I have a question regarding beta extraction. I got a significant activation in a cluster from a paired-t test with a covariate. How should I proceed to get beta values that take in account the covariate?
I have tried the following approach, but the t-test run on the resulting values did not turn significant as expected:
[ul]
[li] I started by extracting “raw” betas for each condition
[/li][li] I included the variance associated with the covariate in each condition by residualizing each condition against the covariate
[/li][li] I added the mean value back to the residuals
[/li][/ul]
In a “normal” 3dttest++ scenario we e.g. test if two groups differ (in beta values) and we get a significant cluster. From this cluster we extract the betas (the same betas the 3dttest used to get a significant cluster) so that we can make nice plots (and if it’s an ANOVA we can do post-hoc). These betas should, of course, also produce a significant difference if we compare them e.g. Excel as well.
But when we run a 3dttest++ with covariates and the covariates are responsible for getting the significant cluster it’s not enough to just extract the betas, we also need to incorporate the covariates in some way. I think it is here that we don’t quite know what to do to be able to “alter” the betas with the covariates to that the significance show up in Excel as well.
Sure, that sounds good. But my point is that 3dttest++ will remove the means of the covariates, unless you tell it otherwise or how to do it. So the covariate series that you fit and remove should have the same de-meaning operation done, first (well, you would probably demean both the data and the covariates before fitting, depending on exactly what regression model you apply, and then add the original data mean back in).
rick
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.
