On Tue, 17 Feb 2004 [EMAIL PROTECTED] wrote: > DB said:
> "Possibly somewhat more to the point (or at least differently...), > Art's question might have been asked in a multiple-regression (MR) > context; in which case, however <d*g> were constructed, it would only > be represented with one d.f. in an analysis predicting a dependent > variable (Y?) from <d*g>." < snip, two paragraphs > > What confuses me is how one gets from a 1 df interaction to a 3 df > interaction by deleting the main effect terms from the analysis. > Clearly these are not tests of the same thing. True. My suspicion is that the key lies in Karl's offhand remark that the analysis he presented was for "unequal cell sizes" in a 2x2 design. Sometimes, when cell sizes are unequal, it is recommended that a two-level factor be coded (+a, -b) where <a> and <b> are not equal and depend on the number of cases for each code. If that also happens with the second factor so that it's coded (+c, -d), then an interaction term formed by multiplying these two codes together will have four different values: +ac, -ad, -bc, +bd. If this is used as the sole "independent variable" in an ANOVA, the program will find that it requires 3 d.f. (For equal cell sizes, the same procedure would assign (+a, -a) and (+c, -c) as codes for the two factors, and the product variable would have only two values: +ac and -ac.) I don't KNOW that SAS does this, but I can imagine that it might. ------------------------------------------------------------ Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
