Dear All,
I don't know whether this question is appropriate for this list or not. My
question is as follows:
I am fitting a non-linear mixed model using the SSlogis function. When the
parameters xmid and scal are both considered as random (let us say model M1),
the p-value for my covariate of interest is greater than 0.05 (non-significant).
However, when only the xmid is considered as random (Model M2), the p-value is
less than 0.05 and my covariate becomes highly significant. The introduction of
random effects of course increases the variability in the fixed effects but does
not affect their Bias as I understand. My question is: Is it then legitimate to
use the model with only xmid as random effects (model M2 since it gives me
significant results)? knowing that the AIC for M1 is lower than that one for M2
meaning that M1 is slightly fitting better the data than M2
Many thanks in advance
Bernard,
---------------------------------
Une boite mail plus intelligente.
[[alternative HTML version deleted]]