Given the enthusiasm (perhaps out of social conformity pressure) many EAs had for the BLM movement last summer, I think we might perhaps consider this a cautionary tale:
From 2014 to 2019, Campbell tracked more than 1,600 BLM protests across the country, largely in bigger cities, with nearly 350,000 protesters. His main finding is a 15 to 20 percent reduction in lethal use of force by police officers — roughly 300 fewer police homicides — in census places that saw BLM protests.
Campbell’s research also indicates that these protests correlate with a 10 percent increase in murders in the areas that saw BLM protests. That means from 2014 to 2019, there were somewhere between 1,000 and 6,000 more homicides than would have been expected if places with protests were on the same trend as places that did not have protests. Campbell’s research does not include the effects of last summer’s historic wave of protests because researchers do not yet have all the relevant data.
The two theories he suggests are:
BLM protests make people distrust the police, so they are less likely to dial 911, and less likely to cooperate with investigations.
BLM protests change the cost-benefit for police, because they face additional risk of vilification/prosecution for intervening, so they spend more time protecting the doughnut shop and less time in dangerous neighborhoods.
I worry that attitudes towards BLM by both the OP and the commenters strongly correlate with reactions to this research, and that the ensuing discussion is going to be highly politicized. It might perhaps be more fruitful if parties on both sides instead try to make a series of explicit and verifiable predictions related to this research, such as whether it will be published in a peer reviewed journal without significant alterations, or whether a similar analysis on the effects of last summer’s historic wave of protests will yield similar results.
I agree it is highly politicized. I feel like this is an asymmetric demand for rigor however. I do not recall many people anyone making this objection last summer when everyone, including EA organizations, was super-keen on BLM.
It’s important to highlight this point from the article:
Just a word of caution before we jump to conclusions.
Also, I think the main findings from the research were extremely interesting in their own right; That the BLM protests were successful in achieving some of their intended aims. The effectiveness of protests are rarely assessed quantitatively so glad to see someone doing this work. It would be interesting (and probably extremely challenging) to do a cost-effectiveness estimate for BLM, considering you have to account for the counterfactual value of 350,000 people etc.
Did they check the sizes of the effects over time since the protest? If BLM achieves their aims and gets the reforms they want, killings by police might remain reduced (but perhaps less so), while policing would become less politicized and scrutinized, so become more effective again.
EDIT: Towards the end of the Vox article:
I think you should modify the title to include “in the short term”, since the increase in non-police killings was temporary. It seems pretty plausible that killings are reduced overall in the longer term.
I disagree—the point estimate for the increase in murders is well above 0 for the entire time period [figure 6]. The effect possibly fades away a little over time, so the confidence interval extends slightly over zero, but that doesn’t mean you can assume it is zero! If you did a statistical test for ‘did the increase in killings reduce over time’ you would not get a significant result.
Red is for murder and blue is for property crime, right?
Even if it’s still above at year 4 , that’s as far as the analysis warrants a conclusion for, and 4 years is still the short-term. You could specify “in the 4 years following the first protest” (although does this figure include police killings?). Your title reads to me as saying the rate will settle higher than without the protests or at least the drop in police killings will never make up for the increase in other murders, neither of which follow, and looking at that figure, neither seems more likely than not. How I read the title, it does’t even seem more likely to be true than false.
I thought there had also been some cautionary tales learned in the last year about widely publicisng and discussing headline conclusions from preprint data without appropriate caveats. Apparently not.