3dMEMA question

AFNI Message Board
1

Hello,
I’m trying to run 3dMEMA for the first time and need some clarification. I have three groups, one condition.

  1. If I understand correctly, I need to use a covariate file because there are more than 3 groups, correct?

  2. If so, are you coding group membership in the covariate file or contrasts?

  3. If you are coding contrasts, will 3dMEMA then just run pairwise comparisons or will it also give me an omnibus effect?

Thanks!
Amanda

2

Amanda,

  1. If I understand correctly, I need to use a covariate file because there are more than 3 groups, correct?

Yes, that’s correct.

  1. If so, are you coding group membership in the covariate file or contrasts?

Through a covariate file using option -covariates.

  1. If you are coding contrasts, will 3dMEMA then just run pairwise comparisons or will it also give me an omnibus effect?

Check out this web page: https://afni.nimh.nih.gov/sscc/gangc/MEMA.html

3

I’ve looked at that page and I don’t understand how the dummy coding works in this context. Is the dummy coding supposed to code group membership or a contrast of interest. Is each row a subject?
Can you provide more explanation?

4

The following page explains various coding strategies and their respective interpretations:

http://www.ats.ucla.edu/stat/r/library/contrast_coding.htm

5

So I understand the dummy coding now and it should be the following for my three groups:
grp2 grp3
grp1 0 0
grp2 1 0
grp3 0 1

For the 3MEMA format, do you list all the files under the same set? How would it know who is in which group?
Also how is this method different from doing 3 separate 3dMEMAs , one for grp1 vs grp2, one for grp1 vs grp3, and one for grp2 vs grp 3?

Thanks!

6

For the 3MEMA format, do you list all the files under the same set? How would it know who is in which group?

If you code your subjects via the following fashion:

   grp2 grp3

grp1 0 0
grp2 1 0
grp3 0 1

you should assign “0 0” for any subject in group1, “1 0” for group2, and “0 1” for group3. Also you need to label the two columns as something like group2-group1 and group3-group1, respectively.

Also how is this method different from doing 3 separate 3dMEMAs , one for grp1 vs grp2, one for grp1 vs grp3, and one for grp2 vs grp 3?

Not exactly the same, but very similar to each other. The reason is that a contrast in a model with all the three groups is tested with the residuals across all the three groups, instead of two as in the contrast between those two groups.