Many thanks for your speedy reply to my query. However, your answer has raised
further questions in my mind.
When I run the conjunction analysis I obviously get the results table with
areas of activation and associated p values at the voxel and cluster level.. My
question is therefore at what level of p value do the results become
"significant" in a conjunction. I.e. if the corrected p value for an
activation is p=0.24 and the uncorrected p<0.001 is it "valid" to report this
as a "significant" result because it has survived a conjunction of 12 subjects
at Karl's suggested threshold (from your reply) even though I had no a priori
hypothesis for this location?
Many thanks again.
Cathy Price wrote:
> Hello Mark
> According to Karl, the following heuristic can be applied.
> The threshold you need to chose for each subject (Pc) should equal the
> threshold you want for the conjunction (Pw), when Pc is raised to the power
> of n.
> So if the uncorrected threshold you want for the conjunction is p<0.001
> then the threshold for each subject is 0.4217 for 8 subjects (approximately
> what you used)
> and 0.1778 for 4 subjects (higher than you used).
> >> 0.001^(1/4) = 0.1778
> Hopefully this stricter threshold with the lower P value will give you more
> sensible results
> for the n=4 group.
> >Dear SPMers,
> >I have a data set of 12 subjects who had 12 run water PET scans. The
> >subjects split into 2 groups (n=8 & 4 for the groups). I would like to
> >see which areas of activation are common to the subjects in each group.
> >I therefore want to do conjunction analyses for each group of subjects.
> >My question concerns appropriate thresholding (p values) for the results
> >from these conjunctions.
> >I have entered the data using the subject x condition interaction design
> >for PET studies with each subject having 6 scans under each condition.
> >I have then done a separate contrast for each subject (-1 1). I then
> >obviously do a conjunction of these individual contrasts including the 4
> >or 8 subjects I am interested in. I have no a priori hypotheses of
> >particular areas that I am expecting differences so suspect that I
> >should be using corrected p values.
> >I have specified low thresholds for the contrast conjunctions (eg.
> >p<0.5) which for 8 subjects give fairly discrete activations in sensible
> >places, but for 4 subjects most of the brain is pale gray. Intuitively
> >I can understand that the fewer subjects I include in a conjunction the
> >more likely activations will be common in the conjunction. What I do
> >not understand is some way to quantify this so that I do not make type I
> >or type II errors interpreting the conjunctions.
> >Help on this would be appreciated.
> >Dr. Mark Daglish
> >Clinical Research Fellow
> >Psychopharmacology Unit
> >University of Bristol
Dr. Mark Daglish
Clinical Research Fellow