[Marsbar-users] Correlation brain activation and variable of interest

[a.vercammen <a.v...@ne...>]

It's a bit of a long story, but hopefully someone will read the whole thing
and be able to help me out with this situation that is confusing me...

In SPM, I constructed a two sample t-test design with a covariate of
interest (testosterone levels). I want to look at the difference in the
relationship between testosterone and brain activity in 5 ROIs between my
two groups. I have done that by including two covariates in my SPM design
(one with the testosterone values for group 1, patients, and one with
testosterone values for group 2, controls).
http://old.nabble.com/file/p32603571/SPM%2Bdesign.png 

Now using the contrast [0 0 -1 1] presumably gives me the desired
interaction (the difference in correlation of testosterone and contrast
values between groups).

Using MarsBar, if I extract ROI values, and specify this contrast, and
estimate my results. MarsBar generates a results table with contrast values,
t values and p values for my 3 ROI. Brilliant: I get some significant
results. Which I take to mean that the correlation between testosterone and
contrast values in the ROIs is significantly different between patients and
controls.

To plot the correlations, I export the summary time courses for the ROIs
(which gives me a contrast value for each individual), and plot them against
the testosterone values in a scatter graph.
In doing this, also compared the correlation values I computed between the
two groups, by using Fischer r-to-z transformations. Now here is the
problem: the p-values I get from doing this, do not match up with the
p-values in MarsBar generated table. The same regions are significant, but
the values are different. What's the difference here? Am I doing something
wrong?

Thanks very much for any clarification on this subject!
-- 
View this message in context: http://old.nabble.com/Correlation-brain-activation-and-variable-of-interest-tp32603571p32603571.html
Sent from the MarsBaR mailing list archive at Nabble.com.