I have 30 patients and 30 controls and am interested in group differences in an MRI measure. This measure was taken in 5 different anatomical structures, and each structure exists bilaterally (there is a left and right structure). I am not interested in the difference in the measure between structures, but would like to know if there are side x group or side x structure x group interactions besides an overall group difference.

A similar paper used a three-factor repeated measures ANOVA: group as the between-subject factor, and then structure and side as within-subject factors. They then did post-hoc two-factor repeated measures ANOVAs to follow up. Another paper looking at just two structures simply did ANOVAs for each structure (group as between-subject, and side as within-subject).

I did the repeated measures analysis in SPSS and in the within-subjects effects and between-subjects effects tables, none of the effects or interactions are significant. However, the multivariate tests table indicates a tract x group interaction.

What exactly does this mean? I was told by another investigator that my significant result is from a MANOVA, and I can use it as long as subsequent analyses are also multivariate. Which post-hoc test would I use to find out which structure(s) are driving the interaction?

Thanks so much!

  • $\begingroup$ It's not clear to me exactly what "repeated measures analysis in SPSS" you ran. A univariate or multivariate model? It would be nice to present your model as well as the details of the results before any help could be offered. By "tract" in "tract x group" interaction do you mean structure? Also, how many subjects do you have? $\endgroup$ – bluepole Jun 16 '14 at 20:14
  • $\begingroup$ Apologies, I myself am not entirely sure what I did in SPSS--I only entered the one measure when defining factors for the repeated measures analysis (under the GLM options, I only see univariate, multivariate, and repeated measures). $\endgroup$ – moose Jun 17 '14 at 14:00
  • $\begingroup$ Also yes, tract = structure, and I have 30 patients and 30 controls. $\endgroup$ – moose Jun 17 '14 at 14:00
  • $\begingroup$ I've performed a similar analysis recently. Since this would most likely involve detailed discussions about the data structure, you may contact me directly (my ID at gmail) if you like. $\endgroup$ – bluepole Jun 17 '14 at 14:52

You would need to run a simple effects analysis (one-way ANOVA) to determine where the significant difference comes from. So, first find significant interaction using MANOVA, then run simple effects analysis using ANOVA and report results.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.