I am puzzled by what -suest- is doing when the routine generating the
estimates is -ivreg-. Even when ivreg is invoked to estimate an ols
regression (with no instruments), use of the 'pscore' option (which
merely generates OLS residuals in this case) causes the suest results
to wildly differ from those of suest after regress--standard errors off
by a power of 10 or so. Please see the do-file results. Whether I
generate resids with predict or ivreg pscore I am using the same
series, but suest after ivreg seems to be doing something very
different than OLS with robust standard errors (which I believe it
should be reproducing, or nearly so, in this case).
Thanks
Kit
---cut here--
webuse auto,clear
regress mpg weight
predict double olseps, res
est store ols
suest ols
regress mpg weight,robust
ivreg2 mpg weight, psc(ivsc)
predict double iveps, res
est store iv
est change iv, scorevars(ivsc)
suest iv
