Repeated measures results - Which to report and why?

[Belzer, Ken]

Hi All,

I would appreciate it if anyone familiar with analyzing data via SPSS from factorial designs can help me interpret/understand some results. I am using SPSS 13.0 GLM repeated measures to analyze data on self-reported sedation from a 3 (drug condition) x 4 (time point, i.e., baseline, 1 hour, 2.5 hours, etc.), x 2 (genotype) experimental study. The first two factors are within-subjects and the latter is between.

In the SPSS output for repeated measures three sets of results are provided under the titles 1) "Multivariate Tests", 2) "Tests of Within-Subjects Effects", and 3) "Tests of Within-Subjects Contrasts".

In "Multivariate Tests"and "Tests of Within-Subjects Effects", I obtain significant main effects of drug condition and time, with no significant interaction effects. Under "Tests of Within-Subjects Contrasts" I obtain a significant quadratic effect of drug, and a significant cubic effect of time (main effects - I think). However, I also obtain a significant two-way quadratic interaction of drug x time, and significant three-way linear and quadratic interactions of drug x time x genotype.

Two questions:
1. Why am I not picking up the interaction effects under 1) "Multivariate Tests" and 2) "Tests of Within-Subjects Effects"?
2. Which of the three ANOVA tables, i.e., 1) "Multivariate Tests", 2) "Tests of Within-Subjects Effects", or 3) "Tests of Within-Subjects Contrasts" should I rely on when reporting reults and why.

Any suggestions and/or insight would be greatly appreciated.

Thanks very much.
Ken

Kenneth D. Belzer, Ph.D.
Research Scientist
Nathan Kline Institute for Psychiatric Research
Geriatric Psychiatry Division
140 Old Orangeburg Road
Orangeburg, NY 10962
Tel: [phone]

[Swank, Paul R]

see below.


Paul R. Swank, Ph.D. Professor
Director of Reseach
Children's Learning Institute
University of Texas Health Science Center-Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Belzer, Ken
Sent: Tuesday, March 06, 2007 1:05 PM
To: [hidden email]
Subject: Repeated measures results - Which to report and why?

Hi All,

I would appreciate it if anyone familiar with analyzing data via SPSS
from factorial designs can help me interpret/understand some results. I
am using SPSS 13.0 GLM repeated measures to analyze data on
self-reported sedation from a 3 (drug condition) x 4 (time point, i.e.,
baseline, 1 hour, 2.5 hours, etc.), x 2 (genotype) experimental study.
The first two factors are within-subjects and the latter is between.

In the SPSS output for repeated measures three sets of results are
provided under the titles 1) "Multivariate Tests", 2) "Tests of
Within-Subjects Effects", and 3) "Tests of Within-Subjects Contrasts".

In "Multivariate Tests"and "Tests of Within-Subjects Effects", I obtain
significant main effects of drug condition and time, with no significant
interaction effects. Under "Tests of Within-Subjects Contrasts" I obtain
a significant quadratic effect of drug, and a significant cubic effect
of time (main effects - I think). However, I also obtain a significant
two-way quadratic interaction of drug x time, and significant three-way
linear and quadratic interactions of drug x time x genotype.

Two questions:
1. Why am I not picking up the interaction effects under 1)
"Multivariate Tests" and 2) "Tests of Within-Subjects Effects"?
************************************************************************
***************************
Because the multivariate test is an omnibus test under the assumpton of
multivariate normality,  equal variance-covariance matrices, and
indepedence between subjects, while the within subjects efects are also
omnibus tests under the assumption of sphericity, normality, and
independence, whereas the contrasts are post test comparisons. Unless
you had a priori hypotheses about quadratic or cubic effects then you
should not look at the contrasts unless you have a significant omnibus
test.
************************************************************************
*****************************
2. Which of the three ANOVA tables, i.e., 1) "Multivariate Tests", 2)
"Tests of Within-Subjects Effects", or 3) "Tests of Within-Subjects
Contrasts" should I rely on when reporting reults and why.
************************************************************************
****************************
It would depend on the data. If you meet the assumptions for the
multivariate test, use it. But the assumptions for the univariate
(within subjects) may be easier to test as well as  meet. Looking at
estimates of the eccentricity should tell you about the sphericity
assumption although if it is not met you can used the Greenhous-Geisser
or Huynh-Feldt corrections.

Any suggestions and/or insight would be greatly appreciated.

Thanks very much.
Ken

Kenneth D. Belzer, Ph.D.
Research Scientist
Nathan Kline Institute for Psychiatric Research Geriatric Psychiatry
Division 140 Old Orangeburg Road Orangeburg, NY 10962
Tel: [phone]