CRAN task view proposal: Mixed models

[bbolker]

Description

Mixed models are a broad class of statistical models used to analyze data where observations can be assigned a priori to discrete groups, and where the parameters describing the differences between groups are treated as random variables. They are also referred to as multilevel, or hierarchical, models; longitudinal data are often analyzed in this framework.

We had already started on this task view before writing the proposal, so a repo and a draft are in progress.

Scope

The proposed task view would include packages for linear, generalized linear and nonlinear mixed model fitting (including convenience wrappers), model diagnostics and summaries, and inferential tasks (e.g. hypothesis testing, prediction and estimation). Packages for a variety of specialized data types and model structures (e.g. censored or zero-inflated data, spatial models) are included, as well as packages containing data sets that are widely used in teaching and learning about mixed models. This proposed TV would only cover models that incorporate continuous (usually although not always Gaussian) latent variables; this excludes packages that handle hidden Markov Models, finite (discrete) mixture models, latent Markov models, etc..

Packages

Please see the current draft. The current set of proposed core packages:

• lme4
• nlme
• brms
• MCMCglmm
• multilevelmod
• geepack
• lavaan
• broom.mixed
• DHARMa
• lmerTest
• glmmTMB
• blme
• rstanarm
• ordinal
• robustlmm

Overlap

There is some overlap with the Agriculture task view regarding mixed models that include a kinship or relatedness matrix. We could point to that resource or vice versa. There is overlap with the Robust which also has a mixed models section, and with the Bayesian Inference task view.

Maintainers

Ben Bolker would be the principal maintainer. Julia Piaskowski (and others) would co-maintain.

[zeileis]

Feedback as CRAN Task View Editor: Wonderful, this is a great proposal. It's very clear what goes in and what not and it's certainly a highly-relevant relevant topic. I'm very much looking forward to having it on CRAN. Details:

• Currently, the "Links" section cannot have subsections. In the Econometrics we simply have prefixes to the links.
• As sections are not numbered in the CRAN style I would probably not rely on three levels of headers but just on two.

Feedback as author: In the additive models section i think it would be appropriate to add gamlss, VGAM, and bamlss. And the GLMM trees in the glmertree package could also be incorporated into the specialized models section. Disclaimer: I'm co-auhor of bamlss and glmertree.

[rociojoo]

@bbolker It looks like a great proposal, thank you! Relevant, well structured, and easy to navigate. The multinomial responses section is empty but I guess you'll fill it soon.

[zeileis]

Both VGAM and R2BayesX support multinomial responses and should support GAMs with random-effect terms.

[tuxette]

Thanks for the proposal that looks good for me too! A few very minor suggestions:

• I don't really get why emmeans is only cited in the Prediction subsection and not in Hypothesis testing. In addition, I would have placed that package in core (but it's a personal point of view so I guess that you can disagree);
• You have a remaining "FIXME" in the section "Multinomial responses";
• You have an unnecessary comma at the end of the section "influential data points". In addition, in this part, titles do not start with a capital letter whereas they do in other parts;
• I think that, as you suggested, you should cite relevant related TV.

[zeileis]

Great, this means that we have endorsement from three CRAN Task View Editors and you can officially move forward.

Ben @bbolker, I suggest that refine the document you already have based on our feedback and then let us know about it. Also, you can try to bring on a couple of further co-maintainers, ideally persons which can add a different view, be it in terms of methodology (e.g., someone from the more "specialized" models), ethnicity, geographic region, etc.

In terms of links to other task views: Once your new task view is ready, we can contact the maintainers of the other task views and suggest that they link back to your view and possibly reduce the discussion of mixed models in their view.

[jpiaskowski]

It doesn't look like VGAM solves or supports mixed models...am I missing something?

[bbolker]

Agreed: I can't find anything. Both Yee et al 2010 and Yee et al 2015 say that allowing mixed models would be a useful extension ...

---

Yee, Thomas. 2010. “VGLMs and VGAMs: An Overview for Applications in Fisheries Research.” Fisheries Research 101 (January): 116–26. https://doi.org/10.1016/j.fishres.2009.09.015. Yee, Thomas W., Jakub Stoklosa, and Richard M. Huggins. 2015. “The VGAM Package for Capture-Recapture Data Using the Conditional Likelihood.” Journal of Statistical Software 65 (June): 1–33. https://doi.org/10.18637/jss.v065.i05.

