Ghent University, Ghent – Belgium
Friday, November 6, 2020
Time: 15:15 (3.15 pm) at EPFL in Lausanne if COVID permits – to be confirmed
Title: Assumption-lean inference for generalised linear model parameters
Abstract: Inference for the parameters indexing generalised linear models is routinely based on the assumption that the model is correct and a priori specified. This is unsatisfactory because the chosen model is usually the result of a data-adaptive model selection process, which may induce excess uncertainty that is not usually acknowledged. Moreover, the assumptions encoded in the chosen model rarely represent some a priori known, ground truth, making standard inferences prone to bias, but also failing to give a pure reflection of the information that is contained in the data. Inspired by developments on assumption-free inference for so-called projection parameters, we here propose novel nonparametric definitions of main effect estimands and effect modification estimands. These reduce to standard main effect and effect modification parameters in generalised linear models when these models are correctly specified, but have the advantage that they continue to capture respectively the primary (conditional) association between two variables, or the degree to which two variables interact (in a statistical sense) in their effect on outcome, even when these models are misspecified. We achieve an assumption-lean inference for these estimands (and thus for the underlying regression parameters) by deriving their influence curve under the nonparametric model and invoking flexible data-adaptive (e.g., machine learning) procedures.This is joint work with Stijn Vansteelandt.
Leibnitz Institute for Prevention Research and Epidemiology – BIPS, Bremen
University of Bremen, Germany
Friday, October 30, 2020
Time: 15:15 (3.15 pm) – zoom meeting
Title : Graphs, Time and Causal Inference
Abstract: In this presentation, I will address key aspects of statistical modelling of and causal inference for events in (continuous) time. This is for instance relevant when some events correspond to some “treatment” and others to important outcomes such as “relapse” or “death”.
First we discuss the visual representation of multivariate dependence structures among events, building on a marked point processes framework, using the concept of local independence and associated graphs (Didelez, 2008). It will be shown how reasoning and inference for systems with latent processes can be facilitated using such graphical representation and a suitable notion of graph-separation, called delta-separation. Secondly, I present a formal notion of causal relations between events (or processes) in time based on a decision theoretic approach (Dawid and Didelez, 2010; Didelez, 2015), and discuss the use of local independence graphs to decide the question of identifiability in systems with unmeasured processes. This will be illustrated with an application to cancer screening in Norway (Roysland et al., 2020). Moreover, we will address the connection to recent developments in causal mediation and competing events in survival analyses (Didelez, 2019; Stensrud et al., 2020).
The presentation will focus on basic principles and concepts rather than technical details.
Dawid and Didelez (2010). Identifying the consequences of dynamic treatment strategies: A decision theoretic overview, Statistics Surveys, 4, 184-231.
Didelez (2008). Graphical models for marked point processes based on local independence. JRSS(B), 70, 245-264.
Didelez (2015). Causal Reasoning for events in continuous time: a decision-theoretic approach. Proceedings of the 31st Annual Conference on Uncertainty in Artificial Intelligence – Causality Workshop (Invited Paper).
Didelez (2019). Defining causal mediation with a longitudinal mediator and a survival outcome. Lifetime Data Anal 25, 593–610.
Roysland, Ryalen, Nygard, Didelez (2020). Graphical criteria for identification in continuous-time marginal structural survival models. In preparation.
Stensrud, Young, Didelez, Robins & Hernán (2020) Separable Effects for Causal Inference in the Presence of Competing Events, JASA (online).