We will continue our exploration of GLMs this week, looking at how GLM’s can be applied to data expressed as zeroes and ones, proportions, rates, and more.
This lecture will be a bit different than others. We will cover many topics in very brief detail. But each of these topics could be its own lecture. Consequently, I will ‘name-drop’ several techniques, simply state what they are and when they might be used, and encourage self-study.
- GLM’s for data expressed as zeroes and ones, proportions, and rates
- Contrast Bernoulli, Binomial, and Beta regression
- In brief: Multinomial and cumulative logistic regression (i.e. what to do when Y is categorical)
- Simulating from the model as a validation step
As always, this class will have a mixture of R code and lecture slides
Beta regression:… excuse the similar-sounding names. Each of the below articles comes at it a bit differently.
- Mangiafico: Beta regression for percent and proportion data
- Zeileis et al: Beta regression in R
- Prabhakaran: Beta regerssion with R
- Cribari-Neto and Zeileis: Beta regression in R
We will use data from the below paper for one exercise. Read if you like:
- Porsmoguer, S. B., Bănaru, D., Boudouresque, C. F., Dekeyser, I., & Almarcha, C. (2015). Hooks equipped with magnets can increase catches of blue shark (Prionace glauca) by longline fishery. Fisheries research, 172, 345-351.
A good reference for how to run various types of glm in R:
Slides available on Speakerdeck