The Mixtape with Scott (Featuring Caitlin Myers) Season 5: Episode 1 of The Odd Couple!

The Odd Couple

The Mixtape with Scott is back. Season 5. Season 5 of the Mixtape with Scott is going to be different, and fun, and different, and creative! It’ll be called The Odd Couple. And it’ll be called “The Mixtape with Scott (Featuring Caitlin Myers)”. It’ll have different naming conventions until Caitlin pick one we like! Let me tell you all about it.

I started the podcast around four years ago as a way of creating an oral history of economics while also tracing out the history of the credibility revolution through Orley Ashenfelter, his students, and the Industrial Relations Section at Princeton. I tacked on a bunch of other things too along the way like “the students of Gary Becker” and “economist in the tech industry”, as well as any number of eddies I wanted to swim in along the way.

And after 130 interviews, I more or less felt like I had tapped my creativity out. I largely came to understand the evolution of causal inference a particular way, which I wrote up across several substacks, as well as added throughout my new book, Causal Inference: the Remix (proofs came to me today in fact). It was very rewarding. Maybe one day I’ll write up the interviews as a book (even Claude Code cannot yet do that), but for now, I’m just ready to move on, as 130 interviews is a lot.

But move on to what? Well, that’s what I want to tell you about now. Today’s episode is the first episode in a season I’m calling “The Odd Couple” featuring the brilliant economist, Caitlin Myers. And the concept is simple:

Caitlin Myers and me will start a research project together which is only performed on the podcast. And we will use Claude Code to do this project on the air. While doing it, we will talk and laugh and share our thoughts about what we are doing. Think of Bob Ross talking while he paints trees. Only instead of trees, it’s estimated dosage parameters of abortion clinic closures’ effect on marriage using continuous diff-in-diff. And instead of a brush, we are using Claude Code who is using R, python and Stata. But other than those trivial details, it is exactly like Bob Ross, or maybe the View.

The Odd Couple featuring Caitlin Myers, Scott Cunningham and Claude Code

Caitlin Myers is the John G. McCullough Professor of Economics at Middlebury College in beautiful Vermont. And she is, at the time of this writing, arguably one of the leading economists working on reproductive policy in the United States, maybe the world. She’s been published a lot on the topic for a very long time, including this article in the Journal of Political Economy, our JHR on abortion clinic closures, and numerous others. You can find it all at her slick website.

She’s also been a contributor to the public good by creating public data repositories. She built this dashboard. She knows where every clinic opened and closed and when, going back decades. She’s meticulously described each and every relevant law regulating abortion access. If you’ve read a paper in the last ten years about abortion services, there’s a good chance a design by Caitlin, or data she helped curate and distribute, was somehow connected to it. Her influence in this space has been massive.

But in addition to being great, she’s also funny, thoughtful, and thinks really well on her feet. Which is one of the reasons I thought it would be great to have her as my research partner and conversation partner on the podcast. Because I think if this concept is going to work, a lot of planets have to align, and I had been thinking for a very long time that if there was such a square peg to fit a square hole, it would be her.

I would say that Caitlin and I are right at that sweet spot of professional acquaintances bordering on friends. That’s the type of person who you make a point to find when you are at a conference and get a drink with even if you aren’t at that moment writing a paper together. It’s that person who you shared a little about your private life with when you were on a car ride together to the airport. It’s that person who you text memes of Beyonce giving out high fives for no good reason. It’s that person you want to send a note to in class saying “Will you be my friend? Circle yes or no”.

No one does this on the air

So the idea of this podcast is that she and I are going to extend an old study of ours with Jason Lindo and Andrea Schlosser published in the Journal of Human Resources called “How Far Is Too Far?” It studied what happened when Texas passed HB2 in 2013 and nearly half the state’s abortion clinics closed overnight. We used the sudden, geographically uneven changes in driving distance to the nearest clinic to estimate the causal effect of access on abortion rates. The punchline was that distance matters, the effects are non-linear, and congestion at the surviving clinics matters too.

