Mixtape Mailbag #4: Matrix Completion, diff-in-diff or synthetic control?

Welcome to this week’s Mixtape Mailbag, a weekly Q&A where readers write in questions and I take a stab at answering them. Sometimes with some help from my friends, sometimes alone. This time it is alone as I need to stand on my own for something I have said.

This week I received an email from someone regarding something I wrote in my book about matrix completion with nuclear norm regularization. I predicted that the availability of code would cause the adoption of matrix completion relative to difference-in-differences. This is a prediction I’ve made before, so I am glad for the opportunity to think about it again.

---

Dear Professor,

I hope this message finds you well. 

My name is AGG, and I am a first-year doctoral student in an European university studying Economics. Please allow me to ask a question that will not take more than a minute of your time. I have been immersed in your book, "The Mixtape," and I have gone deep into the Matrix Completion method by Athey (2021). In your book, you express the belief that when Stata releases the code for this method, its popularity will likely increase. Do you still hold this view? I am currently grappling with a panel dataset comprising approximately 500 observations, and I am uncertain about the extent to which employing the Matrix Completion method would be advantageous compared to more conventional approaches such as Difference-in-Differences (DiD) or the Synthetic Control Method. Your guidance on this matter would be highly valued.

Let me also show you my sincere appreciation for your time and for sharing your knowledge through your book which is very useful.

Best regards,

AGG