I won’t be able to find a source for this, but I remember hearing a discussion on NPR about how hard it was to design studies to test the link between alcohol and cancer. If you rely on rates between people that drink and people that don’t by choice there are way too many other variables. You’re not going to be able to get a large population to act as a control and not drink even though they want to for 20-30 years, and on the other side of the study ideally you’d want them to drink similar amounts of alcohol.

They estimated up to 40% of cancer could be caused by alcohol consumption.