dschoenig
commented
4 years ago I agree with these points, especially the point on stepwise regression, and I think it should not be taught for this course (if at all briefly mentioned as a controversial extension of linear models for which better solutions exist). I think in the larger context of the R workshops, there are better methods presented in the subsequent workshops that deal with the kind of problems that prompts some people to use stepwise. I think this method
Edit: Oops accidentally closed the issue
It would be a good idea to reformulate and critically think about the "Optional Section".
For instance, the current version instructs "Stepwise regression": an approach that has been highly criticized in the literature. I have included a few disclaimers within the first slides of this topic, but ideally, this method should not be there (or one should spend time explaining why it is often not a good idea to use this type of regression).
Unbalanced ANOVA, contrasts and polynomial regression are indeed very relevant topics for the context of this workshop, but their explanation should go a little more beyond the examples shown in the slides. For instance, an explanation contrasting type I, type II and type III sum of squares should be provided when discussing the analysis of variance with unbalanced designs (as in https://mcfromnz.wordpress.com/2011/03/02/anova-type-iiiiii-ss-explained/).
Variation partitioning (note that it is incorrectly referred to as "variance partitioning" within the slides) is probably best instructed if it is put in the context of partial regression, which is not introduced in this workshop. This portion could be kept, but it needs to be extended for its explanation to become more solid.