The Mixtape with Scott

In this week's episode of the Mixtape with Scott season 5, "The Odd Couple" Scott Cunningham (Professor of Economics at Baylor) and Catilin Myers (Professor of Economics at Middlebury College) set out to use Claude Code to get the data for their project studying travel distance to the nearest abortion clinic's effect on marriages in Texas after House Bill 2 shut down half the state's clinics. As they do, they talk about their project, the trappings of having a third party robot as a colleague and RA on this project done on the air, and articulate aloud the prompts as they do them!
Thanks again for all your support! This substack and the podcast are labors of love. Please consider becoming a subscriber at only $5/month!
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

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

Welcome back to The Mixtape with Scott. I’m currently in the process of putting together a new slate of interviews, and while it’s not quite ready yet, I didn’t want to leave you hanging. So in the meantime, I’m re-sharing some conversations from earlier seasons — episodes that I think are worth revisiting or perhaps discovering for the first time.
Today’s rerun is from Season Two, and it’s one of my favorite interviews from that time: my conversation with Ariel Pakes, the Thomas Professor of Economics at Harvard University.
This was such a fun and rich interview. People either know Dr. Pakes very well or only by the letter “P”. He’s a towering figure in industrial organization and structural econometrics, with landmark contributions both theoretical and applied. Among many things, he’s the “P” in the Berry-Levinsohn-Pakes model — BLP — which remains one of the most influential tools for estimating demand in differentiated product markets. That paper — Automobile Prices in Market Equilibrium — published in Econometrica in 1995, has had a ripple effect not just in economics, but well beyond it.
But this interview wasn’t just about methods and models. Dr. Pakes and I talked about basketball, about growing up in a radical socialist youth group, about his early love of philosophy, and his own path through Harvard as a young man trying to straddle economics and philosophy before finding his place. He spoke softly, with depth and reflection, and he offered a glimpse into how he works — by getting himself in way over his head and then slowly, patiently, working his way out. It’s a way of thinking that hasn’t just shaped his own work but has helped shape the rest of ours too.
I hope you enjoy this one as much as I did.


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

Greetings from Cambridge! I’m still mid-move and not fully settled—classes kick off next week and I’m wrapping the last bits of admin—so I’m sharing one more rerun before we close out Season 4.
Today’s guest is a Cambridge neighbor just down the Charles at MIT: Dr. Amy Finkelstein, John Bates Clark Award–winning economist.
If you’re new to her work: Amy is a leading health economist at MIT and coauthor of We’ve Got You Covered (with Liran Einav), a timely book from a couple of years ago. In it, they argue for universal basic coverage that guarantees financial protection from major medical costs, while leaving room for supplemental private insurance—simple, fair, and focused on what insurance is actually for.
In our conversation we cover the Oregon Medicaid Experiment and the ideas that shaped it, plus the arc of her career. I loved this one. Hope you enjoy the rerun.
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

Greetings everyone. I’m still in moving mode, packing up life in Texas and getting ready for a year in Boston. I hit the road on Friday of this week for a three day road trip and am still behind on everything. That means the podcast is still on reruns for now, but I should have a new episode for you next time. This week’s rerun is one I really liked, though—my conversation from two years ago with Steven Berry.
Steven is the Sterling Professor of Economics at Yale and the inaugural Faculty Director of the Tobin Center. His work in industrial organization has shaped how economists think about markets in equilibrium, and his research spans industries from autos to airlines to media. He’s also a winner of the Frisch Medal, a member of the National Academy of Sciences, and one of the field’s most respected voices.
We talked about his path into economics—from the Midwest, to Wisconsin, to a career that’s helped define modern empirical IO. Naturally, we dug into the BLP model, the landmark framework he developed with James Levinsohn and Ariel Pakes that changed how we estimate demand in differentiated product markets. It’s one of those ideas that’s both deeply technical and hugely practical in policy and business.
If you missed it the first time, I think you’ll enjoy hearing Steven reflect on his career, his collaborators, and where the field is headed. Here’s my rerun conversation with Steven Berry.
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