It's generally hard to tell whether long-term problems are getting better or worse from looking at news coverage or hearing about problems. Each gives you some information, but it's hard to tell how it compares to what was happening before. Quantitative data can help give some perspective, however, and I decided to look some into US hate crime statistics.

Since 1992 local police have reported crime information to the FBI under the Uniform Crime Reports program, and then the FBI collates them into an annual report. There's some risk of underreporting, or changing reporting levels over time, but this seemed really hard to look into and I decided to just using the numbers as presented in the reports. I couldn't find an annual table, so I went through the 27 published reports and pulled out the number of incidents from each report's Table 1 (raw data).

There were ~30% more people in the US in 2018, though, so we should look at per-capita numbers instead:

The report breaks down crimes by the motivating bias. The five most common, anti-Black (34%), anti-Jewish (13%), anti-White (12%), anti-Gay (10%), and anti-Hispanic (6%) account for 75% of the total incidents:

The spike in 2001 is coded as "Anti-Other Ethnicity/National Origin" but it's probably anti-Arab with 9/11.

Instead of normalizing by the total population, however, for specific biases it would be clearer to normalize by the population of the affected group. For example, how many anti-Hispanic hate crimes are there compared to the number of Hispanic Americans? Here's what that looks like for the same five most common categories:

This is very sensitive to the size of the target group, however. I've taken the number of Black and Hispanic Americans from the Census (pretty good), the number of Jewish Americans from the Statistical Abstract of the US (ok), and the number of Gay Americans from the General Social Survey (not so good, and other sources can give pretty different numbers).

Statistics are fundamentally a summary, and they can't be our only window to the world. But for questions like, "is this better or worse than it was ten years ago" they're a very powerful tool.

Comment via: facebook

New Comment
6 comments, sorted by Click to highlight new comments since:

Does this control for any changes in reporting? In the UK we have seen a recent spike, but this is because police are now required to check for any hate crime elements when dealing with someone it could apply to, rather than people volunteering the information.

Does this control for any changes in reporting?

No, this is the stats directly from the FBI reports

In the UK we have seen a recent spike, but this is because police are now required to check for any hate crime elements

I think this hasn't changed in the US over this period, but I don't know. If there has been a substantial change it would mean comparing across years wouldn't be very meaningful.

I think it's possible that the initial years in the US saw some increase because getting the system of passing information from the police to the federal government took some time?

I was surprised when I noticed that the categories weren't necessarily mutually exclusive.

They're categorized by bias, not by the identity of the victim. Consider someone harassing Black Hispanic gay man with anti-Black vs anti-Hispanic vs anti-gay vs anti-male slurs.

Yeah, but without a map of the overlap, it's not clear that "Anti-X sentiment" in uncorrelated with "Anti-Y sentiment". At first it seems that if they're tracked separately, a combined score can tell you whether overall things are going up or down. But unless an unlabeled score is available, then if there's overlap, by adding everything together we might be overcounting. (Does the number of "hate crimes" increase or remain the same, if all 4 types of slurs (one of which isn't in the graphs in the post) are employed in a single case?)

Does the number of "hate crimes" increase or remain the same, if all 4 types of slurs are employed in a single case?

Yes. If only one type of slur were used it would be categorized as a "Single-Bias Incident" and included in the breakdowns, while if more than one were used it would be a "Multiple-Bias Incident" and not included in the breakdowns. So for the total number of incidents I'm using the Multiple-Bias count, while for the breakdowns I'm using the Single-Bias ones. The number of incidents categorized as Multiple-Bias has gone up over the years, from initially ~0 to about ~1%, but it's still pretty small compared to the total.

one of which isn't in the graphs in the post

Anti-male isn't in the graph because I'm only graphing the five most common. Looking at the my cleaned data I see 22 anti-male incidents, compared to 1,943 anti-Black, 726 anti-gay, and 485 anti-Hispanic ones.