Levels of alcohol consumption tend to be similar for individuals living in the same household. This may be because: (a) individuals with similar characteristics collect in households (correlated effects); (b) individuals in the same household are influenced by common factors (exogenous effects); and/or (c) the consumption levels of an individual directly influences the consumption levels of other individuals in the same household (endogenous effects). Whichever of these three possibilities is the principal reason underlying household clustering of consumption levels has important policy implications. In this paper we propose a testing strategy to distinguish between the three types of effect in a cross-sectional data-set. Allowing for exogenous or endogenous effects shows that the significant socio-economic gradient in a model containing only individual variables arises because of misspecification. However, because we find significant evidence of correlated effects, we cannot identify whether it is endogenous or exogenous effects which give rise to statistically significant group level variables. The results indicate the possible pitfalls of omitting group level influences.