log linear model for dependence

#1
Hi

I fitted a log linear model (proc genmod dist=negative binomial link=log) to see any dependence/association between 3 categorical variables. I know how to evaluate the best model but i have difficulty to explain the following output.

parameter p-value

intercept <0.0001
var1 0.2558
var2 0.3584
var3 0.0006
var1*var2 0.3945
var2*var3 0.1367
var1*var3 0.1304
var1*var2*var3 0.7634

I can't remove any parameter because the model will not converge.
So only var3 is significant (P-value<0.05).
What can i say for any type of associations between these 3 variables?

Thank you
 

jrai

New Member
#2
Try using proc catmod. Algorithm might converge & then you should be able to remove parameters and do further analysis. Catmod is usually preferred over genmod for loglinear analysis.