The latest research also does a lot of data shenanigans to avoid getting a statistical null for lower quantities. For instance, the famous lancet study that sta

The latest research also does a lot of data shenanigans to avoid getting a statistical null for lower quantities. For instance, the famous lancet study that started this all corrected for the fact that some of the non-drinkers were former drinkers who quit drinking due to health reasons (increasing percieved risk), but didn't correct for the fact that heavy drinkers often lie to medical professionals about their drinking (which would decrease perceived risk). They also counted harms from drunk driving and synergistic harms from alcohol and tobacco use as harms from alcohol alone when both are 100% avoidable even if you drink. And even after doing this consistently they still had p values very close to 0.05 for 1 and 2 drinks per day on average.