Date: Thu, 11 Aug 2005 16:06:37 +0200
Reply-To: erik christiansen <ec@SELVMORDSFORSKNING.DK>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: erik christiansen <ec@SELVMORDSFORSKNING.DK>
Organization: UNI-C
Subject: Same reference group for two variables in a conditional logistic
regression setup
Hi
I have problems with interpreted of estimated odds ration in a conditional
logistic regression setup. I have one variable measuring the time since an
event. I made 5 dummy variables out off the variable, where one of the
dummys where the reference group and that is coded 1 for cases when the
event never occurred. The rest of the dummys where coded 1 if the time since
the event was from 1 to 20 time units ago, 21 to 40 time units ago ect. In
the regression with only the last 4 dummys, we estimate the odds ration for
each of the dummys compared to the group where the event never occurred. We
have results like this:
Categories: odds ratio
Event never occurred 1,00
1 to 20 time units 2,00
21 to 40 time units 1,50
Ect.
I the have another variable, measuring the number of events like the one
from before. I then coded a number of dummys from this variable after same
procedure as before. The reference group is the group where the event never
occurred. Like before. When I do logistic regression I get something like
this:
Categories: Odds ratio
Event never occurred 1,00
First dummy 1 to 5 times 6,00
Secord dummy 6 to 10 4,00
3'th dummy 11 to 15 3,50
Ect.
I would at this time like to control the two variables for each others
effects and therefore perform multiple logistic regression where all the
dummys are in the model at the same time. All the dummys will have the same
reference group and it is here my problems starts. It looks like the first
group of dummys have event never occurred as reference group and this is
okay, but then it looks like the last group have one of it own dummys as
reference group. Is this right? What is the interpreted of the estimates?
Thanks
Erik
| 
