I was wondering if it would be possible to obtain model fit statistics (ideally log likelihood) at the voxel-level from 3dLME. The logic here being, as I add predictors to a multilevel model (e.g. gender, group status, etc.) I would like to identify regions (voxels) where not only does my predictor demonstrate a significant fixed effect, but its addition also creates a more parsimonious model (compared to a model without that predictor). To do this I would like to be able to compare the model fit statistics of multiple models (via chi-square or something of the sort) in a voxel-wise fashion. To my knowledge 3dLME doesn’t contain an option to write out model fit statistics (please correct me if I’m wrong!), but I can image while it’s running lme iteratively in R it would be possible to grab the logLik value along with the coefficients?
Thanks again for all your feedback and adding these options. I am curious to know if 3dLME is fit using REML or ML. I ask because I believe that if we want to compare the loglik, AIC, or BIC for models with different fixed effects, it is required that they were fit using ML, whereas the default for lme is to use REML.
Yes, 3dLME currently uses REML. I’ll look into the situation and see if I can add an option for ML.
I’m curious - if you’re interested in the impact of an fixed-effect variable, wouldn’t it be enough to examine the statistical significance of the variable (e.g., t-stat for the variable in the 3dLME output)?
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.