Viktoria - You sent me a message to an inbox I cannot for some reason reply on. Please re-send to marcantr@us.ibm.com. Thanks and sorry for the inconvenience.
BTW - My reply was going to be that you need to use the SD of the differences, not the SD of one of the groups, for the test you were referring to.
Original Message:
Sent: Wed March 01, 2023 03:13 PM
From: Rick Marcantonio
Subject: Repeated Measures ANCOVA
Hi. I passed this on to the statisticians. Unfortunately, they don't have time to consult on issues that are more theoretical, due to several upcoming, time-sensitive assignments. Their involvement must necessarily be limited to statistical issues particular to the SPSS algorithms.
That said, one of them did get back to me with this:
"It is true that you can't put the same variable in as a covariate and a dependent in repeated measures. You can make a copy of the baseline measurement and include that as a covariate."
I hope that helps to some degree.
------------------------------
Rick Marcantonio
Quality Assurance
IBM
Original Message:
Sent: Tue February 28, 2023 04:39 PM
From: Viktoria Luhaste
Subject: Repeated Measures ANCOVA
1. Is my understanding correct that you can not covary for baseline scores with repeated measures ANCOVA?
2. The analyses require a baseline in the time factor for within-subjects effects, so it can not be used as a covariate in between-subject effects in the same way as you can do in ANCOVA, is this correct, or there is another way to covary for a baseline for between-subject effect?
If I move pre-test scores to covariates it calculates 1-time point as a baseline and I lose the difference from the baseline to the first time point.
And main within-subjects effects become inaccurate.
3. Perhaps repeated measures ANCOVA analyses already contain adjusting for variance in the baseline scores?
4. If I use another covariate e.g age for repeated measures ANCOVA for between-subject effect, I observe that there are no significant interactions between the covariate and three treatment subgroups. Treatment sub-group differences did not change much, however, the main within-subjects effect time for all measures (the main result of this one-group repeated measures study) changes from being significant to non-significant.
Is my understanding correct that because age significantly interacts with the within-subjects effect, I need to first run it without the covariates and report the within-subjects effects and then run sub-group analyses with covariates and report the between-subject effect?
Do I also report about the age as a covariate interaction with treatment effects over time, and how to correctly interpret it?
I will be very grateful for your advice and explanations.
------------------------------
Viktoria Luhaste
------------------------------