1. According to Friston & Penny (NeuroImage, 2003) "Posterior
Probability Maps and SPMs", on page 1246 they mention a threshold of
0.7% (equivalent to percentage of whole-brain mean signal). So I believe
your guess is correct.
2. I am unsure, since the interpretation is different than in
frequentist inference. I should think that you can report 0.90 if there
is precedent in the literature, or if you can reasonably defend that choice.
S Calderbank wrote:
> Dear all
>
> I have conducted some analyses using a conventional frequentist SPM
> approach. I want to make some arguments concerning both a presence
> and absence of an effect in specific regions. I believe that using a
> Bayesian approach (PPM) I can reframe my hypothesis so that I can
> identify the probability of finding an effect of a specific size in my
> regions - thus allowing me to show evidence in favour of "accepting"
> the null hypothesis by showing a low probability of an effect size x.
> I have conducted a second level Bayesian estimation on the SPM.mat
> arising from the frequentist analysis. My questions are:
>
> 1. I am a little confused as to whether there have been changes
> between SPM2 and SPM5/8 in visualising PPMs. When you choose the
> effect size in SPM5 or 8 does the value which appears in the gui (in
> my case 0.01) refer to a 1% change in the global mean? From the
> previous literature it seems that the conventional way of trying to
> demonstrate a null result is to visualise > 1SD of the prior variance
> and report the probability at specified regions. How do I find this
> value?
>
> 2. It seems it is common practice to set the probability value to
> >.95. Is this still the case even in the instance of a one sided
> t-test where one would assume a more conventional criterion would be >.90?
>
> Any help would be very appreciated.
>
> Kind regards
>
> Simon
|