Chernoff Faces are a statistical technique for finding similarities/differences between many variables measured across many observed units. In database lingo, it’s for data that has lots of fields and records.
The method exploits human facial recognition abilities. Similar observed units have similar faces. Dissimilar faces indicate differences.
Here’s some crime rates for the U.S.:
What makes D.C. stand out like that? Relatively more murders, robberies, assaults, and car thefts; relatively less rapes, burglaries, and larcenies.
I’ve lived in Utah, Louisiana, Alabama, and New York. Personally, I’d group Utah and New York as criminally similar, and Louisiana and Alabama as closer to each other than to the first pair. The Chernoff faces tend to confirm this.