Error during command authentication.
Error - unable to initiate communication with LISTSERV (errno=10054, phase=RECEIVE, target=10.0.0.37:2306). The server is probably not started.
Hi Donald,
Thank you very much for your rapid reply.
I do take your point, that as I'm not comparing the within-subject levels to one another, one might consider it akin to performing multiple one-sample t-tests, so it's strictly speaking not really a within-subject effect that I'm testing... GLM_flex sounds like an interesting toolbox and I will give it a try. The idea of the eye contrast was derived from the SPM manual (chapter 30: face group fmri data). As in the example in the manual, my factor is time and the 9 levels are the FIR time bins. I basically would like to assess those voxels that show any form of event-related response, analogous to the example in the manual.
The problem remains though that the same-vein contrast (looking where any of the levels of a within-subject repeated measures factor are different from 0) works with other datasets, so I'm still not sure why it would fail to run on this particular dataset and would appreciate any hints or suggestions.
I apologize in advance if I missed something obvious,
Bianca
Bianca,
The contrast you plan use (eye(n)) to test whether any of the values are different than 0 is not a valid contrast as this would be a between-subject effect and the error term is for within-subject effects. If you use GLM_flex, then you can evaluate the contrast with the appropriate error term.
See below for answers to your other issues.On Fri, Mar 23, 2012 at 12:40 PM, Bianca de Haan <[log in to unmask]> wrote:
Dear all,
When estimating a repeated measures 2nd level model in the latest version of spm8 (r4667, however, the problem remains when switching back to r4290) running on Matlab R2010b, I keep getting the following error message:
Failed 'Model estimation'
Error using ==> eig
Input to EIG must not contain NaN or Inf.
In file "C:\Programme\spm8_R4667\spm_reml.m" (v3791), function "spm_reml" at line 202.
In file "C:\Programme\spm8_R4667\spm_spm.m" (v4515), function "spm_spm" at line 875.
In file "C:\Programme\spm8_R4667\config\spm_run_fmri_est.m" (v4403), function "spm_run_fmri_est" at line 33.
The following modules did not run:
Failed: Model estimation
This error message is preceded by multiple warnings of "Matrix is singular, close to singular or badly scaled. Results may be inaccurate. RCOND = NaN."
Similar problems have been reported on the list (http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind1202&L=SPM&D=0&I=-3&d=No+Match%3BMatch%3BMatches&P=705417 and http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0807&L=SPM&P=R7435&I=-3&d=No+Match%3BMatch%3BMatches), however, my batch does not contain duplicate filenames and changing the spm_reml.m file around line 78 did not solve the problem.
My design is a flexible factorial with factor 'subject' (independence 'yes', variance 'unequal') and my repeated measures factor with 9 levels (independence 'no', variance 'equal'). Under 'main effects and interactions' I specified only the main effect of my repeated measures factor (factor number '2').
Ultimately, I would simply like to test where any of my levels on the repeated measures factor is significantly different from 0 (i.e. eye(9) contrast)
See above comment about this being invalid.
The error remains when I set the variance of the subject factor to 'equal'. However, when additionally specifying the main effect of the subject factor (factor number '1') under 'main effects and interactions' the estimation runs without a problem.
You need to include the main effect of subject in the model to properly do a repeated-measures analysis.
I encounter no problems when performing similar analyses on other datasets, which leads me to suspect that it must be something specific to this particular dataset... The results of preprocessing and the 1st level analyses, however, look fine.
I am at a loss how to solve this problem and would be grateful for any hints or advice that might help me.
Many thanks in advance,
Bianca