Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: ttest var1==var2 for clustered data

From   "Nick Cox" <[email protected]>
To   <[email protected]>
Subject   RE: st: ttest var1==var2 for clustered data
Date   Tue, 22 Mar 2005 14:28:57 -0000

I'd agree that this is not obviously 
a regression problem in so far as the two variables
are likely to be on the same footing. That said, 
some might advocate some model that took errors
in both variables into account. 

The issue of whether two variables
agree lends itself to various other approaches. 
Some graphical ideas are reviewed in 

Graphing agreement and disagreement
Stata Journal 4(3): 329--349 (2004) 

and one numerical measure is described in 
various articles on -concord- in the 
Stata Journal and Stata Technical Bulletin. 

Admittedly, that is not your question, 
and those methods do not take account 
of clustering, but even if a way 
existed of getting credible values 
for a t-statistic and a P-value in your situation, 
they would still be rather drastic summaries
of a problem that might possess a lot
of fine structure.

[email protected] 

Christian Hunkler
> The problem is, we want to perform a mean comparison between 
> two variables
> (paired t-test) and not between two subgroups. The syntax 
> without adjusting
> for the clustering is  "ttest var1=var2", thus using inflated 
> degrees of
> freedom, 'cause no adjustment for the clustering of the data 
> is possible.
> In contrast performing the test with a disaggregated file 
> would result in
> deflated degrees of freedom...
> For our group comparisons we exactly use your proposition.

Rich Goldstein 
> >a t-test is a simple regression where the only predictor is the dummy
> >variable identifying the groups -- so, put the two variables 
> into one long
> >variable; add a group identifier and use regression with the 
> cluster option

Christian Hunkler

> >>we have network data clustered around egos. Now we want to 
> test for mean
> >>differences in variables that contain proxy information 
> provided by ego
> >>about their alteri. In short a t-test comparison of two 
> variables, but the
> >>observations are not independent but clustered.
> >>I already checked the clttest ado, but it does not yet 
> support comparison
> >>of two variables.

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index