I've seen people say 2020 vindicates Peter Turchin. I disagree. Here's his 2010 article that people cite. He doesn't tell you what the DV is, what kind of "violence" we'll see. The riots have killed a few dozen people, no noticeable uptick in murder. 1/n https://www.nature.com/articles/463608a

This, in academia, is called p-hacking. If I want to test whether a treatment makes you more conservative, I can ask you your feelings towards GOP, Trump approval, ten other things. By chance one of them will show a statistically significant effect. 2/n

Turchin's theory tells you "instability" or "violence" will increase. But we're usually not told ahead of time what that means, his theory is good at finding cycles after the fact. Most straightforward definition of violence is how many people are killed, like @sapinker uses 3/n

But that can't be what Turchin means, because total murders and violent deaths are down. Here's a 2012 article where he lists mass shootings as a kind of violence. We haven't seen much of them this year, even before the coronavirus. Again, p-hacking. 4/n

One might've predicted that 2020 would see racial unrest for a different reason. Black Lives Matter took off in the media around 2016, was ignored for four years and then did so again 2020. There's a cyclical nature to this, but it's the American election calendar. 5/n

Another dimension to the p-hacking is you can always shift a year or two around. If you say "my theory says cycles every 50 years," few people will quibble if it appears to be vindicated in 45 or 55 years. 6/n

So in sum, you have maybe a 10 year window, by Turchin's own count, and a lot of definitions of violence, and if any one of those trends appears bad at any point you can claim vindication. You ignore any measures of "violence" that don't get worse. 7/n

Even if you ignore measures of violence like murder, terrorism, and mass shootings, and only look at riots, this still isn't an usually violent time. During the Rodney King riots in 1992, 63 people died, more than in the recent unrest. 8/n https://www.latimes.com/entertainment/arts/miranda/la-et-cam-la-riots-jeff-beall-los-angeles-uprising-20170427-htmlstory.html

So here's another opportunity to engage in p-hacking. I've seen Turchin count number of riots, and he shows they've increased. Why not total deaths(again, most clear definition of "violence"!),or monetary damage? We're only told the specifics of the prediction after the event 9/n

I already mentioned 1992 riots were deadlier than 2020. New analysis shows economic damages are comparable. $1.4 billion in 1992, $1-$2 billion in 2020. Maybe you can adjust the cycles and use the early 1990s, when violent crime and riots both peaked? 10/n https://www.axios.com/riots-cost-property-damage-276c9bcc-a455-4067-b06a-66f9db4cea9c.html