Posted by
Rich Ulrich on
Mar 30, 2020; 4:58pm
URL: http://spssx-discussion.165.s1.nabble.com/heatmap-of-correlation-matrix-w-o-varstocases-tp5738932p5738943.html
What is the story that you see when you look at the
Group x Time means? What is the size of your effect and your power?
Is this exploratory, or confirming some hypothesis?
Do the Groups reflect random assignment, or could there be
differences at Pre that have not been mentioned? That should
influence how strongly you can assert whatever differences
appear to exist.
Having a reportable difference between two groups when they
both improve suggests at least moderate power of analysis and
a decent effect.
This reminds me of a one-group, repeated design that I set up
which suited certain data on Depressed patients:
Intervention (drugs) had a profound and immediate effect on most
Sleep variables. We tested that with a paired t-test, Pre to Post.
Whatever happened after that was apt to follow a linear trend --
Continuing the same direction; falling back toward the Pre mean; or
staying the same. "Linear Trend" on the 4 post-drug Times was the
powerful and informative test.
You could test your groups separately that way, for description.
I'm impressed by the apparent difference in immediate effect, but I
don't know the whole context. It is less useful if the "better" group
is improving from a worse baseline, so the "adjusted means" show
more difference than the raw means.
If that test would be impressive to your audience, then it might be
appropriate to report that result as a first thing. (Was that a Study
Question?) If that is worth reporting alone, then look at the means
and linear trend for the 3 follow-on periods. - If there are Pre differences,
it is also possible to use Pre as a covariate for the Repeated Measures
test on the other. If there is a real and useful and unique treatment
effect, the groups will also show a difference across all three followups.
But (potential) differences at Pre have to complicate the description
of outcome. Does using Pre happen to increase the power? or does
it create the difference when groups end up at (approximately) the
same place? As I started out: Look at the plot of the means.
--
Rich Ulrich
hello everyone,
I am trying to figure out how to best analyze my data. I have pre-treatment, post-treatment, and two follow-up time points for two treatment groups. When I analyze just pre and post treatment points (using MIXED procedure) I find an interaction of group
x time indicating that Treatment A improves more than Treatment B. But if I analyze all four points, both treatments maintain their gains but no more changes after post-treatment for either group and the interaction of group x time (all four time points) disappears.
Would it make sense to still want to report on the interaction from pre- to post-treatment? if yes, what is the best way to do it? Or I should just ignore that interaction if it does not show when all four times are in the analysis?
Any thoughts greatly appreciated!
cheers,
bozena
===================== To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD