Closing my tabs

Hope everyone is having a nice weekend. We’ll be together closing our tabs we’ve kept open this week. By “our”, I mean “my tabs”, the royal plural pronoun of course. It’ll be mixture of this and that spread across depressingly three devices including an Apple Vision Pro where it’s next to impossible to copy and paste but never say never.

Economics

A team that includes economists took a representative sample of around 50,000 folks involved in a 9-year RCT at Facebook. The study had been exposing some to advertising and others not randomly. They then put them through a separate set of controlled studies to see if they could extract evidence of the disutility of using Facebook and being exposed to advertising but found precisely estimated zeroes. NBER here.

Bannerjee won the Nobel prize with Kremer and Duflo for their work in development. Much of this relates to poverty and also the launch of the field experiments more generally in economics. Along with John List, they’re considered pioneers in that respect. But without diminishing that, my favorite paper by Bannerjee is a theory paper on “herding”. It was published in the 1992 QJE, and I had the pleasure of teaching it one time for a now shelved class on economic and social networks. It was absolutely exhilarating to learn from him on that concept. It’s a very generally valuable theoretical model, maybe even a major one, that illustrates the wisdom of the crowds can actually be entirely uninformative. I suspect social media and worries about being canceled only amplifies this phenomena too.

Speaking of herding, this paper by Mohammad Akbarpoura and Matthew O. Jackson has caught my attention and I want to dig into it more. Basically what makes ideas or viruses (I’ll call them both now “things”) spread through a network? The diffusion of some thing throughout a network is fascinating. Historically the focus has been on the structure of the network, but this papers analyzes the timing of the thing being deposited into the network too. They note that heterogeneity in when people (let’s say people) enter the network at all actually amplifies the diffusion. Theres two types — poisson types who regularly enter and sticky types who enter in a different pattern and simply having more of these latter types makes the spread of the thing diffuse more rapidly. Theyre apparently complementary to the Poisson types. This would apply to “bad thing” like sexually transmitted infections spreading through a network to neutral, good or bad ideas (truth versus misinformation).

Callaway and Sant’Anna, known affectionately by Stata users as simply csdid is now available in python and is also known as csdid. In R, it’s just called did and so ask yourself — which is the bolder naming? To name your estimator basically did or to have you now famous diff in diff estimator branded after your abbreviated name.

This isn’t economics but I’m putting it here. A sociologist named Robert Merton put forth a concept called “middle range theory” in the 1940s that is unbelievably helpful and I’ve never heard it until this week. Middle range theories seem to be somewhere above basic facts and below the all encompassing big ideas of science. I asked Cosmos to explain it in three sentences for us here.

Middle-range theory, as developed by sociologist Robert K. Merton, refers to theories that are not as broad and abstract as grand theories, nor as specific and narrow as empirical generalizations. These theories aim to explain specific social phenomena or behaviors within a limited context, serving as a bridge between empirical observations and broader theoretical frameworks. Merton's approach allows for the development of testable hypotheses that can be empirically validated, making it a practical tool for advancing sociological knowledge incrementally.

Examples from economics on the grand theories might be Adam Smith, Karl Marx, John Maynard Keynes, possibly even David Ricardo. But the fact Tyler Cowen doesn’t list Ricardo as a candidate for the greatest of all time has me doubted myself or Tyler’s ability to properly rank economists — but usually one of the two. Middle range theories might be Becker’s work on crime, Coases theory of the firm. Maybe even the collective fusion of quasi experimental and experimental design that we associate with the credibility revolution. So then one wonders where does my body of work fall, and I would say that my body of work on sex work covering the period I studied with Todd Kendall and the period I studied with Manisha Shah was an effort towards middle range theory, but probably most of my other work on abortion policy and drug policy would be best characterized as below middle range theory.

In fact I might say most empirical work today could even be thought of as below middle range theory if we say middle range theory is Coase and Becker. But if that is the case, let’s say, and the profession has shifted entirely or nearly entirely towards empiricism, one wonders what going forward the Nobel Prize will select. Even Card doesn’t win for a specific study — he seems to win for a middle range theory. That is I think Dave Card is winning the prize for the middle range theory we associate with the Princeton branch of causal inference, but more specifically in the realm of empirical labor economics. But now the credibility revolution has won, so I don’t think you can win a Nobel for another middle range theory that’s a body of work only consisting of empirical studies that are not themselves forming a middle range theory. It seems like the Nobel is about that though without saying it. So it does make you wonder — we are alive as the wave of the credibility revolution has crashed. It’s now settled more or less that causal inference is the standard as defined by the Rubin causal model and the practiced approaches we associate with the Princeton/Cambridge school. So then what? Anyway, I’m going to teach history this semester and probably have them read this article and we’ll use it to think about the classical economists. It seems like a helpful model.

This is a book called The Checklist Manifesto which if you haven’t seen, I recommend it. It fits with medical history non-fiction. I remembered it because this week I was working on another “Pedro’s diff in diff checklist” article.

Sean Taylor is chief scientist at Motif and I only know Sean in a weak ties sense — through social media, originally Twitter but now exclusively LinkedIn, and usually just by watching him post. Here he is discussing causal discovery products and experimentation at Motif. I really enjoy reading whatever Sean writes.

