Professor of Statistics
Ghent University, Gent, Belgium
Homepage
Abstract
Causal inference has taken off over the past decade. A seemingly never ending stream of new and complex methods enters the literature allowing to draw causal conclusions from observational data, as long as one is willing to make causal assumptions in context. Many of these methods are derived from basic pillars.
Using the potential outcomes framework, we describe principled definitions of causal effects and of estimation approaches classified according to whether they invoke the no unmeasured confounding assumption (including outcome regression and propensity score-based methods) or an instrumental variable (pseudo randomisation). Starting from (sequential) point exposures, we discuss interpretation, challenges and potential pitfalls. We illustrate application using a “simulation learner”, that mimics an existing study of the effect of various breastfeeding interventions on a child’s later development. This involves a typical simulation component with generated exposure, covariate, and outcome data inspired by a randomised intervention study with intercurrent events. The simulation learner thus generates various (linked) exposure types with a set of possible treatment values per observation unit, from which observed as well as potential outcome data are generated. It thus provides true values of several causal effects.
There will be several hands-on exercises allowing to also work with available R code for data analysis.
References
Goetghebeur, E., le Cessie, S., De Stavola, B., Moodie, E. E., Waernbaum, I. (2020). Formulating causal questions and principled statistical answers. Statistics in Medicine, 39(30), 4922-4948. doi: 10.1002/sim.8741