RsquareAdj not returning $adj.r.squared in versions 2.5+ (including 2.6) #295

LucasMS · 2018-11-06T13:42:55Z

Hi,

I noticed that RsquareAdj do not return $adj.r.squared in the package versions 2.5+ (including 2.6). It works as expected with package version 2.4-5.

Reproducing the error:

Does work

> library(vegan)
> library(devtools)
> data(dune)
> data(dune.env)
> dune.dist <- vegdist(dune) 
> m <- dbrda(dune.dist ~ Management + Condition(Moisture + A1), data=dune.env)
> RsquareAdj(m)
$r.squared
[1] 0.2101923

$adj.r.squared
[1] 0.1568308


> session_info()
Session info -------------------------------------------------------------
 setting  value                       
 version  R version 3.4.3 (2017-11-30)
 system   x86_64, linux-gnu           
 ui       RStudio (1.1.456)           
 language (EN)                        
 collate  en_US.UTF-8                 
 tz       CET                         
 date     2018-11-06                  

Packages -----------------------------------------------------------------
 package   * version date       source        
 base      * 3.4.3   2018-03-08 local         
 cluster     2.0.6   2017-03-16 CRAN (R 3.4.3)
 compiler    3.4.3   2018-03-08 local         
 datasets  * 3.4.3   2018-03-08 local         
 devtools  * 1.13.4  2017-11-09 CRAN (R 3.4.3)
 digest      0.6.18  2018-10-10 CRAN (R 3.4.3)
 graphics  * 3.4.3   2018-03-08 local         
 grDevices * 3.4.3   2018-03-08 local         
 grid        3.4.3   2018-03-08 local         
 lattice   * 0.20-35 2017-03-25 CRAN (R 3.4.3)
 MASS        7.3-47  2017-04-21 CRAN (R 3.4.3)
 Matrix      1.2-12  2017-11-16 CRAN (R 3.4.3)
 memoise     1.1.0   2017-04-21 CRAN (R 3.4.3)
 methods   * 3.4.3   2018-03-08 local         
 mgcv        1.8-22  2017-09-19 CRAN (R 3.4.3)
 nlme        3.1-131 2017-02-06 CRAN (R 3.4.3)
 parallel    3.4.3   2018-03-08 local         
 permute   * 0.9-4   2016-09-09 CRAN (R 3.4.3)
 stats     * 3.4.3   2018-03-08 local         
 tools       3.4.3   2018-03-08 local         
 utils     * 3.4.3   2018-03-08 local         
 vegan     * 2.4-5   2017-12-01 CRAN (R 3.4.3)
 withr       2.1.2   2018-03-15 cran (@2.1.2) 
 yaml        2.2.0   2018-07-25 CRAN (R 3.4.3)

Does not work

> library(vegan)
> library(devtools)
> data(dune)
> data(dune.env)
> 
> dune.dist <- vegdist(dune) 
> m <- dbrda(dune.dist ~ Management + Condition(Moisture + A1), data=dune.env)
> RsquareAdj(m)
$r.squared
[1] 0.2101923

$adj.r.squared
[1] NA

> session_info()
Session info --------------------
 setting  value                       
 version  R version 3.5.1 (2018-07-02)
 system   x86_64, linux-gnu           
 ui       RStudio (1.1.456)           
 language (EN)                        
 collate  en_US.UTF-8                 
 tz       CET                         
 date     2018-11-06                  

Packages --------------------
 package   * version date       source                          
 base      * 3.5.1   2018-08-10 local                           
 cluster     2.0.7-1 2018-04-09 CRAN (R 3.5.1)                  
 compiler    3.5.1   2018-08-10 local                           
 datasets  * 3.5.1   2018-08-10 local                           
 devtools  * 1.13.5  2018-02-18 CRAN (R 3.5.1)                  
 digest      0.6.15  2018-01-28 CRAN (R 3.5.1)                  
 graphics  * 3.5.1   2018-08-10 local                           
 grDevices * 3.5.1   2018-08-10 local                           
 grid        3.5.1   2018-08-10 local                           
 lattice   * 0.20-35 2017-03-25 CRAN (R 3.5.1)                  
 MASS        7.3-50  2018-04-30 CRAN (R 3.5.1)                  
 Matrix      1.2-14  2018-04-09 CRAN (R 3.5.1)                  
 memoise     1.1.0   2017-04-21 CRAN (R 3.5.1)                  
 methods   * 3.5.1   2018-08-10 local                           
 mgcv        1.8-24  2018-06-18 CRAN (R 3.5.1)                  
 nlme        3.1-137 2018-04-07 CRAN (R 3.5.1)                  
 parallel    3.5.1   2018-08-10 local                           
 permute   * 0.9-4   2016-09-09 CRAN (R 3.5.1)                  
 stats     * 3.5.1   2018-08-10 local                           
 tools       3.5.1   2018-08-10 local                           
 utils     * 3.5.1   2018-08-10 local                           
 vegan     * 2.6-0   2018-11-06 Github (vegandevs/vegan@6097c89)
 withr       2.1.2   2018-03-15 CRAN (R 3.5.1)                  
 yaml        2.1.19  2018-05-01 CRAN (R 3.5.1)

