The P-hacking frontier of rationality

Baltimore police have apparently stopped noticing crime.

There’s something called “drive-and-wave” policing, which is when the police cruise around just as much as ever, since there are surely GPS trackers in their cars.  But they make the decision not to get out of their cars to investigate this or that potentially suspicious thing.  They use their discretion on that, and they do it in a way that produces positive statistics.  They’re provably more present in high-crime areas, which is what their bosses need to see and their bosses need to grandstand about.  But it doesn’t produce civil-rights or police-brutality prosecutions of themselves.  That it produces fewer arrests and less protection of the population is a side effect which surely they must casually regret.

This practice is called p-hacking, in science.  Basically, it refers to manipulating the parameters of the experiment, or more broadly, the professional situation, in order to maximize the evidence for the desired outcome.

“Scientists aren’t supposed to desire a particular outcome!” I hear you cry.

Of course they aren’t.  Steering science, even when you can’t see it, makes it cease to be science.  And police are supposed to desire their job to be done properly, requiring a fair amount of investigating potentially suspicious behavior in person.  But when you set up the incentives badly, p-hacking occurs.  Scientists want headlines, which findings of null hypotheses and re-testing others’ results don’t bring, and they want tenure, and they want grants, and that means they aren’t permitted to find things that could be construed as racist or sexist.  The police want not to be hit with a discrimination or police-brutality lawsuit by people who have learned that politicians, with still other incentives, have established that as the fastest way to victory.

So both scientist and cop use professional discretion– in this case, to stop at a certain point in the process– to avoid bad results for themselves.  It’s akin to the agency problem in business, in which CEOs and investment advisors make decisions for others, like churning their accounts, that are defensible on the surface but which are actually taken to serve the purposes of the agent (the advisor or executive)– maximizing commissions, for example, or producing “accomplishments” to point to.

More broadly, then, this is a structural problem with rationality.  When you mix rationality with politics, yes, politics does become more rational.

But rationality, in turn, becomes more political.