Behavioral Data Analysis with R and Python

Part II. Causal Diagrams and Deconfounding

In Part I, we saw how confounding can jeopardize even the simplest data analyses. In Part II, we’ll learn to build causal diagrams (CDs) to represent, understand, and deconfound relationships between variables.

First, Chapter 3 provides an introduction to CDs and their building blocks.

In Chapter 4, we’ll see how to build a CD from scratch for a new analysis. The CD we saw in our ice cream example was very simple by design. But in real life, it can often be complicated to know what variables to include in our CD beyond our cause and effect of interest, and how to determine the relationships between them.

Similarly, removing confounding from our ice cream example was simple: we just needed to include in our regression a joint cause of our variables of interest. With more complex CDs, it can become difficult to know which variables we should include in our regression. In Chapter 5, we’ll see rules that you can apply to even the most complex CD.