Our analyses comprise models that include two different sets of control variables. The standard set includes standardized age, standardized age squared standardized height, handedness (right/left/ambidextrous; dummy-coded), sex (female:0, male:1), current smoker status, former light smoker, former heavy smoker, and standardized Townsend index of social deprivation measured at the zip code level62. To control for genetic population structure, the models also include the first 40 genetic principal components63 and county of residence (dummy-coded)62. A second set of extended control variables includes all standard control variables and in addition standardized BMI, standardized educational attainment64, and standardized weight. To determine whether observations at the extreme ends of the drinking distribution bias the estimates of the relationship between alcohol intake and IDPs, we also estimate a model that excludes abstainers and a model that excludes heavy drinkers (i.e., women who reported consuming more than 18 units/week and men who consumed more than 24 units/week), both with standard controls. For each IDP, we test the null hypothesis that alcohol did not affect the outcome measure via an F-test that compares our model against a model with only the control variables (excluding alcohol intake and related interaction terms).
