Assumption-lean Inference: Generalised Linear Model Parameters

This lecture by the instructor focuses on the challenges of model misspecification in statistical inference, proposing a shift from model-based to estimand-based approaches. The discussion covers the definition of estimands, main effect estimands, interaction estimands, and survival analysis estimands. The lecture emphasizes the importance of choosing estimands that are well-defined and generic, allowing for robust inference. Various criteria for selecting estimands are explored, along with the limitations of plug-in estimators and the benefits of data-adaptive inference. The presentation concludes by highlighting the advantages of assumption-lean inference and the implications for practical applications.