Mira,
yes, you have probably hit the mentioned (or another) bug in covariance
component specification. If you review the model and look at the covariance
components, you will most likely notice that these don't have the
proper matrix
size nscans-by-nscans.
If you send me the batch file with your design specification, I might have a
look into it.
Volkmar
Quoting Mira Bühler <[log in to unmask]>:
> Dear Dr. Penny,
> Dear all,
>
> I`m using the flexible factorial design specification in SPM5 to specify a
> mixed design with a within-subject factor reward_level (4 levels) and a
> between-subject factor group (2 groups).
> When I try to estimate the model the following error occurred:
>
> SPM5: spm_spm (v$Rev: 301 $) 17:19:06 - 03/06/2006
> ========================================================================
> Initialising parameters : ...done
> ...initialised
> Plane 69/69 , block 1/1 : ...done
> Temporal non-sphericity (over voxels) : ...REML estimation
> ??? Error using ==> plus
> Matrix dimensions must agree.
>
> Error in ==> spm_reml at 87
> C = C + Q{i}*h(i);
>
> Error in ==> spm_spm at 860
> [V,h] = spm_reml(Cy,xX.X,xVi.Vi);
>
> Error in ==> spm_getSPM at 210
> [SPM] = spm_spm(SPM);
>
> Error in ==> spm_results_ui at 260
> [SPM,xSPM] = spm_getSPM;
>
> ??? Error while evaluating uicontrol Callback.
>
>
> When specifying a within-subject ANOVA (without the between-subject
> factor)the estimatation of the model is working fine.
>
> I`m wondering whether there is something wrong with my design specification
> or could this be related to the following note of Dr. Will Penny?
>
> Date: Wed, 21 Dec 2005 12:09:16 +0000
> Reply-To: Will Penny <[log in to unmask]>
> Subject: Re: spm5 repeated measures ANOVA
> Comments: To: Martin Kronbichler <[log in to unmask]>
>
> **A note of caution**. There is currently a bug in SPM5 when specifying
> nonsphericity for 2 or 3 way ANOVAs. We are working on a fix.
>
>
> Design specification in SPM5, matlab 71:
> 1) 'Flexible Factorial' design
> 2) Factors: subject (reserved word – no condition vector necessary),
> reward_level, group
> 3) 'Specify Subject or all Scans and Factors': 'Subjects'
> 4) 'Scans' , 'Specify Files' : 4 con images for each subject(reward level
> 1,2,3,4)
> 5) 'Conditions': [1 2 3 4; 1 1 1 1] for all subjects of the first group and
> [1 2 3 4; 2 2 2 2] for all subjects of the second group.
> 6) 'Main effects and interactions':
> Main Effect 1 (reward_level): 2
> Main Effect 2 (group): 3
> Interaction (group*reward_level): 3 2
> Main Effect 3 (subject/constant): 1
>
>
> Any suggestions would be appreciated.
> Many thanks
> Mira
>
>
|