I'd like to use 3dLMEr to analyse if adding certain covariates improve the model fit, that's why I'd like to be able to extract some voxel-wise model goodness-of-fit stats (say, AIC and BIC). I know that you added that feature to 3dLME but I couldn't find it in 3dLMEr.
Because it's a novel covariate metric I need to be able to quantify its gain by means of improvement of the model fit.
Welcome on board!
I'd like to use 3dLMEr to analyse if adding certain covariates improve the model fit
I completely understand the appeal of allowing the modeling process to determine the inclusion/exclusion of covariates. However, I'd like to clarify your objective: Are you aiming to improve the model's predictive accuracy or to make inferences about specific predictors? For the latter, I would advise caution against such an approach for two reasons.
(1) Information criteria such as AIC/BIC are intended to be adopted when comparing two or more models that share the same 'fixed' effects but differ in 'random' effects. It would be a misuse and even an abuse to use them for covariate selection.
(2) The selection of covariates should primarily be based on domain knowledge. Simply relying solely on model indices for decision-making would be akin to letting the tail wag the dog.
While I may not possess exhaustive domain knowledge, I am more than willing to discuss your specific covariates and offer whatever insights I can.