David,

Thank you for your help.

>If you have created a dummy for each category then you can easily do >different regressions with a different dummy left out each time, giving >a different reference category. However, these signficance tests are >not all independent of each other, so you should only test the >comparisons that you are really interested in, and not compare >everything with everything just in case there is a difference >somewhere; that needs the proper use of post-hoc comparison methods.

Now I realize that regression and ANOVA has the same issue in the post-hoc comparison. So hypotheses are very important. I will pay attention to this!

Qinghai