Page 31 - Malcolm Gladwell - Talking to Strangers

them. The toys might, it is true, include models of airplanes dropping heavy bombs which exploded on defenseless towns or
villages; but, as he observed when I reproached him on the subject, it was not part of the Nazi conception of life to be
excessively civilized or to teach squeamishness to the young.” (In case you were wondering, that’s what Nazism was really
about: tough-minded child-rearing.)

3 The law has since been changed. A defendant must be eighteen years old or above to be sent to Rikers.
4 Two technical points about the dueling lists of 400,000 defendants: When Mullainathan says that the computer’s list
committed 25 percent fewer crimes than the judge’s list, he’s counting failure to appear for a trial date as a crime. Second, I’m
sure you are wondering how Mullainathan could calculate, with such certainty, who would or wouldn’t end up committing a
crime while out on pretrial release. It’s not because he has a crystal ball. It’s an estimate made on the basis of a highly
sophisticated statistical analysis. Here’s the short version. Judges in New York City take turns doing bail hearings. Defendants
are, essentially, randomly assigned to them for consideration. Judges in New York (as in all jurisdictions) vary dramatically in
how likely they are to release someone, or how prohibitively high they set bail. Some judges are very permissive. Others are
strict. So imagine that one set of strict judges sees 1,000 defendants and releases 25 percent of them. Another set of permissive
judges sees 1,000 defendants, who are in every way equivalent to the other 1,000, and releases 75 percent of them. By
comparing the crime rates of the released defendants in each group, you can get a sense of how many harmless people the strict
judges jailed, and how many dangerous people the permissive judges set free. That estimate, in turn, can be applied to the
machine’s predictions. When it passes judgment on its own 1,000 defendants, how much better is it than the strict judges on the
one hand, and the permissive judges on the other? This sounds highly complicated, and it is. But it’s a well-established
methodology. For a more complete explanation, I encourage you to read Mullainathan’s paper.