The problem was that many of these population studies classified those who “quit drinking in response to ill-health” as nondrinkers. This is the problem of reverse causation: Instead of abstaining from alcohol consumption leading to poor health, poor health may have led to abstaining. It’s similar to studies showing that those who sit around and watch TV have worse health. Is watching more TV leading to illness, or is illness leading to more TV? As you can see at 1:24 in my video, this is one of the reasons why, if you look at the hierarchy of evidence, where stronger evidence is higher on the pyramid, interventional trials, such as randomized controlled trials, tend to offer better evidence than observational studies of populations, which can suffer from both reverse causation and confounding factors. As a group, light-to-moderate drinkers “display a range of healthy behaviours, such as better diet and more physical activity,” so, for example, they may be more likely to drink their glass of wine with a salad than a cheeseburger, and that’s why the wine appeared protective. It can be hard to do randomized controlled trials, though. For instance, you can’t randomize people to smoke a pack a day for a few decades, so you sometimes have to base your decisions on observational studies. We now have a new tool, however: Mendelian randomization.