Friday, March 16, 2018

Falsus in uno, falsus in omnibus - journalistic innumeracy edition

Arrrghh!!! The innumeracy of the press drives me crazy. From False Confessions, Mistaken Witnesses, Corrupt Investigators: Why 139 Innocent People Went to Jail by Niraj Chokshi. Despite our inclination to free ten guilty in order to avoid falsely convicting one innocent, we still manage to convict a lot of innocent people. Nobody has yet found a way to drive all Type One and Type Two errors to zero.

Seeing the headline, I wanted to know what the distribution among the three causal components might be. What percentage of false convictions are due to coerced confessions, to corrupt investigators, or due to simple process error (such as mistaken witnesses.) Those three causal sources have different remediations so it is worthwhile to know what the causal distribution of false convictions might be.

But does the article provide answers to the question posed in the headline? No. In fact, it is an analytic mess.

Here is the relevant breakdown in the article for the cause of the false convictions of the 139 innocent people:
Exonerations caused by official misconduct: 84

Well over half of the people exonerated last year were initially convicted because of official misconduct, such as officers threatening witnesses, analysts falsifying tests or officials withholding evidence that would have cleared the defendant.

No-crime exonerations: 66

In just under half of the exonerations last year, defendants were wrongfully convicted in cases in which no crime was committed. This included more than a dozen drug possession cases, 11 child sex abuse cases and nine murder cases.
OK. Issue number one. They indicated three causal sources - coerced confessions, incompetence/error, and professional misconduct. Now they are lumping everything into only two categories. One, professional misconduct, carries over but the second category is new; no-crime exonerations. I understand the concept but the journalist is creating a category error. We are supposed to be identifying the causes of false convictions. Those are the categories we are interested in. The category "non-crime exoneration" is descriptive of the crime for which they were convicted. Other crime types would, with non-crime convictions, include violent crime convictions, property crime convictions, victimless crimes, etc. Those categories tell you something about the nature of the crime, not the nature of the cause of the false conviction.

If the person was convicted of a non-crime, you still have the question: Why were they convicted? And the causal categories for false convictions remains the same for non-crime convictions as they are for false convictions for real crimes - coercion, incompetence and misconduct.

Chokshi has categories which don't match what is in the headline and he has heterogeneous categories, one of which does not address the root cause of false convictions.

Issue number two. Chokshi claims there were 139 exonerations but his analytical numerator is 150 (84+66) rather than the 139 in the denominator. Because "professional misconduct" and "non-crime exonerations" are heterogenous categories, it is not possible to answer the question we are interested in.

If this is a representative example of the clear analytical thinking of the New York Times, we would have to conclude that they have innumerate reporters, glib headline writers, and no editors. Which might be the case.

But if we can't trust them to get simple analysis like this correct, then on what basis would be believe that they are able to report on the economy, climate change, poverty, the Mueller investigation, etc. A great example of Gell-Mann Amnesia effect.

UPDATE: Beyond numeracy there is the issue of basic beat knowledge. Do you know enough about what you are reporting in order to write intelligently. Even this low bar is hard to clear sometimes as reported in Scott Kelly’s medical monitoring has spawned some horrific press coverage by John Timmer. His subheading is "Analysis: Don't believe the headlines. And in many cases, the articles below them." I saw a few tweets and headlines on this topic but they seemed so nonsensical that I did not bother to read them. Fortunately, Timmer did and is reporting back that indeed there was plenty of cognitive pollution.

No comments:

Post a Comment