But what we want to do is extend the research design in a couple of ways. First, we want to study the effect that the abortion clinic closures had on marriage. While Caitlin has studied the effect of abortion access on marriages, no one has look at the clinic closures on marriage using, more specifically, the “travel distance design” as I call it. Secondly, we are going to be learning how to estimate treatment effect parameters, as well as what those estimands even mean, using the new conditionally accepted (at the AER — woo hoo fellas!) continuous diff-in-diff estimator by Callaway, Goodman-Bacon and Sant’Anna estimator. This estimator already has over a thousand cites and it’s only just now conditionally accepted — it’s not even really really accepted. It’s like the AER is saying it likes you, but does it really really like you? Not until it’s accepted you does the AER really really like you. Right now it’s a conditional accept which is more like a situationship. Anyway, I’m rooting that these two get hitched, and so we’re going to be using their estimator with this travel distance design to estimate a bunch of estimands that we’re going to learn about together. So that’s fun.

The AI angle

And then third, and maybe the goofiest of all — Claude Code. We are going to do all of this using Claude Code. The hope being that we can wrap our hands around just how to use this thing to do good, and not evil. And I think this is the funnest (most fun?) part because Caitlin is probably the more pessimistic towards AI, whereas I am the most optimistic, which on average means we are aloof to AI. And Claude is probably going to sometimes agree with me, sometimes with Caitlin, and sometimes just want to say we all have a great point. Anyhow, we are going to be doing this project together using Claude Code so that listeners and viewers can better see how we use Claude Code for practical empirical research, and how we go about trying to get it to not jump the electric fence, or if it does, not cause mayhem.

But as I said, Caitlin and I have very different priors on this. I’m the AI optimist and she’s the AI skeptic. While we have both been using Claude Code for months, and we’ve both seen what it can do, and we both agree we’re in the early innings of something that fundamentally changes how research gets done, I think we both have fundamental opinions and concerns that sometimes overlap with each other and other times don’t.

But she is, I think like me, curious to a fault. She wouldn’t be doing this if she weren’t — but she thinks AI is, in her words, an existential threat to humanity. And she is not being dramatic. She means it. And that’s not an uncommon worry among people, nor is it an uncommon position to take that people simultaneously are angry or upset about AI and want to better understand Claude Code’s utility for practical empirical research. That’s just the times that we are in that both of those can be true at the same time for the same person. She’s the person at the table asking the hard questions about what happens when these tools get good enough that the verification problem becomes the only problem.

So you have one person who thinks this is going to be incredible and one person who thinks it might end civilization, and we’re both using the same tool to do the same project. That tension is real, it’s productive, and it’s part of what you’ll hear.

And here’s the thing about podcasting with Claude Code running in the background: there’s a lot of time while it’s working. It’s reading files, writing scripts, compiling things, running pipelines. And during that time, Caitlin and I are talking. About AI, about science, about what we’re seeing in real time on the screen, about the project, about whether what just happened was impressive or terrifying or both, or just about life, about the meaning of being a researcher, about our worries and hopes and where, and so on. And we are joking around and bantering.

It’s like The View if The View had two economists staring at a terminal.

What to expect

Episodes will drop as we work through the project. Some will be data work — the kind of session where we’re elbow-deep in county FIPS codes and file format inconsistencies. Some will be methodological — working through the continuous diff-in-diff framework, figuring out what the identifying assumptions actually require. Some will be the conversations that happen in between — about AI, about the future of empirical research, about what it means to do science in public.

I don’t know how many episodes this will be. I don’t know what we’ll find. I don’t know if the marriage result will be a null or something real or something we can’t interpret. As they say in therapy, it’s about the journey not the destination! This podcast is about the journey, which is to say it’s about the joy researchers get from doing research, not necessarily from completing it. And it’s a podcast of two people talking while they do it.

The Mixtape with Scott is back. Season 5. The Odd Couple. Featuring Caitlin Myers. We're making the sausage, and you're invited to watch

Scott's Mixtape Substack is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Get full access to Scott's Mixtape Substack at causalinf.substack.com/subscribe