Introduction to Causal Inference for Psychologists: Testable and Non-Testable Causal and Statistical Assumptions

Abstract

The main goal of basic research is to answer causal questions. Generally, only the statistical part of this process tends to proceed in a partially formal way and according to clearly defined rules. At the same time, the causal relations are often treated informally or implicitly in a way that is prone to difficult-to-detect errors. This introduction aims to show psychology researchers some of the great benefits of approaching causal issues using a formal theory of causal inference. In this part, I discuss the non-obvious status and role of causal and statistical assumptions in causal inference. After covering, in a simple setting, the general shape of inference from causal assumptions, statistical assumptions, and data to causal effects, I outline, from a contemporary perspective, the limits of applicability of the general linear model. Then, I introduce the formal part of Pearl’s theory that relies on graphs. Using these tools, I show how one can analyze and interpret the results of an experiment on short-term memory search, and I discuss the back-door and front-door adjustments. To present the mathematical part of the theory in an accessible way without overly simplifying it, I illustrate some issues by using simulations written in R.