question re: lmer_crossvalidation_test()

jens_s20

Hi there,

I would like to test whether adding interaction terms to a model improves the model fit (as opposed to testing whether the estimate of an interaction is different from 0). So in R I would consider the significance of a predictor based on model comparisons (e.g., comparing a model including an interaction to one with only main effects). I wonder if it would be a valid approach to take the tested samples from the lmer_crossvalidation_test() output and rerun the model on these samples but using a likelihood-ratio-test to determine whether including an interaction significantly improves model fit.

I hope the question is clear and thanks in advance for any help.

sebastiaan

Hi @jens_s20 ,

Yes, absolutely! In that case, you would use the lmer_crossvalidation_test() purely as a non-circular way to localize the interaction in the time series and then use your own test to actually test for significance.

— Sebastiaan

jens_s20

dankjewel!