About the course | Intended audience | Prerequisites | Content details
About the course
A practical introduction to generalised linear models for researchers analysing biological, experimental, or other research data with non-continuous response variables.
This course teaches how to construct, interpret, and assess generalised linear models using R or Python. The focus is on practical application and interpretation rather than mathematical derivation.
Topics include logistic regression, proportional response models, Poisson regression, negative binomial regression, significance testing, goodness-of-fit, and model diagnostics.
By the end of the course, participants should be able to:
- choose appropriate models for binary, proportional, and count response data
- construct logistic, Poisson, and negative binomial models in R or Python
- plot observed data and fitted model predictions
- assess model significance, goodness-of-fit, and assumptions
- interpret generalised linear model outputs confidently
Teaching is primarily hands-on, with short explanations introducing the statistical ideas needed to apply the methods appropriately.
Back to top Intended audience
This course is suitable for:
- researchers and students who analyse research data
- participants who encounter non-continuous response variables, including binary, proportional, or count data
- people who already have experience with standard statistical models and want to extend this to generalised linear models
- participants with a working knowledge of either R or Python
Prerequisites
Participants should have a solid understanding of statistics, ideally equivalent to completion of a core statistics course covering linear models, regression, hypothesis testing, and model assumptions. A good working ability in either R or Python is required. This is not an introductory programming course.
Related courses: Data analysis in R or Python (or equivalent), Core statistics
Content details
The course covers the following topics:
- Introduction to linear models
Recaps the structure and interpretation of standard linear models, including response and predictor variables, fitted values, residuals, regression coefficients, and model assumptions. This provides the foundation for understanding how generalised linear models extend standard regression approaches.
- Generalising the linear model
Introduces the key ideas behind generalised linear models, including linear predictors, response distributions, and link functions. Participants learn why standard linear models are not appropriate for many non-continuous outcomes and how GLMs provide a more suitable framework.
- Binary response models
Covers logistic regression for binary response variables such as yes/no, success/failure, or positive/negative outcomes. Participants learn how to construct models, interpret coefficients and predicted probabilities, visualise fitted curves, and work with continuous and categorical predictors.
- Proportional response models
Extends logistic modelling to proportional response data. The session focuses on modelling outcomes that represent proportions or grouped binary events, interpreting fitted relationships, and understanding how proportional data differ from individual binary observations.
- Significance testing for GLMs
Introduces statistical significance testing for generalised linear models. Participants learn about deviance, maximum likelihood estimation, Wald tests, and likelihood ratio tests, with emphasis on testing whole models and individual predictors appropriately.
- Goodness-of-fit
Covers approaches for evaluating how well a GLM describes the observed data. Participants learn how to assess model fit using appropriate metrics and how to compare competing models while avoiding misleading interpretations from standard linear-model summaries.
- Checking assumptions
Focuses on diagnostic checks for generalised linear models, including assessing whether the chosen distribution and link function are appropriate. Participants learn how to identify problematic models and when model results should not be trusted.
- Analysing count data
Introduces Poisson regression for count response variables. Participants learn how to model count outcomes, interpret log-link models, generate predictions, and assess whether the assumptions of Poisson regression are reasonable.
- Overdispersed count data
Covers overdispersion in count data and introduces negative binomial regression as an alternative to Poisson regression. Participants learn how to diagnose overdispersion, fit negative binomial models, compare models, and interpret results.