Generalised additive models with R

Course level: Intermediate

Next live dates: 9 - 12 December (enrollment link coming soon).

Course description

Generalised additive models (GAMs) take regression to the next level, allowing a more flexible exploration of our data. Unlike generalised linear models, GAMs do not assume a linear relationship between the outcome and covariate(s). Using smoothing functions, we can explore nonlinear trends, such as the impact of temperature on barbecue food sales, or the relationship between income and happiness. GAMs strike the balance between flexibility offered by machine learning, whilst ensuring that findings are interpretable, and avoiding potential issues with overfitting that can occur with some machine learning methods.

In this online course, participants will learn the theory behind generalised additive models and apply this to real data using R with the mgcv package. Participants will be able to interpret, visualise and communicate results of GAMs, and use diagnostic tools to ensure models are valid and robust. We will explore different types of smooths that can be applied to models, including how they can provide a relatively uncomplicated way to include space and time into models. All analysis will be applied using R, with an emphasis on creating clear, tidy, reproducible code.

The course will consist of a combination of theory-based sessions and hands-on practical exercises to ensure that participants have the tools to apply the approaches taught in this course to their own data immediately.

Outline

Topics covered in this course include:

• Re-introduction of generalised linear models
• Introduction to generalised additive models and smoothing splines
• Fitting GAMs using R
• Interpretation and visualisation of model results
• Model diagnostics and validation
• Extensions of GAMs: multi-dimensional smooths, temporal and spatial applications of GAMs, Bayesian interpretations of GAMs

Target audience

This course is designed for anyone that would like to gain a deeper, more flexible understanding of their data, and explore non-linear, multi-dimensional relationships between variables.

This course does not require any prior understanding of generalised additive models. Participants are expected to be comfortable with loading, tidying, and visualising data in RStudio using the Tidyverse package (I offer a short course on everything you need to know here!). Although this course does begin with a re-cap of generalised linear models, ideally participants will have some prior understanding of model interpretations and diagnostics (if you have never fitted a regression model and want to learn more, consider taking our Introduction to Regression with R course first).

Learning outcomes

By the end of this course, participants can expect to have a robust understanding of generalised additive models and their applications in R. They will understand the different smooth splines available, and how to choose the most appropriate for their application. Participants will be able confidently fit and interpret models, and visualise smooth functions to communicate results. They will be able to diagnose models and check their validity, ensuring reliability of their results. All this will be carried out using reproducible, tidy R code.