Informal sampling of the Fatal Force Database

See a random example

Police shot dead 1,160 people in 2023. What’s a typical shooting like?

We learn about fatal police shootings through two routes. First, dramatic examples get in national news. However, these are clearly very atypical, or they wouldn’t be in the news. Second, we can compile statistics. The Washington Post’s Fatal Force Database aims to be a comprehensive database of every police-involved killing in the United States since its inception in 2015, and includes a variety of information on whether the suspect was armed, their race, and so forth. But when what you really want to know is how often the police should have acted differently, how do you gather statistics on that?

If you visit my Fatal Force Database informal sampler webapp, you’ll see a randomly selected row from the Post’s database. A few details such as name and location are shown on the page; to learn more about the incident, press “Search Google” for a Google search tailored to find news reports about the killing. A second button, “Sample again”, selects a fresh random row from the database.

I’m not sure if there’s an established name for this technique, but I call it “informal sampling”: to get an informal picture of what a population is like, choose a random member of that population, look into that member by whatever means, and then repeat for another random sample. This seems to me to retain many of the advantages of anecdote in getting a gut feel for a population, without being thrown by rare vivid examples in the way that non-statistical methods often are. I have used this technique in other ways too, such as by trying to work out where all my money goes by sampling on where a randomly chosen penny was spent. One of my main motivations in writing this webapp and this blog post was to promote the idea of informal sampling as a technique; I have very often wished for a tool like this in other domains.

Although I encourage you to try this out and get your own picture of these things, I must warn you that it is all very sad.

See a random example