my aim is to run a multiple linear regression on resting state functional connectivity maps from 17 subjects.
Predictors are represented by a 10-dimensional questionnaire.
I’m particularly interested in obtaining R[sup]2[/sup] coefficient, for each predictor, for each voxel. It is clear that I need to take in account dependencies between predictors, in this analysis.
I’m aware that a PCA can help solve this problem, but there is a way to take into account multicollinearity in 3dRegAna (or other programs) in multiple regression?
A related question: is there a method that allows to use independent bootstrapped regressions on input data (something similar to random trees generation)?
I’m aware that a PCA can help solve this problem, but there is a way to take into account multicollinearity
in 3dRegAna (or other programs) in multiple regression?
There is nothing available in 3dRegAna that could handle multicollinearity. You may try PCA first with something like the following before running 3dRegAna:
Supposing that I’m not allowed to reduce dimensionality, because I need one coefficient for each predictor, is there a way to conduct a penalized/regularized regression?
Or, alternatively, what may be the best strategy in this situation?
I cannot think of anything currently available in AFNI to directly handle multicollinearity. You may try the following approaches if you read your data into R:
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.