I have a multinomial ordered response variable (0, 1 and 2) corresponding to population trends (negative, no trend, increasing); 2 continuous independient variables (use and distance) and I am controlling by taxonomy, so I have as random variables family adn genera nested into family. My final model code is:
proc glimmix data=sasuser.Indices_octene ;
class trend family genera;
model trend = use distance distance*distance / dist=multinomial solution;
random family genera(family);
I would like to know:
- Is this a logistic regression with random factors?
- Why can't I get the AIC of the model? (I get it if I delete random factors for the model)
- How could I know if my model is correct? do I have to look at the residuals, and how could I do it?
- Should I wonder about overdispersion? how could I get a measure of that?
I also found that if I use a normal error distribution, the estimates are almost the same when using multinomial distribution, but with the oposite sign. How should I interpret the sign of the estimates in the model with multinomial error distribution?