There seems to be some question in the OP about level of measurement (LOM). I think that it might be a good idea to see whether it makes a substantive difference to check results with different assumptions about LOM.
I would tend to go with the interval level, unless the results of the CATREG or some other program with the ability to compare results with different LOM assumptions showed that a lower level of measure actually worked better.
I would be surprised if some GLM such as ANOVA were not in fact preferable. However, my ______ does not always work right. (what is the term that David Marso uses for "mind reading" via the net?)
Art Kendall Social Research Consultants
On 3/13/2013 9:33 AM, Bruce Weaver wrote: > On 13/03/2013 12:29 AM, Rich Ulrich wrote: > > --- snip --- >> I was treating the 3+ Grades as being the categorization for >> looking at some ordinal outcome. The "question" was about >> how to avoid having 4 instead of 3 groups. > --- snip --- > > Yes, that's how I read it too. From the OP: > > "When performing say a Kruskal Wallis test, I don't want to consider > such patients [i.e., those labeled as "Either 1 or 2"] as a separate > group." > > I don't think we've been told what the outcome variable is. Nor have we > been told why the OP wants to use the KW test. It would not surprise me > at all if ANOVA was preferable to KW. >