What is the difference between the two sets of statistic tests?

Which is recommended for finding out if the difference between groups is more significant than within groups, for non-parametric stable isotopic signature data?

  • $\begingroup$ How exactly is data nonparametric? For those of us who are you know, statisticians and so not necessarily versed in your area, what's 'isotopic signature data' actually consist of? I presume it's relative concentrations of various isotopes in some kind of sample, but more info would help. $\endgroup$ – Glen_b Jun 5 '13 at 7:24
  • $\begingroup$ Since I assume this will be an unfamiliar term to more than just me: Link on ANOSIM. I see that the R package vegan can do this. $\endgroup$ – Glen_b Jun 5 '13 at 7:24
  • $\begingroup$ @Glen_b Although an isotope chemist might not agree, for all intents and purposes you can consider the data to range from -Inf to +Inf. Isotope ratios are in permille units and are the relative ratio of heavy to light isotope in a standard and in a sample. $\endgroup$ – Gavin Simpson Jul 5 '13 at 15:32

ANOSIM was designed for multivariate responses, Kruskal-Wallis for a univariate response. ANOSIM converts the response to distances/dissimilarities and computes rank from those whereas Kruskal-Wallis ranks the observations themselves.

Unless you have multivariate responses (i.e. a matrix for $y$) then I would probably avoid ANOSIM. If you do then I would probably direct you to the adonis() function in the vegan R package as it does a multivariate form of ANOVA but using dissimilarities; it is generally considered a more robust approach than ANOSIM.

You will need to explain more about your specific data and hypothesis for us to be able to comment more.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.