Mixtape Mailbag: Staggered Adoption and Combining Synthetic Controls

Here’s a letter I got a while ago for the Mixtape Mailbag that I’ve been meaning to reply to. I set it aside to finish the book, but it’s a good one—covering two big topics: staggered adoption with synthetic control and extrapolating effects to untreated units.

Before diving in, a quick note: If you have a question, feel free to write me at causalinf@mixtape.consulting and cc my new assistant, priscilla@mixtape.consulting. I anonymize all questions and strip out identifying details, so you don’t have to worry about sharing too much too soon. The goal is to keep the essence of the question while making it useful for everyone.

Mixtape Mailbag is something I’m trying out for paying subscribers—both as extra value and as a way to hear what people are working on. Plus, it helps me think through interesting problems from different angles, which I always enjoy.

Now, let’s get into it.

---

Hi Scott,

First, I want to say how much I enjoyed your workshop. It was one of the best I’ve attended since grad school, and I learned a lot—thank you!

I have two related questions about policy evaluation methods, particularly in cases of staggered adoption.

Question 1: Can I use synthetic control when treatment is staggered?

I’m studying a policy that has been adopted by 21 states at different times since 2013. One state—let’s call it Midwestia—has not yet adopted the policy, but I want to analyze its potential impact. My initial approach was to compare Midwestia to neighboring states that have already implemented the policy. However, I’m now considering whether synthetic control could work, given the staggered nature of adoption.

From what I understand, synthetic control typically works best when there’s a single treated unit with a clear treatment date. Since my setting involves multiple states adopting at different times, would synthetic control still be valid?

Question 2: Can I use synthetic control estimates from treated states to infer the effect in a state that has not adopted the policy?

Since Midwestia hasn’t yet adopted the policy, I was considering an approach where I use synthetic control to model the policy’s effects in the 21 treated states and then extrapolate those effects to Midwestia. The idea would be to construct a synthetic Midwestia from the states that did adopt the policy and estimate what might have happened if it had also implemented it.

Thanks again for your time and insights—I really appreciate it!

Best,

Wondering in Wichita

---

Hi Wondering in Westeria,

First, thanks for your thoughtful question and for engaging so deeply with the material! You’re raising two distinct but related issues:

1. Can you use synthetic control in a staggered adoption setting?

2. Can you use those estimates to extrapolate what would have happened in Midwestia, a state that hasn’t yet adopted the policy?

Let’s tackle these in order.