CART (classification and regression tree) analysis is an easy
alternative. You won't have to make the sick leave rate categorical.
CART would be helpful if you have a lot of variables, and if you are
interested in the relationship of flu vaccination status in combination
with your other variables. I don't know what you mean by "the issue of
within subject variation" but it probably isn't anything to be concerned
about with CART.
-----Original Message-----
From: A UK-based worldwide e-mail broadcast system mailing list
[mailto:[log in to unmask]] On Behalf Of Andrew Dixon
Sent: Tuesday, April 29, 2003 10:44 PM
To: [log in to unmask]
Subject: mixed logistic regression?
Hi all,
May I seek the groups input on this ?
Im interested in the relationship between flu-vaccination status and
sick leave rate in hospital staff. I have the following variables for
each person: Year to be considered (1998-2003), Vaccination status
(Y/N), Sick leave rate (continuous but skewed), Job category (eg
Clinincal contact or not). My thoughts are to make all the variables
categorical (eg sick leave rate would become high/low) and use a
logistic regression with categorigal sick leave level as the outcome
variable and vaccination status and job category as predictors. What
worries me is the year, because of the issue of within subject
variation. (Ive tried to imagine what happens in this case but my head
started to spin). Also it seems a shame to lose all that info on the
variation in sick leave rate by categorising it (but I guess if I want
to use an ordinary regerssion I'll have to use dummy variables and
transformations, and assume linearity). The more I write this the more
worried I'm becoming. Is there some neat way to do this?
Thanks
Andrew
|