Date: Sat, 9 Apr 2005 15:20:42 -0400
Reply-To: Peter Flom <flom@NDRI.ORG>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <flom@NDRI.ORG>
Subject: Re: Interesting data and regression question (long)
Content-Type: text/plain; charset=US-ASCII
Thanks Manon, for your reply.
Some replies to your comments, below
Won't you have a kind of "multiple-testing" problem ? Are you going
fishing with this ?
It's definitely exploratory......whether it's 'fishing' depends on your
definition of 'fishing' :-). OTOH, we have strong reasons to include
all the IVs -they are all things that make sense, substantively, as
opposed to some sort of true fishing through a huge number of IVs
Why don't you just categorize you DV as an ordinal one and do a
polytomous logistic regression (e.g., category 1=0, category 2=1,
category 3=2+) ?
Ordinal is certainly something to consider, but I see two issues. 1)
Either we choose A) a small number of categories, and thus lump people
with (say) 10 partners with those with 100 OR B) we have a large
number of categories, in which case we run into stability
problems......I think these will be worse for an ordinal logistic with a
lot of categories than for a set of binomials, some of which have small
numbers in one of the groups......but that's just my intuition, I don't
have a formal way of showing it.
About the large amount of missing values for you DV, you could create a
category for this and check if this is significant.
I don't have missing values on my DV......I have INACCURATE values, but
everyone gave a number. Sorry if my original post wasn't clear
Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)