Leaving Lucca, Closing Tabs

Well, this was my second visit to Lucca Italy in two year. Last year I came for two weeks, after visiting Collegio Carlo Alberto. I asked ChatGPT last year to recommend somewhere to go for two weeks to be peaceful and write, as I had a gap between Collegio Carlo Alberto and St Gallen, and it recommended Lucca. I loved it. I spent two weeks doing nothing but writing and walking around the city, inside the old renaissance era wall, and along the wall. And this year, I visited IMT School of Advanced Studies where I presented new research on the behavior of AI agents in semi-autonomous empirical research, as well as met with students and faculty. I made new friends, hung out with a new old friend, and relaxed again. But now I head to Belgium to KU Leuven for six days. Three days to rest, three days to present, do a workshop and more or less meet faculty and students. And then I am done — Friday I arrive in Madrid, and my week long vacation in Spain, in San Sebastián, visiting again Amelias’s, begins. I have promised to myself I’ll eat at Amelia’s every year until I’m 60 as part of a “50 things by age 60”. As I only turned 50 in November, I can’t count the other two times I went with my daughters, so technically it starts now. This will be year one.

I think after presenting new diff-in-diff material around five times this summer, the new material has a nice arc. I go through the 2x2, them covariates, then continuous, then Callaway and Sant’Anna in the context of an applied paper that went topsy turbo. I have decided to not do the “all the diff in diffs” for a while, because I think it’s more important to focus deeper on one or two things than go shallow across fifty things. So that’s what I’ve been doing and so far, it’s going good. I almost have the covariate lecture where I want it, but not quite. I’m trying to start teaching the fully saturated regression (two way fixed effects) model for the 2x2 case so that I can lead into outcome regression (Heckman, ichimira and Todd 1997, restud) and then Jeff Wooldridge’s own regression adjustment diff-in-diff under staggered adoption. Then I can cover Borusyak, et al after that, since Jeff shows his model is numerically the same as theirs in a balanced panel without covariates. If I can land all those planes, then I can show the connection between OLS and imputation and regression adjustment while covering the bias of “vanilla” two way fixed effects. I think it’s going to work. I just have to work a little bit harder at it but it’s close.

But let me share some tabs as there were some interesting ones this week.

---

This post by Noah smith about the declining influence he thinks he may have because of the advancing of AI was interesting. Maybe. Maybe not. But even if so, it reminds me that the pursuit of influence has never been my person objective function. I do what I do for two reasons. I feel a sense of mission and calling to love unconditionally the world and help create bridges back to econometricians and design. And I do what I do because it feels good and makes me happy. I think we can overthink it to be honest. Mission and love and service and joy are really only coincidentally related to impact. Get the Cart and the horse out of order and I think you can lose the joy, which is the worst of all.

Does anything I write matter anymore?

I read three stories about the same thing, all in somewhat mainstream media. They are all about how a major hack and release of everyone’s private information is imminent. This one was particularly troubling. Here was the second one. And the Atlantic article, “assume you will be hacked.” My prediction is Boomers and the newest generation will not care, and millennials and Gen X will be troubled, if it rolls out en masse, as Boomers probably have on average poorer email, text message and social media histories, and the newest generation too.

Which again I think the lesson for us all will be to cultivate the courage to be disliked, and become better at separating our life tasks from what is others’ life tasks. Somehow this is both easy to do and easier said than done.

All the editors of the AEA association journals.

Anthropic finds persistent returns to experience (or expertise rather) for AI agentic coding. I’ve been saying that a lot in my talks the last few times. But of course if you use AI agentic coding to undertake tasks you do not remotely understand, then is that really complementing expertise? That seems to be substituting for a lack of expertise.

Twenty-five US states have passed age verification laws for looking at porn online. It has led to blocking of the major porn aggregators including pornhub and its various properties. (I wonder what the HHI is for online porn?). A new NBER study finds some evidence for deterrence — 10% reduction. Which given the base rate, I have to think that’s large. But I didn’t read the paper and I’m not sure if that is their interpretation. It’s just that you need the base as a comparison to know if estimated treatment effects are economically meaningful in absolute value. And confidence bands are maybe useful too.

Interesting graphic of market share in AI, presumably chatbot usage but I’m not sure because I wonder how copilots and agents count. I think it’s worth having another graph showing absolute volume not just market share though, as I would love to see those lines as well. Btw, I think you can see when Anthropic integrated Claude code into the native desktop app (November 2025) not just the command line interface (early 2025). It’s product not just tech that seems to be driving a lot of this.

El Niño is coming And/or is here already.

