Level: Professional (P)

Generalised additive models (GAMs) take regression to the next level, allowing a flexible exploration of data. Unlike linear models, GAMs do not assume a linear relationship between outcome(s) and covariate(s), providing flexibility associated with machine learning, whilst preserving interpretability, and avoiding issues with overfitting.
 
In this course, participants will learn the theory behind GAMs and apply this to real data using R. Participants will interpret, visualise and communicate results of GAMs, and use diagnostic tools to ensure models are valid and robust. We explore different types of smooths, including how they can provide a relatively uncomplicated way to include space and time into models.
 

Learning Outcomes

By the end of the course participants will:

• Have a robust understanding of generalised additive model and their application in R

• Understand what smoothing splines are, the different types that are available, and how to choose the most appropriate for their model

• Be able to confidently fit and interpret generalised additive models

• Know how to extract and visualise smooth functions to communicate results clearly

• Be able to diagnose and check model validity, ensuring results are robust and reliable

• Understand how GAMs can be exploited to model temporal and spatial data, and how they can be extended to have Bayesian interpretations
   

Topics Covered

Day 1: Ensure all participants are comfortable with R software and generalised linear models. Finish the day with a simple example of a generalised additive model.
 
Day 2: Introduction to GAM theory and smoothing splines. Visualisation and interpretation of GAM results.
 
Day 3: Model selection and diagnostics of GAMs, with worked examples and exercises. Finish with theory on extensions of GAMs (including spatial and temporal models). Give a simple example of a temporal model fitted this way

Target Audience

This course is open to anyone that would like to fit flexible, nonlinear models to their data. This could include PhD students, academics, government or other public sector analysts, etc. from any discipline.

Knowledge Assumed

Participants are expected to be comfortable with R and RStudio software, ideally with previous experience of using Tidyverse to load and tidy data. Some knowledge of generalised linear models is useful, although a brief re-cap is included at the beginning of the course. There is no requirement to have any prior knowledge or experience of fitting additive models.

Dr. Sophie Lee

 
Sopie has over 10 years' experience of teaching statistics and R courses to people from a wide range of backgrounds. She is dedicated to producing courses that are accessible to all, encourage active participation, and provide tools that enable attendees to continue learning after the course has finished.
 
Sophie has a PhD in spatio-temporal epidemiology from LSHTM, and an MSc in Statistics from Lancaster University. Her research interests lie in spatial data analysis, planetary health, and Bayesian modelling approaches. She strives to ensure my work is open and reproducible.