Pfeffermann (1993) considers Hausman (1978) test to check whether the
survey design (in particular, weights) is ignorable. The
null-efficient estimator is then a simple OLS (or logistic, or
whatever) estimator, while the consistent under both the null
(non-informative design) and the alternative (informative design that
cannot be ignored) is the usual -svy: - prefix estimator. Stata does
not allow the test of this format, however, and says that -hausman
cannot be used with clustered or p-weighted data-. Was it that Stata
Corp just did not want to mesh econometrics and survey statistics (as
they indeed frequently don't tend to go hand in hand), or are there
any more substantial considerations? I am probably adult enough in
Stata to do all I need by just handling the matrices, but I personally
don't think this is an important restriction (provided of course that
the survey estimator is only used as the consistent one). Moreover, if
Stata wanted to look cool, this test can be made standard for a lot of
-svy- style models.
References:
Hausman J (1978) Specification tests in econmetrics. Econometrica,
Vol. 46 (1978), pp. 1251-1271,
http://www.citeulike.org/user/ctacmo/article/553118
Pfeffermann D (1993) The role of sampling weights when modeling survey
data. International Statistical Review, Vol. 61 (1993), pp. 317-337,
http://www.citeulike.org/user/ctacmo/article/1036965
--
Stas Kolenikov
http://stas.kolenikov.name
*
* 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/
