st: ivreg2 handling of VCEs

[Christopher Baum <kit.baum@bc.edu>]

<>
Correspondence with a user of -ivreg2- (Baum, Schaffer, Stillman, on SSC) concerned the loss of observations from his estimation sample when a -robust- or -cluster(id)- VCE was specified. This turns out to have been caused by the user's dataset being -tsset- by an id and time variable, but the time variable having hundreds of missing observations. In this situation, -ivreg2- considers those observations to be unusable. A -tsset clear- caused -ivreg2- to behave like -regress-. The user was not trying to make use, at this point, of the panel nature of the data.

It makes little sense to have observations in a panel context that are missing one of their identifiers (or at least to imagine that they should enter the computations). That is the assumption that -ivreg2- makes, and I tend to think of it not as a bug, but as a feature.

Kit

Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/