Date: Wed, 17 Dec 2008 06:05:21 -0500
Reply-To: Peter Flom <firstname.lastname@example.org>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject: Re: What to do when parallel slope assumption doesn't meet for
Content-Type: text/plain; charset=UTF-8
Yu Zhang <zhangyu05@GMAIL.COM> wrote
>My first STAT qustion on the L.
>data consist of lab value which are measured at baseline and annual follow-
>up for six boimarkers. no treatment was given during this period. there
>are 11 risk factors collected at baseline.
How many followups do you have? Just one, or multiple ones, from the below,
it appears that you have multiple years? How many people?
>assess change over time for each biomarker by some of these more important
>demographics and risk factors. We may be able to see if any non-
ifiable/modifiable risk factor is influencing change over time.
>I was told to condcut a ANCOVA analysis controlling the baseline lab value
>for each factor. unfortunately, for one biomarker, the parallel slope
>assumption does not meet.
Who told you this? :-)
If you have multiple follow ups, I strongly suggest PROC MIXED.
If you have only one follow up, then ANCOVA is one strategy, but then I am not sure which sloples you want to be parallel. The slopes for different groups (e.g. Male vs. Female) are what you are testing in an ANCOVA, so you certainly don't want to assume that they are parallel.
>I looked up the textbook I used and don't have a clue of how to handle
>Can anyone kindly suggest what i should do? when i was in school, the
>homework data was perfect for fitting ANCOVA model. the real world data
>really drive me crazy!!!
Oh yeah, those real world data are a mess, aren't they?
Peter L. Flom, PhD
www DOT peterflom DOT com