By the way, many thanks for this awesome package and your support to the community!

Cheers,
Lucas

The text was updated successfully, but these errors were encountered:

jarioksa · 2018-11-06T15:29:42Z

This is even documented -- NEWS for the release 2.5-1 say:

    • Adjusted R-squared is not calculated for partial ordination,
       because it is unclear how this should be done (function
       `RsquareAdj`).

We used one way in vegan, but after inspecting alternative ways of defining R²_adj I was no longer sure that the way we used was correct and compelling for partial ordination. Therefore the feature was disabled in partial ordination. Just as soon as I get convinced on the correct way, I can re-implement the feature. Perhaps after I get response to my queries (@legendre @gavinsimpson) the situation can change.

LucasMS · 2018-11-06T15:55:58Z

Hey Jari,

Thanks for the fast reply! I completely missed this release note.

Yeah, I see that is wise not to implement something that you are not sure how it is better done. Would this also apply for AdjRsquare of varpart function?

For instance, considering the example in my post, the adj.r.squared that I get from RsquareAdj(m) using version 2.4-5 is 0.1568308. This is identical to the Adj.R.Square that I get calling varpart for the individual fraction [a] = X1 | X2+X3 obtained with version 2.6:

In 2.4-5

> dune.dist <- vegdist(dune) 
> m <- dbrda(dune.dist ~ Management + Condition(Moisture + A1), data=dune.env)
> RsquareAdj(m)
$r.squared
[1] 0.2101923

$adj.r.squared
[1] 0.1568308

In 2.6

varpart(dune.dist, ~ Management, ~ Moisture, ~ A1, data = dune.env)

#Result 
                      Df R.square Adj.R.square Testable
[a] = X1 | X2+X3       3               0.15683

I am basically mimicking varpart using Condition as in an old tutorial of yours Constrained Ordination: Tutorial with R and vegan (jan, 2012). Now wondering if I should stop using this approach until further reply from @legendre and @gavinsimpson.

Cheers,
Lucas

legendre · 2018-11-06T19:36:15Z

Hello Jari. About the adjusted R-square in partial regression and partial RDA: 1. Please open the "Numerical ecology with R" (2018, orange) book. In the last paragraph of page 234, we explain how the R2.adj is computed in partial RDA. There is a reference to the Peres-Neto et al. (2006) paper. We are happy with that computation method until proven that this calculation method produces results that are incorrect. In the meantime, it goes around the difficulty that R2.adj cannot be computed directly using an equation. Even if only for that, I believe the procedure should be maintained in vegan. 2. What are the "alternative ways of defining R2.adj" that you are referring to in your message? Pierre ========

…

Le 6 nov. 2018 à 10:29, Jari Oksanen ***@***.***> a écrit : This is even documented -- NEWS for the release 2.5-1 say: • Adjusted R-squared is not calculated for partial ordination, because it is unclear how this should be done (function `RsquareAdj`). We used one way in vegan, but after inspecting alternative ways of defining R2adj I was no longer sure that the way we used was correct and compelling for partial ordination. Therefore the feature was disabled in partial ordination. Just as soon as I get convinced on the correct way, I can re-implement the feature. Perhaps after I get response to my queries ***@***.*** <https://github.com/legendre> @gavinsimpson <https://github.com/gavinsimpson>) the situation can change. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#295 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABf7nM-mQ6MyBJlihGoAPG1jN8efjkEEks5usarpgaJpZM4YQeCz>.

