I have encountered a problem in estimating a regression equation using
svyreg. The data I am using is from a national four-stage probability sample
of emergency room visits, with the unit of observation the individual
patient visit (n=20,000). The data is divided into geographical PSUs, which
after some adjustment for multistage sampling and strata that contain a
single PSU, total approximately 200 in number. The fixed effect in which I'm
interested is the individual hospital, of which there are more than 300.
When I run regressions, Stata does not produce an error message, but it
notifies me that it cannot compute an F-statistic for joint significance of
the full set of covariates (as the number of degrees of freedom is
negative). Stata further tells me that "you need to consider carefully
whether any of the reported standard errors mean anything. The theory that
justifies the standard error calculation is asymptotic in the number of
clusters [PSUs minus strata], and we have just established that you are
estimating at least as many paramaters as you have clusters [PSUs minus
strata]."
So, I have 3 questions:
(1) Are hypothesis tests involving my coefficient estimates meaningful?
(2) If the answer to (1) is "no", is there a way to adjust for this while
still retaining the hospital fixed effects?
(3) If the answers to (1) and (2) are "no", is it more appropriate for me to
report the results with fixed effects omitted or with them included?
Thanks.
Martin
Martin Zelder, Ph.D.
Section of Emergency Medicine
Department of Medicine
University of Chicago
5841 S. Maryland Ave., MC 5068
Chicago, IL 60637
phone: (773) 702-5066
fax: (773) 702-3135
*
* 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/
