Pubdate: Fri, 25 May 2018
Source: Chicago Tribune (IL)
Copyright: 2018 Chicago Tribune Company
Author: Cathy O'Neil


Legalizing marijuana makes sense for a lot of reasons, but there's one
valuable thing we'll lose when police stop arresting people for
smoking pot: A sense of just how misleading our crime data are.

Data on arrests and reported crime play a big role in public policy
and law enforcement. Politicians employ them to gauge their success in
making neighborhoods and the entire country safe. Police departments
use them to determine where to deploy more officers to look for more
crime. They are fed into recidivism-risk algorithms, which help judges
and parole boards make decisions on sentencing and release.

This is troubling, because such data can be a terrible proxy for
actual crime. Consider arrests: Only two thirds of murders result in
arrests, which means that the homicide data are missing at least a
third of actual incidents. And murders are unusual in that we
typically have the body, so we know a crime actually occurred. That's
not the case with assaults, rapes, thefts or illegal gun possession.
There's no reason to think that the majority of these crimes lead to
arrests, or that all arrests are related to actual crimes.

Reports aren't much better. People decide whether to report in a
cultural context. For example, they're more likely to do so if they
trust the police, and the level of trust can vary sharply over time. A
year after Donald Trump was elected president, the number of reported
rapes among the Latino population of Houston declined by 40 percent, a
strong indication that people became afraid to report the crimes.
Police often don't take rape victims' reports seriously, a problem
that is probably even worse for male victims.

So how can we get a better understanding of the underlying rate of
crime? Surveys typically don't help: People who get away with
committing serious offenses aren't likely to admit it, even if they're
guaranteed anonymity. The one notable exception is marijuana use,
which -- though still illegal in most places -- is mild and socially
acceptable enough that people are willing to tell the truth. Hence, if
we compare the reported rate of marijuana use to the arrest data, we
can gain some insight into how useful the latter really are.

The picture isn't pretty. The latest government surveys, for example,
suggest that black and white Americans use marijuana at about the same
rate. Yet blacks get arrested about four times more often than whites
- -- and 15 times more often in Manhattan, according to a recent New
York Times analysis. This means that all the policies, policing
strategies and algorithms that use the arrest data will unfairly
target blacks for closer monitoring and harsher sentencing,
perpetuating the ills they are supposed to address.

Given the ample evidence of extreme bias in marijuana arrests, there's
no reason to think that the situation is any better in other areas of
crime. Indeed, from a statistical standpoint, we should assume there
is a bias in all categories. We just can't know how extreme it is,
because we're missing data, and we don't know how much data we're missing.

Worse, when marijuana is legalized, we'll lose our only indicator of
how off-base the available data really are. Overall, decriminalization
is probably a good idea, considering how much devastation the policing
has caused in black and brown communities. But it will eliminate our
most reliable barometer of police racial bias. When the arrests stop,
we'll stop seeing the disparity, but that doesn't mean bias in other
police practices will suddenly end.

What to do? It won't be easy, but there are some ways we can work
toward improving crime data in the longer term. If police strive to
increase public trust and improve follow through, people will be more
likely to report crimes, making the data more reliable. This would
also require more consistent policing across different neighborhoods,
which, for example, means treating black youths the same way one
treats middle-aged white women. The Center for Policing Equity, for
example, is helping police departments identify and challenge their
own biased practices.

In the meantime, it's crucial to recognize how bad our crime data are,
lest we perpetuate the biases they reflect.

- -----------------------------------------------------------------

Cathy O'Neil is a mathematician who has worked as a professor,
hedge-fund analyst and data scientist.
- ---
MAP posted-by: Matt