0
$\begingroup$

In repeated measure for timepoints in different Group, Age and Gender act as a covariate in the model, Age was calculated at each timepoint, but Gender is a static variable for each subject.

m1 <- lmer(Score ~ Group + Timepoint + (1 | Subject))
m2 <- lmer(Score ~ Group + Timepoint + Age + (1 | Subject))
m3 <- lmer(Score ~ Group + Timepoint + Gender + (1 | Subject))
m4 <- lmer(Score ~ Group + Timepoint + Age + Gender + (1 | Subject))

AIC was used to assess the model fit

> AIC(m1, m2, m3, m4)
   df      AIC
m1  5 752.6940
m2  6 758.4476
m3  6 753.5923
m4  7 759.0831

If the AIC was not improving, should these covariates be excluded in the model?

$\endgroup$

1 Answer 1

1
$\begingroup$

If you are going to use AIC as your sole criterion for model selection, then yes. But I wouldn't recommend that, for most cases. You don't say what your dependent variable was or what sort of study this is (e.g. was there random assignment to groups?) but sex and age are pretty important variables in looking at many different DVs.

You also don't give us measures of effect size, but the ones for age and sex might be interesting even if they are small, especially if theory says they should be larger.

AIC is a penalized method of model evaluation - it is one attempt to adjust for model complexity when evaluating model fit. So, your AIC values are saying "the added complexity in the model isn't improving the model fit that much" but that may not be the only thing you are interested in.

$\endgroup$

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.