2-Way Factorial, Non-Parametric, Repeated Measures test.

R

RobH

New Member
#1
#1
Hi
I've got data from a 2x2 fully factorial, repeated measures design. The residuals are non-normal so I am looking for non-parametric equivalent of the 2-way repeated measures ANOVA. I know there isn't one available from the SPSS menus, but has anyone got any SPSS syntax that can perform this sort of test? There is apparently a way of doing this sort of test using the R software, but I can't get an understandable desription from anywhere as to how to do it, and I haven't really got any experience of the R software. So I was hoping someone might be able to help via SPSS.

My current solution to the problem is to use the parametric ANOVA and then 'check' the significant main effect findings using the non-parametric one-way Wilcoxon test. Not particularly kosher from the statistical point of view so I was hoping to find a cleaner method of doing the analysis.

Thanks
Rob
 
Karabiner

Karabiner

TS Contributor
#2
#2
I do not know what exactely you mean by "nonparametric" here, but perhaps
your sample size is large? Then you do not have to care for normality of residuals
too much (cf. central limit theorem).

Kind regards

K.
 
Last edited:
R

RobH

New Member
#3
#3
The sample size is 70. I'm aware of central limit theorem but have read that high sample sizes are not always a protection against skewed distributions, and that using parametric statistics against non-normal distributions can lead to a lack of power or an inflated Type 1 error rates. Indeed this may be the case with my data as I have noticed that the non-parametric correlational tests (i.e. Spearman's Rank) appear to have greater power than the parametric equivalents (Pearson's) when applied to the data I have.

By non-parametric I mean a test that does not rely on the data (or it's residuals) being normally distributed. Perhaps 'Robust method' is a better term.
 
Last edited:
C

Cynthia_MSc

New Member
#4
#4
Hello,
I have the same problem only I have 2 groups with n=12 and n=14. My distributions are not normal and I have outliers. If I try to elminate the outliers, I have other outliers, so that is not helpful. I only have basic statistical understanding. However, I need a non-parametric test for 2 groups, 4-way repeated measures. Is this possible?
Cynthia
 
You must log in or register to reply here.