Cornell University TREESPEAR Graduate Research Associate Dingyi Li presented his research on air pollution avoidance behavior at the North American Summer Meetings of the Econometric Society in Miami, FL.
For his research on air pollution avoidance behavior, Dingyi is using sophisticated econometric techniques, machine learning, and big data to analyze how air pollution affects travel mode decisions in China, where the daily average concentration of fine particulate matter (PM2.5) is about six times of the World Health Organization guideline. In 2013, China launched a nation-wide, real-time air quality monitoring and disclosure program, the first-of-its-kind in history. Dingyi is comparing the effects of air pollution on travel mode in 2010, prior to the reform; as well as in 2014, after the reform, to distinguish and untangle the different channels through which air pollution may influence behavior.
Dingyi is analyzing how air pollution affects travel mode decisions in China using a large and detailed hourly household-level data set on hourly household-level travel mode decisions in Beijing, and hourly air pollution, weather, wind speed, and wind direction in and around Beijing. He is using atmospheric chemistry and instruments used in the previous literature to form a set of potential instruments for air pollution to address its endogeneity. He is then using machine learning, LASSO regressions, and atmospheric chemistry to select his instruments for air pollution from among his large set of potential instruments. His machine-learning-based method of IV selection also serves as an innovative data-driven method for analyzing air pollution, air pollution transport, and air pollution spatial externalities.
Air pollution is endogenous to travel decisions for two main reasons. The first reason air pollution is endogenous is due to unobserved variables. For example, the error term contains the number of days a person is staying in the home, which is influenced by pollution. The second reason air pollution is endogenous is due to simultaneity: since travel modes that rely on fossil fuel vehicles emit air pollution, aggregate travel mode and travel decisions influence air pollution. The endogeneity of air pollution can be only partially addressed by the day of a week and hour of a day fixed effects.
Dingyi uses atmospheric chemistry from cities around Beijing or above Beijing at high altitudes to form a set of potential instruments for air pollution to address its endogeneity. Since the set of potential instruments is so large that there may be weak IVs in the 2SLS, Dingyi is conducting a 0-stage lasso regression to pick the strong instruments. He is conducting falsification tests to further validate the instruments selected by LASSO. For example, the cities for which wind instruments are picked by LASSO should be large and polluted. Similarly, if wind from a particular city is picked by LASSO, then pollution from that city should cause an increase in other cities downwind of that city in between that city and Beijing. Dingyi's machine-learning-based method of IV selection also serves as an innovative data-driven method for analyzing air pollution, air pollution transport, and air pollution spatial externalities.
In addition to the North American Summer Meetings of the Econometric Society, Dingyi has also presented his research at the World Conference of Spatial Econometrics Association (SEA).
For further reading: