SPSS/ANOVA: Repeated measures post-hoc tests

General Questions
#1

Esoteric stats question.

SPSS/PASW v. 18 if it matters, although I think this has been a feature for awhile.

If you are running a between-subjects design, then you may select from about 18 post-hoc tests to run if you find significance. However, if you want to run a repeated measures (within subjects) design, you will find that that whole section is grayed out with no explanation.

If you go to Options > Display means for FactorX, then you can do “confidence interval adjustments.” There are only three choices here, “LSD (none),” “Bonferroni,” and “Sidak.” Googling suggested these are also post-hoc tests, even if not labeled as such.

My questions:

Are these three in fact post-hoc tests, usable in the same manner?

Why doesn’t SPSS let you use the rest of them with repeated design? Are they not appropriate with repeated measures, or is the program just being difficult?

#2

It’s been a while, but as I recall because of the way the sum of squares is calculated, post hoc analyses are not appropriate.

I wouldn’t call Bonferroni, et al tests, they really are just adjustments for the fact that you’re performing multiple tests on the same subject.

#3

Thanks, that’s a good starting point. Anyone else know about that first part? crickets

You’re right that many of them aren’t tests. However it seems that some are more testlike, as you calculate a new test statistic and compare it to a new critical value.

#4

SPSS glm is pretty limited in what types of things you can do. Ideally, you’d be using SAS proc mixed. If you’re stuck with SPSS though, you can do MANOVA instead, it allows for additional tests and makes no assumptions about homogeneity. Again, I haven’t used SPSS in a while, but a few versions ago, there was no point and click way to do MANOVA - you have to write it out.