Thread by @JohnFPfaff

This is probably one of the best discussions of crime stats and analysis I've read in a long time, maybe ever--academic or popular.

• Confusing and contradictory data
• All sorts of spurious correlations
• Critical spillover effects to measure
• It's... so hard to parse

https://twitter.com/NewYorker/status/1215305611883270146

We really don't talk abt how these data issues matter. Change from the UCR to NCVS, and entire results just... disappear.

But papers never engage in this debate. Just trot out the UCR, add a footnote abt its weaknesses, and then... act like it's just fine. (I've done it too.)

I mean, we talk all the time about how the "crime boom" from the 1960s to the 1980s helped fuel mass incarceration and the politics of punishment.

But did crime boom? Or did it collapse?

In the 1960s and 1970s, the UCR and NCVS seem to be looking at two different countries.

(The best explanation I've heard for the UCR/NCVS came from a Twitter egg years ago. The NCVS is a national survey, the UCR the aggregate of local police reports. Perhaps, the egg said, the NCVS picked up national trends, while the UCR was detecting sharp-but-v-local increases.)

(In other words, perhaps things OVERALL didn't get worse in the US, but where crime WAS getting worse things got A LOT worse. Given that the NCVS is a survey, it likely has harder time get responses from those areas, which would contribute to the divergence.)

Anyway, our crime stats are a mess. They are always released late, and when they are released it can be hard to know what they are saying, and hard to use.

Like, really hard. It's hard.

So thank you to @NewYorker for a fantastic piece on this.