Hi - I *think* that if you want to treat this as a fully "paired" analysis then you will have to discard subjects who don't have results for all 4 conditions. If you don't discard them, and instead do an unpaired analysis (eg a simple ANOVA), then you will probably lose sensitivity overall. Cheers. On 27 Apr 2012, at 12:37, Anne Klomp wrote: > Hi all, > > I've found similar topics but none answered my specific problem. > > I've done a study in which I've scanned 12 subjects on 3 different days, and 2 scan sessions on each day. > So in total 6 scans per subject. > > There are 3 baseline sessions, all followed by a 'challenge' scan where the subjects were under influence of a drug challenge or placebo challenge. > > I started with doing paired t-tests on the baseline scans and their following challenge scan to see the effect of the drug or placebo on each separate day. > > Now however, I want to compare these 'drug effects' to see if they are reliable in time (and different from placebo effect). > For this, I want to do a third level analysis: a t-test on the results of the paired t-tests. > (So, what is the drug effect of day 1 and how does this differ from the drug or placebo effect on day 2) > > The problem is as follows: > I have some missing data sets in one of my tasks because of poor task performance. > > I understand that this is not a problem with the paired or 'triple' t-test or in a repeated measures ANOVA because the EV's with the subjects means will correct for this. > > But what if I use the cope files of the paired t-tests for the t-test to see differences in the drug effect between days? > Can I only use those subjects from which I have all scans, so those that were both in the paired t-test of day 1 and in the paired t-test of day 2? > > And what if I would do a 2x2 ANOVA, with the four scans of 2 consecutive days and thus looking at pre/post challenge and day1/day2 with of course special interest in the interaction term? > Do I then need to include only the subjects from which all four scans are present or could I also include subjects with missing scans since you also define EVs with the subjects means and thus 'tell' the program which scans are missing? > > How does FSL deal with the within-subject and between-subject variances in these cases? > > Many thanks in advance for clarifying this! > --------------------------------------------------------------------------- Stephen M. Smith, Professor of Biomedical Engineering Associate Director, Oxford University FMRIB Centre FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [log in to unmask] http://www.fmrib.ox.ac.uk/~steve ---------------------------------------------------------------------------