If I want to use multiple imputation, it seems like this is the first step:

Code:
mi set mlong
mi register imputed y1 x1 x2 x3
mi impute chained (regress) y1 x2 x3 = x1
And then if I want to run a series of models progressively adding covariates, do I need to re-impute for only the covariates I want in each respective regression? Or can I impute all covariates, and then run my regressions (below) all using that same imputed dataset? This is perhaps a conceptual question as much as a Stata-specific one.

Code:
mi estimate: regress y1 x1
mi estimate: regress y1 x1 x2
mi estimate: regress y1 x1 x2 x3