On 16 Dec 1999 09:55:51 -0800, [EMAIL PROTECTED] (Burke
Johnson) wrote (with no control on line length) :
< snip -- a GLM prediction model ... a main effects
only model >
" Here's my question: Do you suggest using dummy
coding (0,1) or effects coding (1,0,-1) for the
categorical variables included in the model?
" The reason I'm asking is because dummy coding
does not always give the same result for a factorial
design as does ANOVA and effects coding, and,"
<break>
- What I think of as "effects coding" is: saving DF by
using just *one* variable with the means, instead of using
a bunch of dummy variables. Indeed, for unbalanced data
(cells without equal Ns) there are *two* choices of coding,
the raw means or Adjusted means -- if you can figure
a rationale for the latter, which *would* match certain
of the factorial results. My version of "coding" does have
the advantage of making sense of the recommendation --
as others have posted, the essential ANOVA results do
not depend on the values of the dummies, so long as they
cover the same DFs of variation.
" Pedhazur recommends using effects coding rather than
dummy coding in the factorial case. Do you know if
the choice of dummy or effects coding matters for a
main effects only model with multiple categorical
and quantitatively scaled predictor variables?"
- You may want to look more closely at Pedhazur, because
I think you are mis-understanding *something*. For one,
if it has "effects coding" in my sense, it won't be exactly
(-1,0,1). I don't find anything like what you say, in my
copy of Kerlinger/Pedhazur (1973)
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
