Read on Twitter
So, this paper is finally published. Here comes a longer-than-usual thread. (1/18) https://doi.org/10.1080/15614263.2021.1878894
In it, @DrJLozada and I revisit the data used for a study published in The Lancet which concluded that “police killings of unarmed black Americans have effects on the mental health among black American adults in the general population.” (2/18) https://doi.org/10.1016/S0140-6736(18)31130-9
The data used for that paper indicated there were 303 incidents involving unarmed Black people killed by police officers from 2013 to 2016. We believe, given the authors’ research question, 91 should have been removed or recoded prior to analysis. Or at least justified. (3/18)
Broadly these include:
-Drivers/passengers who fled police officers, crashed & died
-Incarcerated ppl who died in jail/prison custody
-People killed by off-duty cops (e.g., domestic murder-suicides)
-People who *were* armed or were attempting to gain control of a cop’s gun (4/18)
-Drivers/passengers who fled police officers, crashed & died
-Incarcerated ppl who died in jail/prison custody
-People killed by off-duty cops (e.g., domestic murder-suicides)
-People who *were* armed or were attempting to gain control of a cop’s gun (4/18)
As well as:
-People whose deaths weren’t attributed to officer actions
-People who had a toy/replica gun or object resembling a weapon
I concede that these last two are very debatable. But the authors never discussed the nuances of the data they analyzed. (5/18)
-People whose deaths weren’t attributed to officer actions
-People who had a toy/replica gun or object resembling a weapon
I concede that these last two are very debatable. But the authors never discussed the nuances of the data they analyzed. (5/18)
When we removed/recoded these 91 incidents, the effects decreased in magnitude and were no longer statistically significant. (6/18)
If this sounds familiar, it’s because we posted this as a preprint on SocArXiv in January 2020. We submitted to The Lancet on January 8th. They desk rejected us on January 24th, and a few hours later, the authors of the original paper had posted a reply to our preprint. (7/18)
In it, they show the results of 128 models, which is every way to slice up the incidents we disputed (2^7=128, where 2=the possible ways to code each incident [armed/unarmed] and 7=the different categories of incidents we disputed). (8/18) https://osf.io/preprints/socarxiv/h6y5w/
Point estimates were all in the hypothesized direction, but ~40% were non-significant. More important, this approach implies that all 128 combinations had similar construct validity. I don’t believe so, but interested readers can decide for themselves. (9/18)
We appealed The Lancet’s decision. They agreed to consider it, then proceeded to string us along for several months, ignoring our emails asking for updates. Finally, on August 4th, we emailed them and said we were withdrawing the paper and sending it elsewhere. (10/18)
In less than an hour, we got a response saying “for this manuscript this is a decision that we cannot reconsider.”
Unbelievable. (11/18)
Unbelievable. (11/18)
Anyway, the way we frame discussions about police killings absolutely matters. E.G., I wrote about the trouble with blanket use of the phrase “police violence” here. (12/18) https://doi.org/10.1371/journal.pone.0236158
The authors of the original paper warned against privileging official accounts of these events over community-generated information. That’s fair. But all the more reason to discuss the nuances of the data and be transparent about how everything is coded. (13/18)
For example, Sandra Bland died in jail three days after she was stopped, and her death was ruled a suicide. Her death should’ve been removed from the analysis. (14/18) https://www.nytimes.com/2019/05/08/us/sandra-bland-texas-death.html
And video shows Jason Harrison advance toward officers at close range with a screwdriver before he was shot. Was it a reasonable or necessary shooting? We could definitely debate that. But there’s no debating whether he was armed or unarmed. He was armed. (15/18)
According to @fatalencounters, police were involved in 1856 civilian deaths in 2018. BJS estimates that 28.9 million people were stopped by police at least once that year. If we accept this estimate, then a civilian death occurred in .006% of police interventions in 2018. (16/18)
Fatal Encounters doesn’t have an “armed/unarmed” field as far as I know, but WAPO’s data indicate 384 of 5,608 *fatal shootings* involved an unarmed civilian (6%). It’s simultaneously a rare phenomenon and one that occurs with sad and predictable regularity. (17/18)
Anyway, our only motive here was to better understand what the data say about this important issue. We list each of the 91 disputed incidents in the appendix. If you can’t access the article, I’ve got a non-paywalled version up on my website. (18/18) https://jnix.netlify.app/publication/37-ppr-bor-replication/
