Joe- The SAS/Stat manual (part II) on pages 928-930 discuss this at some length.

You should specify the e option on the contrast statement. Specifying the options E E1 E2 and E3 on the model statement to see what functions are estimable. Note that a clear understanding of type 1, type 2, and type 3 sums of squares really are needed to make sense out of the estimates.

If you wish to use the interaction effects to construct main effects you should probably use the model:

y= a*b rather then the one you used.

The linear contrasts you wish to specify are then estimable- however they are identical to the main effects generated by the AOV table.

Joseph A Bonito <jbonito@U.Arizona.EDU> wrote in article <Pine.A41.3.95q.970211121917.64548A-100000@f1n4.U.Arizona.EDU>... > Hi, > > I have question about specifying interaction contrasts in proc glm. > If I have two class variable called a and b, each with two levels, how > does sas treat or order the means. > > For example, > > proc glm; > class a b; > model y=a b a*b; > constrast 'label' a*b 1 1 -1 -1; > run; > > Obviously, this tests the main effect of a. However, when I specify this > constrast, I get an message saying that SAS is unable to estimate effect > for this contrast. > > Thanks in advance > > Joe Bonito > > Joseph A. Bonito > University of Arizona > email: jbonito@u.arizona.edu > http://www.u.arizona.edu/~jbonito > phone: 520-621-4800 > fax: 520-621-5504 > >