>
> Here's Paige Miller's table, updated by Tom Stewart, with Data Desk results
added:
>
> Package Type of SS SS due to A SS due to B
> SAS 6.07 Type I 75.6000000 2.3333333
> Type II/III 66.6666667 2.3333333
> SYSTAT 66.667 2.333
> MINITAB 7.2 Seq SS 75.600 2.333
> Adj SS 66.667 2.333
> SPSS Unique 66.667 2.333
> Hierarchical 75.600 2.333
> Experimental 66.667 2.333
> NCSS GLM ANOVA 66.66666 2.33333
Data Desk GLM sequential 66.6667 2.33333
GLM partial 75.6000 2.33333
(p.s. StatSoft claims in their pamphlet that Data Desk lacks a full GLM...)
> STATISTICA Specific Effects 77.23189 11.26667
>
Nevertheless, I think I can "explain" Statistica's results. 77.23189 is the
correct partial (type III) SS for A when B is nested in A. 11.26667 is the
correct partial SS for B when A is nested within B.
(Data Desk's interactive GLM interface made it easy for me to alter the
model and watch the instantly updating ANOVA table until I found this
result -- about 3 minutes of work. I doubt I could have done it if Data
Desk's interface or GLM capabilities had been as poor as StatSoft
claims...).
Of course, these two SS's cannot coexist in the same model, so it seems odd
to report them together. Most of the other packages in the table would
forbid a model of A(B) + B(A) -- preferrably before any attempt was made to
compute it. And if Paige did not specify any nesting...
-- Paul Velleman
P.S. Thanks to all who have emailed support to me during the past
discussion. I've been waiting for the rhetoric to cool down before posting
again, but I thought I could help here with a real statistics answer.