3dMVM: Help with interaction

Hi!

Running a factorial analysis using 3dMVM since my groups are uneven. First off, does my interaction set-up look OK? We have a patient group were some get treatment and some get placebo. We want to see the effect of the treatment when they are presented with two stimuli (one that should be affected by the treatment). I.e. the contrast between stim1 and stim2 should vary depending on real or placebo treatment if the treatment worked.

#!/bin/bash
3dMVM -prefix Group_Results -dataTable @table.txt -jobs 10
-mask mask+tlrc
-bsVars Treatment
-wsVars Stim
-num_glt 3
-gltLabel 1 Treatment
-gltCode 1 ‘Treatment: 1real -1placebo’
-gltLabel 2 Stim
-gltCode 2 ‘Stim: 1stim1 -1stim2’
-gltLabel 3 TreatmentXStim
-gltCode 3 ‘Treatment: 1real -1placebo Stim: 1stim1 -1stim2’

The data table:

Subj Treatment Stim InputFile
S01 real stim1 stats.subject_01_REML+tlrc.BRIK’[stim1#0_Coef]’
S01 real stim2 stats.subject_01_REML+tlrc.BRIK’[stim2c#0_Coef]’
S02 placebo stim1 stats.subject_02_REML+tlrc.BRIK’[stim1#0_Coef]’
S02 placebo stim2 stats.subject_02_REML+tlrc.BRIK’[stim2#0_Coef]’

etc

What I get is the F-score for Treatment and Stim. But I guess what I’m really interested in is the Interaction? Since I want to know if they react differently in e.g. insula (stim1-stim2) depending on if they got the real treatment or not. Why don’t I get an F-score for the interaction? Can you really say that that the interaction takes the variance from the other Factors into account when you use the t-score presented?

What is really the interpretation of the F-scores and the t-score for the interaction?

Stim F: Taking in the variance explained by all the other levels there is still an effect of stim1 vs stim2?
Treatment F: similar?
Interaction t-score? Not sure what that would mean?

Thanks!

Why don’t I get an F-score for the interaction?

With the current setup in your script, you should have an F-stat for the interaction Treatment:Stim. You don’t see it in the output?

-gltCode 3 ‘Treatment: 1real -1placebo Stim: 1stim1 -1stim2’

This t-test should give you essentially the same as the F-stat for the interaction: F is the t-stat squared.

Can you really say that that the interaction takes the variance from the other Factors into account when you use the t-score presented?

The default in 3dMVM is type 3 (marginal) sum of squares, which means that each effect is computed after all other variances are accounted for.

Thanks!

Oh, so the gltCodes are not necessary, they only provide me with the t-scores…

No, I do not get an F-score. Or I don’t think I do. I get, except the t-scores;

The Intercept (how should I intepret this and can I use it for something?)

The F-scores for Treamtent and Stim (These sub-briks are called Stilm F-score and Treatment F-score).

Two sub-briks for the interaction. One is the T-test, the other only says “Interaction” or something similar and it’s without the “F-score” notation and I cannot threshold when displaying it. It simply says N/A when I try to set p to e.g. 0.05. I would guess this is the F-sub brick for interaction but it does not have the same name as the others and I cannot select a p-score… So something is strange with that sub-brick

Thanks! If needed I can send you the resulting file tomorrow if this makes little sense.

Robin

Could you post the output of the following command?

3dinfo -verb outputFrom3dMVM

Sorry, the interaction was there (sub-brick 3). But still confused about the Interaction without the label. Here is the output:


