0
$\begingroup$

I am using LMMs and GLMMs(where necessary) to model carbon pools between different treatments. i.e specifically intrested in contrasts.

These pools are divided between different groups e.g. diferent soil horizons. My question is, is it more valid to create individual LMMs for each group or examine differences with an interaction model including all groups? i.e. carbon ~ treatment*group +...

What implications does this have for interpretatrion particularily when certain groups can be modeled with LMMs and others (and the pooled data) require GLMMs?

$\endgroup$

1 Answer 1

2
$\begingroup$

In general, the interaction model would enable you to test if a treatment works better for one group than for another. With separate groups you would only see that there is a difference in treatment effect, but not if such difference is statistically significant.

The fact that some groups can only be modeled by LMM's and others only with GLMM's would make the interaction method impossible. But also the "separate groups" method would be complicated: how to compare the treatment effect in a linear model with the treatment effect in e.g. a logistic model?

$\endgroup$
2
  • $\begingroup$ yes, the "seperate groups" would mean that they are essentially analysed seperatly and the treatment effects would not be compared at all. $\endgroup$ Apr 5 at 7:56
  • $\begingroup$ It seems that you would have no choice then. $\endgroup$
    – BenP
    Apr 5 at 7:58

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.