Please accept this as an admission of overcommittment, rather than laziness, but I posted something on bluesky and realized immediately afterwards that this can probably be easily tested.

If someone wants to take it upon themselves to regress wins over player-games lost to injury, I’d be most gracious. If they further wanted to interact that variable with total payroll expenditures (player payroll only, please), that would go further towards really testing the hypothesis. While I don’t tend to think there is much to be intuited from correlation coefficients, I would be curious to know how much the R-squared increases when you run a regression strictly over payroll and the lagged wins and then subsequently add player-games lost to injury to the independently variables. The delta on R-squared could be charted over time. There are other metrics that could be applied to try to control for overall talent, but real question is how accurately could you predict the final standings in a sports league if all you knew was player expenditures and injury luck, and if this has changed over time.

I’ll happily sit on a masters or undergraduate thesis committee for anyone who pursues this!

(Not sure there is enough meat on the bones for a PhD thesis, but happy to be proven wrong)

For those who missed the big story last week, it turns out that Blake Lively’s director and co-star, Justin Baldoni, feared that he was going to be publicly outed as an abuser and subsequently instructed his publicity team to start a preemptive disinformation campaign against her. The story is hot because the cache of subpoenaed text messages are the seeming definitition of “overwhelming evidence” and “receipts”, the victim is a prominent woman, and the activities in question are heinous. Which is all true, but I’m interested because 1) it seems to have really, honestly worked and 2) is was relatively cheap and easy, all to an extent that even surpised the alleged perpetrators.

We all know about Russian disinformation efforts at this point, but those are are the products of a government agency. They have enormous resources at their disposal. This internet campaign to pre-emptively attack and discredit a woman who is the (alleged) victim of gratuitous harassment was carried out by a small band of publicists, agents, and their team of assistants. This isn’t a billion dollar operation. This isn’t even a million dollar operation. This is a project carried out over cronuts and text messages by middle brow entertainment business aspirants looking to climb the ladder in between improving their scores at Orange Fitness.

What I’m saying is that disinformation scales faster and easier than I would have ever guessed and I don’t think I’m alone. A couple reddit threads, instagram and tik-tok posts, and tweets, all posted by accounts run and backpocketed by the publicity agency for precisely these purposes, and within hours the world has turned on a single human in a wave of disapprobation. A woman, you’ll recall, who had done absolutely nothing that would seemingly be able to give traction to public shaming.

This is a massive technology shift. If there is a final lesson to 2024, it’s we don’t know what’s real and what’s manufactured news. Worse still, those who would proclaim to be the least trusting are generally those that are the easiest to mislead, falling down endless rabbit holes of conspiracy theory and fabrication. Those conspiracy theories are fun to laugh at (and I suspect even more fun to believe with your whole heart), but I don’t think conspiracy theory falsehoods are the only plague going forward. It’s going to be joined by a growing trend of informational nihilism, an inabiilty to trust any news or information source.

It’s not that hard for me to imagine a swirling, vicious online discourse between left and right wingers, each fully enveloped in their cozy echo chambers of conspiracy and confirmation bias, while their more moderate (and numerous) peers simply drop out of the conversation entirely, unable to see the bits and bytes flying back and forth as anything more than unverifiable noise.

What happens to a democracy when the median voter believes in nothing?

Nothing says “Christmas cheer” like David Hume and empiricism. I am at EconLog this week with

Rediscovering David Hume’s Wisdom in the Age of AI

In our era of increasingly sophisticated artificial intelligence, what can an 18th-century Scottish philosopher teach us about its fundamental limitations? David Hume‘s analysis of how we acquire knowledge through experience, rather than through pure reason, offers an interesting parallel to how modern AI systems learn from data rather than explicit rules.

In his groundbreaking work A Treatise of Human Nature, Hume asserted that “All knowledge degenerates into probability.” …

Furthermore, I explain why this could have implications for the limits of AGI, if LLMs learn from experience and are limited in the number of datapoints they can observe. It is also a follow-up to my summer post: Is the Universe Legible to Intelligence?

Did you really think I was going to write a post this week? Sorry, this week is for far flung family and nutritionally disastrous cookies.

If you simply must have an economic observation, here you go: if you don’t gain weight during the holidays you’re probably too debt averse. Consume now, pay later. It’s worth the vig.

The reply culture on Bluesky is starting to get nasty. I know this is either ironic or churlish coming from someone who wanted more tension on Bluesky (I swear I just wanted people arguing about research and papers in a fruitful manner). Maybe I am in fact just getting what I asked for (oops). So what exactly can we do about said reaping of cursed sowing?

