Scott's Mixtape · Economics & Policy
TIER 4 Wed, 27 May 2026 04:54:23 +0000
Greetings from beautiful Madrid! ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ | | ---|---|--- | | | Forwarded this email? Subscribe here for more --- # Claude code 53: Journal Cartels and AI Disclosure | | scott cunningham --- | May 27 --- | --- --- | | | --- | | --- | | --- | | --- | | READ IN APP --- Greetings from beautiful Madrid! What an amazing city. The lively night life, the evening walks on the sidewalks from the restaurant to my hotel, nothing like it. | | ---|---|--- I went to the Mercado de San Miguel last night for dinner at Botin, the world's oldest restaurant. That part of the city is so interesting and lively. I will explore it more after CodeChella concludes. | | ---|---|--- | | ---|---|--- | | ---|---|--- Here is a video I took on the way home. Today is day three of CodeChella. We've spoken so far about the 2x2 (day one, morning), covariates (yesterday, day two) and some hidden curriculum stuff. During the hidden curriculum, Dan Rees gave a provocative talk about the journals' policies on AI use, and I have been thinking about it since as we as a group had a lively discussion, both during the talk, and also when we broke out to meals and hang out time. I asked Caitlin Myers if maybe we could talk about the AI policies of various journals on the air of our next podcast, as I also just wanted to use it to openly parse through word by word precisely what is and is not being said. But until then, I wanted to think out loud a bit myself here. I don't know if you recall this article I shared a month or so ago written by a woman at the Atlantic who had been banned from Hinge, but it was really interesting from an economic point of view. And I promise this is relevant to journals' Ai policies. But this woman noted that she, and a few others she'd interviewed, had been banned from hinge for violating terms of service. She had been unable to get to the bottom of just what she had done, and couldn't find anyone to talk to either. The thing that struck me, though, was the market concentration. One company -- Match -- owns I believe 75 of the online dating platforms. They do not own Bumble or Grindr, but otherwise it's fairly monopolized. I calculated an HHI of something like 4500, far above the 2500 mark that usually gets the dept of justice involved. What this means is that when you get banned on Hinge, you are effectively banned on more or less the entire marriage market itself given that online dating has become since around 2017 the modal way that heterosexual couples, at least, meet. So think about it -- banned on Hinge, banned on every dating app, banned therefore from online dating, banned from nearly the dating market pretty much _en Toto_. That is quite the punishment. Match is a publicly traded company. It is not the government and it does not owe anyone due process, or less putative punishments or anything for that matter. It answers to shareholders and legal regulatory bodies which have apparently largely left them alone despite their successful monopolizing of the marriage and dating market. Why do I bring this up? What does online dating cartels have to do with the AI policies at Econ journals you might ask? Well, in the sense that the journals are largely controlled by just a few publishers. It's nontrivial calculation to really drill down to exactly how concentrated the market for scientific manuscripts is because it differs a lot by field. From what I could find, in 2022 the top 5 publishers controlled 61% of the web of science material, the top 10 controlled 75% and the top 20 83%. The ordering I found for 2022 was: Elsevier, Springer Nature, MDPI, Wiley (incl. Hindawi), Taylor & Francis, Frontiers, Sage, IEEE, Oxford University Press, American Chemical Society. Well, there apparently exists a massive number of journals by the American chemical society, which I am nearly certain is not economics journals. And this matters because the HHI could be some number as a whole, and that not really capture what it is effectively within fields. The journals that one can publish in for the field of chemistry and biology are not relevant to an economist. So to figure out concentration for economists, you have to narrow in a bit more. I found An, Williams, & Xiao (2026) which appears to be the most authoritative recent IO-economics paper on this. I gave Claude that paper and asked him to try and come up with an implied HHI and make a beautiful figure. And this what he found. It appears that Elsevier owns almost half of the relevant economics journals followed by Wiley. So together, they own around 61%. Claude used a threshold of 1800, not the 2500 I noted above, and deduced that the market was 'highly concentrated' with an HHR of 2,430. This concentration by Elsevier may be somewhat more recent as their paper showed article-share trajectories from 2009-2018 and they found that Elsevier's share jumped from ~30% to ~45-48% in 2011 when _Value in Health_ moved from Wiley to Elsevier. Claude Code used the post-2011 (current state) shares for the HHI. _Value in Health_ is not something I have heard of before, but they apparently publish things like cost-utility analyses of cancer drugs, QALY measurements, and other