I forgot that Sylvia Nassar, author of A Beautiful Mind about John Nash, leaked Mankiws contract for his principles textbook back in 1995 for a NYT article. But she did. He got around $1.4m advance for that textbook, worth probably double that adjusted for inflation. He was 39. Mankiw also got tenure at Harvard and made full Professor a mere two years after getting there after graduating from MIT. He was only 29. Even if you drop the cites on Mankiws textbook, he still has over 100,000 cites. It’s incredible when you think about it how Mankiw is a household name for so many economists for reasons that aren’t even related to his scholarship and yet his scholarship has been profoundly impactful, arguably even Nobel worthy. He’ll be a guest this semester on season fours podcast. I booked him before Jon Hartley but Jon interviewed him first dang it.

Not economics, but I’ll put it here. This is an article about gender bias in scientific work. But it’s coauthored by people who don’t agree. So it’s called an “adversarial collaboration.” They had to write a collaborative document getting uniform agreement among a team who didn’t agree about gender bias. That’s wild and intriguing. I’d like to see more of this in economics, maybe in the JEL and/or JEP.

Frederico Masera is a professor at the University of New South Wales and he’s been knocking balls over the fence right and left since he graduated. You can see his online vita vita here under the Research tab. Apparently I was looking at his webpage this week.

Culture

Disney Plus cancels The Acolyte. “No soup for you!” they said. Man did it sting and on two levels. Level one: now I don’t see this story’s arc, and boy was I in love with season one. I thought it was the best Star Wars story I’d ever seen or read literally. Level two: I am apparently the only human alive who even liked it let alone enthusiastically loved it. Fans wouldn’t watch it. I am alone in the universe.

Speaking of alone, John Penniman teaches a religious course at Bucknell that sounds amazing to me. It’s on solitude and loneliness in religious traditions and the syllabus frankly looks like a genius wrote it and it’s a class I may just try to follow remotely by simply doing the entire assignment myself. It’s a master class also in making a good syllabus. I badly need to up my syllabus game and I won’t embarass myself by sharing my lame ones. You maybe remember John also because he wrote the best tweet I’ve ever seen in my life. “The late 1900s”. Omg. I may have to make this into a tee shirt.

But seriously, there’s maybe two characters in the Christian tradition who you might say live solitary lives, but I think in fact there’s really only one truly. You usually think it’s the monk living in the monastery, but in reality monks live in community and follow a common rule. They’re living in solitude somewhat but they’re doing it together which is kind of a contradiction. It’s the hermit, as a friend reminded me last night, and the hermitage that is the solitary one in religious traditions. But there are modern secular versions of the hermit and I think it’s an ambiguous character. Jon Krakauer’s book Into The Wild is about a young man named Christopher Mccandless who I think fits that description. He lived in the wilderness in Alaska and I won’t spoil the story but you should read it. Thoreau might be another example.

But there are examples of people for whom the solitude is associated with mental health in problems, but complex ones. For instance the Unabomber might also be considered a modern day hermit. And in Japan, there is a seemingly large phenomena of hermits but that we might call “recluse” or “shut ins”. I am watching a show on Apple TV called Sunny which is a mystery set in a near future Japan. Almost like near future as in next year maybe. The Japanese phenomena is called “hikkomorri” and Penniman’s syllabus covers that too. It’s not a strange phenomena because it involves young men who refuse to go out but it seems far more extreme and associated with almost unimaginable, oppressive levels of loneliness and self loathing. Pennimans syllabus also gets into social media and the internet and its relationship to loneliness and social isolation, so it sounds like it’s the full spectrum of that expression — the spiritual, psychological and the social.

Artificial Intelligence

Not much on AI this week.

OpenAI now offers you the ability to fine tune ChatGPT-4o. I have more confidence now having worked with my team using the different ChatGPT APIs that this is easier to navigate than I used to think, but not doing it yet.

And if you’re interested, me and two students are trying to evaluate ChatGPT-4’s ability to predict election results using only (a) its training data and (b) 100 articles from that day. But we are experimenting with how the results differ if you ask the election results to be read off in a story about the election results as told by four voices: 1) an “independent trustworthy reporter”, 2) Fox News, 3) MSNBC or 4) the BBC. We’ve been collecting the data on this for around a week and a half and will be reporting it here and on a shiny app, with the data itself stored at a repo once we’ve gotten further along. But the lines you see here are averages from 100 trials per day per distinct voice pulled using different APIs using two different LLMs for two separate tasks. It’s been quite eye opening. We will in the end have available the text of all 100 articles from every day’s pull as well as state by state measures of the average Harris/Trump victories. Insofar as the words pulled each day are independent with respect to potential outcomes, then I think reasonably one could say that the words are causing the deviations in trends from Fox News distinctive voice and the others. I’m still working on how we will harness the words in order to see as even if there’s a causal effect in there, there’s going to be so many words to categorize that it’ll suffer from the contamination bias that Kolesar, Goldsmith-Pinkham and Hull detail in their forthcoming AER on contamination bias with multiple treatments. But for now, the goal is to actually get as much of this automated and make the repo public, the shiny app operational, and figure out how to use data wrapper and store the results as an interactive map on here too.

Update on Psychedelics

I’ll put this behind the paywall because it’s not general interest enough. But it gets into causal inference as it relates to studying psychedelics, which is hard even — and maybe weirdly enough even moreso — using randomized controlled trials. A lot of this is me just explaining the facts as I understand them about the recent FDAs decision not to approve Lykos new drug application for MDMA assisted therapy. But I’m responding to an excellent new opinion piece in the NYT. It’s the most helpful thing I’ve read yet, and I try to respond (or do respond). If it’s the only thing you read, though, read that NYT opinion piece as it’s very fair, extremely helpful and not dull. But that’s it for the free readers. I’m going to now run to the ATM to get the tree guy some money.