Dear All,
Sorry if this question may not be appropriate to this list. I have the
following question about the significance of effects in NLME. I fitted data with
nlme using the SSlogis as a specification for the conditional mean of my
dependent variable Y. When xmid and scal were considered as random, my covarite
of interest X was found to have a significant effect on xmid but not on scal.
However, when I assume that only xmid is random, the effect of my covariate X
was found to have a significant effect on both xmid and scal. Of course the AIC
and the residuals were both smaller in the firts model. Thus my question: Which
model should I select: the one that fits better to the data (i.e Model 1) or
the one that shows significant effects of my covariate (Model2). Or in another
way, Is it legitimate to use model 2?
Many thanks in advance,
Bernard
---------------------------------
[[alternative HTML version deleted]]
