Date: Fri, 20 Aug 2004 17:01:31 -0400
Reply-To: Richard Ristow <wrristow@mindspring.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Richard Ristow <wrristow@mindspring.com>
Subject: Re: Repeated measures control for baseline scores
In-Reply-To: <000a01c4844c$c9b24160$6aa1d781@ed.ac.uk>
Content-Type: text/plain; charset="us-ascii"; format=flowed
I'm hesitant, being far from the best mathematical statistician on the
list (Hi, Marta!), but nobody else has responded --
At 07:24 AM 8/17/2004, Marie-Claire wrote:
>I am wanting to compare 3 experimental groups with a control group on
>measures of cognitive performance at 2 time points. I have previously
>analysed this data using a repeated measures ANCOVA in SPSS (covarying
>for differing times between the 2 assessment points between groups),
>with group as a between-subject variable and cognitive perfromance at
>time 1 and at time 2 as the two levels of the within-subject variable.
On the face of it, this makes perfect sense.
>I have been advised recently however to amend this analysis by
>controlling for differences between the groups at baseline, such that
>all groups are equated on cognitive performance at time 1 and from
>here any changes to time 2 can be seen across all groups (main effect
>of 'time') or differentially in one particular group relative to all
>others (group*time). My question is, are baseline scores not already
>held constant in the repeated measures ANCOVA model, or do I need to
>redo all analyses using a linear regression model in SPSS instead?
I'm assuming your "time 1" is the baseline. If it is not, address the
question by adding it to the analysis as a third level of the
within-subject variable. But if "time 1" is the baseline, I very much
believe you're right: the within-subject effect has two degrees of
freedom, of which a useful representation is the mean of the two
values, and their difference.
| 
