|Date: ||Fri, 21 Sep 2007 08:43:14 +0000|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Doug Morse <morse@EDOUG.ORG>|
|Organization: ||Vanderbilt University|
|Subject: ||Re: ANCOVA - slopes are heterogeneous - what next?|
ok, dorsey & soeken (1996) agree with you, and i suspect other articles
talking about the johnson-neyman technique will as well. it seems,
essentially, that J-N technique mostly provides a way of determining what
regions have significant differences between groups (in the presense of
nonparallel regression lines) and what regions do. this of course is
basically the same thing you were saying, that is, include the interaction
term(s) and take them into account when interpreting.
all good to know. it's shocking to me how many texts/sources strongly say
"stop" when this assumption isn't met -- especially in light of the fact that
there's such a straightforward and reasonable/logical workaround. very
confusing and misleading. oh well. thanks to all for the help!
dorsey, susan g., and soeken, karen l. (1996). use of the johnson-neyman
technique as an alternative to analysis of covariance. nursing research,
On Thu, 20 Sep 2007 05:42:10 -0400, Peter Flom
> Doug Morse <morse@EDOUG.ORG> wrote
> >sure, i understand that GLM underlies all the regression and anova models.
> >just because i can run a particular model, however, doesn't mean that it's
> >appropriate to do so (as we've all encountered, i'm sure).
> Oh, OK. That's good. You'd be surprised at how many people don't know that ANOVA and Regression are the same thing. Certainly the model can be inappropriate, even with an interaction. I find it hard to imagine a situation where the interaction itself makes it inappropriate, but there could be one.