|Title:||Using Machine Learning Methods to Improve Regression Discontinuity Designs|
|Principal Investigator:||Opper, Isaac||Awardee:||RAND Corporation|
|Program:||Statistical and Research Methodology in Education–Early Career [Program Details]|
|Award Period:||2 years (07/01/2020 - 06/30/2022)||Award Amount:||$224,999|
|Type:||Methodological Innovation||Award Number:||R305D200008|
Co-Principal Investigators: Engberg, John; Johnston, William
The purpose of this project is to develop an approach that incorporates machine learning methods into a regression discontinuity design to improve precision and reduce bias in the treatment effect estimates. The new approach will use ridge regression to obtain residuals of the outcome, with the residuals then used in a typical regression discontinuity analysis. After initial development, the researchers will extend the software to calculate standard errors in a less computationally intensive manner, to estimate treatment effect heterogeneity, and to allow for random effects at the classroom level and/or the school level.
The design of the study features simulations to compare the new technique to existing methods and an analysis of the New York City administrative data in order to provide an applied demonstration. This project will result in journal publications of the theoretical results, conference presentations, a how-to guide for the new method, and software for conducting the new method in R, Stata, and potentially Python. The research team will test the user-friendliness of the software packages on graduate students.