different result for permutest (vegan, R) and permdisp (PERMANOVA/Primer)

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

different result for permutest (vegan, R) and permdisp (PERMANOVA/Primer)

ellenpape
Dear all,

As I want to switch for my multivariate data analyses from PRIMER/PERMANOVA
to R, I am comparing all tests I am doing in the former with those
performed in the latter program, to see whether results agree.

I performed a test for the homogeneity of multivariate dispersions in both
programs (R: betadisper + permutest, Primer: PERMDISP, both using
centroids), and I have very different results, i.e. very different P
values. The P value of the permutest (in R) is 0.0013; the one of PERMDISP
is 0.30. I am really puzzled by this difference, as:

- I have used the same pre-treatment for the data, i.e. standardize (divide
all entries by row totals, with 1 sample per row) and square-root transform
-I have used Bray-Curtis dissmilarities in both cases; however in
Primer/Permanova you calculate a resemblance matrix and thereon you perform
the PERMDISP. In R you calculate Bray-Curtis distances, which you then
submit to betadisper and then you can test significance of differences
using permutations with permutest (if you don't want to use ANOVA)

I have looked at the Bray-Curtis resemblance values in Primer/Permanova and
I have converted them to distances (Tools-> Dissim), but I get different
values for Bray-Curtis distances (though they all seem to be higher for
Primer)! I have pasted the values in Excel so you can have a look for
yourself. I think this might be the reason for these underlying differences?


One thing I also saw was,  in the case of my dataset, where I have two
groups of 3 observations each, permutest had to perform a complete
enumeration of permutations (=720-1=719), but that the PERMDISP in
Primer/PERMANOVA still gave as output that it did 9999 permutations, which
then seems inpossible to me!

Thank you for any advise!

Bets regards
Ellen
_______________________________________________
R-sig-ecology mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
Reply | Threaded
Open this post in threaded view
|

Re: different result for permutest (vegan, R) and permdisp (PERMANOVA/Primer)

Cade, Brian
Ellen:  Not sure why these differences would occur but note that with two
groups of 3 observations each there are only 6!/(3!3!) = 20 possible
permutations.  Not sure where your 720 came from.  Also, I would not expect
a permutation test for homogeneity of dispersions to be very useful with
such small sample sizes.

Brian

Brian S. Cade, PhD

U. S. Geological Survey
Fort Collins Science Center
2150 Centre Ave., Bldg. C
Fort Collins, CO  80526-8818

email:  [hidden email] <[hidden email]>
tel:  970 226-9326


On Wed, Nov 23, 2016 at 5:40 AM, Ellen Pape <[hidden email]> wrote:

> Dear all,
>
> As I want to switch for my multivariate data analyses from PRIMER/PERMANOVA
> to R, I am comparing all tests I am doing in the former with those
> performed in the latter program, to see whether results agree.
>
> I performed a test for the homogeneity of multivariate dispersions in both
> programs (R: betadisper + permutest, Primer: PERMDISP, both using
> centroids), and I have very different results, i.e. very different P
> values. The P value of the permutest (in R) is 0.0013; the one of PERMDISP
> is 0.30. I am really puzzled by this difference, as:
>
> - I have used the same pre-treatment for the data, i.e. standardize (divide
> all entries by row totals, with 1 sample per row) and square-root transform
> -I have used Bray-Curtis dissmilarities in both cases; however in
> Primer/Permanova you calculate a resemblance matrix and thereon you perform
> the PERMDISP. In R you calculate Bray-Curtis distances, which you then
> submit to betadisper and then you can test significance of differences
> using permutations with permutest (if you don't want to use ANOVA)
>
> I have looked at the Bray-Curtis resemblance values in Primer/Permanova and
> I have converted them to distances (Tools-> Dissim), but I get different
> values for Bray-Curtis distances (though they all seem to be higher for
> Primer)! I have pasted the values in Excel so you can have a look for
> yourself. I think this might be the reason for these underlying
> differences?
>
>
> One thing I also saw was,  in the case of my dataset, where I have two
> groups of 3 observations each, permutest had to perform a complete
> enumeration of permutations (=720-1=719), but that the PERMDISP in
> Primer/PERMANOVA still gave as output that it did 9999 permutations, which
> then seems inpossible to me!
>
> Thank you for any advise!
>
> Bets regards
> Ellen
> _______________________________________________
> R-sig-ecology mailing list
> [hidden email]
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>
>

        [[alternative HTML version deleted]]

_______________________________________________
R-sig-ecology mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology