 |
 |
Hi Stephen, Yes I did. Like I mentioned in the last post, I think the results for some of these suspicious clusters may be influenced by an outlier subject whose PEs are consistently way above or below all other subjects. I'll have to try and re-run things without that subject and see - unless you guys have any other ideas, that is. Cheers Sent from my iPhone > On Oct 24, 2014, at 4:04 AM, Stephen Smith <[log in to unmask]> wrote: > > Hi - did you demean your confound EVs? > Steve > > >> On 23 Oct 2014, at 19:14, Paul Beach <[log in to unmask]> wrote: >> >> Hi Anderson, >> >> I think I figured it out, though using a different method than fslglm. I just plugged the PEs into my spss spreadsheet that already had my nuisance EVs and did a univariate analysis (GLM) with the nuisance EVs as the only parts of the model, saved the standardized residuals, and plotted those against the EV of interest. Things look somewhat better, thought not by much. >> >> I'm more concerned now that an outlier may have lead to some of these seemingly spurious results. One subject's values are consistently way off, compared to the general trend of others (for both of their RS scan results). Removing them from the plot pretty much flattens out the linear trend, regardless of whether I'm looking at the full PEs or the residuals. >> >> In any case, here's my design.png file. >> <design.png> >> >>> On Thu, Oct 23, 2014 at 9:47 AM, Anderson M. Winkler <[log in to unmask]> wrote: >>> Hi Paul, >>> Could you show your original design and contrasts? >>> Thanks >>> Anderson >>> >>> >>>> On 23 October 2014 11:09, Paul Beach <[log in to unmask]> wrote: >>>> Thanks for the advice, Anderson. >>>> >>>> Could you possibly give me a hint on how to do that? >>>> >>>> >>>> Cheers >>>> >>>> Sent from my iPhone >>>> >>>>> On Oct 23, 2014, at 3:20 AM, Anderson M. Winkler <[log in to unmask]> wrote: >>>>> >>>>> Hi Paul, >>>>> >>>>> Yes, you'd need to take the nuisance EVs into account. Perhaps consider regressing them out in fsl_glm, then using the residuals as the dependent variable in the scatter plots. >>>>> >>>>> All the best, >>>>> >>>>> Anderson >>>>> >>>>> >>>>>> On 22 October 2014 11:00, Paul Beach <[log in to unmask]> wrote: >>>>>> FSL folks, >>>>>> >>>>>> I've obtained some seemingly significant clusters from GICA/dual regression/randomise. However, when I extract each subject's average cluster PEs from the appropriate IC from DR-stage 2 and plot them against the behavioral covariate of interest (the one included in my GLM in the first place) several of the scatter plots show no linear trends (ie the scatter is somewhat homogeneous or horizontal in distribution). >>>>>> >>>>>> Several of the clusters are in regions I would have expected, a priori, so this is a little concerning. >>>>>> >>>>>> My question: How should I interpret this? Does this suggest my results are spurious? Or is simply plotting them like this too simplistic a measure, considering I corrected some a few nuisance variables in my GLM? >>>>>> >>>>>> >>>>>> Thanks for your advice, >>>>>> Paul >> >> >> >> -- >> Paul Beach >> DO/PhD candidate - Year VI >> Michigan State University >> - College of Osteopathic Medicine >> - Neuroscience Program >> - MSU Cognitive and Geriatric Neurology Team (CoGeNT) > > > --------------------------------------------------------------------------- > 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 > --------------------------------------------------------------------------- > > Stop the cultural destruction of Tibet > > > >