AFNI 3dMVM: Interpreting Self–Other contrast and interaction with behavioral covariate

AFNI Message Board
1

Hi Gang,

I’d love to run a few things by you as I interpret a 3dMVM model.

This is a cross-sectional study in a DID patient sample. Participants completed a task in which they viewed two types of faces (self, other) and rated the familiarity of each face. The difference in familiarity ratings between self and other faces represents the magnitude of a Self Advantage in familiarity ratings (SelfAdv), where positive values indicate that the self face was rated as more familiar.

The study spanned two scanners (Scanner), and the dataset also includes some pilot data that had a slightly different design than the main study (Design).

My primary interests are:

Differences in brain activity for self vs. other faces, and

Whether this difference varies as a function of the magnitude of the Self Advantage in familiarity ratings.

To test this, I ran the following 3dMVM model:

3dMVM -prefix MVM-DID-SelfAdv -jobs 2 \
      -bsVars  "Scanner+Age+Design+SelfAdv" \
      -wsVars "Face" \
      -qVars "Age,SelfAdv" \
      -qVarCenters "42.61,-.01" \
      -SS_type 3 \
      -GES \
      -resid Residuals \
      -num_glt 9 \
      -gltLabel 1 'Self' -gltCode 1 'Face : 1*Self' \
      -gltLabel 2 'Other' -gltCode 2 'Face : 1*Other' \
      -gltLabel 3 'S-O' -gltCode 3 'Face : 1*Self -1*Other' \
      -gltLabel 4 'S*SelfAdv' -gltCode 4 'Face : 1*Self SelfAdv :' \
      -gltLabel 5 'S*SelfAdv-H' -gltCode 5 'Face : 1*Self SelfAdv : .37' \
      -gltLabel 6 'S*SelfAdv-L' -gltCode 6 'Face : 1*Self SelfAdv : -.37' \
      -gltLabel 7 'S-O*SelfAdv' -gltCode 7 'Face : 1*Self -1*Other SelfAdv :' \
      -gltLabel 8 'S-O*SelfAdv-H' -gltCode 8 'Face : 1*Self -1*Other SelfAdv : .37' \
      -gltLabel 9 'S-O*SelfAdv-L' -gltCode 9 'Face : 1*Self -1*Other SelfAdv : -.37'

First, does this overall model specification look appropriate given my main questions?

I also had a few specific interpretation questions:

1. -gltLabel 3 'S-O' -gltCode 3 'Face : 1*Self -1*Other'

Is it correct that this contrast tests the difference in brain activity between self vs. other faces, while controlling for Scanner and Design and holding Age and SelfAdv at their centered values (i.e., the sample mean)?

2. -gltLabel 7 'S-O*SelfAdv' -gltCode 7 'Face : 1*Self -1*Other SelfAdv :'

Based on examples in the 3dMVM documentation, it seems like this contrast tests whether the Self vs. Other brain activity difference varies as a function of SelfAdv. Is that the correct interpretation? If so, could you help unpack exactly what the coefficient represents?

  1. Interpretation of negative values

When I look at the results for gltLabel 7, all significant clusters have negative values.

Is the correct interpretation that as the behavioral Self Advantage decreases, the neural difference between Self vs. Other increases in those regions?

Or is there a better way to conceptualize that effect? For example, how would I determine whether the change is being driven primarily by the self condition, the other condition, or both?

Thanks so much for your time and help—I really appreciate it.

Best,
Lauren

2

Hi Lauren,

does this overall model specification look appropriate given my main questions?

What is the nature of the factor "Design"? For centering, it could be centered around the mean or a conceptually interpretable value (e.g., 0) depending on the specific questions you want to address.

In general, make sure that the inclusion of all covariates can be justified as discussed in this blogpost.

1. -gltLabel 3 'S-O' -gltCode 3 'Face : 1*Self -1*Other'

Is it correct that this contrast tests the difference in brain activity between self vs. other faces, while controlling for Scanner and Design and holding Age and SelfAdv at their centered values (i.e., the sample mean)?

Yes, that's correct.

2. -gltLabel 7 'S-O*SelfAdv' -gltCode 7 'Face : 1*Self -1*Other SelfAdv :'

Based on examples in the 3dMVM documentation, it seems like this contrast tests whether the Self vs. Other brain activity difference varies as a function of SelfAdv. Is that the correct interpretation? If so, could you help unpack exactly what the coefficient represents?

  1. Interpretation of negative values

When I look at the results for gltLabel 7, all significant clusters have negative values.

Is the correct interpretation that as the behavioral Self Advantage decreases, the neural difference between Self vs. Other increases in those regions?

Or is there a better way to conceptualize that effect? For example, how would I determine whether the change is being driven primarily by the self condition, the other condition, or both?

In this case, each face is hypothesized to be associated with Self Advantage in familiarity ratings. The estimate from -gltCode 7 shows the difference in the amount of association between the two face types. To parse the difference further, I suggest that you add one more estimate:

... -gltCode ? 1*Other SelfAdv :' \

Then you would be able to check what the sign of the association difference means by combining three estimates: the newly added one, -gltCode 4, and -gltCode 7.

Gang Chen