Hi,
Currently assessing NDVI data (vegetation index, continuous data) against land type (4 groups, categorical variables). My data is not normal, or has equal variance (transforming hasn't worked). Is there a non-parametric alternative to one-way ANOVAs that let you look at 4 groups at once? I looked at Kruskal-Wallis (in MiniTab) but it won't let you put in all of the variables at once (for the purpose of post-hoc tests) like ANOVA does.
Thanks!
Hello Jessica,
Do consider bootstrap/resampling or exact/permutation tests as a viable alternative to non-parametric tests when you have concerns about normality and/or homogeneity. As well, rank-based non-parametric tests (such as Kruskal-Wallis, or Mann-Whitney) do not free you from the homogeneity requirement.
Most statistics packages offer bootstrap options.
Finally, any post hoc test such as the usual anova options is really nothing more than a set of repeated pairwise comparisons. So, running pairwise Mann-Whitney or Kruskal-Wallis tests as follow-ups is conceptually no different, save for the absence of a built-in adjustment for multiple comparisons (which you can always invoke on your own; e.g., Bonferroni adjustment, Benjamini-Hochberg method, etc.).
Good luck with your work.
I don't know about MiniTab, but, yes, the Kruskal-Wallis test can handle a one-way design with four groups. An appropriate post-hoc test is the Dunn test (1964), that compares the individual groups.
see page 124 onwards
http://www.mim.ac.mw/books/Phillip%20I.%20Good%20Introduction%20to%20Statistics%20Through%20Resampling%20Methods%20and%20R%202013.pdf
- Badges
- Science topic
- Similar topics
- Indexes