Hi Rosemary,

I agree that "never" is possibly interesting, but that already exists as a category. If you have a variable measuring how many cigarettes people smoke, you have a two-part variable, the first part is Never-NotNever, the second part, only relevant to the nevers, is how often they smoke.  But I wouldn't really consider this to be equivalent to a categorical variable, or a transformation to a categorical variable. The analysis of these is sometimes done in two-part (or hurdle) models, where first you predict if someone does smoke, and then if they do, you predict how much they smoke. But these are for count variables, which have weird distributions - it's got a big pile up at zero, and if you ignore them, it's kind of normal.

(Your other issue is very similar to James's, so I'll talk about that in a separate email).

Jeremy