Two in three Americans who called themselves vegetarian had eaten meat, chicken or fish when asked what they ate the day before

Studies of vegetarian diets usually start with a simple question: are you a vegetarian? The answer sorts people into two groups, and everything that follows, from cholesterol to bone density to mood, is compared between them. A paper published in the International Journal of Epidemiology on 6 October 2026, by researchers at Charité in Berlin, the German Federal Institute for Risk Assessment and Boston University, asks how much that first answer can be trusted, and what happens to the results when it can't.

They used the US National Health and Nutrition Examination Survey (NHANES) for 2007 to 2010, which asked adults whether they considered themselves vegetarian and, separately, took two 24-hour dietary recalls in which people listed everything they had eaten and drunk the day before. The study counted a day as vegetarian if it had no meat, poultry or seafood. Of 11,180 adults, 2.2% said they were vegetarian. More than two in three of them (68%) reported meat, poultry or fish on at least one recall day, and 52% on both. Going the other way, 1.6% of self-declared non-vegetarians reported none in their recalls, which is not surprising over a day or two. Measured against the recalls, the vegetarian question had a sensitivity of 32% and a specificity of 98%.

The team then looked at four outcomes linked to vegetarian diets in earlier research: low bone density, high total cholesterol, depression and poor bowel health. They ran each analysis with the self-report, with the recalls, with both combined, and with three statistical corrections for misclassification (a method called probabilistic bias analysis, repeated 100,000 times per scenario). The answers moved around. For depression, self-reported vegetarians had lower odds than everyone else (odds ratio 0.76), but defining vegetarians by the recalls gave higher odds (1.14), and the corrected estimates ranged from 1.09 to 1.49. For low bone density, one correction more than doubled the apparent effect, from 1.11 to 2.16. The mistakes weren't random either: how well the label matched the recalls differed between people with and without each condition, which statisticians call differential misclassification and which can push results in either direction.

None of this says whether a vegetarian diet is good or bad for any of these outcomes, and the authors say so: that wasn't the question. Most of the estimates, before and after correction, had wide intervals that included no effect at all. Two recall days are not a gold standard; someone who eats fish once a week may simply call themselves vegetarian, and the data are more than 15 years old. The point is narrower and useful: 'vegetarian' means different things to different people, and when a study relies on the label alone, its findings may say more about who chooses the label than about what is on the plate. When you next read that vegetarians have more or less of something, it's worth checking how the study decided who counted.