I have a question regarding the comparison of the following two multilevel models:

  1. Null model: outcome.nullmodel <- lmer(outcome ~ 1 + (1 | ID), data=multileveldata)

  2. Random slopes model: outcome.rs <- lmer(outcome ~ 1+ predictor_L1+(1+predictor_L1|ID), data = multileveldata)

I read in some articles that a random slopes model of that kind was compared to a null model/intercept only via deviance/likelihood-ratio test. But is it appropriate to use the deviance test here? I am not sure about that because the models differ in the fixed AND the random effect. Therefore, I would assume they are non-nested and should be compared using information criteria such as AIC and/or BIC instead of likelihood ration test. Or am I wrong?

Thank you very much in advance for your support!


1 Answer 1


The models are still nested: you get model 1 by setting the coefficient of predictor_L1 and the variance component for predictor_L1 and the covariance component to zero. That's just three parameters.

It's a slightly non-standard likelihood ratio test, because the null value of the variance component is on the boundary of the parameter space, but there isn't an issue with nesting.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.