On Sat, Nov 3, 2012 at 8:31 AM, Jens Kronschnabel
<[log in to unmask]> wrote:
> Dear SPMers,
>
> I think I have now read all the threads on conjunction analysis, but I am
> still unsure how to set it up properly. I tried all constellations and
> results vary a lot. First of all, I'd like to use the "global" null option
> as offered by SPM, because the "conjunction" option will be to conservative
> for my data (including similar options that use imcalc to compute common
> regions/ overlaps).
> The correct way for one group of subjects that underwent several conditions
> seems to use a 2nd level one-way ANOVA (not the within-subject one) and to
> do the proper adjustments for sphericity.
>>>> You are doing a within-subject ANOVA, you need to select the flexible factorial or one-way within-subject ANOVA. Other methods will not be correct.
However I am not sure on
> sphericity. Levels of the factor are probably dependent, so is it correct to
> set the independence option (stats.factorial_design.des.anova.dept = ) to NO
> (= 1)? I am even less sure about the variance option
> (stats.factorial_design.des.anova.variance =), and this even seems to be
> more critical. What would be the appropriate setting here?
>>> Dependence is Yes. Variance can be equal or unequal, either are acceptable. I generally use equal because the covariance between any 2 levels should be the same as should the variance of each level.
My data look nice
> if I use equal variance, and less expected if I use unequal variance.
> What is the analoguous way to set this up for two groups of subjects that
> underwent the same, say 4, conditions. Ideally, I'd like to obtain SPMs for
> both groups separately and then also the group comparison for the
> conjunction. I think I know how to define all the contrasts needed, but
> again I do not know how to properly set up the model. Use of full-factorial
> or flexible factorial? What independence and variance options should be
> applied to the group and condition factors?
>>>> Flexible factorial with group, condition, subject specified AND in the model OR GLM Flex. group --> variance unequal, dependence no; condition --> variance equal, dependence yes; subject --> variance equal, dependence no. From the flexible factorial you can get the condition effect, condition effect with in each group, group*condition effects. You cannot get group comparisons of a single condition or the average of conditions though. There have been many previous posts on the limitations of the GLM for between-subject effects.
>
> Any help is highly appreciated.
> Thank you,
> Jens
>
>
>
>
> --
> Jens Kronschnabel, Ph.D. student
>
> Department of Child and Adolescent Psychiatry, University of Zurich
> Neumünsterallee 9, CH-8032 Zürich, Switzerland
>
> +41 43 499 2650
> [log in to unmask]
> --
|