Hello Statalisters,
I have the following problem and I would appreciate
your suggestions:
I want to conduct a meta-analysis of published data.
The outcome of interrest is a quantitative continuous
measurement which is assumed to be normally
distributed (let's call it y). Each article reports
means, SDs (of y) and sample size for two or more
independent samples treated with a different technique
(A, B, C ...). I want to compare mean values of y
between B and A, C and A etc. I thought that it would
be a good idea to work with regression coefficients
and their SEs (beta_BvsA, SE_beta_BvsA etc.). I know
that if there are only two arms (say A and B) one can
use simple formulas to obtain mean difference and its
SE or equivalently a regression coefficient and its
SE, or use the corr2data command to calculate these
estimates using t-test or regression ( I remember a
long thread about corr2data vs. drawnorm, sufficient
statistics etc.). The question is if it is right to
use the corr2data command to create three (or more)
independent samples for A, B and C arms with given
mean, SE and N to obtain via regression two betas and
two SEs (B vs. A and C vs. A). I know that I could use
only two arms for each meta-analysis ( B vs. A, C vs.
A and so on) but I think that this would induce some
multiple comparison error so I decided that I have to
calculate all possible betas of each study ( B vs. A,
C vs. A and so on) simultaneously.
Any thoughts?
Nikos Pantazis
Biostatistician
__________________________________
Do you Yahoo!?
Yahoo! Search - Find what you�re looking for faster
http://search.yahoo.com
*
* 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/