 |
 |
Hi Josh, Yes, sure. In this case it's simple: split the 4D file with the regressor in 2 parts, one for each group (use fslsplit or fslroi for this; fslroi is simpler). Then multiply one of them by -1, and concatenate them back (with fslmerge). All the best, Anderson On 10 October 2014 19:09, Joshua Balsters <[log in to unmask]> wrote: > Hi Anderson > > Is it possible to do that with a voxel wise EV as well? > > Best > Josh > > > On Friday, October 10, 2014, Anderson M. Winkler <[log in to unmask]> > wrote: > >> Hi Josh, >> For the interaction, create a new EV that is the elementwise product of >> EV1 and the continuous EV that is to be tested. >> The contrast should contain 1 or -1 for this new EV, and 0 for all others. >> All the best, >> Anderson >> >> >> On 10 October 2014 16:22, Joshua Balsters <[log in to unmask]> wrote: >> >>> Thank you again for you help. One more question if that's ok ... >>> >>> Using the design matrix you proposed how would I code a contrast to look >>> at the interaction between group and one of my continuous covariates such >>> as age? >>> >>> Thanks >>> Josh >>> >>> On 9 October 2014 09:05, Anderson M. Winkler <[log in to unmask]> >>> wrote: >>> >>>> Hi Josh, >>>> No need to use -D (and using it may cause rank-deficiency). >>>> All the best, >>>> Anderson >>>> >>>> >>>> On 8 October 2014 21:29, Joshua Balsters <[log in to unmask]> wrote: >>>> >>>>> Thanks so much for your help. One last question, would you use -D with >>>>> this new design matrix or should it not make any difference? >>>>> >>>>> Best >>>>> Josh >>>>> >>>>> On 8 October 2014 21:26, Anderson M. Winkler <[log in to unmask]> >>>>> wrote: >>>>> >>>>>> Hi Joshua, >>>>>> >>>>>> 1) Yes, you can just drop the "-e design.grp" and all subjects will >>>>>> automatically fall into the same group. >>>>>> >>>>>> 2) The design is fine, but the contrast would be different and less >>>>>> intuitive. Instead, a suggestion is: >>>>>> >>>>>> EV1: Code as +1 or -1 for each group, and mean-center if the number >>>>>> of subjects isn't the same on each; >>>>>> EV2-EV9: code as 0 or 1 for each centre (no need to omit any centre; >>>>>> the design won't be rank-deficient this time); >>>>>> EV10-EV12: use for age, FIQ and FD, mean centered. >>>>>> >>>>>> The contrasts are then: >>>>>> C1 (A>B): 1 0 0 0 0 0 0 0 0 0 0 0 >>>>>> C2 (A<B): -1 0 0 0 0 0 0 0 0 0 0 0 >>>>>> C3 (global mean): 0 1/8 1/8 1/8 1/8 1/8 1/8 1/8 1/8 0 0 0 >>>>>> C4 (mean A): 1/2 1/8 1/8 1/8 1/8 1/8 1/8 1/8 1/8 0 0 0 >>>>>> C5 (mean B): -1/2 1/8 1/8 1/8 1/8 1/8 1/8 1/8 1/8 0 0 0 >>>>>> >>>>>> 3) sure! >>>>>> >>>>>> All the best, >>>>>> >>>>>> Anderson >>>>>> >>>>>> >>>>>> On 8 October 2014 18:14, Joshua Balsters <[log in to unmask]> >>>>>> wrote: >>>>>> >>>>>>> Hi Anderson >>>>>>> >>>>>>> Thank you so much for your quick reply. >>>>>>> >>>>>>> 1) If I don't use the -e flag then will it automatically permute >>>>>>> across all subjects or do I need to change the .grp file to all 1's? >>>>>>> >>>>>>> 2) I've attached an image of the design matrix but I can give you >>>>>>> the exact variables if you'd prefer. I've modelled all the centres except 1 >>>>>>> so the overall mean isn't modelled. This was to avoid the message for rank >>>>>>> deficiency of the model. >>>>>>> >>>>>>> 3) thanks for this! I'll use these as well. >>>>>>> >>>>>>> Best >>>>>>> Josh >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 8 October 2014 18:30, Anderson M. Winkler <[log in to unmask] >>>>>>> > wrote: >>>>>>> >>>>>>>> Hi Josh, >>>>>>>> >>>>>>>> For (1): In the group indicator, use all