3

I ran a MANOVA on 3 DVs against 2 IVs (2 levels each factor, so 2x2) and found no interaction effect, but I found a main effect for one of the IVs. Would it make sense to follow up with a discriminant analysis, especially since the univariate results were not significant?

EDIT to include research question: My 3 DVs are ratings on essay quality of 138 students. I want to know whether the presence or absence of one or both guidance tools (that are the two factors) significantly influences these ratings (control group receives no guidance). Since the univariate ANOVAs didn't yield any significance, I would like to know which of the 3 ratings (if any) are influenced more by the interventions.

ttnphns
  • 51,648
  • 40
  • 253
  • 462
iamnarra
  • 213
  • 2
  • 7
  • This is impossible to answer unless you explain what your research question is and what you are trying to prove or to investigate. – amoeba Feb 10 '17 at 14:56
  • @amoeba good point, I have just edited my post with the research question. Basically I would like to know which of the essay quality ratings was influenced by the intervention(s) the participants received (or didn't receive). – iamnarra Feb 10 '17 at 15:01
  • 1
    Yeah, sure, then follow up. See http://stats.stackexchange.com/questions/131241 and links therein. – amoeba Feb 10 '17 at 15:03
  • @amoeba great, I just needed to make sure since I have 2 IVs, so the discriminant analysis would only involve one IV that had a significant F value from MANOVA. – iamnarra Feb 10 '17 at 15:05
  • Conceptually it would be better to use both IVs. I don't know if this is possible with your software, but it *is* possible in principle, as I explain in the linked answer and in some answers that are linked from there. – amoeba Feb 10 '17 at 15:07
  • @amoeba thanks! Looks like it can be done with R. But would it make sense conceptually to do the posthoc even with the IV that *isn't* significant? – iamnarra Feb 10 '17 at 15:18
  • I see what you mean. Hard to say, I guess one could argue either way. It will probably not matter too much. Note that you are now doing exploratory analysis, so you are free to try both ways and see what you get. It's also a good idea to plot your data! – amoeba Feb 10 '17 at 15:20
  • @ttnphns I am not sure I understand what you mean. Does that mean I should do discriminant analysis on both factors, then, even if the other factor yielded no significant main effect based on the MANOVA (which also didn't have a significant interaction effect)? – iamnarra Feb 10 '17 at 16:27
  • @iamnarra, please forget that comment, it was actually imprecise, I erased it now. If MANOVA test (such as Wilk's) is significant there is sure at least one significant discriminant. – ttnphns Feb 10 '17 at 16:35

0 Answers0