Yunie Uyen Le

Economist, Ph.D & Research Scientist

Research Interests:

Applied microeconomics with an emphasis on using causal inference to identify the impacts of laws and policies on issues related to public safety, education, development, and mental health. I’m also interested in how the changes in laws and policies alter the behaviors of individuals in the decision-making process.

Working Papers and Projects

The Unintended Consequence of Locally Centralized Online Marketplace (Job Market Paper)

Market centralization has been shown to improve market efficiency and stability by reducing the cost of searching and increasing the matching quality for both sides of the market. Existing literature has focused on examining the benefits of centralized markets on different aspects of society. However, the unintended consequence of centralized markets has only been studied in a limited number of environments, including municipal solid waste, disease transmission, and female safety. But, centralized markets have had a considerable impact on market exchanges that could result in non-violent criminal behavior. Using crime data from 1991 to 2014, I examine the effect of Craigslist - a locally centralized online marketplace - on bike theft, which accounts for nearly $77 millions in costs per year. In addition, since Craigslist became available in each location at different time periods, the introduction of Craigslist yields heterogeneous treatment effects across groups of locations based on the year of entry. I exploit the semi-parametric differential-timing difference-in-differences method introduced by Callaway and Sant'Anna (2020) to estimate the Craigslist effect. I also decompose the ordinary difference-in-differences with two-way-fixed-effect estimates using Goodman-Bacon (2021) decomposition to show that these estimates are biased. I find that bike theft increased about 11% in Craigslist service areas compared to non-service areas, while other types of property and violence crime are not affected by the entry of Craigslist.

Pandemic Safeguards and Household Safety (with A. Assamidanov, S. Cunningham, G. DeAngelo, and R. Thornton)

A flurry of research examined the effect of COVID-19 related shelter-in-place as well as school and daycare closure orders on family violence. There is considerable variation in the data samples used in previous work, which include geographic units included in the analysis and time frames studied. Previous analyses utilizing difference-in-differences analyses often produce conflicting results. In this work we include the most comprehensive data set from the United States, which includes nearly three years of data from 30 cities, to examine the effect of COVID-19 related orders on family violence to ensure that our conclusions are not reached due data selection issues. We also improve on previous analyses by utilizing the decomposition in Goodman-Bacon (2021) and the estimator in Callaway and Sant'Anna (2020) to account for the differential timing in the implementation of COVID-19 related orders. Our analysis concludes that school closure significantly increased the number of child abuse calls by 70% from the mean. We detect no effect for shelter-in-place or daycare closure orders. We also document a reversal in the direction of our main coefficient estimates when using Callaway and Sant'Anna (2020)'s estimator to measure the effect of daycare closure relative to two-way fixed effect.

Job Mismatch and the Rise of Suicide (with G. DeAngelo, M. Garner, and S. Nishioka)

Suicides present one of the most serious public health crises that the United States faces. Suicide contributes 1.7\% of all deaths in the United State, making it the tenth leading cause of death. Much of the existing literature has focused on research that examines the impact of job loss on mental health and suicides. Using U.S. data from 1990 to 2015, however, we show that suicide rates increased even in the regions that added a substantial number of jobs. In this research, we attempt to resolve this puzzle by focusing on the change in the manufacturing share of employment, instead of job losses, to examine the impact of job mismatches on suicide. To ensure we have uncovered the causal relationship between job mismatches and suicides, we utilize an instrumental variable strategy where we instrument the manufacturing share of employment with the exposure to the manufacturing shock created by the Permanent Normal Trade Relationship agreement with China. We refer to this as tariff gap. This research finds that regions experiencing a 10-percentage-point decrease in the manufacturing share of employment leads to a 0.73 increase in deaths by suicide per 100,000 residents in the working-age population, which is equivalent to 1,460 deaths for a working-age population of 200 million. The main effects of our analysis are largely driven by regions experiencing non-decreasing job prospects, but significant losses in manufacturing jobs.

Cooperative Games: Fairness of Matching Games and Applications of Matching Theory

In recent years, among the field of cooperative game theory, matching theory is getting a lot of attention due to its high practicability. Since its first introduction in 1962, matching theory has been developed from a beautiful, yet simple, theory to highly practical applications in market design. One of the concerns about cooperative games is the fairness of the games. Is the matching fair to all players from both sides of the market? This paper discusses the fairness of matching games and surveys the developments of matching theory and its applications in solving resource allocation problems in different fields.

Bayesian Blocks: Applications in Big Data and Segmentation (with R. Shiroma, S. Deo, and K. Lenk)

The idea of this project is to model arrival-time data with the ultimate goal being to model the shape of gamma ray burst radiation intensities. We assume that gamma ray bursts follow some underlying radiation intensity that varies over time. We might expect to see intensity to start off constant for some time(background radiation), then a spike of radiation intensity at the start of the burst, then a gradual reduction in intensity thereafter. This intensity comes in the form of individual photons hitting a sensor in a gamma ray telescope. The photon data can be collected and analyzed in a variety of ways, mostly involving some form of binning photons. We would therefore like to capture the shape of this intensity over time over multiple piecewise intensity functions. The current method, Bayesian blocks developed by Jeff Scargle, captures this gamma ray burst shape by representing it as a step function over time (piecewise constant blocks). Our method takes it one step further and generalizes Bayesian blocks to allow for non-constant blocks. If we can model differently shaped blocks rather than patterns of piecewise constant blocks, we can hopefully better identify gamma ray bursts in the data.