> For example, if you had a study which compared moderate drinkers to non-drinkers, but not carve out "cannot drink due to other medication" from the second group, you get a more discriminatory result from the study.
This is in fact what seems to have been underneath all those studies 10-25 years ago that seemed to show that non-drinkers have worse health outcomes than moderate drinkers. The problem is that "non-drinkers" includes a lot of people who can't drink because they're already sick or because they're alcoholics and drug addicts with long histories of abuse (and so on).
Once you control for that group non-drinkers do better than moderate drinkers.
Oh really? That sounds super plausible. Do you have a good source for that? I'd love to dig a bit deeper on this as the whole "drinking once a day is good" thing is something that comes up a lot.
From what I understand, it's hard to know exactly how non-drinkers are different (besides not drinking), but the nail in the coffin for the moderate drinking-endorsing studies which tried to correct for all the factors they could think of, was the rise of Mendelian randomization studies.
Some people have gene variants that interfere with alcohol digestion, effectively making hangovers come much faster and harder. Because these people drink less than other people for a well-understood reason, and are otherwise like everyone else (the gene variants are well mixed into the populations where they occur), it can be used as a causal wedge to pry apart the health contribution from alcohol from the contribution of basically anything else.
And when they did this, the supposed health benefits of alcohol, which were already smaller the more reasonable things you controlled for, vanished entirely.
You can look up meta studies of Mendelian randomization studies on alcohol use on Cochrane and so on of if you want to dig deeper into it.
There has been some pushback, trying to argue that if you don't assume the effect of alcohol to be linear (I.e, don't assume a normal dose-response relationship as you would by default on pretty much anything else), then there's too little data to tell. I don't buy that, as you can probably guess.
This is in fact what seems to have been underneath all those studies 10-25 years ago that seemed to show that non-drinkers have worse health outcomes than moderate drinkers. The problem is that "non-drinkers" includes a lot of people who can't drink because they're already sick or because they're alcoholics and drug addicts with long histories of abuse (and so on).
Once you control for that group non-drinkers do better than moderate drinkers.