Thread by @JohnFPfaff

1. The 2018 Nat’l Crime Victimization Survey is out. And here’s the main take-away.

Violent victimization is up (except for robbery), but with only one exception the increases are NOT statistically significant.

The one increase is for rape, tho, and it’s big: it almost doubled.

2. The NCVS measures crime in various ways: total incidents, total victims (the prevalence) and just “serious” violence.

For all three measures, we see roughly similar patterns: robbery down, other non-rape violence up but weakly, rape up by about 66% to 80%.

3. Now the NCVS is just that—a survey—so there’s sampling error. That’s why the report acknowledges that the increases outside of rape are not statistically significant.*

But here are the standard errors. The rape increase is statistically significant.

* For those unfamiliar with “statistical significance.” It’s a… tricky term. But the basic idea is: if you were to pick ten ppl out of the same pool of 100 ppl several times, the average weight per sample would differ each time, just by chance.

** So when we see a diff btw samples, have to ask: is this a fluke from sampling, or is there something more “real” going on here?

There’s a lot of math and problematic assumptions, but “statistically significant” means “we think it’s not just chance.”

Ok, back to the NCVS:

4. The NCVS has a footnote on for “rape/sexual assault” saying to look at the methodological appendix, but nothing there suggests a change in sampling between 2017 and 2018.

Now the UCR changed how it defined rape a few yrs back, but this doesn’t appear to be a product of that.

5. So it’s something else.

The usual response to big shifts in rapes in the police-based UCR is to say it reflects a change in willingness to report.

But the NCVS is designed to avoid this: it is based on interviews, not police reports.

Did THAT willingness to talk change?

6. A near-doubling in actual rape victimization seems like something that would have produced more media attention.

But I just looked at reporting rates, and those seem to have cratered in 2018, so maybe it didn’t show up for the police, so it didn’t make the news?

7. On the one hand, an increase of this magnitude automatically raises some red flags. Crime isn’t often something given to huge one-year jumps, especially with levels as high as they are.

On the other hand, don’t want to dismiss this too quickly. If real, it’s serious.

8. The full report is here: bjs.gov/content/pub/pd…

And correct me if I’m wrong, but doesn’t the UCR usually come out before the NCVS? Sort of surprised that this broke first.

9. One last technical note. The NCVS did substantially increase the sample size in 2016, in an effort to create more-local estimates. This makes some pre-post comparisons tricky.

But even then: that shouldn’t explain the 2017-18 jump.

10. To make sure this idea sticks with this thread, there could be a reporting issue, one linked to #metoo, if that leads to women being more willing to discuss sexual violence in any situation (which surely seems plausible).

https://twitter.com/TamasoJ/status/1171433111302705154

11. Hm. About those reporting rates. It looks like 2017 is the outlier, not 2018. Here are previous years’ reporting rates. None gets to 40%, most in mid-to-low 30s. So 2018 is low, but not exceptionally so (and higher than 2016); 2017 seems really high.

12. I thought to check those prior years bc of this: nytimes.com/2019/01/06/nyr…

Now, dates here don’t work well for the #metoo hypo. BdB points to #metoo for 2018, and #metoo didn’t start until late 2017—and 2017 has the big spike.

But still: for reporting, 2017 is the outlier.

13. Ok, one more NCVS thing, abt the politics of punishment.

Many reform efforts have framed decarceration as “look, as prison pops have shrunk, so has crime, so it’s a good thing to do!”

This is… risky. It suggests—wrongly!—that if crime goes up, maybe… prisons should too.

14. It’s a framing that opens up the door for those who favor tough-on-crime to point to any rise in crime and say “well, you had your shot, but crime is up. Time to be tough again.”

Which is bad policy for a host of reasons. But this is a real risk:

15. This isn’t a hypothetical fear. This is exactly what happened in Alaska.

AK passed a big reform bill, crime went up mostly due to a recession and opioids, and the reforms—even those that had not yet gone into effect!—were rolled back.

governing.com/topics/public-…

So, the NCVS:

16. Look at how the BJS chose to frame the past few years of NCVS data. It’s objectively true, sure… but so is saying that there has been a small increase while things stay at historic lows.

It’s framed, though, as a reversal.

17. And maybe it is a reversal. Both the NCVS and the UCR have gone up a bit over the past few years, even while overall rates and levels stay low.

Maybe it’s a trend. Maybe we’re bouncing around a temporary or permanent status quo.

But the framing here is not irrelevant.

18. This framing—emphasizing the small short-term increase instead of the bigger long-run decline—is exactly the framing that undermines a “look, crime keeps falling with prison populations” argument.