Alcohol consumption is near an ideal case to study for Mendelian randomization, since there are fairly common, well-distributed gene variants which make alcohol

Alcohol consumption is near an ideal case to study for Mendelian randomization, since there are fairly common, well-distributed gene variants which make alcohol consumption a lot less enjoyable. Since this a known reason for drinking less, and the people with the variants are otherwise similar to everyone else, we can disentangle it from all the other things which affects how much you drink - not just things we can guess, like health issues and dietary customs, but things we haven't even thought of. So there's really no reason to try to use studies like this to draw casual inferences about if (or "for who", nice garden of forking paths you have there) some low level of alcohol consumption may still be healthy. We have fundamentally better tools. And those tools have, in later years, essentially sunk the old idea of a j-shaped alcohol harms curve. There is no benefit at low levels.