Introduction to modern Generalised Additive Models in R

Event Overview

Generalized Additive Models (GAMs) are an extension of traditional regression models, and have proved to be highly useful for both predictive and inferential purposes in a variety of scientific and commercial applications. One reason behind the popularity of GAMs is that they strike a balance between flexibility and interpretability, while being able to handle large data sets. The taught part of the course will provide an overview of GAM theory, methods and software, while the hands-on sessions will make sure that the attendees will be ready to start doing GAM modelling in R as soon as the course is over.

The course is now fully booked. To be added to the reserve list please contact Nicole Augustin.

Please note that we will contact you separately for payment to confirm your place. 

Fees:

University of Edinburgh stuent/staff*: £60.00* 

Higher Education*: £260

Industry*: £500

*Note the fee includes membership to the British and Irish Region of the International Biometric Society.  Current members will receive a £60 discount -- please state your BIR membership status when you book.  Students may join for free!

Instructor: Prof Simon Wood

Simon Wood is Chair of Computational Statistics in the School of Mathematics, University of Edinburgh. He is the author of "Generalized Additive Models: An Introduction with R" and the R package mgcv for generalized additive and other smooth regression models.  

Topics Covered: 

Basics of basis penalty smoothing. GAMs. Empirical Bayes framework for inference and smoothness selection. Cross-validation approaches. Mixed GAMs, distributional regression and other extensions. The mgcv package. Practical model specification, checking and visualization. Model building and further inference. 

Learning Outcomes:

- Understand the basic theory underpinning regression modelling with spline like smoothers. 

 - Use mgcv to fit various GAMs to data, including mixed model extensions, GAMs for  location scale and shape (distributional regression), and some functional data analysis.

 - Be able to check modelling assumptions.

- Compare and critique competing models  

- Justify your modelling choices

Tentative daily timetable:

09:30 - 10:30  Lecture

10:30 - 11:00  coffee break

11:00 - 12:30  Lecture/Hands on session

12:30 - 13:30  Lunch

13:30 - 15:00  Lecture/hands on session

15:00 - 15:30  break

15:30 - 17:00  Lecture/hands on session

Target Audience:

Anyone with some statistics training who is aware of the advantages of nonlinear modelling could benefit from attending. Fields where this may be most popular are: science, insurance, finance, public health, epidemiology, psychology, econometrics.

Assumed Knowledge:

Attendees should be comfortable with using R. They should understand generalised linear models, though this can be intuitive and doesn’t have to be mathematically rigorous. They do not need to have used mgcv before.  

This event is sponsored by the British and Irish Region of the International Biometric Society.