Troubleshooting
Problem
I've performed a doubly-multivariate repeated measures analysis and find that there are two different tables of multivariate test results in my output. Can you explain the difference between these two tables?
Resolving The Problem
The first table entitled Multivariate Tests carries the true doubly-multivariate test statistics. The averaged results appear in the table entitled Multivariate that is listed under the heading Tests of Within-Subjects Effects. Averaged results are like those you would have obtained had you set up a multivariate model with a data set that is configured for the univariate approach to repeated measures (e.g., a split-plot approach).
Here is an example to help clarify things. Let's say we have a data set in which ten subjects were assessed on two measures at each of three time points.
The doubly multivariate setup for the data would include 6 variables and 10 cases. Let the measures be named A and B and let TIME be indexed by T1, T2, and T3. We will use this mnemonic device and name the variables AT1 to AT3 and BT1 to BT3. The following doubly-multivariate repeated measures commands would produce both doubly-multivariate results and averaged results.
GLM
at1 at2 at3 bt1 bt2 bt3
/WSFACTOR = time 3 Polynomial
/MEASURE = a b
/WSDESIGN = time .
The following GLM Multivariate (not repeated measures) command syntax would give you these same doubly-multivariate repeated measures results in the Multivariate Test Results portion of Custom Hypothesis output.
GLM
at1 at2 at3 bt1 bt2 bt3
/MMATRIX = ALL 1 -1 0 0 0 0;
ALL 0 1 -1 0 0 0;
ALL 0 0 0 1 -1 0;
ALL 0 0 0 0 1 -1 .
The univariate setup for this hypothetical data set would have three rows per subject (i.e., one for each time) and four variables. The variables could be named subject, time, A, and B. The GLM (Multivariate) command syntax for this setup would be as follows.
GLM
a b BY time subject
/DESIGN time subject.
The results of this analysis would replicate the averaged multivariate results from the original doubly multivariate run in the table entitled Multivariate Tests. Note that these tests (also known as Scheffe's multivariate mixed model tests) assume a type of multivariate sphericity for which SPSS does not currently offer a test. The fully or doubly multivariate tests do not require this assumption to be met. Note also that when there are only two levels of the within subjects factor, the two sets of tests are the same.
Related Information
Historical Number
19602
Was this topic helpful?
Document Information
Modified date:
16 April 2020
UID
swg21476707