Hi

James M. Clark
Professor of Psychology
204-786-9757
204-774-4134 Fax
[EMAIL PROTECTED]

>>> "Marc Carter" <[EMAIL PROTECTED]> 04-Apr-07 9:26:35 AM >>>
I am also
getting a sense that the field is moving away from factorial ANOVA and
more into somewhat more straightforward regression models with
interaction terms in them.  Pedagogically, it seems to me that
regression's an easier conceptual model than ANOVA (though
quantitatively they're the same model).  Why aren't we teaching that?
Why are we teaching them things that are so badly out-of-date?

In my honours stats class I do first teach regression and then anova (I have 
the luxury of a full year course).  And when we start anova I demonstrate the 
equivalence between the two approaches.  But it is incorrect to think that 
everything is as easy to do via the regression approach as via the anova 
approach (which Marc correctly points out are equivalent).  One major area that 
anova wins hands down is with repeated measures designs, a common situation in 
psychological research. In addition to the initial (relatively minor) problem 
of generating subject vectors for large numbers of subjects (or using subject 
means and adjusting df, as per Kerlinger), factorial and follow-up analyses for 
within-s factors generally involve unique error terms as well as partitioning 
of the numerators.  This is less readily accommodated by regression than by 
anova.  There are undoubtedly other topics (e.g., the post hoc procedures that 
started this discussion) that are generally less available in regression 
packages than anova programs.

Take care
Jim



---
To make changes to your subscription go to:
http://acsun.frostburg.edu/cgi-bin/lyris.pl?enter=tips&text_mode=0&lang=english

Reply via email to