--- Pierre Legendre ------------------- Courriel: Pierre.Legendre@UMontreal.CA Département de sciences biologiques, Université de Montréal, C.P. 6128, succ. Centre-ville, Tel.: (514) 343-7591 Montréal, Québec H3C 3J7, Canada Fax : (514) 343-2293 http://www.numericalecology.com/ ** Nouveaux livres -- ** Legendre & Legendre 2012: Numerical ecology, 3rd English ed (Elsevier). ** Borcard, Gillet & Legendre 2014: Numerical ecology with R, Chinese ed. ** Borcard, Gillet & Legendre 2018: Numerical ecology with R, 2nd English ed.

jarioksa · 2019-01-21T14:48:15Z

@legendre : A little bit of history. We have RsquareAdj for CCA, but that function did not explicitly handle partial CCA. However, it returned adjusted R² also for partial models using the same simulation method as for non-partial models. This was inconsistent with adjusted R² for RDA and with varpart. It is possible to implement consistent adjusted method for CCA, but it is pretty complicated. When I had a look at the issue, I had my doubts on the correct way of doing these things – both for the CCA and for RDA – and decided to disable R_adj² for both temporarily. Soon after that some non-vegan matters took all my attention, and I forgot to settle the issue. I'll go through my papers and possibly I'll mail a working paper to you.

The previous method was similar as in your NEwR book, and it can be re-instated by reverting a commit. Then we should still decide what to do with the partial CCA: (1) leave it NA, like now, or (2) implement the longer method that is consistent with RDA, but requires two permutation cycles.

I have made this issue a milestone for 2.5-4, and we can have a new release (fixing issue #299) with merged changes as soon as we decide what to change (back).

The commit history and my comments can be found with git.

This reverts commit 32fd2e0. see github issue #295

… varpart" This reverts commit 6ac3365. see github issue #295

jarioksa · 2019-01-27T12:55:50Z

I have now re-instated adjusted R² for pRDA, and also added a consistent version of the same for pCCA on branch restore-partial-R2-adj. Both of these are also consistent with varpart. This does not mean that I would be certain that these should be implemented in this way (also concerns varpart).

legendre · 2019-01-27T19:44:48Z

Thank you, Jari, for putting back the calculation of adjusted R-squares in pRDA function. [[ Why do we call it pRDA? There is no pRDA funciton in vegan! But I know what you are talking about: pRDA is the pseudo-function that you sent me once and that I included in Table 11.5 of "Numerical eoology" (2012 edition), with your permission.]] Thank you for having also fixed the the goodness problem. These is very good news for our R book and for users. Pierre ========

…

Le 27 janv. 2019 à 07:55, Jari Oksanen ***@***.***> a écrit : I have no re-instated adjusted R2 for pRDA, and also added a consistent version of the same for pCCA on branch restore-partial-R2-adj. Both of these are also consistent with varpart. This does not mean that I would be certain that these should be implemented in this way (also concerns varpart). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#295 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABf7nL9ixJ8psFd3d_D1DVbPRBx9w1LRks5vHaHXgaJpZM4YQeCz>.

--- Pierre Legendre ------------------- Courriel: Pierre.Legendre@UMontreal.CA Département de sciences biologiques, Université de Montréal, C.P. 6128, succ. Centre-ville, Tel.: (514) 343-7591 Montréal, Québec H3C 3J7, Canada Fax : (514) 343-2293 http://www.numericalecology.com/ ** Nouveaux livres -- ** Legendre & Legendre 2012: Numerical ecology, 3rd English ed (Elsevier). ** Borcard, Gillet & Legendre 2014: Numerical ecology with R, Chinese ed. ** Borcard, Gillet & Legendre 2018: Numerical ecology with R, 2nd English ed.

LucasMS · 2019-01-28T09:17:19Z

Thank you @jarioksa for the followup and re-implementation of the function.
Thank you both @jarioksa @legendre for the availability and discussion.

Cheers,

Lucas

jarioksa · 2019-01-28T17:52:50Z

@legendre , there is no pRDA function in vegan, but there is rda function that can also do pRDA (the method) as well as PCA. Just as well, function cca can do CA, CCA and pCCA, and function dbrda can do PCoA, dbRDA and p-dbRDA.

legendre · 2019-01-28T19:57:30Z

Thanks Jari. Now I will know what ansewer to give to students who often ask me why the function in Table 11.5 in the "Numerical ecology" (2012) book is called pRDA. I remember the reaction of Cajo when he first saw it: he wrote back and described the funciton as "superb programming". Pierre ========

…

Le 28 janv. 2019 à 12:52, Jari Oksanen ***@***.***> a écrit : @legendre <https://github.com/legendre> , there is no pRDA function in vegan, but there is rda function that can also do pRDA (the method) as well as PCA. Just as well, function cca can do CA, CCA and pCCA, and function dbrda can do PCoA, dbRDA and p-dbRDA. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#295 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABf7nGab6_YgV-XStF1lj4MgOVS7brgyks5vHzj0gaJpZM4YQeCz>.

