Dear Prof. Clarke and Dr. Pailanir,
I hope this PR finds you well.
My name is Diego Ciccia, I am an incoming PhD student at Northwestern-Kellogg and I am currently part of Prof. Clément de Chaisemartin's RA team at Sciences Po Paris.
Before presenting the details of my PR, I would like to take the occasion to thank both of you for the sdid package. sdid has been a central estimation tool during my predoc and I have looked back so many times to its coding as a way of learning and improving on my own programming.
In this PR, I propose an extension to sdid, allowing for the estimation of dynamic effects. The extension comes in the form of a new command, sdidevent, fully integrated in your GitHub repo. You can find a formal description of the estimators in this short note. In a few words, I show that cohort-specific event study coefficients similar to those of Borusyak et al. (2024) can be obtained by disaggregating $\hat{\tau}^{sdid} a $. A weighted average of such estimators across cohorts yields event study estimators $\hat{\tau}^{sdid}_ \ell$, which can be, in turn, aggregated into the sdid estimator $\widehat{ATT}$. Overall, the extension works by reaggregating the cohort-specific estimators in event time. Yet, the resulting estimators are robust to differential adoption timing and can be useful to study the dynamic evolution of treatment effect in designs where the use of sdid is appropriate.
As for the package, this PR pushes the first viable version of the sdid_event command. As of now, it comes with 2 inference options (bootstrap and placebo) and an option to retrieve cohort-specific dynamic effects. I have also written an help file (where I list both of you as co-authors of the module due to the heavy reliance of sdid_event on sdid) and all the other ancillary files to download the extension from GitHub.
Let me know what you think. I welcome any feedback from your end and I am open to any revision.
Best,
Diego
Dear Prof. Clarke and Dr. Pailanir, I hope this PR finds you well. My name is Diego Ciccia, I am an incoming PhD student at Northwestern-Kellogg and I am currently part of Prof. Clément de Chaisemartin's RA team at Sciences Po Paris. Before presenting the details of my PR, I would like to take the occasion to thank both of you for the sdid package. sdid has been a central estimation tool during my predoc and I have looked back so many times to its coding as a way of learning and improving on my own programming. In this PR, I propose an extension to sdid, allowing for the estimation of dynamic effects. The extension comes in the form of a new command, sdidevent, fully integrated in your GitHub repo. You can find a formal description of the estimators in this short note. In a few words, I show that cohort-specific event study coefficients similar to those of Borusyak et al. (2024) can be obtained by disaggregating $\hat{\tau}^{sdid} a $. A weighted average of such estimators across cohorts yields event study estimators $\hat{\tau}^{sdid}_ \ell$, which can be, in turn, aggregated into the sdid estimator $\widehat{ATT}$. Overall, the extension works by reaggregating the cohort-specific estimators in event time. Yet, the resulting estimators are robust to differential adoption timing and can be useful to study the dynamic evolution of treatment effect in designs where the use of sdid is appropriate. As for the package, this PR pushes the first viable version of the sdid_event command. As of now, it comes with 2 inference options (bootstrap and placebo) and an option to retrieve cohort-specific dynamic effects. I have also written an help file (where I list both of you as co-authors of the module due to the heavy reliance of sdid_event on sdid) and all the other ancillary files to download the extension from GitHub. Let me know what you think. I welcome any feedback from your end and I am open to any revision. Best, Diego