Dr Friston: you specify below data with a grand mean of 100-- does this
apply to an analysis in which grand mean scaling was set to 100-- even
though the raw data do not [usually] have this mean?
Equally true if ANCOVA or PS were used?
and a further clarification: it appears what you propose applies to a given
voxel or region, not over the SPM as a whole--true?
At 01:35 PM 06/26/2000 , you wrote:
>Dear Otto,
>
>> I have a quick question: performing spm analysis of PET data comparing
>> two groups (17 and 7 subjects) and not detecting any significant
>> differences, what is the power for my null findings? In other words,
>> how confident can I be that I did not missed effects larger than say
>> 10% difference?
>
>In theory you should be able to do this using the standard error of the
>contrast of parameter estimates at a representative voxel. After
>plotting the contrast of interest the variable SE in working memory
>represents the standard error of that contrast. The probability of
>detecting an activation of A% at a specificity of defined by a T
>threshold u (given the standard deviation of the parameter estimate is
>SE) is (I think):
>
>1 - spm_Ncdf(u*SE,A,SE^2)
>
>This is the power or sensitivity.
>
>For example if your corrected threshold is 4.2, SE = 2.6 and the grand mean
>of your data was 100, then the power to detect a 10% activation would be:
>
>1 - spm_Ncdf(4.2*SE,10,SE^2) = 0.3617
>
>i.e. 36% power. I am not sure how rigorous this is (I am not very
>expert in this) but you could certainly check this with your local
>statistician.
>
>
>I hope this helps - Karl
>
********************************************************
Christopher Gottschalk, MD *
Assistant Professor of Neurology & Psychiatry *
Yale School of Medicine *
*
Mailing Adress: *
VAMC [116-A] tel [203] 932-5711 x4329*
950 Campbell Avenue FAX 937-4791*
West Haven, CT 06516 *
********************************************************
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|