Date: Fri, 27 Mar 2009 06:09:25 -0400
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: ANCOVA versus Multilevel Random Coefficient Models/HLM/mixed
Content-Type: text/plain; charset=UTF-8
leah rubin <lhr1807@GMAIL.COM> wrote
>We conducted a randomized clinical trial with 4 treatment groups (3 active
>treatments and placebo) and baseline and 1-year.
>Our primary interest was in examining differences in linear rates of change
>between each of the active treatment groups and the placebo group on select
>outcome measures from baseline to 1-year. I conducted a multilevel random
>coefficient model and included a random intercept to account for individual
>differences at baseline. Is this statistical method better than conducting
>ANCOVAs where the final score is the dependent measure and the baseline
>value as the covariate? Any input and/or pertinent references to this issue
>would be greatly appreciated.
First, I am surprised the multilevel model worked with only two time points.
Second, with only two time points, you cannot really address your question - you
have no way to assess whether change was linear or not.
Third, in general, multilevel models are superior to the ANCOVA method, but see above.
Fourth, a good reference that discusses pros and cons of ANCOVA and MLM is Hedeker and Gibbons,
Longitudinal Data Analysis. I'm not sure where my copy is, or I'd provide some quotes.
Peter L. Flom, PhD
www DOT peterflomconsulting DOT com