We have a design with several levels of nestedness: lichen species richness sampled at N vs. S expositions on trees located on northern and southern versants in different macroregions.
We have tried to run a PERMANOVA analysis using the function adonis() in vegan. But I am not sure to have understood how and if it is possible to set more than one level of nestedness in the script, so that we could estimate the effects of all factors (region, versant and aspect).
Any suggestions? Thanks!
Hello. Apparently, it should be done in the same notation of a typical anova. I've found this reply from Oksanen in the R-forge forum
"Expansion of the formula to the model matrix is done in base R: it is not a vegan issue. basically, ~ A/B (nested) and ~ A*B (interaction) give the same general models, but different coefficients. Neither has anything to do with permutation. The formula defines the systematic pattern, and the 'strata' argument defines how data are permuted, or the random structure. If you have a nested error structure, so that you do not want your data be shuffled over classes (strata), you should define strata in your permutation. I have no idea if that is your case: probably you know.
cheers, Jari Oksanen "
https://r-forge.r-project.org/forum/forum.php?thread_id=2758&forum_id=194&group_id=68
You could try perMANOVA in PC-ORD but I think there would be the same problem of having only two levels of blocking factors/treatments.
The PRIMER software with the PERMANOVA+ package should be able to handle multiple levels of nestedness without issue, although it does require going outside of R.
adonis () uses the standard notation in R: matrix data ~ factor1 *factor2* ..... , in the same way anova for example.
Have you tried putting your multiple levels as arguments in the "strata" portion adonis?
Thanks Rachel: I know it works with PRIMER (or with PCORD, as suggested by Peter). till now it seems to be the easiest solution, but I have not anymore access to that software...
@Peter, I have just tried to put multiple arguments in the "strata" but it seems it doesn't work.
This is the 'normal' call with a single nested variable:
adonis(Y ~ region*versant*aspect, data=var, permutations=999, method="bray", strata = var$region)
Finally I have not a clear idea where to define the second level of nestedness...
@Alain, do you mean that adonis can recognize by itself the nestedness by putting the factor in the right order?
Thanks everybody
I have struggled with adonis for 2 days. I also want to do an analysis with multiple nested factor. Unfortunately it seems for me that the strata argument has no effect on results
But when I wrote the nested structure in simple R languange, I got a result different from the result of case using strata. For example:
adonis( Y ~ factorB, strata=factorA)
Here, factorB should be nested within factorA.
adonis( Y ~ factorA/factorB, strata=NULL)
Here, factorB is nested within factorA.
Anywy, the previous formulation is equivalent with this formulation:
adonis( Y ~ factorA + factorB + factorB%in%factorA, strata=NULL)
and with this formulation:
adonis( Y ~ factorA + factorA:factorB, strata=NULL).
Actually, my problem is that I would expect the same result for these two formaultions:
adonis( Y ~ factorA/factorB, strata=NULL)
adonis( Y ~ factorB, strata=factorA).
And I would expect a different result for these two formulations:
adonis( Y ~ factorB, strata=factorA)
adonis( Y ~ factorB, strata=NULL).
But unfortunately, this did not happend with my data, and nor the example data in the help page of the adonis function ('help(adonis)').
Hello. Apparently, it should be done in the same notation of a typical anova. I've found this reply from Oksanen in the R-forge forum
"Expansion of the formula to the model matrix is done in base R: it is not a vegan issue. basically, ~ A/B (nested) and ~ A*B (interaction) give the same general models, but different coefficients. Neither has anything to do with permutation. The formula defines the systematic pattern, and the 'strata' argument defines how data are permuted, or the random structure. If you have a nested error structure, so that you do not want your data be shuffled over classes (strata), you should define strata in your permutation. I have no idea if that is your case: probably you know.
cheers, Jari Oksanen "
https://r-forge.r-project.org/forum/forum.php?thread_id=2758&forum_id=194&group_id=68
Hello,
I just came across the following:
..."a new function enables testing two-factor nested models, nested.npmanova() in the BiodiversityR package (Kindt and Coe, 2005)"... in Zeldich et al. 2012. The book is available here: http://booksite.elsevier.com/9780123869036/content/Workbook.pdf
Greetings,
g.
Hi Paolo,
Probably is too late, because previous responses already noted that "adonis" uses the standard notation in R, but I think that this information could be useful for futures doubts related with this analysis.
Looking for information to this package, I have found a good example with more factors in the Vegan Tutorial (pp. 32-33) http://cc.oulu.fi/%7Ejarioksa/opetus/metodi/vegantutor.pdf
Cheers,
Jesús.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics