Group Level analysis with between subjects variable that varies within condition

AFNI Message Board
1

Hello!

Hope your week is going well. I am trying to run a group level analysis where I have a within subjects variable (Reputation) and a between subjects variable (Aggression), but my between subjects variable also varies within subjects (Reputation). Essentially, I have different reputations (Mean, Nice, UnpMean, UnpNice) and there is an average aggression value for each of those (aggression to mean condition, aggression to nice condition, etc.).

Typically, when I’ve run these analyses in the past, my between subjects variable was something like age which wouldn’t vary within conditions. So I would set up the GLT like so

Reputation: 1Mean Age :
Reputation: 1Mean - 1*Nice Age :

However, in this instance, it seems to me that this would not accomplish the same goal as:

Reputation: 1Mean Aggression to mean :
Reputation: 1Mean x aggression to mean - 1*Nice x aggression to nice.

How can I set up my GLT to accomplish this aim? I have pasted a subset of my script below. Thank you!

3dMVM
-prefix 3dMVM.Aim1FBxAgg.062222
-jobs 11
-bsVars ‘Aggression’
-qVars ‘Aggression’
-wsVars ‘Reputation’
-num_glt 13
-gltLabel 1 FB.Mean-Nice -gltCode 1 ‘Reputation : +1Mean -1Nice’
-gltLabel 2 FB.Uncertain-Mean -gltCode 2 ‘Reputation : +0.5UnpNice +0.5UnpMean -1Mean’
-gltLabel 3 FB.Uncertain-Nice -gltCode 3 'Reputation : +0.5UnpNice +0.5UnpMean -1Nice’
-gltLabel 4 FB.AllCertain -gltCode 4 ‘Reputation : +1Mean +1Nice’
-gltLabel 5 FB.AllCertain-Uncertain -gltCode 5 ‘Reputation : +1Mean +1Nice -1UnpNice -1UnpMean’
-gltLabel 6 FB.Mean-Nice_Agg -gltCode 6 ‘Reputation : +1Mean -1Nice Aggression : ’
-gltLabel 7 FB.Uncertain-Mean_Agg -gltCode 7 ‘Reputation : +0.5UnpMean +0.5UnpNice -1Mean Aggression : ’
-gltLabel 8 FB.Uncertain-Nice_Agg -gltCode 8 'Reputation : +0.5UnpMean +0.5UnpNice -1Nice Aggression : ’
-gltLabel 9 FB.Nice_Agg -gltCode 9 ‘Reputation : +1Nice Aggression : ’
-gltLabel 10 FB.Mean_Agg -gltCode 10 'Reputation : +1Mean Aggression : ’
-gltLabel 11 FB.Uncertain_Agg -gltCode 11 ‘Reputation : +1UnpMean +1UnpNice Aggression : ’
-gltLabel 12 FB.AllCertain_Agg -gltCode 12 ‘Reputation : +1Mean +1Nice Aggression : ’
-gltLabel 13 FB.AllCertain-Uncertain_Agg -gltCode 13 ‘Reputation : +1Mean +1Nice -1UnpMean -1UnpNice Aggression : ’
-dataTable
Subj Reputation Aggression InputFile
s1562 Nice 2.86 $subjdata/s1562/s1562.results.IndividualLevel062122/stats.s1562+tlrc’[FB.Nice#0_Coef]’
s1562 Mean 4.83 $subjdata/s1562/s1562.results.IndividualLevel062122/stats.s1562+tlrc’[FB.Mean#0_Coef]’
s1562 UnpNice 3.5 $subjdata/s1562/s1562.results.IndividualLevel062122/stats.s1562+tlrc’[FB.UnpNice#0_Coef]’
s1562 UnpMean 3.25 $subjdata/s1562/s1562.results.IndividualLevel062122/stats.s1562+tlrc’[FB.UnpMean#0_Coef]’
s1571 Nice 2.36 $subjdata/s1571/s1571.results.IndividualLevel062122/stats.s1571+tlrc’[FB.Nice#0_Coef]’
s1571 Mean 4.54 $subjdata/s1571/s1571.results.IndividualLevel062122/stats.s1571+tlrc’[FB.Mean#0_Coef]’
s1571 UnpNice 3.0 $subjdata/s1571/s1571.results.IndividualLevel062122/stats.s1571+tlrc’[FB.UnpNice#0_Coef]’
s1571 UnpMean 2.7 $subjdata/s1571/s1571.results.IndividualLevel062122/stats.s1571+tlrc’[FB.UnpMean#0_Coef]’
s1577 Nice 3.17 $subjdata/s1577/s1577.results.IndividualLevel062122/stats.s1577+tlrc’[FB.Nice#0_Coef]’
s1577 Mean 4.04 $subjdata/s1577/s1577.results.IndividualLevel062122/stats.s1577+tlrc’[FB.Mean#0_Coef]’
s1577 UnpNice 2 $subjdata/s1577/s1577.results.IndividualLevel062122/stats.s1577+tlrc’[FB.UnpNice#0_Coef]’
s1577 UnpMean 3 $subjdata/s1577/s1577.results.IndividualLevel062122/stats.s1577+tlrc’[FB.UnpMean#0_Coef]’ \

2

Can you show the contingent table for the two factors of Reputation and Aggression? And what are your effects of interest?

3

Hi Gang,

Hopefully what I have below is what you’re asking for!

For reputation, there are three levels (Nice, Mean, Unpredictable) and for aggression, it is a continuous variable, but there will be a measurement for Nice, Mean and Unpredictable. So average aggression to Nice, average aggression to Mean, and average aggression to Unpredictable.

My effects of interest are brain response to nice condition x aggression, brain response to mean condition x aggression, and brain response to unpredictable condition x aggression.

Please let me know if I’m off the mark and you need something else. Thank you!

4

First of all, if I understand your data structure accurately, I believe that “Aggression” is a within-subject quantitative variable because it varies within subject. Second, use 3dLMEr because 3dMVM cannot handle within-subject quantitative variable. Lastly, does Aggression correlate with Reputation to some extent? If so, you may want to center Aggression for each level of Reputation (remove the mean Aggression per each Aggression level from each subject).

5

Hi Gang,

Thank you, yes you understand it correctly! And it does correlate with reputation for some subjects. And just to clarify that be the overall mean aggression (e.g., take every subject’s average aggression across conditions and use that to demean?).

6

just to clarify that be the overall mean aggression (e.g., take every subject’s average aggression across conditions and use that to demean?)

No, remove the mean (across subjects for each condition) from each subject. Hope this is clear.

7

Yes it is, thank you!

8

Hi Gang,

One other quick question. If I remove the mean (across subjects for each condition) from each subject, will 3dLME then demean a second time while it’s running? Thank you!

9

will 3dLME then demean a second time while it’s running?

Yes, it would. To avoid the problem, add the following to your 3dLME script

-qVarCenters 0 \

10

Thank you!