External Email - Use Caution
I am performing a longitudinal LME analysis in which participants were scanned anywhere from 2-6 times over the course of several years. There are 4 groups denoting participants: Controls (C, coded as 0), group 1 (G1, coded as 1), group 2, (G2, coded as 2), and group 3 (G3, coded as 3). Additionally, I have included two covariates (age and number of previous concussions). With a single random effect (intercept, all 1s), my design matrix is:
*Yij = intercept + time + age + numPrevConcussion + G1 + time*G1 +G2 + time*G2 + G3 + time*G3*
I am primarily interested in time*group interaction differences between G1 and G2 (G1 - G2) and G1 and G3 (G1 - G3).
I first tried single row contrasts separately (i.e. CM.C = [0 0 0 0 0 1 0 -1 0 0] for G1-G3, and CM.C = [0 0 0 0 0 1 0 0 0 -1] for G1 - G3), which produced no significant results. I then created a multi-row contrast of the two contrasts above (i.e. CM.C = [zeros(2,5) [-1 0 1 0 0;-1 0 0 0 1]]), which also did not reveal significant findings. Only when I created a 3-row contrast (CM.C = [zeros(3,5) [1 0 0 0 0;1 0 -1 0 0;1 0 0 0 -1]]) did I get significant results.
I'm a bit unsure how to properly interpret these contrasts and wanted to get clarification on my thinking. Is the 3-row contrast assessing the null hypothesis of no group differences in the rate of change over time among the four groups? If so, does the lack of findings in the 2-row and single row contrasts indicate that there are no interaction differences between G1 and G2, and between G1 and G3?
Thank you