At 12:13 18/05/2005, you wrote: >Hi guys > >I'm trying to understand how PPm works and the question is how and from >where do you get prior values??? > >Thank you > > >Bernie Dear Bernie, PPMs are based on a hierarchical model of the following form (y = data, beta = parameters, e = error, index in brackets = level): First level: y = X(1)*beta(1) + e(1) Second level: beta(1) = 0*beta(2) + e(2) The first level is just a normal GLM characterising effects at a single voxel. At the second level, beta(2) is the average effect over voxels and e(2) is its voxel-to-voxel variation. As the parameters of interest, i.e. the parameters at the first level, beta(1), reflect regionally specific effects, one can assume that they sum to zero over all voxels. This corresponds to using a shrinkage prior (i.e. zero mean) at the second level; the variance of this prior is implicitly estimated by estimating the variance of e(2). This empirical prior can then be used to estimate the posterior probability of beta(1) being greater than some threshold at each voxel. Or, in more simple terms: The first level of the hierarchy corresponds to the experimental effects at any particular voxel and the second level comprises the effects over voxels. The variation in a parameter (or contrast), over voxels, can be used as the prior variance of that parameter (or contrast) at any particular voxel. Don't confuse this hierarchy with the usual "first level" and "second level" models which one uses to refer to single subjects and group analyses. Here the hierarchy is over voxels, not over single subjects and groups. PPMs can operate both on single-subject data as well as on a set of contrast images. Best wishes, Klaas _____________________________________ Dr Klaas Enno Stephan Wellcome Dept. of Imaging Neuroscience 12 Queen Square, WC1N 3BG, London, UK phone: +44-207-8337485 fax: +44-207-8131420 web: http://www.fil.ion.ucl.ac.uk/~kstephan/