Weekend open-to-closed tabs: from Glasgow to Madrid
Two weeks down and six weeks to go. Last week I did three days of teaching of six hour days on panel data at the university of Zurich. And this week I did five days of six hours of lecturing also on panel data in Glasgow. I have two days before I start back up in Madrid at CodeChella — the third annual iteration of it. But this morning, right now anyway, I remain in the unbeatable Glasgow.
I think this is fourth or fifth time to come to Scotland. Twice to Stirling for a summer workshop for grad students from all over the UK, and twice to Glasgow. Glasgow is one of my favorite cities in the UK. It’s restful, yet lively too. Green and urban. It has all the amenities of urban life, which at bare minimum is a large variety of coffeeshops. A lot of sidewalks. Buildings. Old and new.
My name is scott Cunningham which is a pretty in-your-face Scottish name. Through my dad’s side there appears to be the Scottish line. My great great grandfather was named William Wallace Cunningham. But that said, I know very little of my ancestry. I’ve been a bit homeless ever since I moved from Mississippi the summer before the 8th grade, and have never since felt much connection to anyone or anything other than my closest friends, my immediate family, stories, songs and the supernatural.
I have felt a very strong connection to Boston though some of that is from making new friends. Finding new life. Doing things for the first time in a long time which I had put off doing. When I meet people from New England who are dyed in the wool lovers of New England, I don’t feel jealousy so much as I am deeply happy for them. That they get to have that bondedness with the land, with their communities — I am happy for them. And I have met more people like that up in Boston than I was expecting. Of course I meet people like that, too, all over the south, especially in texas.
But now it’s time to leave Scotland. I have to pack up my room, and then head to the airport. My flight isn’t until 5, but I like airports and besides, I don’t want to ask strangers again if I can break in line so I don’t miss my plane. I plan to get there early, wait in line, and instead, let everyone else break in front of me. Paying it forward.
In the meantime, here’s a few links open here and there. I do t have a ton of things because when I wasn’t teaching this week, I was basically passing out from fatigue. But here is what I have. Thanks again for all your support.
Tyler Cowen says the big disruption from AI isn’t the mass unemployment, which to be honest won’t happen, but rather the way that our interaction with reality will change as well as the navigating of our inner and outer lives. The labor market disruptions are to be determined, and predictions are extremely model dependent, and there are no shortages of models, but the disruptions with inner and outer life — that’s here now and for me it has been profound.
A new documentary about the John Carpenter classic “The Thing” is out. I didn’t know it was at its release a commercial and critical failure. It’s actually now considered a masterpiece — hence the documentary about its making.
On that note though — The new Mandalorian and baby Yoda movie is somewhere between a colossal failure and masterpiece.
An actress from game of thrones joined a wellness cult and has written about it. If there’s one cult I am 100% certain could never sink their hooks into me, it is a cult about being healthy.
People from Prince Harry’s old circle continue the time honored practice of throwing their old friend and kin under the bus in exchange for 15 min on the podcast circuit.
If you book at either of these cottages near the crown jewel of American luxury hotels, the beautiful Ocean House on the coast of Rhode Island, next door to Taylor swifts house, you get the fourth night free. These are not cheap but on the other hand, they’re also sleeping like 10+ people so maybe split it.
This new synthetic control survey article by Robert Pickett, Jennifer Hill and Sarah Cowan looks really interesting. I like survey articles with an opinion and a strong point of view that also make new contributions beyond being merely an assembly of practices. The one thing that struck me was the claim that in fact omitting covariates introduces biases in their extensive simulations. They also caution against using pre treatment RMSPE as the guide for model selection. I started it last night and will leave it open for now, but check it out.
I met someone in Boston who once said to me “legalized cannabis, legalized sports betting and video games have harmed a generation of young men”. It started a long conversation. Ever since that conversation, I feel like I see story after story supporting parts of the sports betting part.
First instance, the more extreme cases tend to make the Google news algorithm but check out this story from NYT The Athletic about the texas tech QB who has a lifelong major gambling addiction going back to high school.
But even beyond anecdotes, there’s some evidence that it has had harmful effects on the financial security of Americans. This NBER working papers authors found that sports betting led to more bets, more credit card debt, and concentrated on the poorest families.
And a similar finding was reported by Federal Reserve economists as well.
The economist writes about Al Roth’s new book on morality and economics, extending a line of his historic research that some of us associate a bit with his “Repugnance as a Constraint on Markets” piece. My copy is waiting for me on my doorstep back home in Boston.
