Reply herewith.
Nick
[email protected]
Patrick Royston:
Curiously enough I had some correspondence some months ago about
-ivreg-. The latest version of -micombine- (version 1.1.1, 28nov2005)
distinguishes between supported and unsupported commands, and gives a
warning message in the case of an unsupported command being run. -ivreg-
is one of the latter. As Roy Wada implies, in such cases -micombine-
blindly applies Rubin's rules of combination of parameter estimates. The
user must take responsibility for the validity or otherwise of the
results. However in all such cases I have seen, the results look very
reasonable.
Such intuition could be supported by some simple simulation studies in
which a proportion of missing values are introduced at random into a
complete dataset, -ice- is applied followed by -micombine- with the
desired command, and the results are compared with the command applied
to the original, complete data. Of course we would not expect the SEs
of parameter estimates to be the same, but there should be essentially
no bias in the imputation-based estimates.
-micombine- v 1.1.1 is also available with the recent update to
-ice- in the Stata Journal.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
