a-random-american.github.io• Jul 4, 2026• 1 min read
Show HN: See a Random AmericanA little project for America's 250th anniversaryThis is kind of the worst possible side project, because it sounds very unimpressive but it's actually pretty tricky. You would think, as I did, that because the American populace is one of the most heavily surveyed, most reliably censused people on earth, that generating demographically and statistically accurate random Americans would be fairly easy. For one or two variables, that's true. We have extremely good data on, for instance, the number of men versus the number of women in the country. But combining variables gets much harder because of the way probabilities intersect. For example, ~30% of people are Democrats and ~10% of people are from California, but substantially more than 30% * 10% of people are California Democrats because P(Democrat | Californian) is higher than P(Democrat). So getting in age, gender, state, race, and political affiliation all based on good-quality data without making independence assumptions was a headac
