Date: Thu, 18 Aug 2005 18:01:47 0500
ReplyTo: "Swank, Paul R" <Paul.R.Swank@uth.tmc.edu>
Sender: "SPSSX(r) Discussion" <SPSSXL@LISTSERV.UGA.EDU>
From: "Swank, Paul R" <Paul.R.Swank@uth.tmc.edu>
Subject: Re: Ancova results interpretation
ContentType: text/plain; charset="usascii"
If the regression lines are not parallel, then at some point they will
cross. In the vicinity where they cross, there will be no significant
difference between the groups. Thus, you must carefully evaluate the
results to ensure an accurate interpretation. One way to address the
problem to to determine for what values of the covariate do the groups
differ. This can be done artificially by centering the covariate at
different values and determining if the groups differ. This procedure
does take advantage of chance, like post test comparisons and so may
need a correction for the probabilities depending on the number of
points you investigate.
Paul R. Swank, Ph.D.
Professor, Developmental Pediatrics
Medical School
UT Health Science Center at Houston
Original Message
From: SPSSX(r) Discussion [mailto:SPSSXL@LISTSERV.UGA.EDU] On Behalf Of
emaguin@acsu.buffalo.edu
Sent: Thursday, August 18, 2005 2:59 PM
To: SPSSXL@LISTSERV.UGA.EDU
Subject: Ancova results interpretation
All,
I have a couple of questions about interpreting ancova results. The
basic design is one between factor with two levels (group) and pretest
score as covariate. I checked the group by covariate interaction and
found it to be significant as well as the group main effect. So i have
the case where the regression line differs across the two level of
group. I'll report that both the main effect and the group by covariate
interaction is significant.
I have to confess that i can't remember seeing a report where a group by
covariate interaction was significant. What else beyond the above
statements needs to be said? More generally, what is done after a group
by covariate interaction shows up significant. More to the point, I
don't see that this alters my conclusion that posttest scores differ by
group after controlling for pretest and allowing for different
regression slopes.
I'd like to show estimated means. I don't seem to be get them in GLM.
I'm using the command ../emmeans table(group).
Is that correct or am i missing something? I understand that i can hand
calculate them from the unstandardized regression coefficients but can
GLM do it?
Thanks, Gene Maguin
