 |
 |
Dear claus, list, we noticed something very similar in a 2nd level ANOVA the way you describe using the betas for hrf + dhrf/dt regressors from the 1st level, on data from a simple visual stimulation fMRI paradigm. We did additional tests on the betas for the hrf and dhrf/dt regressor alone, and had a look at the distribution of the Betas over subjects. We then noticed that the betas for hrf are quite comparable over subjects, but the betas for the dhrf/dt were not, they were all over the place. Together with the notion that timing of BOLD responses is pretty variable over subjects (see: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=pubmed&dopt=Abstract&list_uids=15050587&query_hl=2 ), we came to the following conclusion: adding dhrf/dt regressors might help at the single subject level, but you simply add a lot of unexplained variance to your 2nd level ANOVA, and hence your F-test is less significant (although roughly showing the same regions). Which means using both betas for hrf and dhrf/dt is not very useful at the 2nd level. Instead, I stumbled upon this paper, which shows a promising aproach for 2nd level tests on 1st level betas from a hrf + dhrf/dt model: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=pubmed&dopt=Abstract&list_uids=15110015&query_hl=11 Perhaps our considerations can help you in some way. regards, Bas -------------------------------------------- Dr. S.F.W. Neggers dept. of Psychonomics,Helmholtz Institute Utrecht University Heidelberglaan 2 3584 CS, Utrecht, room 17.09 the Netherlands Tel: (+31) 30 253 4582 Fax: (+31) 30 2534511 E-mail: [log in to unmask] Web: http://www.fss.uu.nl/psn/web/people/personal/neggers/ -------------------------------------------- -----Oorspronkelijk bericht----- Van: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]Namens Claus Lamm Verzonden: maandag 10 oktober 2005 14:39 Aan: [log in to unmask] Onderwerp: [SPM] t-test on hrf for model with derivatives Dear SPM community, I recently reanalyzed data I had modeled before using the hrf only, this time using a model with hrf+temporal and dispersion derivative. According to the suggestions by Rick Henson (e.g. ftp://ftp.fil.ion.ucl.ac.uk/spm/data/rfx-multiple/rfx-multiple.htm) I set up a rfx 2nd level analysis for this. I then used an F-contrast (1 0 0; 0 1 0; 0 0 1) testing for whether any of the three regressors explains some variance in my data. In addition, I calculated a T-contrast (1 0 0) for the canonical hrf only. In general, the pattern of results looks very similar for the two contrasts. However, what I do not understand is why the t-contrast shows some voxels and clusters which fail to be significant at all in the F-contrast. Shouldn't it be the other way round (if at all)? I.e. that I see activation in the F contrast which I do not see in the T contrast (with the reason for this being that the F test also uses variance explained by the derivatives)? Thx a lot for helping me with this claus