I once had some experience in SPSS/statistical testing, but now I really need it, I cannot seem to find the appropriate tests. I hope anyone can assist me on this one.
I've got 2 IV's (both nominal on two levels), 2 DVs' (both scale). Since non of the data are normally distributed, so I can not run a MANOVA, is that correct?
N=53, but I've got some random missing data.
If anyone can tell me how to (safely) complement this data, maybe I could run the MANOVA after all (since higher N per cell can make MANOVA robust against violation of the multivariate assumption)?
I am really interested in the interaction effects between my two IV's. Is there any way you can tell me, please?
(I'm using SPSS)
I'm very sorry, but I'm not familiar with that test...should I? Or is there another name for it in the statistical software I am using (which is SPSS)?
I thought about ranking the scores and doing a nonparametric test, but there appears to be no nonparametric test that also gives me the interaction effects between my two IV's. And that interaction effect is one of the most important parts of my design/hypothesis.
PERMANOVA. This is available throught PRIMER-E V6. There used to be a free FORTRAN version available, you might have soem luck if you look it up on line; but is was pretty tricky getting the data formatting correct. BEst bet is to track downs the add on to PRIMER.
The earth is round: P<0.05
Hold on hold on, I'm just a simple student in the social sciences, only required to know some basic statistics stuff. Besides that, I'm from Europe and I've discovered that some of the underlying assumptions (although I always thought mathematics could be considered the only science having global rules) vary from continent to continent. For instance: my Australian text book says univariate normality is required for MANOVA, the British text book I used, only required multivariate normality.
Is there any (fairly simple) method I can use to get some results out of my data, only having access to SPSS?
as far as what your options are, i believe that what Dragan said it's a good option if you're stuck with SPSS... i've heard there's good research out there backing the idea of doing the normal parametric analysis on the ranks of the data and actually getting pretty solid conclusions, although i might be wrong...
one thing, though... the assumption of multivaraite normality is **NOT** on the variables themselves but on the residuals... have you checked those? if the residuals are normally distributed you don't really care about the distribution of your other variables...
And ahm, I'm not quite sure what residuals are, so I surely don't know how to check them...
Thank you very much for your answer, it was the most understandable to my question (until now). (No offense to all the others, of course. Thank you all for responding so quickly)
Zimmerman, Donald W.; Simplified interaction tests for non-normal data in psychological research. British Journal of Mathematical and Statistical Psychology, Vol 47(2), Nov, 1994. pp. 327-335.
Zimmerman, D. W., & Zumbo, B. D. (1993). Relative power of the Wilcoxon test, the Friedman test, and repeated-measures ANOVA on ranks. Journal of Experimental Education, 62, 75-86.
and if you type "rank transformations" in PSYCInfo i'm sure there's gonna be more stuff out there... careful, though. i haven't seen these implemented in MANOVAs and i know rank transformations can get real fishy (or so other authors have found)... if you were to REALLY use the non-parametric version of a MANOVA, PERMANOVA would be the way to go though...
now, what is this for? is it like a school project or something you're working on towards publication? i dunno, i'm just wondering whether you're over-complicating things for yourself here...
Advertise on Talk Stats