I have ran a 3dLME on my data. My factorial design includes a within subject factor (Agent), a between subject factor (Order) and a Covariate for which subjects have a value for each of the 2 levels of the within subject factors. Now I have two questions that are crucial to correctly report the results of this analysis:
When a correspondent analysis (Mixed Linear Model) is carried in SPSS, one can test different models without random effects or with different combination of random effects (only slope, only intercept, both) and the SPSS output will report certain criteria to predict the best model (AIC, BIC). Is it possible to do something similar in AFNI? Also, why it does not let me run the analysis without random effects specified (I have tried both: -ranEff NA and no ranEFF option at all)?
How can I correctly report the betas that are taking the covariate in account?
Model comparison is currently not implemented in 3dLME due to the complexity at the voxel level.
Also, why it does not let me run the analysis without random effects specified (I have tried both: -ranEff NA and no ranEFF option at all)?
The current implementation in 3dLME requires a minimum setting with a random intercept. For voxel-wise analysis, it’s hard, it not impossible, to vary the model or the random-effects components from voxel to voxel.
How can I correctly report the betas that are taking the covariate in account?
How are you modeling the random effects? With a random intercept or both random intercept and slope? Does the average covariate value differ much between the two levels of the within-subject factor?
How are you modeling the random effects? With a random intercept or both random intercept and slope?
I have started with both random intercept and slope, but then I have tried also with each on its own (only slope, only intercept) and the one with only intercept seems to be a bit different (less activation in all clusters) from the other two which do not differ. How do I peek the ‘right’ one? Should be my decision theory-driven? Or, for instance, can I extract the betas from a significant cluster and run a LME in SPSS to predict the best model and generalize that approach to all voxels?
Does the average covariate value differ much between the two levels of the within-subject factor?
Not really, I have run a t-test and they don’t differ (p=0.4).
The issue of random intercept and slope: With voxel-wise analysis, it’s hard (and not worthwhile) to fret about the possibility that one model is better than the other at some voxels and vice versa for the other way around. So, forget about model tuning and comparison, just adopt the model with both random intercept and random slope, and be done with it.
Not really, I have run a t-test and they don’t differ (p=0.4).
Forget about the p-value obsession. Think about the crucial question: do you have some prior knowledge as to whether one condition is expected to have a higher covariate value than the other condition? If so, it may make sense to perform centering for the covariate within each condition. If not, do global centering.
In case my prior knowledge would bring me to assume that the covariate would be affected also by the between subject factor, should I do demeaning 4 times?
In case my prior knowledge would bring me to assume that the covariate would be affected
also by the between subject factor, should I do demeaning 4 times?
I guess that, after I add the demeaned data into the table, I don’t need to use the qVarCenters option, right? (Which means that it will still use the global mean as centering value).
I have two related questions:
How do I test the difference between levels of each factor? My design is Agent (within: A,B) x Order (between: 1,2).
Order 1: A1 B1
Order 2: B2 A2
Should I be interested in comparing A1vsB2 and B1vsA2, how should I proceed? With two independent 3dMVM? (Since ttests is not possible because A1 and B2 are different subjects?).
I am interested also in the conjunctions of the different conditions. In this case, how should I proceed? I guess I first need to get 4 contrasts between each condition and baseline (how?) and then perform the conjunction with the step function? Is that correct?
after I add the demeaned data into the table, I don’t need to use the qVarCenters option, right?
Right. Or you could simply set it as ‘qVarCenters 0’.
Should I be interested in comparing A1vsB2 and B1vsA2, how should I proceed?
If those contrast make sense to you, do it with 3dttest++ as a two-sample t-test (the covariate can be added too).
I am interested also in the conjunctions of the different conditions. In this case, how should I proceed?
I guess I first need to get 4 contrasts between each condition and baseline (how?) and then perform
the conjunction with the step function? Is that correct?
The conjunction among those 4 conditions? If so, what you’re describing is the conventional approach, but personally I’m not a fan of dichotomization.
If those contrast make sense to you, do it with 3dttest++ as a two-sample t-test (the covariatecan be added too).
If I understood correctly, I cannot do that, since each group (e.g A1 and B2) includes different subjects and will have a different covariate file. 3dttest++ do not handle two Covariate files, right?
The conjunction among those 4 conditions? If so,
what you’re describing is the conventional
approach, but personally I’m not a fan of
dichotomization.
I see. What would the glt code look like for such contrast against baseline?
each group (e.g A1 and B2) includes different subjects and will have a different covariate file.
What do you mean by different covariate file? Is it about the same variable? If so, why cannot you put the covariate values in one file for 3dttest++?
As for the conjunction analysis, are you trying to find out the brain regions where all the four conditions show strong evidence for activation? If so, just try the conventional approach and see if it meets your expectation.
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.