Estimating remote work’s effect on air pollution

  • Group project
  • I helped lead the design and completion and performed substantive troubleshooting on all aspects.
  • Goal: Isolate the effect (linear coefficient) of remote work’s effect on PM2.5 air pollution in U.S. metropolitan areas
  • Method: Compared multiple approaches to weight the covariates in a Double ML model
    • Ultimately selected random forest (aka ranger)
  • Result: a 50 percent increase in remote work could reduce the amount of PM2.5 in the air by about 25 percent of the EPA standard.
  • Data from EPA and Census Bureau
  • Coded in R
    • tidycensus
    • mlr3
    • DoubleML
  • Github for version control