Dear Felix,
one way to get rid of the dependency of degrees of freedom is the use of effect size (e.g. for t statistic) or the coefficient of determination (e.g. for F statistic). This is commonly used for meta analysis.
There exists a function to convert spmF maps in VBM8 to coefficient of determination (threshold and transform spmF maps) that might be of some help. However, maybe a mathematician might give more helpful hints, because this is absolutely not my field of expertise...
Best,
Christian
On Tue, 4 Feb 2014 17:01:10 +0000, Felix Schlegel <[log in to unmask]> wrote:
>Please have a look at this example: http://i.imgur.com/3KEReVZ.png
>
>I want to compare 3 different models (left, middle, right) and two different groups (top, bottom).
>The maps show unthresholded F-values ranging from 0 to >20.
>However, I think the values cannot be compared, because for example the upper row has much less subjects, therefore the F-values are generally lower.
>If I set the colormap limits from 0 to 40, the bottom row would look much nicer (not everything yellow as it is now) but not much would appear in the top row. Also the model of the F-map on the right has more parameters and therefore seems to have generally higher F-values.
>
>Either I use separate colorbar limits for each map, which may seem like cheating, or apply some kind of normalization to F-values to account for different number of subjects and ANOVA levels...
>
>Are there some references or guidelines? In all the papers I find, somehow miraculously the same colorbar and limits always match every map...
>
>Thanks
|