 |
 |
Dear Bianca, (see below) B. de Haan wrote: > Dear SPM list, > > So far I've had no replies yet to my question outlined below and I have no > idea how to solve this problem. > > I still do not seem to be able to perform RFX 2-sample t-tests in spm2 anymore. > > If anybody has ideas concerning things I could try to (hopefully) solve this > problem, I would be extremely greatful. > > I don't know if it's relevant here, but I'm using spm2 with Matlab (tried > both R13 and R14) on a Windows 2000 PC. > > Many thanks in advance, > > Bianca de Haan > > > On Fri, 18 Jan 2008 16:30:00 +0000, B. de Haan > <[log in to unmask]> wrote: > >> The problem described below persists (I have no also reproduced the same >> error on every computer I could get my hands on), however, I have managed to >> narrow the potential source of the problem down a bit: >> >> The 2-sample t-test aims to compare group 1 (10 first level .con images) >> with group 2 (10 first level .con images). >> >> When I run this second level 2-sample t-test it produces valid looking beta >> images for both groups (beta1 for group1 and beta2 for group2). The produced >> con image also looks good and corresponds to the difference between beta1 >> and beta2. The resMS image also looks believable. However, the spmT image >> produced from the comparison of the con image to the resMS image contains >> only the value 0, which should not be possible. >> >> Hopefully this additional information will allow someone to find out what >> goes wrong here.... Unfortunately it is very important that I get this >> 2-sample t-test to work again on spm2. >> >> Many thanks in advance, >> >> Bianca de Haan >> >> >> >> On Thu, 17 Jan 2008 18:29:33 +0000, B. de Haan >> <[log in to unmask]> wrote: >> >>> Dear SPM-list, >>> >>> I am having some problems when trying to reanalyze old data with spm2: >>> >>> two-sample t-tests between two groups suddenly seem to produce only empty >>> spmT images (the whole spmT_000x image is 0). I've been using spm2 for years >>> and have never encountered this problem. >>> >>> >>> 1) This problem is specific to two-sample t-tests between the 2 groups; >>> one-sample t-tests in spm2 on each of the 2 groups work fine and produce the >>> same results they did a year ago. >>> >>> 2) This problem is also specific to spm2; the same two-sample t-test between >>> the 2 groups using spm5 works fine and produces the same results as I got a >>> year ago with spm2. >>> >>> 3) 2-group ANOVA in spm2 (should produce the same results as 2-sample >>> t-test) again only produces empty spmT images. >>> >>> 4) I have re-downloaded and installed spm2 and tried the two-sample t-test >>> on both Matlab6.5 and Matlab7.1. All to no avail. >>> >>> 5) The 2-sample t-test design matrix from the current analysis looks >>> radically different from the design matrix from the analysis I performed a >>> year ago and I do not seem to be able to reproduce the design matrix from a >>> year ago in spm2 anymore (I tried every combination of the 2-sample t-test >>> settings possible). This may be a clue that there could be a bug in the specification of the nonsphericity options in a version of SPM2. SPM uses the nonsphericity estimates (ie. error covariances) to `whiten' the design before doing Ordinary Least Squares. SPM then shows the whitened design matrix. So assuming everything else is constant (ie. you have the same data - exactly the same scans), then this could be the problem. However, as mentioned above, if I do the same 2-sample >>> t-test in spm5, I obtain the same results as I did a year ago again, >>> therefore I feel confident that my 'old' results are valid. >>> >>> >>> I am pretty much at my wits end as to why I no longer can perform 2-sample >>> t-tests in spm2. I would really like to stick with spm2 for this old dataset >>> (am just trying to do a few small additional analyses in response to >>> reviewer's comments). I don't know why a reviewer would have a problem with you using SPM5 for this. I'm afraid we don't have the manpower to fix old versions of SPM. I advise you to use SPM5. If you'd like my input in your responses to reviewers here I'd be happy help. Best, Will. Does anybody have any idea what might be going on here >>> and how to solve this? >>> >>> Many thanks in advance, >>> >>> Bianca de Haan >>> ======================================================================== >> ======================================================================== > > -- William D. Penny Wellcome Trust Centre for Neuroimaging 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/