ones. This column in >>>>>>>> randomise is related to exchangeability groups, and here you want all >>>>>>>> subjects (all!) in the same group. >>>>>>>> >>>>>>>> For (2): It sounds the overall mean is modelled across the >>>>>>>> regressors for the scan centres, so demeaning shouldn't be necessary. But >>>>>>>> to be sure, please, include your design in the email. >>>>>>>> >>>>>>>> For (3): There are extra flags: --vxl and --vxf, one to specify the >>>>>>>> 4D file, the other with the position they occupy in the design matrix. >>>>>>>> >>>>>>>> Hope this helps! >>>>>>>> >>>>>>>> All the best, >>>>>>>> >>>>>>>> Anderson >>>>>>>> >>>>>>>> >>>>>>>> On 8 October 2014 15:32, Joshua Balsters <[log in to unmask]> >>>>>>>> wrote: >>>>>>>> >>>>>>>>> Hi >>>>>>>>> >>>>>>>>> I wanted to check a couple things about my use of Randomise... >>>>>>>>> >>>>>>>>> My design matrix is a two sample t-test with covariates of no >>>>>>>>> interest (8 categorial regressors for different scan centres and 3 >>>>>>>>> continuous regressors which I've demeaned before entering them into the >>>>>>>>> design matrix). The inputs are Z-transformed resting state data. >>>>>>>>> >>>>>>>>> 1) I get an error saying "Design matrix uses different groups (for >>>>>>>>> different variances), but these do not contain separable EVs for the >>>>>>>>> different groups... Ignore if you're intending to use the groups file to >>>>>>>>> define exchangeability blocks for randomise". If I understand correctly >>>>>>>>> this is for repeated measures rather than comparing two independent groups. >>>>>>>>> Should I ignore this message? >>>>>>>>> >>>>>>>>> 2) as I don't model a constant should I be using the -D flag as >>>>>>>>> well? >>>>>>>>> >>>>>>>>> 3) for a voxel-wise regressor like grey matter volume, is it >>>>>>>>> enough to specify the file when setting up the design matrix or are then >>>>>>>>> any extra flags I need to use when running randomise? >>>>>>>>> >>>>>>>>> Thanks for your help! >>>>>>>>> Josh >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> >>>>>>> >>>>>>> -- >>>>>>> Joshua Henk Balsters >>>>>>> Postdoctoral Fellow >>>>>>> >>>>>>> Neural Control of Movement Lab >>>>>>> Department of Health Sciences and Technology >>>>>>> ETH Zurich, Switzerland >>>>>>> >>>>>>> Y36 M 12 >>>>>>> Winterthurerstrasse 190 >>>>>>> 8057 Zuerich >>>>>>> >>>>>>> Tel: +41 44 63 56101 >>>>>>> email: [log in to unmask] >>>>>>> http://angelaschoellig.academia.edu/JoshuaBalsters >>>>>>> >>>>>> >>>>>> >>>>> >>>>> >>>>> -- >>>>> Joshua Henk Balsters >>>>> Postdoctoral Fellow >>>>> >>>>> Neural Control of Movement Lab >>>>> Department of Health Sciences and Technology >>>>> ETH Zurich, Switzerland >>>>> >>>>> Y36 M 12 >>>>> Winterthurerstrasse 190 >>>>> 8057 Zuerich >>>>> >>>>> Tel: +41 44 63 56101 >>>>> email: [log in to unmask] >>>>> http://angelaschoellig.academia.edu/JoshuaBalsters >>>>> >>>> >>>> >>> >>> >>> -- >>> Joshua Henk Balsters >>> Postdoctoral Fellow >>> >>> Neural Control of Movement Lab >>> Department of Health Sciences and Technology >>> ETH Zurich, Switzerland >>> >>> Y36 M 12 >>> Winterthurerstrasse 190 >>> 8057 Zuerich >>> >>> Tel: +41 44 63 56101 >>> email: [log in to unmask] >>> http://angelaschoellig.academia.edu/JoshuaBalsters >>> >> >> > > -- > Joshua Henk Balsters > Postdoctoral Fellow > > Neural Control of Movement Lab > Department of Health Sciences and Technology > ETH Zurich, Switzerland > > Y36 M 12 > Winterthurerstrasse 190 > 8057 Zuerich > > Tel: +41 44 63 56101 > email: [log in to unmask] > http://angelaschoellig.academia.edu/JoshuaBalsters >