[R-sig-ME] R Consortium call for funding

[Ben Bolker]

https://www.r-consortium.org/announcement/2018/09/25/fall-2018-isc-call-for-proposals

"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"

The mixed-model ecosystem is admittedly a small part of the R
ecosystem, but I (biasedly) think it's an important one.

If people have ideas & opinions about how a chunk of money on the
order of $10,000 could be valuably spent to improve the mixed-model
ecosystem in a way that would be appealing to a very broad audience of
useRs, please discuss.

The deadline for submitting a proposal is midnight PST, Sunday October
31, 2018.


cheers
Ben Bolker

[Andrew Robinson]

Hi Ben,

I do really miss the variance functions of nlme, and the temporal and
spatial autocorrelation handling.

Warm wishes,

Andrew

Post by Ben Bolker
https://www.r-consortium.org/announcement/2018/09/25/fall-
2018-isc-call-for-proposals
"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"
The mixed-model ecosystem is admittedly a small part of the R
ecosystem, but I (biasedly) think it's an important one.
If people have ideas & opinions about how a chunk of money on the
order of $10,000 could be valuably spent to improve the mixed-model
ecosystem in a way that would be appealing to a very broad audience of
useRs, please discuss.
The deadline for submitting a proposal is midnight PST, Sunday October
31, 2018.
cheers
Ben Bolker
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

--
Andrew Robinson
Director, CEBRA, School of BioSciences
Reader & Associate Professor in Applied Statistics Tel: [phone]
School of Mathematics and Statistics Fax: (+61) 03
8344 4599
University of Melbourne, VIC 3010 Australia
Email: ***@unimelb.edu.au
Website: http://www.ms.unimelb.edu.au/~andrewpr

[[alternative HTML version deleted]]

[Lize van der Merwe]

Thankyou.
I would like to see the functionality of nlme and lmer4 combined, especially
regarding variance-covariance structure.
Regards
Lize van der Merwe



-----Original Message-----
From: R-sig-mixed-models <r-sig-mixed-models-***@r-project.org> On
Behalf Of Ben Bolker
Sent: Tuesday, 25 September 2018 22:51
To: r-sig-mixed-***@r-project.org
Subject: [R-sig-ME] R Consortium call for funding


https://www.r-consortium.org/announcement/2018/09/25/fall-2018-isc-call-for-
proposals

"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"

The mixed-model ecosystem is admittedly a small part of the R ecosystem,
but I (biasedly) think it's an important one.

If people have ideas & opinions about how a chunk of money on the order of
$10,000 could be valuably spent to improve the mixed-model ecosystem in a
way that would be appealing to a very broad audience of useRs, please
discuss.

The deadline for submitting a proposal is midnight PST, Sunday October 31,
2018.


cheers
Ben Bolker

_______________________________________________
R-sig-mixed-***@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[Martin Maechler]

Post by Ben Bolker

Peter Claussen via R-sig-mixed-models

Personally, I’m trying to duplicate some standard repeated measures analysis from SAS and am missing Kenward-Roger and Satterthwaite corrections for nlme. I had some concerns about lmeTest, but I haven’t looked at that in a while - I’m a bit more concerned with structure error covariances.

I assume you mean 'lmerTest' above, the CRAN package originally
mostly by Rune Haubo Christensen
https://cran.r-project.org/package=lmerTest

I think its Satterthwaite (and Kenward-Roger via CRAN package
'pbkrtest') 'df' approximation computations have become quite
reliable, after Rune's recent refurbishing, see also
https://htmlpreview.github.io/?https://github.com/runehaubo/lmerTestR/blob/master/pkg_notes/Satterthwaite_for_LMMs.html

-- as always at least as long as you refrain from fitting much
overparametrized models -- which you should even though some
authors tell you to do so.


Martin Maechler
ETH Zurich

Cheers,

Post by Ben Bolker
https://www.r-consortium.org/announcement/2018/09/25/fall-2018-isc-call-for-proposals
"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"
The mixed-model ecosystem is admittedly a small part of the R
ecosystem, but I (biasedly) think it's an important one.
If people have ideas & opinions about how a chunk of money on the
order of $10,000 could be valuably spent to improve the mixed-model
ecosystem in a way that would be appealing to a very broad audience of
useRs, please discuss.
The deadline for submitting a proposal is midnight PST, Sunday October
31, 2018.
cheers
Ben Bolker
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[Manuel Ramon]

I totally agree with you, Ben. I have to admit that all the tidyverse world
has suppose a great improvement in the way I work with data, but in the
end, almost all my analyses conclude with the nlme/lme4 packages. So I
think it is worth investing funds and time on it.

As suggested by others, the inclusion of the variance functions from nlme
would be very useful. Also, some of the capabilities of the mixed.models in
Julia language in terms of computation time and data size would be very
welcome, but this latter it is probably very difficult (almost impossible)
given that they are to different platforms.

In any case, thanks for the initiative and I hope it will go ahead.

