Reading focus
Syllabus: How to Read Psychological Science
Begin with a simple causal diagram and explain each arrow in words. Then follow how adding another variable can change the interpretation of an association. Use the glossary to name the different roles variables can play; understanding the causal story comes before reproducing a statistical calculation.
Why you’re reading it. Learn to make the assumptions behind a causal claim visible without having to fit a statistical model.
A question to consider. Draw two possible explanations for an association between study time and exam performance. What comparison could distinguish them?
Why this reading
A correlation becomes a causal claim only with additional assumptions. Rohrer makes those assumptions visible through diagrams of possible relationships. The paper gives readers a language for asking what an analysis controls, what a comparison leaves unresolved, and why adding more control variables can sometimes make an inference worse.
Before you read
A correlation is an association between measured variables; a causal effect concerns what would change if one variable were changed through an intervention. A confounder is a common cause of variables whose relationship interests us. In a directed acyclic graph, circles or labels represent variables and arrows represent assumed causal influences. The arrows are assumptions to examine, not causal discoveries made by drawing the graph. Read simple examples before worrying about the terminology of statistical models.
Read for
- Explain the education-and-income example in words, then identify the extra variable that could account for part of the association.
- Compare a causal chain with a common-cause fork. What different stories could produce an observed association?
- Notice the assumptions about time, omitted causes, and measurement that make a diagram useful.
- In the discussion of statistical control, ask why a variable's causal role matters more than simply its availability. The more technical sections distinguish confounders, colliders, and mediators.
- Read the conclusion for the limits of observational inference and the reasons to compare plausible alternative models.
Consider afterwards
- Draw two possible explanations for an association between sleep and academic performance. What observation or intervention would help distinguish them?
- What would you need to know before accepting the claim that an association survives adjustment for all important variables?
Editorial context
This is a methodological tutorial, not an experiment establishing a particular causal relationship. Its diagrams clarify the implications of an assumed structure; they do not demonstrate that the structure is correct. Rohrer also distinguishes choosing a causally appropriate adjustment from estimating an effect well. A credible graph does not remove problems of measurement, sampling, or model specification. The paper's conclusion supports using observational evidence with explicit assumptions, rather than treating either correlation or randomization as a complete argument on its own.