Inquiry Regarding Palm Analysis Results

[qiuziru]

Dear Anderson and Palm-team,

I hope this message finds you well. I am currently using PALM to examine group*time differences in a TBSS task. My participants are divided into Group A and Group B, and they have related parameter data at time1, time2, time3, and time4. I used the following approach to run PALM (illustrated here with two participants from each of Group A and Group B):

design matrix：
time-first time-second time-third Gtime-first Gtime-second G*time-third sub1 sub2 sub3 sub4
-1 -1 -1 -1 -1 -1 1 0 0 0
0 1 0 0 1 0 1 0 0 0
0 0 1 0 0 1 1 0 0 0
1 0 0 1 0 0 1 0 0 0
-1 -1 -1 -1 -1 -1 0 1 0 0
0 1 0 0 1 0 0 1 0 0
0 0 1 0 0 1 0 1 0 0
1 0 0 1 0 0 0 1 0 0
-1 -1 -1 1 1 1 0 0 1 0
0 1 0 0 -1 0 0 0 1 0
0 0 1 0 0 -1 0 0 1 0
1 0 0 -1 0 0 0 0 1 0
-1 -1 -1 1 1 1 0 0 0 1
0 1 0 0 -1 0 0 0 0 1
0 0 1 0 0 -1 0 0 0 1
1 0 0 -1 0 0 0 0 0 1

eb:
1 1 1 1 2 2 2 2 3 3 3 3 4 4 4 4

t-contrasts:
0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0

F-contrasts:
1 1 1

palmconfig.txt:
-i myinput.nii
-m mask.nii
-d Dmatrix.csv
-t t-contrasts.csv
-f F-contrasts.csv
-eb eb.csv
-n 5000
-within
-whole
-fonly
-T
-tfce2D
-save1-p
-o myresults
-accel tail
-npc
-nouncorrected

However, the obtained results indicate that the group*time interaction effect is not significant, which is inconsistent with my expectations. To further investigate, I utilized the randomise tool for pairwise comparisons. Surprisingly, for Group A, no significant differences were observed at time2, time3, and time4 compared to the baseline (time1). In contrast, for Group B, significant differences were evident at time2, time3, and time4 compared to the baseline (time1), and the range of differences appeared to gradually expand. Additionally, I conducted a non-paired test for the baseline between Group A and Group B, and the results showed no significant differences. These findings align more closely with my initial expectations.

I would appreciate your valuable insights on whether there might be any errors or issues in my approach to running PALM. Your guidance on this matter would be highly valuable.

Thank you in advance for your assistance.

Best regards,
Ziru Qiu

[andersonwinkler]

Hi Ziru,

That can happen, particularly with TFCE. TFCE does increase power but it also alters the relationship between t and F statistics.

Having said that, you design is remarkably similar to the one used in this paper: https://doi.org/10.1126/scitranslmed.aad5651

A suggestion is that you followed a similar analysis strategy, using NPC to combine across timepoints, in a multivariate test. NPC itself is described here: https://doi.org/10.1002/hbm.23115

That said, if you decide to retain your current design (a repeated measures 2x4 anova), then you don't need the option "-npc".

Hope this helps!

All the best,

Anderson