Oops! The danger of email packages proposing your addressee while you
type a few characters...
In a recent discussion of "how to use suest after ivreg / ivreg2" I
proposed that suest could be tricked by giving it a score variable
(which in the case of regression, OLS or IV, is just the residual
vector properly computed). That appears not to work -- not even for OLS
regressions estimated by the ivreg command, so I would not follow my
previous advice. suest has special code to handle 'regress', as the
help file indicates, but when ivreg is invoked, that special code is
not available. Strangely, suest does not complain or emit any sort of
error; it just produces some very strange results, as the do-file in my
previous message will show.