Submitted by Kirill Borusyak on
In Borusyak, Jaravel, and Spiess (2023), we have developed a new "imputation" estimator for difference-in-differences models. We have provided a Stata command implementing this estimator (did_imputation, available from the SSC archive); a basic implementation for R has been developed by Kyle Butts (https://github.com/kylebutts/didimputation). The methodology has since been used in a number of empirical analyses in economics and related disciplines.
However, both Stata and R commands have limitations and need improvement to benefit the economics research community:
- The Stata command performs poorly in the presence of many control variables. This is because of an inefficient way in which the underlying iterative least squares algorithm has been implemented. A very similar algorithm has been implemented much better in the reghdfe command (using Mata language within Stata). The main goal here is to adopt the algorithms used in reghdfe and improve the performance of did_imputation.
- The R implementation only features the most basic syntax of the command. It needs to be rewritten to allow most functionality available for Stata users, while maintaining reasonable computing speed.
I am therefore looking for one or two students with good general coding skills and some experience with programming in Stata and R, respectively.
Under my supervision, understand the difference-in-differences imputation methodology and implement it in Stata and/or R.
Coding skills and experience + familiarity with programming in Stata and/or R.
Experience in Stata as a script language only, e.g. to run regressions, is not ideal. However, if you have coding experience in other languagues + basic familiarity with Stata + willingness to learn more, that may work.