PROC CATMOD: type I or type III?

[bncoxuk]

Hi Sir,

The PROC CATMOD is a good method to run multinomial logit models (response variable with more than 2 categories). The estimation is usually based on maximum lilihood and a ANOVA table is produced. I am wondering if the AVOVA table is estimated based on Type I Sums of Squares or Type III. Just cannot find any reference for this.

[StatDave]

Maximum likelihood estimation does not involve sums of squares.