Regression by composition (Online)

Date: Thursday 11 April 2024, 2.30PM
Location: Online
Section Group Meeting

Share this event

This is a joint event between Oxford Local Group & Medical Section Group. All welcome! 

Our speakers will be Prof. Rhian Daniel and Prof. Daniel Farewell from Cardiff University.

Across our two talks, we will introduce a new framework for statistical modelling that we call regression by composition. To motivate our work, Rhian will begin with binary regression, and discuss model comparison within the class of generalized linear models. Reasoning causally from first principles, we find that some appealing models fall within this class, while other simple setups do not correspond to a generalized linear model for any choice of link function. Daniel will then introduce a broader class of regression models in which linear predictors have "group actions" (think "families of transformations") on probability distributions. Group actions may flexibly be composed together, producing old friends (generalized linear models, transformation models, quantile regression, ...) and some lesser-known but no-less-useful relatives. It turns out that regression properties like collapsibility relate directly to algebraic features of group actions.

This is the registration page for attending the event online, if you wish to register to attend in person please go here.
Rhian Daniel and Daniel Farewell, Cardiff University
Contact Christiana Kartsonaki for RSS Medical Section and RSS Oxford Local Group
Members - free to attend 

Non members - £10