My situation – I have 5 study variables and ~7 noise variables, and a continuous count Y variable. I’m building models where I don’t use them to predict, I just need to know the coefficients and their significance.
I have built 2 models – 1 a glm regression, and one a mixed affect model.
Here is my challenge – the mixed effect model has a much better fit (by aic) but many of the coefficients are non significant. The glm has a worse fit but is significant for every coefficient I care about.
I’m tempted to focus on the glm bc it is more useful (and all the components pass partial F tests) but it feels weird to use the much less fit model?
Thx for any help