Dataset File:    bp_anaticor+tlrc
Identifier Code: XYZ_88YH5z4k40605p65tuJ85V  Creation Date: 
Template Space:  TT_N27
Dataset Type:    Func-Bucket (-fbuc)
Byte Order:      LSB_FIRST [this CPU native = LSB_FIRST]
Storage Mode:    BRIK
Storage Space:   3,456,000 (3.5 million [mega]) bytes
Geometry String: "MATRIX(3,0,0,-79.5,0,3,0,-79.5,0,0,3,-63.5):54,64,50"
Data Axes Tilt:  Plumb
Data Axes Orientation:
  first  (x) = Right-to-Left
  second (y) = Anterior-to-Posterior
  third  (z) = Inferior-to-Superior   [-orient RAI]
R-to-L extent:   -79.500 [R] -to-    79.500 [L] -step-     3.000 mm [ 54 voxels]
A-to-P extent:   -79.500 [A] -to-   109.500 [P] -step-     3.000 mm [ 64 voxels]
I-to-S extent:   -63.500 [I] -to-    83.500 [S] -step-     3.000 mm [ 50 voxels]
Number of values stored at each pixel = 10
  -- At sub-brick #0 '(Intercept) F' datum type is short:            0 to         32767 [internal]
                                        [*   0.00192918]             0 to       63.2135 [scaled]
     statcode = fift;  statpar = 1 15
  -- At sub-brick #1 'Treatment F' datum type is short:            0 to         32767 [internal]
                                      [*   0.00120118]             0 to       39.3591 [scaled]
     statcode = fift;  statpar = 1 15
  -- At sub-brick #2 'Stim F' datum type is short:            0 to         32767 [internal]
                                 [*   0.00109601]             0 to       35.9129 [scaled]
     statcode = fift;  statpar = 1 15
  -- At sub-brick #3 'Treatment:Stim F' datum type is short:            0 to         32767 [internal]
                                           [*   0.00127682]             0 to       41.8375 [scaled]
     statcode = fift;  statpar = 1 15
  -- At sub-brick #4 'Treatment' datum type is short:       -32767 to         32767 [internal]
                                    [*   0.00305185]          -100 to           100 [scaled]
  -- At sub-brick #5 'Treatment t' datum type is short:       -27952 to         32767 [internal]
                                      [*  0.000191463]      -5.35178 to       6.27368 [scaled]
     statcode = fitt;  statpar = 15
  -- At sub-brick #6 'Stim' datum type is short:       -32767 to         26305 [internal]
                               [*   0.00305185]          -100 to       80.2789 [scaled]
  -- At sub-brick #7 'Stim t' datum type is short:       -28475 to         32767 [internal]
                                 [*   0.00018289]      -5.20778 to       5.99274 [scaled]
     statcode = fitt;  statpar = 15
  -- At sub-brick #8 'TreatmentXStim' datum type is short:       -32767 to         32767 [internal]
                                         [*   0.00305185]          -100 to           100 [scaled]
  -- At sub-brick #9 'TreatmentXStim t' datum type is short:       -32767 to         26081 [internal]
                                           [*  0.000197399]      -6.46819 to       5.14838 [scaled]
     statcode = fitt;  statpar = 15

What is the TreatmentXstim sub-brick? That is not thresholdable. Each glt has one of these sub-bricks. I.e. 4, 6 and 8. Also, what is the intercept?

Thank you!

What is the TreatmentXstim sub-brick? That is not thresholdable.

Sub-brick #3 ‘Treatment:Stim F’ is the F-stat for the interaction between Treatment and Stim: the label Treatment:Stim is an R notation for interaction. What do you mean that it’s not “thresholdable?” The information from 3dinfo indicates that the sub-brick #3 is an F-stat with degrees of freedom (1, 15)

statcode = fift; statpar = 1 15

So you should be able to see p-value underneath the slider bar.

Each glt has one of these sub-bricks. I.e. 4, 6 and 8.

OK, you mean that each t-stat is associated with a specific effect estimate. Yes, for t-stat that is true. However, a composite hypothesis, the corresponding F-stat does not have a particular effect estimate, and that’s why you don’t see a separate sub-brick associated with each F-stat.

Also, what is the intercept?

It’s the overall average across all factor levels (e.g., across both treatment groups and both stimuli in your case).

I mean that sub-BRIK 4, 6 and 8 are not threshodlable. The interaction (sub-BRIK 3) is fine. 4, 6 and 8 are all the sub-BRIK right before the t-score sub-briks.

Thanks!

I mean that sub-BRIK 4, 6 and 8 are not threshodlable. The interaction (sub-BRIK 3) is fine. 4, 6 and 8
are all the sub-BRIK right before the t-score sub-briks.

For each t-test, there is an effect estimate (e.g., BOLD response in percent signal change) associated with it. And those sub-bricks #4, 6, and 8 are the effect estimates (not statistics) that show the physical measurement or brain response amplitude.