-sex schooling.

We scrutinize the dynamics of job loss and reemployment corresponding to the COVID-19 pandemic and the policy reactions from the different governments using a range of high-frequency data on employment and unemployment from diverse countries: Australia, Italy, South Korea, and the United States. Using the information on demographic and socioeconomic characteristics, we can show how COVID-19 affects the labor market in a social stratification framework that exploits population subgroups sorting different jobs. We find greater declines in employment for young, low-educated, and single individuals in most countries, while the size of impact varies across countries. This decline in employment affects individuals’ employment status in three parts: recently unemployed, absent from work, and out of the labor force. However, the changes in the three-unemployment statuses are different between nations. In addition, this paper shows that the pre-COVID-19 sorting of workers into occupations and industries can explain disparities in labor market outcomes during COVID-19 along demographic lines. This paper finds that the face-to-face index is positively correlated to the recently unemployed, but the remote work index is negatively correlated, which provides a potential channel of labor market disparities results from the pandemic. For example, the youngest group who experienced the largest decline in employment are more likely to have a job requiring face-to-face interaction but not allowing remote work in 2019. Finally, our OLS estimation results show that occupational and industrial characteristics account for a large portion of labor market changes.

This paper analyzes the differential fertility effect of maternity leave by a woman's educational level and the type of maternity leave. Identification is based on changes in the opportunity cost of children by the provision of maternity leave. Using NLSY97 data and a single-equation probit model, I find that fertility decisions are strongly correlated with the provision of maternity leave, and the magnitude of the effect is larger for highly educated women than for less educated women Second, unpaid maternity leave may yield a stronger fertility response than paid maternity leave because of the wage-fringe trade-off. Third, this paper shows that the estimated effect of maternity leave in the single-equation probit model might be subject to selection bias. In bivariate probit models with an instrumental variable, I find no evidence that my single-equation estimates are subject to selection bias for working women with a graduate degree. However, the MLE estimates of bivariate probit models imply that there might be upward selection biases in my single-equation estimates for less-educated women. 

This paper uses model selection and machine learning methods to compare methods to estimate the average treatment effect in high dimensions. I focus on regression adjustment with Lasso, approximate residual balancing, and the causal forest methods under the unconfoundedness assumption, which requires randomness of the treatment assignment conditional on pre-treatment variables. By comparing the estimation results from each method, my research can help us after which method is best for a high dimensional experiment in which minimal assumptions are satisfied. In simulation studies, the approximate residual balancing methods perform better than other methods, even in high dimensions. This paper also applies these methods to real data from Angrist and Lavy (2001). The results show that the estimates of the causal forest method are almost the same as the estimates from Angrist and Lavy (2001) in which they use an individual-fixed effect estimation.

This paper investigates the enforcement effects of the “Law on the Improvement of the Mobile Terminal Distribution System (Mobile Terminal Distribution Act),” which was implemented in October 2014. Despite the nationwide attention to the legislation, an empirical analysis of the market effects of the Mobile Terminal Distribution Act is yet to be done. Using a rich dataset obtained from handset retail stores authorized by telecommunication agencies, we perform both an ordered probit MLE and a panel regression. We find the following facts. First, Mobile Terminal Distribution Act has effectively enabled consumers to choose less expensive plans. The reduced amount of subsidy for handset purchases seems to be the main reason for this phenomenon, as Mobile Terminal Distribution Act originally intended. Second, we find no statistically significant reductions in the prices of newly launched handsets. While the prices of some products have actually dropped, our analysis shows that it is not due to the implementation of the Mobile Terminal Distribution Act. 

Using data from the Korean Labor and Income Panel Study (KLIPS), this paper tests four hypotheses on the source of the marriage premium for South Korean men and women: the selection, productivity, specialization, and favoritism hypotheses. Fixed effects regression results show that men experience a significant marriage premium, while women face a significant marriage penalty, resulting in a significant gender differential in the returns to marriage. Our results are generally consistent with all four hypotheses, with the selection and favoritism hypotheses receiving stronger support than the productivity and specialization hypotheses. We conduct instrumental variable estimation throughout our analyses using the zodiac sign of the birth year to test the endogeneity of marital status, presence of children, and spouse’s labor activities, all of which are subsequently suggested not to be endogenous.

This paper examines the long-term relationship between health expenditures, both government and private, and selected health outcomes, infant mortality, and child mortality based on the interactive fixed effects model using data from 160 countries for the years 1995, 2000, 2005, and 2010. The interactive fixed effects model allows unobservable variables like health technology to be captured by the interaction between country fixed effects and time effects. With this model, I estimate the elasticity of infant and child mortality with respect to government and private health expenditures. Results from this paper are at variance with previous studies asserting the importance of increasing public and private health expenditures on infant and child mortality. This paper also provides new evidence of the importance of taking into account the level of country-specific factors interacting with time effects for estimating more accurately the effect of health expenditure on health outcomes. 

This project examines the differential effect of school quality on academic achievements within and between school districts in South Korea. Exploiting the unique educational system and datasets of school characteristics in South Korea, I find that school quality has a positive effect on academic achievement and that its magnitude is much larger for students with low-level previous test scores. Under the random assignment system, this paper suggests that study-related facilities and additional chances to study after regular classes are important factors in improving students’ performance in addition to the traditional school quality measurements (e.g., teacher-pupil ratio and financial spending by school). Once the random assignment is relaxed, I can find a positive relationship between school quality and student background or “ability”. Though exploiting limited individual-level data, this paper supports the argument that high-quality schools attract more students (whether they have increased ability or not) and construct network formation through which the effect of school quality on academic achievements becomes stronger.