[log in to unmask] wrote:
> Dear Will and Karl
>
> A colleague of mine have conducted a set of fMRI experiments where she
> would like to demonstrate the absence of an interaction effect. She
> started of demonstrating that there were no significant voxels in a second
> level one-sample t-test of the interaction contrast at a p-value of 0.001
> uncorrected. I convinced her, that what she would really like to do was to
> demonstrate that the effect with say 95% probability was smaller than some
> value. And here we are having pushed the ->Bayesian button! While the
> second threshold we are asked to specify is easy to decide on (it has to
> be 0.05) we have a problem in deciding what would be a fair threshold to
> use for the effect size (gamma ?). In your paper "Posterior probability
> maps and SPMs" the text on figure 5 says "Height threshold P=0.95, effect
> size 0.7%", but this layout is not similar to that in SPM2. Here instead
> the text only mentions the Height threshold, and the PPM has a
> superscript e.g. 0.03. But how does the value in the superscript relate to
> the % signal change value, does 0.03 mean 3% or 0.03%?.
0.03 means 0.03% of the global mean value.
> And the other
> question would it be fair to set this value to say 10% of the effect size
> of the main effect as determined in the Cbeta from a separate analysis
> (one sample t-test) of the main effect?
>
Yes. This would be reasonable.
Best,
Will.
> I am looking forward to hearing from you
>
> Best regards
> Torben
>
>
>
> P.S. I am unable to post to the list from this email account, but you are
> of course most welcome to forward your answer to the mailing list.
>
> Torben E. Lund
> Danish Research Centre for MR
> Copenhagen University Hospital
> Kettegaard Alli 30
> 2650 Hvidovre
> Denmark
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|