[zeileis]

Sorry for the confusion. I thought Thomas had added it at some point but apparently not. Thanks for checking!

[jpiaskowski]

I'm happy to relocate kinship-augmented mixed models from Agriculture to this CTV once it is published, and point folks the mixed model CTV.

[strengejacke]

Thanks for pointing to this proposal. I would suggest adding following packages:

• clmm
• mixor

In the section Model presentation and prediction I would add:

• parameters
• modelsummary

In the section Prediction and estimation I would add

• marginaleffects

The packages from Vincent are quite actively developed and cover a lot of models.

[bbolker]

The last three are added.

Is there a clmm package, or is it a function within ordinal (which is already listed)? mixor has been archived (quite a while ago, on 2021-07-31). (The Environmetrics task view does include archived packages (the "archived" tag appears to be autogenerated since the code just has r pkg(...) as usual), but I think they're not supposed to be included ...?

[strengejacke]

Yes, my fault, I think clmm() is the function from pkg ordinal. And I agree that archived pkgs should not be included in the task view.

[zeileis]

Since the re-launch of the CRAN Task View Initiative archived packages do not need to be removed from task views immediately. Instead one can allow a certain period of grace (currently 60 days) before an issue is created calling for action, if necessary removal from the task view (after 100 days). See https://github.com/cran-task-views/Environmetrics/issues for some examples.

As this policy is still new, we haven't fully automated everything and are still trying to refine the process.

In any case, archived packages should not be included in a new task view unless it is very likely that the packages will be unarchived very soon.

[bbolker]

As far as I know we've resolved all of the issues discussed above. Current draft is at https://github.com/bbolker/mixedmodels-taskview/blob/main/MixedModels.md . What's the next step? Migrate that document to a new repo under this organization? Prepare an appropriate README?

[tuxette]

As for me, the current version is OK. You have to be endorsed by at least 3 editors and @zeileis will then explain how to migrate the TV.

[zeileis]

If possible, please clean up the repository a bit. I just had a quick glance and my impression was that not all files are needed going forward. (But what is still needed can remain.)

Then please make me a co-owner of the task view. Then I can streamline the README and repository information, the task view file, and transfer the repository to the cran-task-views group.

Finally, you can check whether everything is in order and ready for release.

When I have the "go" from you, I release everything on CRAN.

[bbolker]

As discussed in e-mail, I have so far failed to give you sufficient (?) privileges - repositories owned by personal accounts (which this one is) rather than organizations have only two access levels, "owner" and "collaborator".

I could transfer the repo to the ctv organization, I think:

https://docs.github.com/en/repositories/creating-and-managing-repositories/transferring-a-repository

update: tried to transfer but don't have permission to create a public repo on cran-task-views. I guess I could make it private and try again ...??

update: transferred ownership to @zeileis. Hope that works!

[zeileis]

Thanks @bbolker and apologies for the confusion regarding the transfer. For future reference: The co-admin settings are only possible for repositories hosted by organizations. Otherwise the easiest solution is to transfer the repository to one CRAN Task View Editor directly (typically me). This can be done via:

Settings > Collaborators > Public Repository > Manage

and then

Danger Zone > Transfer ownership

to "zeileis".

[zeileis]

I have completed the transfer and standardization of the task view now, see:

https://github.com/cran-task-views/MixedModels/

• I use our standard README.md template and tweaked a few formatting issues.
• In the intro paragraphs I added cross-references to the Cluster and TimeSeries task views which partially cover some of the relevant packages.
• I copied the Bioc link from the intro to the links section as well.
• I omitted the comment that bamlss relies on JAGS because this can only be used optionally. Usually, the package uses its own samplers.

Please check whether everything is ok for you.

Last detail: The list of maintainers and GitHub contributors should be synchronized (Julia is missing in the latter). And I wouldn't list the maintainers as contributors in the first paragraph of the intro (all their names are prominently in the header).

[bbolker]

Thanks, last details fixed now (everything looks fine). Presumably this automatically gets built/pushed to https://cran.r-project.org/web/views/ at some point?

[zeileis]

Yes, now! Task view online at

https://CRAN.R-project.org/view=MixedModels

and announced on Twitter at

https://twitter.com/AchimZeileis/status/1582397599390519303

Thank you all for your contribution, we're very happy to have this on CRAN!