Karla Diaz-Ordaz, London School of Hygiene & Tropical Medicine, Department of Medical Statistics, UK

18 January 2023, 12:00 – 13:00

Venue: LT 2.41, Fry Building, School of Mathematics, University of Bristol, UK

Organiser: Daniel Lawson (Bristol)

Treatment effect heterogeneity is often studied to aid treatment or policy decisions. In this context, methods for data-adaptive (i.e. machine learning) estimation of conditional average treatment effect (CATE) of a binary exposure on an outcome have gained popularity recently. The resulting CATE, however, may be a complicated function of the covariates. To properly understand “treatment recommendations” made based on CATEs, it is desirable to explore which variables are driving such treatment effect heterogeneity. Variable importance measures may guide decision makers as to which variables are most important to consider when making treatment decisions, and can help stratify populations for further study of subgroup effects.

Here Karla presents a novel “treatment effect variable importance measure” developed for CATEs. Her proposal is developed within the counterfactual framework and is model-agnostic, unlike other variable importance metrics currently used, which are dependent on algorithmic architecture.

Joint work with Oliver Hines and Stijn Vansteelandt.

In cooperation with the Jean Golding Institute, University of Bristol

More information and registration on the Bristol Data Science Seminar Series