[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Test diff in R^2 from metareg models
Good question Wendy, and excellent timing! Afraid I don't have a good
answer at the moment, but try me again in a couple of weeks and i might
well do. I'm going to a meeting about meta-analysis software next week (i
think someone you know may be there too..) and one of the topics will be
meta-regression. I've been thinking a bit about ANOVA-type output for
meta-regression and your question is clearly related. If I remember i'll
report back to the list. If i don't please remind me off-list.
My feeling is you ought to be able to do this using by computing change in
a "residual" Q statistic and referring it to a chi-squared distribution,
but i don't myself think meta-regression programs (including -metareg- in
Stata at the moment unfortunately) calculate the Q the right way for this
purpose for random-effects models at present.
One thing you could do at present though is use -test- (or -testparm-)
after -metareg- to give a Wald test of the block of variables added. Will
give an F-test unless you use the 'z' option to -metareg-, in which case
you'll get a chi-squared test. At present I think i'd go for the F-test,
but i may have changed my mind in two weeks...
At some point i plan to add a 'block' suboption to the 'permute' option in
-metareg-, which would arguably be the best way to do this.
--On 15 August 2005 10:21 -0500 "Garrard, Wendy M."
I have a series of three meta-regressions, where blocks of variables
were added, resulting in an initial R^2 of 27%, the second block adding
13% to R^2, and the third block adding 25% to R^2. This is a weighted
regression with random effects (i.e., metareg).
I want to report whether the change in R^2 is significant for blocks two
and three, but am uncertain how to obtain a test (F- or Chi-sq?) of R^2
change in Stata, especially since it for metareg with random effects.
* For searches and help try: