About

This is a collection of lecture notes that accompany the APTS week 4 module on Causal Inference, and are also used for the StatML Module on Causal Inference.

Aims of Course

This course aims to

introduce the basic concepts of causal learning (reasoning, modelling, and inference);
to enable you to read more advanced ‘causal’ papers.

The focus will be on formulating causal (research) questions, understanding sources of (avoidable and unavoidable) bias, as well as some basic methods. These will include g-methods, propensity scores, and causal discovery.

History of Causal Inference

The topics of causality and causal inference are very broad. They have developed and evolved quite separately in different fields: philosophy, sociology, psychology, epidemiology, econometrics, computer science, mathematics.

As a consequence there is lots of different terminology, approaches, accepted assumptions, experimental designs, and types of data sources. Only in the last few years has some convergence has emerged across fields. This is in spite of the fact that statistical causality is absolutely fundamental to many research questions in data-driven science.

Philosophy

Philosophical, moral or other abstract understandings of the meaning of ‘causation’ or ‘causality’ will not concern us here. Leave the philosophy to the philosophers! We are interested in a narrow view of causality that is relevant to scientific enquiry: in other words, a form of causality that we can implement. For us, a causal effect represents a contrast between two outcomes (or distributions) between different experiments we might perform. That is, it considers the effect of (possibly hypothetical) interventions.