| This is extremely low quality data, if you can even call it data. They used survey results from the Nurses Health Study (NHS), which are decades long studies, still running, where participants fill out a food frequency questionnaire once every few on how often they ate various foods. Not only do these questionnaires rely on memory, they rely on memory of a perception, because people don't actually measure how many cups of broccoli they eat each day. It's the worst quality of data. It's a memory of a guess, there is no measurement involved whatsoever, which is why I hesitate to even call it data. Not only is the data itself low quality, it is compounded by lifestyle factors that are impossible to fully correct for. As an example, people that eat red meat are more likely to smoke and have diabetes, more likely to eat fast food, more likely to exercise less, etc etc. Eating foods that are perceived as healthy is correlated with healthy lifestyles. Butter has been "unhealthy" for years, so it is almost certainly correlated with an unhealthy lifestyle. The nurses health study has led to many incorrect conclusions for these very reasons: - It suggested hormone replacement therapy reduced heart disease risk, but randomized trials later proved it increased it due to confounding factors like healthy user bias. - It linked vitamin E to lower heart disease risk, but trials found no benefit, showing the association came from health conscious behavior. - It linked beta carotene to reduced cancer risk, but trials revealed supplements could increase lung cancer risk in smokers, not prevent it. Many weak associations from the study, like diet and disease, were overstated as causal but didn’t hold up in trials due to confounding and noise. If you are interested in how often studies like this are wrong, see "Why Most Published Research Findings Are False" by Ioannidis (spoiler: 70%). |