Hi Marco,
I know nothing about the RWLS toolbox, but it does sound like it's a
bug there, rather than in the rest of SPM. Based on the comments in
the (standard, not RWLS) spm_spm.m,
xX.nKX - design matrix (xX.xKXs.X) scaled for display
I can think of a couple of things to try:
Check whether xX.xKXs.X has complex elements, or whether the display
scaling is just messing things up -- I expect the former,
unfortunately, but if it turns out it is just the latter, then you
could try loading the SPM.mat after estimation, removing the complex
values in xX.nKX and then resaving; the results might then work
alright, although you'd have to ignore the displayed matrix.
If xX.xKXs.X is complex, double check that the original xX.X is what
you expect, and then investigate SPM.xX.K and SPM.xX.W, and if the
latter is complex, then also look at SPM.xVi.V. If that is where the
problem appears, then it sounds like ReML has gone awry, even if the
estimation appeared to go alright. I can't think of much to suggest
then though, other than contacting the RWLS toolbox authors...
Best,
Ged
marco tettamanti wrote:
> Dear list,
> I am encountering a problem with a first-level statistical analysis
> using the RobustWLS toolbox. The model estimation runs fine, I can also
> define a set of statistical contrasts via the command line (using
> MakeCon), but when I attempt to display the contrast effects clicking on
> the "Results" button, I get an "Image CDdata can not be complex" error
> (see error message attached below). Oddly, this only happens for a
> subset of subjects.
> I have found a previous mailing-list thread on this subjects, including
> the following response:
>
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0703&L=SPM&P=R10863&I=-3
>
> in which Ged Ridgway suggested to enter the following command in matlab:
>
> any(~isreal(SPM.xX.nKX));
>
> When I do this, I find that the nKX structure is indeed complex for the
> subjects in which I encounter the problem and that it is not complex for
> the "good" subjects.
>
> Concerning the RWLS toolbox, it doesn't make any difference whether I
> enter the unsmoothed data or not: the "good" subjects always run fine,
> whereas the "bad" subjects always give the "complex data" error. Also,
> rWLS convergence does not seem to be a problem.
> However, an "equivalent" statistical model estimation in SPM5 (i.e.
> outside the RWLS toolbox) does not produce any "complex data" error,
> even in the bad subjects.
>
> The problem seems to be architecture and software independent: it occurs
> both on 32bit and 64bit linux, with several SPM5 revisions (826, 748,
> 665) and with two different matlab releases (2006a, 2007a)...
>
> Does anyone have an idea of what's going on?
> Thank you so much,
> Marco
>
>
> SPM5: spm_results_ui (v$Rev: 816 $) 10:26:22 - 06/06/2007
> ========================================================================
> ??? Error using ==> image
> Error using ==> image
> Image CData can not be complex..
>
> Error in ==> spm_conman at 960
> hDesMtxIm = image((varargin{2}.nKX+1)*32);
>
> Error in ==> spm_conman at 929
> spm_conman('ImDesMtx',xX,H.hDesMtxAx) %-Depict DesMtx
>
> Error in ==> spm_conman at 812
> [F,cF] =
> spm_conman('Initialise','on',SPM,STATmode,n,Prompt,Mcstr,OK2chg);
>
> Error in ==> spm_getSPM at 278
> [Ic,xCon] = spm_conman(SPM,'T&F',Inf,...
>
> Error in ==> spm_results_ui at 274
> [SPM,xSPM] = spm_getSPM;
>
> ??? Error using ==> [hReg,xSPM,SPM] = spm_results_ui;
> Error using ==> image
> Error using ==> image
> Image CData can not be complex..
>
> ??? Error while evaluating uicontrol Callback
>
>
>
|