SMARTbiomed summer school: “Causal inference, statistical genetics, and machine learning in common disease epidemiology and biology”
Module: Introduction to causal inference I
Module summary
This module serves as an introduction to the foundational concepts in causality and causal inference, providing participants the background reasoning needed to form estimands and discuss estimators. Over the course of two days, participants will consider what it means for an association to be causal, learn how to use potential outcomes to write a causal estimand without a link to a model, and gain an understanding of how to reason about and discuss causal effects in terms of directed acyclic graphs. Additionally, they will learn about identification of a causal estimand and how to clearly state the assumptions needed for identification. By the end of this module, participants will have a working knowledge of potential outcomes, DAGs, identification, and identifying assumptions. In addition, all participants should be able to clearly reason why an estimated association may or may not have a causal interpretation.
Prerequisites
A basic knowledge of R.
Module content
- Counterfactuals and causal language
- DAGs
- Identification and identifying assumptions
Required Software:
R + RStudio + CRAN package DAGitty
Module: Introduction to causal inference II
Module summary
This module builds upon the foundational knowledge from "Introduction to causal inference", focusing on estimation methods rather than language or concepts. Over two days, participants will explore key methodologies such as standardization (G-computation), inverse probability of treatment weighting, and simple double-robust estimators. Additionally, the module will introduce methods for using an instrumental variable for partial and point identification, estimation methods when using an IV, and the basic concepts for mediation analysis.
Prerequisites
Basic knowledge in R and the outcomes from the module “Introduction to causal inference,” especially the concept of what is causal versus what is associational, the writing of a causal estimand in potential outcomes and defining a causal model with a DAG.
Module content
- Understand and implement single time-point standardization.
- Understand and implement simply IPW
- Understand and implement simple parametric DR estimation
- Understand IV point and range (partial) identification and know that we are estimators for both
- Understand the basic concept of mediation and know there are differing schools of thought
Required Software
R + RStudio + CRAN packages stdreg and stdreg2
Teachers
Professor Erin Evelyn Gabriel, Section of Biostatistics, University of Copenhagen
Erin Evelyn Gabriel is one of the co-directors of the SMARTbiomed Pioneer Centre and is a biostatistician with hands-on experience from the National Institutes of Health in the USA, where she remains affiliated. She primarily works on methods for causal inference, with special focus on methods for nonparametric bounding of partially identified and unidentified causal effects. She leads a research group at the University of Copenhagen focusing on methods in causal inference. Although she has previously worked in infectious disease research, she has recently begun working in common complex disease areas.