Pedagogical and structural issues

[pedrohbraga]

There are a few pedagogical issues in this workshop.

They are mainly reflected by an inconsistency in the content between parts. The workshop, for instance, does not follow a linear narrative.

I think that this workshop would be greatly improved if it followed a structured form, where the systematic and random components are clearly stated for every variation of GLM with respect to their families and available link functions.

By repeating the random and systematic components in the GLM structure for each type of distribution [i.e., glm(edm; Link function); e.g., glm(binomial; logit), glm(binomial; probit)], we can help participants understand what is changing between the different models.

Participants and presenters usually find this workshop dense. This is mainly because there is a lack of exercises and challenges, as well as there is a need for learning strategies across the workshop. I find it that, in many moments, the material lacks theory. For instance, mathematical representations are essential to the explanation of generalized linear models. While they are missing in some parts, in others they appear (partially), without appropriate explanation. I do not agree that the theory of this workshop should be more condensed.

One suggestion I thought that could help give more room for this workshop would be to: i. move the content related to the generalized linear mixed model (GLMM) from this workshop (#7) to the one on linear mixed models (workshop #6), and ii. switch positions between both workshops. We could also extend this idea to the generalized additive model workshop (GAM; currently, workshop #8). This proposition could follow as: participants will first learn linear models, generalized linear models, generalized additive models, and then learn the mixed models and their application in the three types of models.

[pedrohbraga]

Update: I have moved the GLMM portion to the workshop on mixed models, but the issues with the linear narrative are still present within this workshop.

[pedrohbraga]

Update: the issue of linearity and clarity has been improved in commit 17bc6db79fcff2aff8c4186148f8080b6995047a.

It is still needed to clearly identify and standardize the presentation of random and systematic components in the GLM structure for each type of distribution [i.e., glm(edm; Link function); e.g., glm(binomial; logit), glm(binomial; probit)], so participants understand what is changing between the different models.

Nevertheless, I made an effort to add more theory and to build-up from general linear models to generalized linear models, their equivalency and the other variations.