pharmaeconomics topics. It's apparently ranked 5th in health policy. If you drop it, then you're dropping a pharmaeconomics journal, but using the mechanical criteria that the authors used, Claude backed out around 2,430. If you drop them, you get closer to 1,700. | | ---|---|--- Why do I say all this? Because Elsevier and Wiley have AI policies, and if you were to violate them, you would be violated from 61% of all economics journals. You wouldn't be banned from the association journals, mind you, such as the AEA journals (e.g., AER, AEJ) or the university journals like the University of Chicago journals (e.g., JPE, JOLE), all of which now dominate the top 10 journals in economics. But those are prestige outlets, and hardly the bread and butter of the modal economist's career. These AI policies, therefore, function as the constraint on the demand side of scientific manuscripts. And whatever you think of it, the fact remains that if you got banned for violating AI policy in one of them, it could cripple your publication options. And not just you -- it would cripple your publication options of your coauthors, and thus handicap you even more. I haven't read the AI policies closely yet but I will. They do both require disclosure of the use of AI, and I want to understand just precisely what that will mean, because if it means that the use of AI for coding is prohibited, then I will have some hard decisions to make. But if it is simply requiring that the use of AI in research be disclosed in the methods section, which was how Dan made it sound, then I will be doing that. In the backdrop of my reasoning are a few things. The first is a principle about gossip. If you would not say something when the person is present, then you should not say it when they are not present. Likewise, if I believe that it is appropriate to use AI in research, then I need to be open about it and simply move forward in life embracing that. That does not require insisting that others share my views, but it does require living with integrity and being fine with others judging me and disagreeing with me. So, once I have had a chance to carefully read the AI policies of the major journals, I will likely be articulating what my position will be on here thus making this more or less my preregistration of my own opinion. But some of this will need to be worked out with coauthors. Because if someone is using AI to do research on a team, then they owe it to the team to also reveal that use because that can create problems for others if the whole team, and not just one part of the team, is treated as a unit. If that is the case, then I must disclose to them. What you decide to do is in a sense no one's business than your own. But this is the larger point I have been thinking about. It is their journals. So no one caught violating the AI policy of the journal really has any recourse legally because the policies are clearly spelled out. No one forces any of us to send our papers to a particular journal, but if we do, then we are implicitly accepting the journal's policies and terms and conditions. And if we mislead, then we must be prepared to be punished if discovered. And so what will the punishment be? Who knows. Others guesses are as good as mine. But I think we live in a particularly hostile moment in history. Social media has convinced me that a large swath of people, including academics, delight in the suffering of others. So my hunch is that given the difficulty of catching anyone lying about the use of AI, then Beckerian reasoning about punishment emerges as a real possibility. Becker noted in his 1968 article, _Crime and Punishment: An Economic Approach_ , that when the probability of detection is very low, the optimal fine grows. And Gary King, in my interview with him, said something similar about scientific communities: they give huge rewards for unique discoveries and they give gigantic punishments for breaking the rules of the community, including permanent exile. So, that's really it. I think if we talk solely about the supply side of AI, focusing on the productivity gains, automatous research, and so forth, we may become inadvertently short sighted as the market is determined by supply and demand, and the journals are the demand side, and they are concentrated by a few firms. Two firms in particular own 50-60% of the market depending on how you cut it. This is just something I think that is important to begin to immediately thinking long and hard about because probably like many of you, I have only been using Claude Code for research around 6 months. I started using it in mid-November, and have not slowed down since. But have I studied the AI policies of the journals? No I haven't. Do I need to? Yes, I very much do. Do I need to think carefully through what I believe? Yes I do. Should I tell the truth about my use of AI in my work? I absolutely should. Will I? I will. Scott's Mixtape Substack is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber. Upgrade to paid You're currently a free subscriber to Scott's Mixtape Substack. For the full experience, upgrade your subscription. Upgrade to paid --- | | | Like --- | | Comment --- | | Restack --- (C) 2026 scott cunningham 910 North 17th Street, Waco, Texas 76707 Unsubscribe