research

• The Effect of Air Pollution on Human Capital Development: Evidence from South Korea
  

  Air pollution is known to affect health outcomes across all age groups, through direct inhalation of toxic matters. While the effect of pollution matters on extreme health outcomes are extensively studied, little is known about its causal effect on academic performances of students. I analyze the effect of air pollution on academic performances of Korean elementary/middle/high school students using an instrumental variable (IV) approach. I use wind direction as source of variation to evaluate the causal effect of air pollution, exploiting unique geography of Korea where pollution matters from China exogenously vary depending on wind direction. I find that increase in the level of particulate matter increases the share of students who underperform in Math/Korean/English subjects, and decrease the share of students who overperform in all subjects.

  (Extended Abstract)

  
  
• Causal Inference Under Covariate Shift via Worst-case Subpopulation Treatment Effects
  with Hongseok Namkoong
  Conference version appeared in Conference on Learning Theory (COLT), 2020
  Journal version under review
  

  Across medicine, economics, and tech, marginalized groups are heavily underrepresented in datasets. Standard causal approaches that estimate the average treatment effect ignore underrepresented subgroups under such biases. We propose a notion of worst-case treatment effect that guarantees uniformly good performance across all subpopulations of a given size. We develop an estimation approach that allows flexible use of ML models to estimate treatment effects in both randomized and observational settings. Our method is agnostic to particular demographic groupings, and guards against brittle findings that are invalidated by unanticipated covariate shifts.

  
  
• The Effect of Delayed Retirement Credit on Social Security Claiming and Employment (in progress)
  with Mark Duggan, Irena Dushi,  and Gina Li
  

  Understanding how individuals respond to incentives for delaying Social Security (SS) claiming is important from a program design perspective, due to insolvency concerns arising from rising program costs, increased life expectancy, rate of retirement among baby boomers, and decreasing birth rates. We investigate the effect of delayed retirement credit (DRC), an upward adjustment in monthly benefits that works as an incentive for individuals to delay their Social Security claiming. In aggregate, individuals do not seem to be changing their claiming behavior in response to change in DRC, despite the significant increase in benefits of claiming later. With micro SSA data, we hope to further shed light on how DRC might affect claiming and working behavior of older individuals.

  (Project Summary)

work experience

• Lyft
  Research Scientist Intern, 2020
  • Investigated long term heterogeneous treatment effects of Lyft's coupon program, and performed subgroup analysis for riders with disparate treatment effects
  • To infer long term business impacts, developed a surrogate modeling approach that predicts long term outcome based on short term outcome, a more flexible improvement to the default long term multipliers


• MIT Economics
  Research Asisstant, 2015-2016
  • Managed a randomized control trial studying the effect of full college tuition scholarship on academic achievement of students attending post secondary institutions in Nebraska
  • Applied semi-parametric statistical methods to estimate the causal effect of the scholarship


• National Bureau of Economic Research
  Research Asisstant, 2013-2015
  • Applied diff-in-diff approach to estimate spillover effect of rent decontrol in Cambridge, MA in 1996
  • Acquired and cleaned data on crime and city budget from the Cambridge Police Department