f you’ve managed to make it this far in life without quite figuring out what demographic boxes to tick on the Census form, a recent working paper published by the National Bureau of Economic Research might have some clues for you. According to the Washington Post, University of Chicago economists Marianne Bertrand and Emir Kamenica “taught machines to guess a person’s income, political ideology, race, education, and gender based on either their media habits, their consumer behavior, their social and political beliefs, and even how they spent their time.” The results are, at times, surprisingly granular and seemingly nonsensical: Owning a pet and a flashlight is a strong indicator of whiteness, as is a proclivity towards Vlasic pickles and Stove Top stuffing mix. Others are disappointing, like the fact that the attitude most indicative of whiteness is approval of police striking citizens.

Bertrand and Kamenica used an algorithm to filter through data from three decades-long national surveys with sample sizes between 669 and 22,033 respondents. According to the Washington Post article, the economists “omitted variables that would have been a dead giveaway” to keep it fair. (So when predicting whether a respondent was liberal or conservative, “they wouldn’t allow the algorithms to consider the answer to ‘Which political party do you support?’”) In general, the consumer behavior associated with, say, income or education or gender make sense. In 2016, the strongest indicators of high income were travel within the continental United States and owning a passport, while the best predictor of liberal views was not owning a fishing rod. Not buying women’s hair product was the easiest way to predict whether someone’s male and strangely enough, having seen Gone Girl or watched Love It or List It in 2016 was a pretty reliable way to guess whether someone was more educated.