pakillo / lm-glm-glmm-intro Goto Github PK

con easystats

https://easystats.github.io/report/index.html

• https://github.com/richierocks/gpk
• https://gitlab.gwdg.de/aleha/paperplanes
• http://blog.yhat.com/posts/7-funny-datasets.html

See https://www.sciencedirect.com/science/article/pii/S0304380008002305

https://github.com/ACCLAB/dabestr

https://easystats.github.io/performance/reference/check_model.html

https://easystats.github.io/see/articles/performance.html

https://github.com/goodekat/ggResidpanel

Particularly with smallish samples

See https://twitter.com/carlislerainey/status/1686389777225113601 & https://doi.org/10.1017/psrm.2021.9

Very nice write-up on solving convergence problems in GLM/GLMM by M. Clark: https://m-clark.github.io/posts/2020-03-16-convergence/

See https://tjmahr.github.io/plotting-partial-pooling-in-mixed-effects-models/

mammal sleep and paperplanes datasets have some complications (e.g. non-linear patterns) that probably advise against using them for such a very introduction to linear models...
Instead switched from iris to my simulated trees dataset (e48427c). But keeping both branches for the record (iris and trees)

https://bbolker.github.io/mixedmodels-misc/glmmFAQ.html

https://datalorax.github.io/equatiomatic/

See http://www.fromthebottomoftheheap.net/2016/06/07/rootograms/

nuts dataset? http://vincentarelbundock.github.io/Rdatasets/doc/COUNT/nuts.html

Table generated with xtable, stargazer...

especially to interpret model with interactions, etc

same group, different group, etc

and maybe switch to Metropolis beamer theme?

Or some other example: Many at Stat2data package: https://cran.r-project.org/web/packages/Stat2Data/Stat2Data.pdf

https://stats.idre.ucla.edu/r/dae/multinomial-logistic-regression/

https://stats.idre.ucla.edu/r/dae/ordinal-logistic-regression/

https://vincentarelbundock.github.io/marginaleffects/

to produce tables and plots summarising models
https://modelsummary.com/

https://modeloriented.github.io/modelStudio/

https://vincentarelbundock.github.io/marginaleffects/articles/svalues.html

In the first example (Heght ~ DBH)

trees dataset may be too biological for some

e.g.

https://www.andrewheiss.com/blog/2022/09/26/guide-visualizing-types-posteriors/

https://www.datacamp.com/groups/education/

https://stats.idre.ucla.edu/r/dae/logit-regression/

https://m-clark.github.io/posts/2019-05-14-shrinkage-in-mixed-models/

https://www.tjmahr.com/plotting-partial-pooling-in-mixed-effects-models/

Many definitions! Use some standard readily available in CRAN packages, e.g.

• r2 from sjstats
• r.squaredGLMM from MuMIn

• https://github.com/seananderson/glmm-course

https://github.com/Pakillo/pacotools/blob/master/R/gfitLogReg.R

• Share slides/handout earlier.
• Share live script in class through Dropbox or similar so that students never get lost with code (see http://environmentalpolicy.ucdavis.edu/blog/2015/09/394). Casto also looks nice for that (works better in Chrome than Firefox).
• Ask questions/live polls in class to detect misunderstandings and unknowns as early as possible (e.g. using https://getkahoot.com/ or https://polleverywhere.com)
• Have voluntary homework for those students who want to have a go at home.
• Omit mixed models?
• Have more examples of molecular/lab biology.
• Change Iris by a more exciting dataset (e.g. pigeon racing, cup stacking speed, etc http://blog.yhat.com/posts/7-funny-datasets.html; belly button biodiversity http://navels.yourwildlife.org/). Or the mammal sleep dataset. Or employment trajectories of Spanish B.S. after graduating. Or this dataset on paper planes.
• Explain better that logistic regression is a non-linear model: effect of continuous predictor depends on where we are in the logistic curve (maybe also mention Gelman's divide-by-4 rule). Hence interpretation of allEffects with categorical and continuous predictor is not straightforward (plots much better).
• Give some tips about model selection, increasing model complexity, choosing model with or without interactions, etc
• Put more emphasis on calibration plots (obs-predicted)
• Emphasize y ~ Normal(a + bx, sigma2) terminology over y = a + bx. More robust and extrapolable to other distributions later

See lme4 manual (http://lme4.r-forge.r-project.org/book/Ch2.pdf) and Bolker's article.
Nested random effects (1|a/b) are actually equivalent to (1|a) + (1|b), as long as levels in b are clearly specified so they can't be confounded. See also https://m-clark.github.io/mixed-models-with-R/extensions.html#crossed-vs.nested

https://doi.org/10.1073/pnas.1207156109 (HT F. Harrell)