Our preprint “Regression by Composition” is now available at https://rss.org.uk/RSS/media/File-library/Events/Discussion%20meetings/rbc.pdf , and will be read to the Royal Statistical Society on March 24th, 2026.
Regression by Composition is a new, unified and modular framework for regression modelling, in which models are constructed by composition of group actions. This approach expands model possibilities beyond conventional regression frameworks such as GLMs. We also believe that it has substantial conceptual and didactic advantages.
This work was in part motivated by earlier discussion on this forum (https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/). In an earlier paper published in JRSSA, we discussed how this new approach relates to choice of effect measure, https://academic.oup.com/jrsssa/article/187/3/636/7755773
We would very much welcome comments and questions, either on this forum, or in the form of written contributions submitted to JRSS-B.