Hi Berk,
Once I faced a similar problem in which I wanted to compare the mean
of a variable y for two (unpaired) groups, say A and B, without
imposing the iid assumption.
My solution was to run a regression of y on dummies for A and B
(noconstant) and allow the "robust cluster" options to take care of
the non-iid errors.
Since you don't exactly say what your variables A and B are, and
whether you have individual data (vs. only means and sd's), it is hard
to know if my suggestion would help...
cheers,
robert
On 4/27/05, Nick Cox <[email protected]> wrote:
> This problem is discussed, among other places,
> in Rupert Miller's book "Beyond ANOVA". As
> you say, the issue is getting a more accurate P-value.
> I think you can do that if you also have a serial
> correlation, i.e. the dependence arises because
> these values are from a time series. I am not aware
> that the procedure is canned as a Stata program, but
> it should yield to calculator-style manipulations.
>
> Nick
> [email protected]
>
> Berk Sensoy
>
> > Thanks for your answer, but I don't think it will work because ttest
> > assumes that variables in each sample are iid.
> > Essentially, because observations of each variable are correlated, I
> > have fewer effective observations than actual observations, so ttests
> > understate the standard error of the mean.
> >
> > Anyone know how to test mean(A)=mean(B) when cov(ai,aj) != 0 and
> > cov(bi,bj) != 0?
>
> *
> * 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/
>
*
* 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/
