 |
 |
Hi Jie-Yu, Please, see below: On 17 September 2015 at 20:33, Jie Yu <[log in to unmask]> wrote: > Dear Expert > > I am using PALM to address the different variances within group. Could you > please check the following designs? > > So there are 4 groups: male patient (EV1) male control (EV2) female > patient (EV3) female control (EV4) > > I then get the mean of ALL participants' ages and demean each participant > with the resulting (EV5) > > I then use GLM_gui to generate .con, .grp and .mat files for input into > PALM. > > (1) Design 1 > > Group EV1 EV2 EV3 EV4 EV5 > 1 1 0 0 0 20 > 2 0 1 0 0 30 > 3 0 0 1 0 40 > 4 0 0 0 1 50 > > Contrasts: EV1 EV2 EV3 EV4 EV5 > group by sex interaction > 1 -1 -1 1 0 > -1 1 1 -1 0 > group effect > 1 -1 1 -1 0 > -1 1 -1 1 0 > sex effect > 1 1 -1 -1 0 > -1 -1 1 1 0 > age effect > 0 0 0 0 1 > 0 0 0 0 -1 > > Looks fine. > (2) Design 2 > > This time, I'd like to know group x age interaction. > > I think it's impossible to find out both group effect and group x age > interaction in one model, so it's necessary for me to use 2 models? > > Please do let me know how to test group x age interaction in PALM…. > It is possible to test all in a single model: split the age in 4 groups, and create contrasts that follow the same logic as you already have: Group by age: 0 0 0 0 1 -1 1 -1 0 0 0 0 -1 1 -1 1 Sex by age: 0 0 0 0 1 1 -1 -1 0 0 0 0 -1 -1 1 1 Alternatively, the design can be coded as: EV1: intercept EV2: code as +1 and -1 for group, then mean-center EV3: code as +1 and -1 for sex, then mean-center EV4: product of EV2 and EV3, for the interaction. EV5: age, mean-centered EV6: product of EV2 and EV5, for the interaction between age and group EV7: product of EV3 and EV5, for the interaction between age and sex EV8: product of EV4 and EV5, for the interaction between age and the further interaction between age and sex (perhaps difficult to interpret this one, you may not want to test, but still include the EV in the model). The contrasts are then +1 or -1 for the respective EVs to be tested, with zeroes for all others. > > Finally, I think the code for PALM in MATLAB is this: > > palm -i 4d.nii -d design.mat -t design.con -eb design.grp -vg auto -ise -T > -m Mask.nii -o Output > Looks perfect. > > Has PALM underwent any dramatic change recently? (I used PALM in June, now > I am redoing the analysis…) > It has evolved but the features you are using in the command above should be same. All the best, Anderson > > > > > Thanks so much!!! >