4

I am conducting a GLMM. I currently have two independent variables (1, 2) and a dependent variable () - all factors (either 0 or 1).

m <- glmer(y ~ x1+x2+x1:x2 + (1 | participant), data = mydata, family = binomial)

I am mostly interested in the x1 main effect and in the interaction between x1 and x2.

When all (x1 x2 x1:x2) in the model, only the two main effects are significant (the interaction is N.S.).

Is there something I can do in order to fit a better model?

Thank you :)

Your help is much appreciated!

Sharon
  • 81
  • You can check whether the assumptions of the binomial model are not violated. You can also try other classification models beside GLMM and then compare the models, for instance, using cross-validation. – Jakub Sep 08 '20 at 11:38

1 Answers1

4

Is there something I can do in order to fit a better model?

I would recommend not seeking to find a "better model".

You have 3 variables besides the outcome. One is a grouping factor that you are fitting random intercepts for. The outcome is binary so you are fitting a logistic model with 2 covariates and the interaction between them. There is nothing wrong with this as a model for these data, based on the information given. There is nothing you can really do unless you have more data and/or more variables (competing exposures). Perhaps there is really no interaction in the population. Presumably you did a power calculation before collecting the data, but if not, perhaps the dataset is too small to find the interaction of interest. What if you had found a "significant" interaction, but the effect size was tiny and meaningless ?

Robert Long
  • 60,630
  • @Sharon does this answer your question ? If so, please consider marking it as the accepted answer. If not then please let us know why – Robert Long Sep 26 '20 at 18:18