 |
 |
Dear Victoria, The recommended model would be, for each participant, to compute the average effect for factors 3 and 4 (with a [1 1 1 1] contrast at the first level or with ImCalc 'i1+i2+i3+i4' with the four contrast images you already have) and enter these in an unpaired two-sample t-test, collapsing over factor 2. Best regards, Guillaume. On 04/04/17 20:55, Victoria Klimaj wrote: > Hi Donald, > > Thanks for the reply! > > The full-factorial we ran had 4 factors with 2 levels each: > factor 1: participant sex (male, female), > factor 2: participant sexual orientation (heterosexual, homosexual) > factor 3: stimulus type (pictures, videos) > factor 4: stimulus preference category (preferred vs. non-preferred) > (participants were viewing images of men and women; so preferred images > were images of men for homosexual men and heterosexual women, and images > of women women for heterosexual men and homosexual women) > > First, we wanted to see whether there was a difference in how all women > vs. all men responded to the stimuli regardless of stimulus type or > stimulus preference category. In our study, we are using erotic stimuli, > so the basic question was: do women respond differently than men to > erotic stimuli? > (We included factors that were not relevant to this question as the full > factorial was also set up to test a few other hypotheses) > > We found many clusters in the male > female t-test within the full > factorial, and one cluster (in the DLPFC) in the female vs. male t-test. > > Out of curiosity, we ran a separate, basic t-test outside the full > factorial looking at female > male for the same question (group 1: > heterosexual and homosexual female participant files for both their > preferred and nonpreferred picture and video responses; group 2: > heterosexual and homosexual male participant files for both their > preferred and nonpreferred picture and video responses). We included the > same number of files as the full factorial, and the same threshold (.05 > FWE). When we ran this, the DLPFC cluster was no longer visible for > female > male, and instead we saw other clusters in the contrast. That > was what confused us--which is the more correct/most accurate contrast? > The t-test within the full factorial or outside of it? > > Hope that information helps--please let me know if there's anything that > I didn't clarify > > Kind thanks, > > Victoria > > On Mon, Apr 3, 2017 at 4:23 PM, MCLAREN, Donald > <[log in to unmask] <mailto:[log in to unmask]>> wrote: > > Please provide more details about what your models are and what you > are trying to test. > > Best Regards, > Donald McLaren, PhD > > > On Mon, Apr 3, 2017 at 4:56 PM, Victoria Klimaj > <[log in to unmask] <mailto:[log in to unmask]>> wrote: > > Hi SPM listserv, > > I found a single significant cluster for a t-test within a full > factorial (looking at one direction of one of four factors, > specified as 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1). > > However, I did not find this same cluster when I ran an > equivalent t-test using the same files outside the full > factorial. Instead, other clusters showed up. > > Does this suggest I am doing something wrong, or does it make > statistical sense that the only significant cluster in a t-test > within a full factorial might not show up in a regular t-test? > What is the reason this might happen? > > Thanks for any help! > > Best, > > Victoria > > > > > -- Guillaume Flandin, PhD Wellcome Trust Centre for Neuroimaging University College London 12 Queen Square London WC1N 3BG