Print

Print
Hi,

You are correct, it won't change the analyses unless you look at copes that involved empty EVs.  They are meaningless since they replace the PE images involved in the contrast estimate with 0's in the case of empty EVs.

It is changing the degrees of freedom, but I'm assuming you have a time series with over 100 time points, so I bet this is negligible.  Also, the DF may not be used in the next level analysis, so this may not matter anyhow (unless I'm forgetting something).

What is nice about the empty EV option is that it allows you to keep your "cope" numbers consistent across subjects.  So, for example, if I have 2 EVs and I run the following contrasts
C1=[1 0]
C2=[0 1]
C3=[1 -1]

cope numbers (1,2 & 3) will only be consistent across subjects if I use the empty EV option.  So, if a subject has an empty EV for EV1, COPE2 will be valid and will still be called COPE2.  If I simply left EV1 out, it would be called COPE1 (since I'd only have one contrast in this case).

IMPORTANTLY, if a subject is missing an EV involved in the contrast estimate it is UP TO THE USER to remove that subject from higher level analyses.  I cannot stress this enough since I've run into people who figured FSL magically removed these subjects in higher level analyses.  Although FSL is pretty awesome, this would be a little tricky to do :)  It is a pain to keep track of this yourself, but not impossible if you use some smart coding.

If you forget to omit these subject FSL will NOT issue an error message.  There are cope images that will be used in the analysis, but they will be meaningless, since a PE image of all zeros will be used in all cope calculations.

Cheers,
Jeanette


On Tue, Feb 15, 2011 at 1:05 PM, Amy Winecoff <[log in to unmask]> wrote:
Hi all,

I have been trying to figure out exactly how the empty EV option works. I ran two analyses, one where the empty EV option was used and the other, where the empty EV option was not. That is to say, one analysis modeled the evs A, B, C, and D(empty) while the other modeled A, B, and C only. Subtracting both the cope and varcope images between the two analyses produced no differences, so it seems as though running the analysis with the empty ev option checked estimates the model as though there were not an ev there at all. My only question is that the degrees of freedom appear to be different between the two analyses. I am correct in assuming that this doesn't have any consequences for the results?

Thanks!

Amy Winecoff