Within subject tests fail in repeated measures analysis
Jan 29, 2014 11:10 AM(1379 views)
I am having trouble with a simple repeated measures analysis. I have 8 subjects each measured at 10 time points. The subjects are assigned to one of 4 groups, ie 2 subjects per group. I am trying to test if there is a difference between groups over time. I have run a Manova, selected repeated measures, but the output does not include full statistics for the within subjects tests. Instead, some of the spaces are filled by a '.' instead, as if something has failed, or there is something wrong.
Please can someone help me work out what the problem is.
This is an issue of degrees of freedom. Although the univariate repeated measures effects are estimatable (and will be displayed if you check "univariate tests also") the multivariate analysis of variances tests will not have enough observations (more precisely, degrees of freedom) to test the effects you're interested in. Ignoring for the moment the effect of group (which I know is important), with 2 more subjects the multivariate F will have sufficient degrees of freedom for error to be calculated (but just barely - 1df for error in that case). Estimating the group effect (and interaction with time) consumes degrees of freedom, so in order to estimate all the effects you want you need more subjects. I would suggest at least 16, 4 subjects for each condition.