Hi,
I'm afraid you can't have as many EVs as inputs, since that leaves nothing in the residuals and hence nothing to estimate the variance from, which is crucial for all the stats. If all you want to do is model the difference between two runs then you can have one EV of the form [1 -1 0 0] (as a column). However, this is a very small number of inputs to try and run a mixed effects model on, so you will also need to use a fixed effects analysis in this case.
All the best,
Mark
On 21 Jan 2014, at 16:56, sidarta <[log in to unmask]> wrote:
> Dear FSL experts,
>
> I'm quite new to FSL and fMRI analysis. I have a series of data of one subject scanned at different sessions. My intention is simply to see the difference between scan1 and 2, scan2 and 3, etc. In my Full Model setup, I assign each of the low-level FEAT as one EV:
> EV1 EV2 EV3 EV4
> - Input 1: 1 0 0 0
> - Input 2: 0 1 0 0
> - Input 3: 0 0 1 0
> - Input 4: 0 0 0 1
>
> so that in contrast setting I just put [1 -1 0 0], say.
>
> After running the high level FEAT, I realized there is a problem with my stats report. The p-values are all NaN. I opened the log file and it seemed there are warnings and RESEL = NaN (as attached). I don't really understand the root cause, perhaps my Full Model Setup is wrong. All low-level FEAT seems fine!
>
> Thank you
> sidarta
> <NaN pvalue.png><log file.png>
|