We could use some tools of social control to use on police
You may have noticed I’ve not been writing much recently. That’s because I turned in the latest draft of my book, and then I promptly took a short vacation from writing. In fact I ensconced myself in a ridiculous crochet project:
which is supposed to be a physical manifestation of this picture proof:
which I discussed a few months ago.
Anyhoo, I’ve gotten to thinking about the theme of my book, which is, more or less, how black box algorithms have become tools of social control. I have a bunch of examples in my book, but two of the biggies are the Value-Added Model, which is used against teachers, and predictive policing models, which are used by the police against civilians (usually, you guessed it, young men of color).
That makes me think – what’s missing here? Why haven’t we built, for example, models which assess police?
If you looked for it, the closes you’d come might be the CompStat data-driven policing models that measure a cop by how many arrests and tickets he’s made. Basically the genesis of the quota system.
But of course that’s only one side of it, and the less interesting one; how about how many people the policeman has shot or injured? As far as I know, that data isn’t analyzed, if it’s even formally collected.
That’s not to say I want a terrible, unaccountable model that unfairly judges police like the one we have for teachers. But I do think our country has got its priorities backwards when we put so much focus and money towards getting rid of the worst teachers but we do very little towards getting rid of the worst cops.
The example I have in mind is, of course, the police that shot 12-year-old Tamir Rice and didn’t get indicted. The prosecutor was quoted as saying, “We don’t second-guess police officers.” I maintain that we should do exactly that. We should collect and analyze data around police actions as long as children are getting killed.
mathbabe 
The value of evaluation depends on the values that are evaluated. If we think the value of education lies in grades and scores rather than the lives of people then our attempt to measure added values has failed from the get-go.
LikeLike
While not addressing use of force, Connecticut is doing an excellent job of collecting and analyzing data about traffic stops.
http://www.courant.com/news/connecticut/hc-state-report-finds-significant-racial-disparities-for-police-traffic-stops-in-20150407-story.html
LikeLike
Quibble: You said, “Basically the genesis of the quota system.” but I suspect quotas existed long before 1995 (https://en.wikipedia.org/wiki/CompStat#Origins). Certainly the urban legends of traffic tickets just for quotas probably go back to late Roman times. The idea behind ComStat was to try to apply some rationality at a system wide level to where resources should be applied instead of allowing every precinct commander make independent decisions based on fiefdom. (The leak isn’t in my end of the boat.)
In theory and practice, the system is effective but, and you probably aren’t aware (snark) but sometimes, members of bureaucracies try to game the system.
LikeLike
Yes sorry. Should have specified modern nyc quota system.
LikeLike
Here’s an example of crowd sourced collection of LEO interactions. Might have (sarcasm)a little bit(/sarcasm) of bias.. “How three teenagers invented an app to police the cops: The high-school students want citizens to rate their interactions with officers” http://www.economist.com/news/united-states/21684687-high-school-students-want-citizens-rate-their-interactions-officers-how-three
LikeLike
The Guardian newspaper in the UK has been collecting data on US police shootings. The statistics are not official, of course, but the fact that a newspaper has done it and publishes updates highlights the absence.
LikeLike
We know that police are capable of realistic threat assessment in critical situations, since they do that all the time when the factors of bias and stereotype do not impair their faculties.
One of the first statistical jobs I ever had as grad student was a study of how personality and cultural stereotypes affected the diagnoses of medical practitioners when presented with identical sets of symptoms in different people. That was a whole paradigm back then and I’m sure there’s been a ton of analogous research in all walks of life.
Some communities of practice have been pretty good at applying research in that vein and others to improve training and practice, others apparently not so much. Maybe someone should study that.
LikeLike
Agree, to a point. One of the challenges this approach would face is purely political: One of the two political parties in the U.S. is eroding science by word and deed, and thus, scientific-based decision approaches are affected. The critical bit is that with every iteration in approach to ‘solve’ a social issue, scientists know there will, and indeed, should, be more, rapid, iteration on the approach. Political processes are not rapid. We would likely be saddled with ‘scientific’ approaches that are known to not be optimal for long periods of time, further giving fuel to the meme of needing to reduce gov’t because it’s incompetent/a poor use of funds. The need for rapid, continued evaluation and change, likely through the input of added data sources, and improvements in analysis methods, should be factored into the approach.
LikeLike
I hesitate to bring it up, but 538 has had some excellent articles about how poorly numbers on police killings are collected.
In the height of ironies, most of these agencies are breaking the law; there’s been a federal law requiring data collection for a decade or two, but most states flat-out ignore it, with no repercussions.
Police killings really can’t be prosecuted by the same D.A. that they work with daily, that’s obvious. (Even if the grand jury being fed crappy info often gets the blame.)
LikeLike
“New York Police Commissioner and Predecessor Spar Over Accuracy of Crime Data”
LikeLike
>>The prosecutor was quoted as saying, “We don’t second-guess police officers.” I maintain that we should do exactly that.
Well, then you are challenging the monopoly of force that the modern state has achieved. You need to consider the alternative.
LikeLiked by 1 person
There are challenges with data with sparse “adverse events”, but it’s worth noting that the NHS in the UK publish the mortality outcomes of individual consultants:
https://www.nhs.uk/service-search/performance/Consultants
The key communication tool used from a statistical perspective is the funnel plot (see e.g. pages 15 and 38 here
https://www.rcseng.ac.uk/surgeons/research/surgical-research/docs/national-vascular-registry-report-on-surgical-outcomes)
LikeLike
“We should collect and analyze data around police actions as long as children are getting killed.” : http://mappingpoliceviolence.org/
LikeLike