Demographic Analysis with… Google Street View? And #DeepLearning!?!

The car you own is a good predictor of your income, education, race, and voting pattern. Oh, not you in particular, of course not! But analyze enough of you — say, across 200 cities in the US — and the predictions turn out to be eerily accurate.

Gebru et al. do exactly this in this excellent paper — https://goo.gl/HbnK34 — using cars extracted from Google Street View images, 50 Million of them across 3068 zips and 39,286 voting precincts.

The results themselves aren’t as important (°) as the way they derive it. Cars need to be classified by make, model, and trim level (can you tell the difference between a 2007 and 2008 Honda Accord? Based on a tweaked tail-light?) something that ends up involving 2,657 buckets — #DeepLearning really excels at this. Also, given that this is all just from Street View images, you could, plausibly, do the same with other factors such as house spacing, type, and, heck, the shrubbery 🤣

(And no, I have no idea who would own a Gucci AMC Sportabout)

(°) The results are interesting though — e.g. if the number of sedans in a city is higher than the number of pickup trucks, that city is likely to vote for a Democrat in the next presidential election (88% chance); if not, then the city is likely to vote for a Republican (82% chance).