Speaker: Susanne Dandl, University of Zurich
Title: "A Nonparanormal Approach to Marginal Inference"
Abstract: Treatment effects for a novel therapy are typically measured by comparing marginal outcome distributions across study arms. While proper randomisation in randomised trials allows their estimation from observed outcome distributions, covariate adjustment is recommended to increase precision. However, for noncollapsible measures like odds or hazard ratios in logistic or proportional hazards models, conditioning on covariates changes the effect interpretation, and different covariate sets yield incomparable estimates.
In this talk, I introduce a novel nonparanormal model formulation for adjusted marginal inference allowing the estimation of the joint distribution of outcome and covariates. This approach offers not only the marginal treatment effect of interest for a wide range of outcome types but also an overall coefficient of determination and covariate-specific measures of prognostic strength. I will show how this method enhances precision of the marginal, noncollapsible effect - both theoretically and through empirical results.
Relevant preprint (under review): https://arxiv.org/abs/2503.01657
This will be a free online seminar. To register to attend, please click here: https://cam-ac-uk.zoom.us/webinar/register/WN_xwbSLwLITemQeb8b-jk-Fg