Unit 9

Introduction

Welcome to Week 9 of BIO144!

Course book

This week we will be covering Chapter 9: Generalised linear models for binary data of the course book.

Code walkthrough

The R code of this week’s chapter is in a code walkthrough script, with a video walkthrough of it. Both are linked at the top of the course book chapter. Work through them before the practical: the last section of the script (“Practical toolbox”) shows code that this week’s practical uses but the chapter does not.

Learning objectives

These learning objectives, together with the content of the course book chapter, define what you need to know for the exam.

After this unit, students will be able to:

  1. Identify binary response variables (e.g. presence–absence, success–failure) in described biological studies.
  2. Distinguish individual-level binary data from aggregated binomial data (number of successes out of a number of trials).
  3. Explain why linear regression on proportions, or Poisson regression on the number of successes, is unsuitable for binary and binomial responses.
  4. Explain the logit link and the relationship between probability, odds and log-odds.
  5. Convert between probabilities, odds and log-odds.
  6. Use R to fit a logistic regression with glm().
  7. Use R to fit a binomial GLM to aggregated data, specifying the numbers of successes and failures.
  8. Interpret coefficients as changes in log-odds, and back-transform them to odds ratios.
  9. Use R to calculate predicted probabilities and plot the fitted relationship.
  10. Use deviance-based tests to judge whether explanatory variables are associated with the response.
  11. Check a binomial GLM before interpreting it, using the GLM checklist, and recognise which checks are hard to interpret for individual-level 0/1 data.
  12. Use the ratio of residual deviance to residual degrees of freedom to check aggregated binomial data for overdispersion, use a quasibinomial model as a pragmatic fix, and recognise that this check is not informative for individual-level binary data.
  13. Recognise separation in a logistic regression from its symptoms (e.g. very large coefficient estimates and standard errors).
  14. Recognise a correct and complete statement of results from a logistic regression.

Lecture

Please attend the lecture. If for any reason you can’t, please watch the podcast, which is available on OLAT.

Homework, Practical, and Weekly Quiz

Available here. You will need to be on a UZH network to access this link, so if you are not on campus, you will need to connect to the UZH VPN first.