nonparametric alternative ANCOVA
Hi,
I'm running an ANCOVA and I'm looking for a nonparametric alternative since some of the assumptions were violated. I noticed that JASP offers the Kruskal-Wallis test under 'Nonparametrics'. Is that indeed considered a valid alternative? On various fora, I found the Kruskal-Wallis test is often suggested as an alternative for repeated measures ANOVA, but not ANCOVA, for which ordinal logistic regression or more complex alternatives are suggested. Hope you can help!
Thanks,
Tim
Comments
Hi Tim,
The Kruskal-Wallis test is the nonparametric alterantive for ANOVA, while the Friedman test is the nonparametric alternative for Repeated Measures (RM) ANOVA. Both the Kruskal-Wallis test and the Friedman test are only available for a single predictor variable unfortunately, and cannot adjust for a covariate. If the covariate is very crucial, then indeed the ordinal logistic regression could indeed be of interest (go to Regression -> Generalized Linear Model -> set Family to "Other" and Model to "Ordinal logistic regression"). Make sure you then add the covariate to the null model in the model tab, and you can then focus on the coefficients table. Specifically, the coefficients for the different levels of your categorical predictor; if these deviate from 0, that indicates a differences with the references group (the first occurring level of your categorical predictor). Please let me know if anything is still unclear!
Cheers
Johnny
@JohnnyB
Hi. Is the method you described above fully non-parametric? I ask because the results result appears to depend on the scaling of the covariate. For example, one gets a different result if the covariate = q than if it is q^2 or log(q).
R