randtest and fourthcorner statistics for partial RLQ? Error in UseMethod("randtest" / "fourthcorner.rlq") : no applicable method for 'randtest' applied to an object of class "c('within', 'dudi')"

[maurinka]

How can I evaluate global significance of a partial RLQ (wca.rlq), and calculate the contributions of the individual variables to the RLQ axes just as described by Wesuls et al. (2012, doi: 10.1111/j.1654-1103.2011.01342.x)?

i.e. what would be the corresponding functions for:

nrepet = 999
testrlq.env <- randtest(rlq.env, modeltype = 6, nrepet = nrepet)

randtest(wrlq_crop, modeltype = 6, nrepet = nrepet) # doesnt work
Error in UseMethod("randtest") :
no applicable method for 'randtest' applied to an object of class "c('within', 'dudi')"

and

testQaxes.comb.env <- fourthcorner.rlq(rlq.env, modeltype = 6, typetest = "Q.axes", nrepet = nrepet, p.adjust.method.G = "fdr", p.adjust.method.D = "fdr")
testRaxes.comb.env <- fourthcorner.rlq(rlq.env, modeltype = 6, typetest = "R.axes", nrepet = nrepet, p.adjust.method.G = "fdr", p.adjust.method.D = "fdr")

fourthcorner.rlq(wrlq_crop, modeltype = 6, typetest = "R.axes", nrepet = nrepet, p.adjust.method.G = "fdr", p.adjust.method.D = "fdr")
Error in fourthcorner.rlq(wrlq_crop, modeltype = 6, typetest = "R.axes", :
Object of class 'rlq' expected

Help very much appreciated @sdray

[sdray]

Sorry for not replying before but was very busy these last days. The method is not properly implemented in ade4 and you should use a small trick. I give you some code from the original paper showing how to run the test:

## we assume that species abundances are stored in lnspec, traits in trait. Run  standard rlq and results stored in rlqTR
## habitat[,1] is a factor with a covariable that we want to remove the effect with withn-class rlq

wit1 <- within.rlq(rlqTR,habitat[,1], scannf = TRUE, nf = 2)

#then fourth-corner on the site scores of within RLQ
four1<-fourthcorner(wit1$lR,lnspec,trait,modeltype=2, nrepet=99) ## the trick : use output of within-RLQ wit1$lR instead of original environmental variables

summary(four1)
plot(four1,type = "G")
plot(four1,type = "D")

four2<-fourthcorner(wit1$lR,lnspec,trait,modeltype=4, nrepet=99)
summary(four2)
plot(four2,type = "G")
plot(four2,type = "D")

four3 <- combine.4thcorner(four1,four2)
summary(four3)
plot(four3,alpha=sqrt(0.05),type = "G")
plot(four3,alpha=sqrt(0.05),type = "D")

print(four3)

[maurinka]

thank you very much for you reply.

am I correct to assume that fourthcorner2 with $lR instead of R, and the respective test (model 2 and 4) correspond to the randtest function implemented for the RLQ?

`# corresponds to randtest:
Srlq_partial2 <- fourthcorner2(wrlq_crop$lR, dat$spe,
dat$traits,
modeltype = 2, p.adjust.method.G = "none", nrepet = nrepet)
Srlq_partial4 <- fourthcorner2(wrlq_crop$lR, dat$spe,
dat$traits,
modeltype = 4, p.adjust.method.G = "none", nrepet = nrepet)

model 6:

Srlq_partial6 <- fourthcorner2(wrlq_crop$lR, dat$spe,
dat$traits,
modeltype = 6, p.adjust.method.G = "fdr", nrepet = nrepet)

Srlq_partial2$trRLQ
Srlq_partial4$trRLQ
Srlq_partial6$trRLQ
`

To test the significance of the variables with the RLQ axes, I used forthcorner.rlq() for the RLQ

the fourthcorner() D2 test-statistic resembles the output of forthcorner(RLQ, typetest = "Q.axes"). (so I assume, it corresponds. How can I obtain the output for the "R.axes"? I want to obtain the table corresponding to table 5 in the original publication.

[sdray]

yep. so for rlq, the idea is to apply pca (without centering/scaling) on outputs of within class analysis performed on environmental variables:

pca_env <- dudi.pca(env_table)
wit_env <- wca(pca_env, habitat[,1]) # remove the effect of covar on env. table
pca_wit <- dudi.pca(wit$tab, scale = FALSE, center = FALSE, row. w= ca_abund$lw) # create a pca object that do nothing (but is accepted in randtest for rlq)
rlq(pca_wit, ca_abund, pca_trait)

Cheers

[maurinka]

sorry, i dont get it. id would be amazing (and probably more efficient) of we could quickly talk. can I send you my phonenumber by email?