It’s a beautiful love story. But Boston residents have loved the Scottish visitors in town for the World Cup, and the feeling has been mutual. Michelle Wu, Bostons mayor, has proposed Boston and Glasgow to be sister cities even. Check out how much beer Scottish drank for instance — three times as much as st Patrick’s day! I love Boston with all my heart. All of it. Every last drop. I love every freezing cold temperature in the winter, I love every sports team, I love every inch of the city.

Will Taylor swift get married at her house in Rhode Island adjacent to the ocean house or at Madison square garden? Unclear.

World Cup visitors have discovered Waffle House and ranch dressing, and like the rest of us, fallen in love. You are a skeptic about dipping your pizza in ranch dressing until the moment you do it, and then you think what have I been missing all my life?

Trump loves McDonalds. You know what I love? I love Taco Bell. I love Taco Bell so much that I literally cannot even drive by one else I stop. So I have to take the long way just to avoid the Taco Bell by ex-wife’s house. My good friend once told me Taco Bell goes sued because the amount of cardboard particles inside the meat was too high. I said “man I love cardboard apparently.” Also maybe I dreamt that conversation. Cardboard? In meat?

Elon bought Cursor for $60 billion making four MIT undergrads in the early to mid 20s billionaires. One of the things I noticed at Harvard this last year was that very thing — students starting startups. It must be such a surreal feeling to know with near certainty that the probability someone in this classroom will invent some common place thing that makes them a billionaire and which will become a very common thing is probably 1.0.

Jersey Mikes has dethroned Chick-Fil-A for fastest service. I call fake news. That’s impossible. No one has faster service than Chick-Fil-A. It’s the scale of the logistics involved. I want to see Jersey Mike with a constant line to the street and yet somehow people are in line in around 5 min. Then I’ll believe it and even then I won’t.

Commodore makes a dumb phone. Technically all I want in a phone anymore is Duo, Google Maps, Spotify, and probably the ability to text no more than five people. No apps.

Remembering the 51-day siege that ended with nearly everyone dying in a massive fire. The Branch Davidian massacre. You know it actually didn’t happen in Waco. You have to drive about a half hour to get there. There’s still an active church there. And I’ve gone inside. They despise the government, and especially the Clintons. They flew a MAGA flag alongside some Christian flags when I last went, and weirdly enough, inside the church along the walls were lots of things about the Reformers Martin Luther, John Calvin, and John Knox. I think also Seventh Day Adventist tradition too, as David Koresh and the Branch Davidian was a sect off a schismatic family tree. It’s worth going because you can see the old bus and also places where the children were trapped. It’s strangely sad and minimalistic without any impressive monuments.

On the Borders closed in Waco texas. It mentions other closings too. You know what they don’t suggest as a possible cause though? GLP-1. I want to say I saw a paper about this. I don’t see how you can have a weight loss drug so popular, where the primary mechanism is appetite cessation, and not cause the restaurant industry to retract.

One of the really cool things last week was visiting DIW Berlin. I highly recommend visiting Berlin.

Inside the 24 hour whirlwind that led to the White House slapping Anthropic with export controls over Mythos and Fable. Honestly not sure I am all that upset about that at all though. Especially in light of the earlier stories about hacking success rates already rising. But it’s only a matter of time. Society will adapt, but not without some serious growing pains. Again, I want a commodore dumb phone.

Who is responsible for the first event study in diff-in-diff? Was it Bob Lalonde et al in this old AER article on displaced workers? Weirdly enough, the earlier draft didn’t have it but the AER did. I wish I knew the story,

My new book, Causal Inference: The Remix, pre-order link is up on Amazon. I have no idea if any have been pre-ordered because I messed up the author page years ago and can’t figure it out and don’t want to figure it out. The fewer number of things for me to obsess over, the better. When I get back, I’ll be working around the clock to get the free version online.

Tim Ferris’s, author of several best selling self help books, says AI is killing the self help industry.

Schema-Guided Reasoning is basically a research workflow for AI agentic research. I mean it’s also more general than that, too, but it’s for sure for us. I’ve been working on my own workflow for empirical research for two months now and I think the bones are there. When I’m done, I’ll share. But for now, read that.

Has AI ruined the job market? Who knows.

Higher order polynomials, staggered adoption, and moving beyond parallel trends.

International student enrollment plummets in many places creating insolvency issues for some schools. That, plus the enrollment cliff, and federal cuts is going to have a profound and unprecedented reshaping of high education.

Quera Jewlery in Spain. And here.

And with that, I say ciao ciao! Not a ton of links this week, but enough I got to close them. See everyone next week!