A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Correlational research examines the relationship between two or more naturally occurring variables without manipulation. The correlation coefficient (r) measures the direction and strength of a linear relationship, ranging from −1 to +1. Correlational designs are useful for prediction, studying variables that cannot be manipulated ethically, and examining naturally occurring phenomena. The fundamental limitation is that correlation does not imply causation — directionality and the third-variable problem both threaten causal interpretation.
For a given correlation (e.g., ice cream sales and drowning rates), generate plausible third-variable explanations and alternative causal directions. Practice computing and interpreting r from small datasets.
You already know what variables are — measured characteristics that take on different values across observations — and how to read a scatterplot. Correlational research is the formal extension of that understanding into a research design: you measure two or more variables as they naturally occur and ask whether variation in one tends to accompany variation in the other. No manipulation, no random assignment, just measurement and observation. This makes it fast, ethical (you can study things you cannot ethically cause), and ecologically valid, but it comes with a built-in limitation that every researcher must understand before drawing conclusions.
The correlation coefficient (*r*) summarizes the linear relationship between two variables with a single number ranging from −1 to +1. The sign tells you direction: positive means the variables tend to move together (more education → higher income); negative means they move in opposite directions (more stress → less sleep). The magnitude tells you strength: values near ±1 indicate tight linear clustering on the scatterplot; values near 0 indicate scatter with no apparent trend. The coefficient is symmetric — the correlation between A and B is identical to the correlation between B and A. This symmetry is a clue to the central limitation.
The reason correlation does not imply causation has two distinct parts, both of which threaten any causal story you try to tell from correlational data. The first is the directionality problem: even if A and B are causally connected, the correlation cannot tell you which way the arrow points. Ice cream sales and drowning rates are positively correlated — but ice cream does not cause drowning. Both are caused by a third variable (summer heat and swimming). This is the third-variable problem (also called confounding): some unmeasured variable Z may cause both A and B, producing a correlation that has nothing to do with any causal relationship between them. The classic public health version: neighborhoods with more hospitals have higher death rates. The confound is severity of illness — sicker people go to hospitals, and some die. Hospitals do not cause death; the underlying illness causes both hospital admission and mortality.
Correlational designs are not weak or second-rate — they are often the *right* design. You cannot randomly assign people to poverty, childhood trauma, or genetic profiles, so the only ethical way to study their effects is to measure them as they occur. Correlational methods are also invaluable for prediction: even without knowing the causal mechanism, a strong correlation lets you forecast. Credit scores predict loan default. SAT scores predict first-year GPA. The prediction works even if the causal story is complicated. The key is to state clearly what the design can and cannot support: it can establish that a relationship exists, estimate its direction and strength, and support prediction — it cannot rule out confounders or establish the direction of causation. Those require experimental manipulation, longitudinal design with temporal precedence, or statistical controls with strong theoretical grounding.