Regards,
Manuel

Post by Ben Bolker
https://www.r-consortium.org/announcement/2018/09/25/fall-2018-isc-call-for-proposals
"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"
The mixed-model ecosystem is admittedly a small part of the R
ecosystem, but I (biasedly) think it's an important one.
If people have ideas & opinions about how a chunk of money on the
order of $10,000 could be valuably spent to improve the mixed-model
ecosystem in a way that would be appealing to a very broad audience of
useRs, please discuss.
The deadline for submitting a proposal is midnight PST, Sunday October
31, 2018.
cheers
Ben Bolker
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[[alternative HTML version deleted]]

[Mollie Brooks]

Lize and Manuel brought up covariance structures and speed, so I wanted to let you all know that the glmmTMB developers have been working towards these goals (https://cran.r-project.org/web/packages/glmmTMB/vignettes/covstruct.html <https://cran.r-project.org/web/packages/glmmTMB/vignettes/covstruct.html>). It’s also easy to add additional structures to glmmTMB. We could use some help testing the covariance structures (https://github.com/glmmTMB/glmmTMB/issues/344 <https://github.com/glmmTMB/glmmTMB/issues/344>). We recently found a bug that could cause problems in models with multiple types of covariance structures, but it has been fixed if you install the development (i.e. Github) version of lme4 and the fix_covstruct_order2 branch of glmmTMB (https://github.com/glmmTMB/glmmTMB/tree/fix_covstruct_order2 <https://github.com/glmmTMB/glmmTMB/tree/fix_covstruct_order2>). These should both be on CRAN soon.


In my opinion, the biggest need for improvement is to provide predictions and coefficients with confidence intervals on a meaningful scale when a nonlinear link function is used. This comes up repeatedly on this list (e.g. earlier this month https://stat.ethz.ch/pipermail/r-sig-mixed-models/2018q3/027237.html). The solution will probably involve marginalizing over random effects, but non-parametric bootstrapping while resampling the levels of random effects could also be useful.

cheers,
Mollie

Post by Manuel Ramon
I totally agree with you, Ben. I have to admit that all the tidyverse world
has suppose a great improvement in the way I work with data, but in the
end, almost all my analyses conclude with the nlme/lme4 packages. So I
think it is worth investing funds and time on it.
As suggested by others, the inclusion of the variance functions from nlme
would be very useful. Also, some of the capabilities of the mixed.models in
Julia language in terms of computation time and data size would be very
welcome, but this latter it is probably very difficult (almost impossible)
given that they are to different platforms.
In any case, thanks for the initiative and I hope it will go ahead.
Regards,
Manuel

Post by Ben Bolker
https://www.r-consortium.org/announcement/2018/09/25/fall-2018-isc-call-for-proposals
"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"
The mixed-model ecosystem is admittedly a small part of the R
ecosystem, but I (biasedly) think it's an important one.
If people have ideas & opinions about how a chunk of money on the
order of $10,000 could be valuably spent to improve the mixed-model
ecosystem in a way that would be appealing to a very broad audience of
useRs, please discuss.
The deadline for submitting a proposal is midnight PST, Sunday October
31, 2018.
cheers
Ben Bolker
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[[alternative HTML version deleted]]

[Phillip Alday]

I'm a bit late to the game on this one, but I would second the votes for
flexible covariance structures as in nlme. Perhaps this could be done by
investing funds into the flexlambda work?

Regarding two other suggestions:

1. Julia-like speed. There are bridges between Julia and R, but this
still doesn't help me once I have a fitted model in Julia and want the R
mixed-models ecosystem (effects, car::Anova(), lmerTest, etc.) to
examine the model. It should however be possible to construct a merMod
object from the fit in Julia. Tools for doing this would be quite nice.
(This also seems like relatively low-hanging fruit for a Google summer
of code type project.)

2. Better confidence intervals and predictions for non-linear links. I
think some parts of this are implemented in the effects and emmeans
packages. In addition to a more extensive/complete implementation, this
seems like something where additional documentation and worked examples
comparing conditional and marginalized coefficients would be useful,
potentially as part of the GLMM FAQ.

Best,
Phillip

Post by Manuel Ramon
I totally agree with you, Ben. I have to admit that all the tidyverse world
has suppose a great improvement in the way I work with data, but in the
end, almost all my analyses conclude with the nlme/lme4 packages. So I
think it is worth investing funds and time on it.
As suggested by others, the inclusion of the variance functions from nlme
would be very useful. Also, some of the capabilities of the mixed.models in
Julia language in terms of computation time and data size would be very
welcome, but this latter it is probably very difficult (almost impossible)
given that they are to different platforms.
In any case, thanks for the initiative and I hope it will go ahead.
Regards,
Manuel

Post by Ben Bolker
https://www.r-consortium.org/announcement/2018/09/25/fall-2018-isc-call-for-proposals
"What can you do to improve the R ecosystem and how can the R Consortium
help you do it?"
The mixed-model ecosystem is admittedly a small part of the R
ecosystem, but I (biasedly) think it's an important one.
If people have ideas & opinions about how a chunk of money on the
order of $10,000 could be valuably spent to improve the mixed-model
ecosystem in a way that would be appealing to a very broad audience of
useRs, please discuss.
The deadline for submitting a proposal is midnight PST, Sunday October
31, 2018.
cheers
Ben Bolker
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[Trevor Walker]

It would be nice to have some of the ASReml utilities for flexible
variance-covariance models. Such as heterogenous residuals for different
levels of fixed effects. Another is coruh structures for nested random
effects (uniform covariance but heterogenous variances along the diagonal).

Trevor

[[alternative HTML version deleted]]