Yes, this is problematic indeed. SPM manual might need some updates/extensions. In Chapter 10 some general information is given for "Independence" several times. A reader might conclude that "two-sample t-test" "one-way ANOVA" "Full factorial" are appropriate for repeated measurements when setting independence to "no". I'm not sure why this option hasn't been disabled so far where misleading.
Concerning "Full factorial", the example of the manual is about a single factor with three levels corresponding to canonical response, temporal and dispersion derivative. This model is discussed in chapter 30 "Face group fMRI" in detail for a group of 12 subjects. However, the model has 33 df within-group... As a sidenote, this model is compared to the results of a one-sample t-test based on the canoncial response only, concluding: "Note also how this F-contrast (Figure 30.5) produces more signicant results than the corresponding F-contrast in the model with the canonical HRF shown in Figure 30.2. This suggests signicant additional information in the two derivatives of the canonical HRF." Now, how much of these results are due to variance explained by the derivatives and how much due to inflation of df?
On 6 November 2013 09:36, "benjamin becker" wrote:
> Dear Helmut, dear Donald & SPMers,
> unfortunately I have to agree with Helmut - I often discuss the validity of SPM-repeated measures analyses with my colleagues & we
> therefore search the literature for papers describing analysis approaches that might be useful suitable for our data. However, incorrect
> models are still often reported (including group-effects from flexible factorial designs or -as already mentioned by Helmut- full factorial
> designs for within-subject data).
>
> Best regards,
>
> Ben
|