test alcohol consumption rates and suicide and homicide

[kayfisher]

No description provided.

[ellisp]

For suicide the problem is we've already burned up all our degrees of freedom (4 max, with 35 observations), so if we do anything more we're into p-hacking territory. Homicide is probably ok though.

We can mitigate by using a lasso / elastic net regularization.