What Is Causal Inference?

Preface

Causal inference, the process of determining the effects of phenomena, lies at the heart of our ability to understand why things happen, and to predict the result of any action. Businesses need to understand how to turn data and information into causal knowledge and decisions. This report provides data scientists and analysts who are interested in causal inference with an introduction to the suite of techniques available, along with a principled way of thinking about causality.

There are many ways to slice the causality space. We introduce randomized control trials (RCTs), well-needed techniques from econometrics, and key aspects of causal graph theory. We made this decision at the expense of detailed discussions on other aspects relevant to causality (such as Bayesian inference, probabilistic programming, generative models, causal machine learning, the technical stack for causal inference, and products like Optimizely that are useful for causality) in order to provide the necessary foundation for causal thinking for working data professionals.

After an introduction to causality, we discuss the key techniques from econometrics and causal graph theory. Techniques from econometrics include:

• Randomized control trials, the gold standard for causality that are used in industry in the form of A/B-tests

• The constant-effects model, which helps to deal with the common problem in RCTs of all things not being equal across the groups we’re comparing

• Regression, which helps ...