I don’t have any genious suggestions in the face of a very difficult exercise in tradeoffs. On the one hand we have the status quo of an open forum where we incur the cost of jerks and interlopers poisoning the conversation. On the other hand we could set up barriers to entry around the conversation, turning Bluesky into a very large #EconSky slack channel with hundreds (thousands?) of economists, policy professionals, and journalists engaging in a conversation. This sounds great at first blush, but the idea of finding new and innovative ways to make economics an even more insular club of insiders does not appeal to me. The costs go beyond that, though, because once you decide to wall something off, mechanisms have to be put in place to admit new members (and kick out misbehavers). Those mechanisms come with their own set of problems, including the costs borne by those who must see to the administering and oversight of those mechanisms.

So what’s the answer? I don’t have a silver bullet, but I am a big fan of trivial costs of entry that will only affect those attempting to enter “at scale” i.e. troll farms. Some sort of third party registration using .edu, .gov, and other profession email addresses. Maybe a google scholar or RePec connection. Basically, anything that will take 5 minutes for professionals to accomplish. Just enough that registering 100 accounts becomes costly for troll farms and repeatedly registering banned accounts becomes too much of a hassle for independent anonymous jerks. Such a thing could work for a professionally accredited jerks as well. If getting blocked by 3 people removes you from the register, then you have to go back and do the 5 minute registration over again. A tiny cost, sure, but I suspect a lot of jerks, after being removed 3 times, will simply take the hint or decide they can’t be bothered.

Yes, I know this means that the laws of ironic comeuppance will strike me down on Bluesky at some point, but if it protects the network from turning into Twitter I’ll take the hit.

Newly published research from Finlay et al takes the deepest dive yet on how the costs of the criminal justice system impact people’s lives going forward. Leveraging the new (and phenonomenal) integrated data from CJARS, the authors look at 9 (!) separate discontinuous increases in the fines and fees associated with misdemeanor and felony convictions. The paper is exceptionally well executed, connecting criminal and earnings records to estimate a pooled seemingly unrelated regression of those 9 separate treatments. They observe null effects on future convictions, earnings, and living conditions. So does this mean we can soak those convicted for every penny in their pockets without consequence? No, I don’t think so (and I strongly doubt the authors do either). Does it mean that people, such as myself, need to soften their calls to stem the growing tide of law enforcement as local regressive taxation scheme? Maybe in some cases, but I do think additional context matters here. A couple quick comments, in no particular order:

1. Of the 9 increased fine and fee treatments, 4 are small (≤$65), 5 are large (≥$200). Four of the large increases are explicitly raising the fines and fees of traffic offenses (DUIs). It’s not unfair to summarize the legislated treatments here as mild for those more likely to be in a state of poverty since you have to have access to a car to receive the larger treatments.
2. There’s always a little bit of a Rorschach test with RD designs, even when the differences are or are not statistically significant. In this case we observe null effects, but it sure seems like something happens with convictions in the first 100 days after reform and then it returns to trend (see below in Panel A). That feels like a system updated the de facto rules to accomodate the new de jure. As for earnings, it’s always tough when slopes change sign (Panels B and C), but the differences aren’t significant.
3. In the subgroup analysis there are two significant increases in recidivism, most notably a 4.7% increase in recidivism for those in the lowest predicted income quartile. This isn’t an enormous effect, but when it comes to what I consider to be a regressive tax, then focusing on the lowest income quartile isn’t an exercise in p-hacking, it’s to some degree the point of the endeavor. Combine that with the fact that the overweighting of the high magnitude treatments on driving offense can be expected to attenuate the potential effect on treated individuals with low incomes, a 4.7% increase in recidivism doesn’t seem that small anymore.

This is good research, but like most contributions it isn’t the last word. The growing use of fines and fees as revenue sources is a complex and, in many ways, adaptive system that exists to generate revenue for local governments whose revenue apparatuses hamstrung in countless ways, frequently struggling to keep the lights on (whether those lights should necessarily be on is a whole separate question). When they’re looking for that revenue, many (but not all) will arrive at the conclusion that they can only get so much much from poor people. Sometimes, such as with traffic offenses, the poorest individuals that do get caught in this system aren’t necessarily the intended targets, but rather the collateral damage. When we’re looking for that collateral damage, it’s important to know both where that damage is occuring and where it is being mitigated by local adaptation, particularly that which exists outside of the laws as written.