--- Pierre Legendre ------------------- Courriel: Pierre.Legendre@UMontreal.CA Département de sciences biologiques, Université de Montréal, C.P. 6128, succ. Centre-ville, Tel.: (514) 343-7591 Montréal, Québec H3C 3J7, Canada Fax : (514) 343-2293 http://www.numericalecology.com/ ** Nouveaux livres -- ** Legendre & Legendre 2012: Numerical ecology, 3rd English ed (Elsevier). ** Borcard, Gillet & Legendre 2014: Numerical ecology with R, Chinese ed. ** Borcard, Gillet & Legendre 2018: Numerical ecology with R, 2nd English ed.

This merge re-enables calculation of adj-R2 for partial RDA that was removed in vegan release 2.5-1. It also implements adj-R2 for partial CCA in a way that is consistent with RDA and varpart. Before 2.5-1 release CCA was inconsistent. See github issue #295

This reverts commit 32fd2e0. see github issue #295 (cherry picked from commit 943408c)

… varpart" This reverts commit 6ac3365. see github issue #295 (cherry picked from commit 6201d51)

jarioksa · 2019-01-29T11:44:46Z

Closed with merge commit 40584a1 which re-enables RsquareAdj for partial RDA and implements consistent permutation-based calculation for partial CCA.

jarioksa added this to the 2.5-4 milestone Jan 16, 2019

jarioksa mentioned this issue Jan 26, 2019

Release 2.5-4 before Feb 8, 2019 #305

Closed

jarioksa pushed a commit that referenced this issue Jan 27, 2019

Revert "disable adjusted R2 for partial ordination (pRDA, pCCA)"

943408c

This reverts commit 32fd2e0. see github issue #295

jarioksa pushed a commit that referenced this issue Jan 27, 2019

Revert "use natural way of calculationg RsquareAdj for cca: conflicts…

6201d51

… varpart" This reverts commit 6ac3365. see github issue #295

jarioksa pushed a commit that referenced this issue Jan 29, 2019

Revert "disable adjusted R2 for partial ordination (pRDA, pCCA)"

5ab1393

This reverts commit 32fd2e0. see github issue #295 (cherry picked from commit 943408c)

jarioksa pushed a commit that referenced this issue Jan 29, 2019

Revert "use natural way of calculationg RsquareAdj for cca: conflicts…

d8c541f

… varpart" This reverts commit 6ac3365. see github issue #295 (cherry picked from commit 6201d51)

jarioksa closed this as completed Jan 29, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

RsquareAdj not returning $adj.r.squared in versions 2.5+ (including 2.6) #295

RsquareAdj not returning $adj.r.squared in versions 2.5+ (including 2.6) #295

LucasMS commented Nov 6, 2018

jarioksa commented Nov 6, 2018

LucasMS commented Nov 6, 2018 •

edited

Loading

legendre commented Nov 6, 2018 via email

jarioksa commented Jan 21, 2019

jarioksa commented Jan 27, 2019 •

edited

Loading

legendre commented Jan 27, 2019 via email

LucasMS commented Jan 28, 2019

jarioksa commented Jan 28, 2019

legendre commented Jan 28, 2019 via email

jarioksa commented Jan 29, 2019

RsquareAdj not returning $adj.r.squared in versions 2.5+ (including 2.6) #295

RsquareAdj not returning $adj.r.squared in versions 2.5+ (including 2.6) #295

Comments

LucasMS commented Nov 6, 2018

jarioksa commented Nov 6, 2018

LucasMS commented Nov 6, 2018 • edited Loading

legendre commented Nov 6, 2018 via email

jarioksa commented Jan 21, 2019

jarioksa commented Jan 27, 2019 • edited Loading

legendre commented Jan 27, 2019 via email

LucasMS commented Jan 28, 2019

jarioksa commented Jan 28, 2019

legendre commented Jan 28, 2019 via email

jarioksa commented Jan 29, 2019

LucasMS commented Nov 6, 2018 •

edited

Loading

jarioksa commented Jan 27, 2019 •

edited

Loading