decorana yields different outputs for different versions of vegan

[jannes-m]

Dear Jari,
first of all, thanks for your fantastic package which I have probably been using for more than ten years.
I have stumbled across some inconsistency regarding the output of decorona. Specifically, decorana returns different results since vegan package version 2.5-1 compared to package version 2.4-6.

So, first of all, let's download and install vegan 2.4-6 and subject the varechem dataset to a DCA. I save the decorana output in a directory created by rappdirs.

devtools::install_version("vegan", version = "2.4-6",
                          repos = "http://cran.us.r-project.org")
#> Downloading package from url: http://cran.us.r-project.org/src/contrib/Archive/vegan/vegan_2.4-6.tar.gz
#> Installing package into '/home/jannes/R/x86_64-pc-linux-gnu-library/3.5'
#> (as 'lib' is unspecified)
library("vegan")
#> Loading required package: permute
#> Loading required package: lattice
#> This is vegan 2.4-6
packageVersion("vegan")
#> [1] '2.4.6'
# attach varespec data
data("varespec")
# compute DCA
dca_1 = decorana(varespec, iweigh = 1)
# save it to a "temporary" file
saveRDS(dca_1, file.path(rappdirs::user_data_dir(), "dca_1.rds"))

^{Created on 2019-01-16 by the reprex package (v0.2.1)}

Secondly, I have terminated R and RStudio and opened RStudio again (pls note that a simple R session restart is not sufficient, instead one has to properly close R and RStudio). This time, I will install vegan from CRAN, run again a DCA on the varespec dataset and compare the results between vegan package versions 2.4-6 and 2.5-3.

devtools::install_cran("vegan")
#> Installing 1 packages: vegan
#> Installing package into '/home/jannes/R/x86_64-pc-linux-gnu-library/3.5'
#> (as 'lib' is unspecified)
library("vegan")  # 2.5.3
#> Loading required package: permute
#> Loading required package: lattice
#> This is vegan 2.5-3
packageVersion("vegan")
#> [1] '2.5.3'

# attach varespec and run the same DCA again
data("varespec")
dca_2 = decorana(varespec, iweigh = 1)

# attach first dca result
dca_1 = readRDS(file.path(rappdirs::user_data_dir(), "dca_1.rds"))

# compare DCA results 
all.equal(scores(dca_1), scores(dca_2))
#> [1] "Mean relative difference: 1.786744"
par(mfrow = c(1, 2))
plot(dca_1)
plot(dca_2)

^{Created on 2019-01-16 by the reprex package (v0.2.1)}

It turns out that the results differ substantially. So the question is which result is correct? Looking at the NEWS, is it possible that a bug was introduced when increasing the performance of decorana in package version 2.5-1?

[jarioksa]

Yes, this is a bug, and it concerns downweighted analysis (argument iweigh = 1): the row and column sums must be evaluated after downweighting. The bug is fixed in commit 0de6bc4 and the fix will be ported to vegan_2.5-4.

[jannes-m]

Thanks for the quick answer and the quick fix!!