Is there a non-parametric test, like a Kruskal-Wallis, but with uneven blocks?

Question

I have a continuous response variable, nominal independent, and want to block by site, though the numbers at sites are unequal. Is there any alternative to a permutation test? If not, does anyone know of an automated process for designing a permutation test for a rookie like me?

Thanks very much

Answer 1

The proportional odds ordinal logistic model is a generalization of the Kruskal-Wallis and Wilcoxon 2-sample tests that can handle many different setups including covariate adjustment.

Answer 2

I assume you are testing for a difference between site means?

Kruskal-Wallis does not require the block sizes to be the same. It's actually quite a good test. Having said that, though, it is a general-purpose test. If you are testing for differences from a site that is a control, then Fligner-Wolfe is a possibility. Other tests with more specific null hypotheses, e.g., ordered means, also exist.

My reference is Hollander and Wolfe, Nonparametric Statistical Methods, probably a little out of date by now as I have the 2nd edition.