Regression by composition

On Feb 2nd at 3pm UTC, my collaborators and I will give a talk at the Berlin Epidemiology Methods Colloquium. This is joint work with Daniel Farewell and Rhian Danewell (who will both join me as speakers) and Mats Stensrud. Registration is open at the BEMC website .

We will propose a new unifying framework for statistical models, which we call Regression by Composition (RBC). RBC includes as special cases all generalized linear and generalized additive models, as well as many other models. Among the advantages of RBC is that it allows link functions based on the switch relative risk. The RBC framework also facilitates conceptual insight about central issues in statistical modelling, including collapsibility and the nature of homogeneity assumptions.

When we started this project, the primary motivation was to generalize regression models to allow link functions that are consistent with Sheps’ preferred variant of the relative risk. However, the solution we came up with took on a life of its own, and turned into a unified modelling framework that has advantages which go significantly beyond the switch relative risk.

We are very excited about the novelty and applicability of these ideas, and certainly hope to see all our statistically oriented readers in the audience. The BEMC presentation is going to be more technical than my talk at LHSTM, but I believe applied epidemiologists and other data scientists will be able to follow the argument.

Ps. A recording of my talk at LHSTM is now available (scroll to talk 4)