Dear all!
We are trying to implement palm in a mixed model in order to test a BMI*time(point) interaction being BMI at baseline a continuous variable with one value per participant and time-point, considered as time since the baseline scan.
Participants had a different number of time points (1,2 or 3) and the time between scans differ across participants.
We believe that palm is an adequate tool to correct for multiple comparisons but we are doubtful regarding the correct exchangeability block that we have to set. Do we have to permute each participant observation together, as a block, but only between participants sharing the same number of observations? Are we missing something else? I attach the eb and an example of our database to illustrate our doubts.
We would greatly appreciate any input. Thank you very much in advance.
int BslTime BslTime*BMI BMI(Z) sex bslAge(Z)
sub1_tp1 1 0 0 1.66 0 1.40
sub1_tp2 1 2.80 4.65 1.66 0 1.40
sub1_tp3 1 5.80 9.62 1.66 0 1.40
sub2_tp1 1 0 0 1.16 1 0.14
sub2_tp2 1 2.78 3.23 1.16 1 0.14
sub2_tp3 1 5.56 6.45 1.16 1 0.14
sub3_tp1 1 0 0 -0.87 0 -1.10
sub3_tp2 1 2.72 -2.3 -0.87 0 -1.10
sub4_tp1 1 0.00 0.00 0.39 0 -0.88
sub4_tp2 1 2.67 1.03 0.39 0 -0.88
sub5_tp1 1 0 0 0.82 0 0.42
sub6_tp1 1 0 0 0.95 0 -0.32
Contrast 0 0 1 0 0 0 (interaction contrast)
EB
sub1_tp1 -1 1 1
sub1_tp2 -1 1 1
sub1_tp3 -1 1 1
sub2_tp1 -1 1 2
sub2_tp2 -1 1 2
sub2_tp3 -1 1 2
sub3_tp1 -1 2 1
sub3_tp2 -1 2 1
sub4_tp1 -1 2 2
sub4_tp2 -1 2 2
sub5_tp1 -1 3 1
sub6_tp1 -1 3 2
|