I’m having some issues with my 3dLME command and would be grateful for some guidance. The design has:
1 between-subject factor “Group” with two levels: “control”, “patient”
1 within-subject factor “Category” with two levels: “shape”, “iaps”
1 within-subject covariate “Corr”
When I run file_tool on the script I get 0 bad characters.
When I run the script I get the following output:
During startup - Warning message:
Setting LC_CTYPE failed, using “C”
Read 515 items
Error in if (len%%wd != 0) errex.AFNI(paste(“The content under -dataTable is not rectangular !”, :
argument is of length zero
Calls: process.LME.opts
Execution halted
Irene, one possibility is the label for the input file column. Change “Inputfile” to “InputFile”, and see how it goes. Currently the program is quite finicky about the letter case.
Of all the synthax variants I tried I hadnt thought of the space before the colons because with 3dmvm it works fine without the space.
The script now run uneventfully.
I wanted to ask you one last question, regarding whether you think the analysis is setup correctly. We are interested in looking at group differences in BOLD during “iaps” given the covariate of interest “Corr” (i.e. Do BOLD and covariate correlate differently across groups?). Therefore the glt 4.
Of all the synthax variants I tried I hadnt thought of the space before the colons because with 3dmvm
it works fine without the space.
3dLME and 3dMVM share exactly the same syntax for GLT and GLF specifications.
I wanted to ask you one last question, regarding whether you think the analysis is setup correctly. We are
interested in looking at group differences in BOLD during “iaps” given the covariate of interest “Corr” (i.e.
Do BOLD and covariate correlate differently across groups?).
The model seems fine in general. Is ‘Category’ a within-subject variable? Do the groups differ substantially in “Corr”?
The model seems fine in general. Is ‘Category’ a within-subject variable?
Yes. In reality it has 4 levels.
Do the groups differ substantially in “Corr”?
I compared the scores between groups for each level in the within-subject variable “Category”. When doing that I find no significant difference between the groups, except for one level (0.04).
When comparing the scores between groups across all within-subject levels, I find a significant difference between the groups.
How do you think is best to proceed? Shall I include -qVarsCenters ‘1.0532’, with 1.05532 as the overall average across “Group” and “Category”, or shal I run it without qVarsCenters?
Also, would it be justified to run independent 3dmvms with covariate on each within-subject level separately instead of having them all in the same model? Our main interest is to see group differences in the relationship between BOLD & covariate at each level, and not necessarily to compare the levels.
would it be justified to run independent 3dmvms with covariate on each within-subject
level separately instead of having them all in the same model?
Yes, I think it would be more preferable to analyze the data at each level of the within-subject factor using 3dMVM or even 3dttest++. Regarding centering, think about how you would interpret the analysis results: Do the groups intrinsically differ in terms of the covariate? If they do, center around a value (mean or some meaningful value) within each group; otherwise simply center around an overall value.
resulted in a significant cluster. If I understand correctly the glt4 contrast compares the beta-covariate correlation between groups.
When I pulled the betas from the significant cluster and run the correlation analysis in SPSS there were no significant correlations
for either groups. Could you help me understand why?
Also, following your suggestion, I also tried to run the analysis on one of the levels of the within-subject factor and I get “model test failed!”.
Do you see any mistakes in my script/table? I have N=25 participants per group.
If ‘Corr’ is a within-subject quantitative variable, consider
-ranEff ‘~1+Corr’ \
If I understand correctly the glt4 contrast compares the beta-covariate correlation between groups.
More accurately, GLT #4 looks for the difference of association between beta and ‘Corr’ between the two group when Category is fixed at ‘iaps’.
When I pulled the betas from the significant cluster and run the correlation analysis in SPSS there
were no significant correlations for either groups. Could you help me understand why?
I’m not familiar with SPSS. If ‘Corr’ is a within-subject quantitative variable, how do you handle it in SPSS? You may try to extract one voxel, and try it in SPSS and see if you can duplicate the result from 3dLME.
More accurately, GLT #4 looks for the difference
of association between beta and ‘Corr’ between the
two group when Category is fixed at ‘iaps’.
Am I right in interepting association as correlation? If not could you please explain what you mean with association?
My main struggle is to find ways to make this 3-way interaction interpretable.
How would you suggest to proceed in order to understand what this significant cluster reflects?
If ‘Corr’ is a within-subject quantitative variable, consider
-ranEff ‘~1+Corr’ \
Unfortuantely the model still failed. I noticed that when adding an additional within-subject level in the table the same script run fine.
This setup wont be the appropriate way of performing the analysis since the Covariate is within-subject but I thought that this information might help with figuring out where the error might be. Any ideas why?
Am I right in interepting association as correlation?
Yes, they are the same thing.
My main struggle is to find ways to make this 3-way interaction interpretable.
How would you suggest to proceed in order to understand what this significant cluster reflects?
For the controls I get basically all the clusters I had for the initial 3-way GroupiapsCorr interaction. The t-scores are all negative for these clusters.
For the patients I dont get anything.
Can I conclude that the controls have a significant negative correlation between iaps and Corr, whereas the patient dont?
Can I conclude that the controls have a significant negative correlation between iaps and Corr, whereas the patient dont?
Your conclusion is about right, but your wording can be a little bit more accurate. First, the negative correlation itself between iaps and Corr in the control group is not necessarily “significant” (the value can be large or small). Instead, you have strong statistical evidence (so-called statistical significance) for the negative correlation between iaps and Corr in the control group. Second, it is not necessarily true that the patient group does not have significant negative correlation between iaps and Corr; rather, you don’t have strong statistical evidence (e.g., statistical significance) for the correlation between iaps and Corr in the patient group. Therefore, your description “For the patients I dont get anything” is not really correct. In other words, the correlation in patients is likely not all zero, but you cannot differentiate them from zero with strong evidence or confidence. Lastly, you may want to interpret these two results together with the GLT #4:
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.