9

I have to analyze a factorial design with five factors (one of them nested in another one) and numeric responses. I would like to perform a nonparametric ANOVA, but of course I can't use neither Kruskall Wallis nor Friedman test (I have replicated measures). Is there a command or a code in R that could help me? Thank you! Stefania

  • 1
    What is the reason for a non-parametric approach? – chl Oct 15 '10 at 19:01
  • @chl The reason why I need a non-parametric approach is that my responses don't satisfy the hypothesis of the model ANOVA (non gaussian residuals, heteroskedasticity, strange dispersion of residuals versus fits) and no transformation is useful (i've also tried tried with a Box-Cox transformation with bad results). Thank you everybody for the answers!!! -- Stefania Oct 16 '10 at 7:49 – chl Sep 11 '12 at 09:13

4 Answers4

4

Tukey's Median Polish is implemented in R as medpolish {stats}. See Chapter 6 in Venables and Ripley

whuber
  • 281,159
  • 54
  • 637
  • 1,101
  • I seem to remember that Median Polish can be used for a two-way layout; how does it extend to 5 factors, including nesting? – chl Oct 15 '10 at 19:19
  • 1
    @chl Median polish can work with as many factors as you might care to handle. Tukey does some three-way examples (by hand!). – whuber Oct 15 '10 at 19:37
4

The vegan package implements permutation testing for distance based ANOVA, which should work with multi-way, repeated measures data.

ars
  • 12,160
  • 1
  • 36
  • 54
  • do you have a function command for doing that? – EDi Mar 29 '11 at 18:58
  • @EDi Lookup `adonis` in the on-line help for vegan. (And [Jari Oksanen](http://cc.oulu.fi/~jarioksa/softhelp/vegan.html) for more info on the package itself.) BTW, this is a pretty elegant answer, you can vote it up like I did a long ago :-) – chl Mar 31 '11 at 21:10
1

You might check out the ezBoot() function in the ez package for bootstrapping confidence intervals on your effects of interest.

Mike Lawrence
  • 12,691
  • 8
  • 40
  • 65
0

Pierre Legendre has some Code on his homepage: nest.anova.perm.R (D. Borcard and P. Legendre): Nested anova with permutation tests (main factor and one nested factor, balanced design).

EDi
  • 1,091
  • 10
  • 21