I’m starting down my own path of building an AI agentic research process sometimes called a harness. In the process, I’m reading what others are doing there too. I’m sure in the next six months, we will see more and more about harnesses. I think harnesses are the least shareable, compared to skills. They’re more like philosophies of organization. But in that sense, they’ll also be quite intuitive to many researchers who have well thought out production processes. Here is one harness like page I found called RISE — Research Information Systems Engineering.
Jon Roth’s paper on how the various diff-in-diff estimators yield somewhat perplexing pre trend coefficients is published. My book coming out this summer is of the unusually strong opinion (for me) to always do long difference calculations with a fixed baseline for pre treatment coefficients. Why? A few reasons.
One, OLS specifications do that and it has squatter rights since estimating event studies with OLS was what people have been doing for decades. When you drop a year to estimate event study leads and lags with 2 groups and T time periods, that specification is a series of 2x2s for each pre and post treatment using the omitted year as the baseline. So it’s literally not possible to do the “short difference” with OLS — not in a single spec anyway.
Second, event studies are falsifications of the parallel trends hypothesis, and parallel trends is a post treatment counterfactual conjecture which states the long difference in expected Y(0) outcomes for the treated group is approximately that of the comparison group. That claim uses the t-1 as the baseline for all of those parallel trend claims. Well, a reasonable falsification of that claim is to do a series of long differences 2x2s on pre period using the same t-1 period as the baseline.
Third, even the slightest imbalance in pre trends will look flat in one but not the other. The only time they are identical is if you have absolutely zero coefficients — on the zero line. So we need guidance and the arguments are stronger for long differences of the same form as the parallel trend form itself which always uses t-1 for identification.
Okay enough of that. More arguments for having Claude and other agents make html exhibits and artifacts, not markdowns. I’m going to be going in that direction too because the harness workflow I’m building is creating a research dashboard and it’s all html.
Not sure if this link and the above link about the effectiveness of html is the same or not, but here’s the one from Anthropic making that same claim.
Nature had an article about semi autonomous agent based scientific discovery that I think did get into a harness they made wrapping the LLM but I haven’t read it yet, and honestly, I think I just need to focus and not keep reading more. But yes here’s more about that.
Tanya Wilson is a talented microeconomist at the University of Glasgow.
Buried in this piece about Bruce Dern at Cannes talking about Tarantino and Brad Pitt having a tense exchange on the set of Once Upon a Time in Hollywood is a reference to a new thing based on Pitts character coming to Netflix written by Tarantino, directed Fincher and starring Pitt. It’ll be out this November apparently.
I’ve been surprised every time I’ve read that younger folks dislike AI, but it may be real or at minimum a nontrivial vibe in the economy, and it may be related to labor market anxiety.
Lots of stuff going around about fertility and smart phones. Here and here.
This NBER working paper is pretty fascinating. Falling college graduation among men has created an imbalance in the marriage markets segmented by education. College educated women sort into matching with non-college educated higher earning men. This result was actually found in an old restat coauthored by Kerwin Charles, but focusing more on black women and incarceration. But the general pattern was the same. Well, this NBER paper I started out discussing said that the effects do not stop with merely the cross-education matching. Because college educated women are then marrying the higher earning noncollege educated men, the non-college educated women are facing a much larger marriage squeeze.
You can do so much with two sided gale-Shapley-Roth-Sotomayor marriage market models to be honest.
They’re re-releasing the old Transformers animated movie from the 80s. It was very polarizing and shocking originally as many know. I saw it in the theater as a little kid in Mississippi and was genuinely stunned. I didn’t like it at all. I wouldn’t say I was traumatized so much as it ruined transformers for me, which was my main toy and comic book at the time. Anyway, it’s doing some apology tour cash grab thing.
Edith Whartons letters to her former boyfriend and clandestine secret lover. We named our youngest Edith after Edith Wharton, but when I sent her this, she didn’t care! Kids.
Speaking of harnesses and AI for research, here’s an NBER paper about that but focused on making datasets.
Jewlery in San Sebastián. I particularly love these colorful fun looking pearl earrings.
More AI agent based harassing for research purposes.
And with that, I bid adieu. I have to pack this messy room. I’m going to leave you with some bad bunny though. Listen to it with me! Have fun everyone and be safe. Look both ways before you cross the street and remember today may be our last day so make it count.