Guide
Correlation vs Causation: What the Data Really Shows
Statistics · Guide · By DailyTools Editorial Team · July 31, 2026 · 3 min read
Two variables moving together does not prove one causes the other. This guide explains confounders, spurious correlations, and what evidence supports causation.
Statistics
When two variables rise and fall together, it is natural to ask whether one causes the other. Statistically, correlation measures the strength and direction of a linear association — but association alone cannot prove causation. Misreading correlation as cause is one of the most common errors in data interpretation, from news headlines to business analytics.
What correlation actually measures
The Pearson correlation coefficient r ranges from −1 to +1. Values near +1 indicate strong positive linear association; near −1, strong negative association; near 0, little linear relationship. Correlation describes how well a straight line fits the scatter of points — nothing more. The Correlation Calculator computes r from paired data so you can quantify association.
Why correlation is not causation
- Reverse causation — Y might cause X instead of X causing Y
- Confounding — a third variable Z drives both X and Y
- Coincidence — spurious correlations from random noise or cherry-picked data
- Non-linear relationships — r near zero can hide strong non-linear patterns
Confounding in real examples
Ice cream sales correlate with drowning deaths — not because ice cream causes drowning, but because both rise in summer (temperature is a confounder). Controlling for season or temperature changes the story. Good analysis asks what other variables might explain both trends simultaneously.
Evidence that supports causation
Establishing causation typically requires controlled experiments (randomized trials), temporal precedence (cause before effect), a plausible mechanism, dose-response relationships, and replication. Observational studies can suggest hypotheses but rarely prove causation on their own. Regression and R² describe fit — they do not replace experimental design.
Practical checklist before claiming cause
- Is the correlation strong and statistically significant?
- Could a third variable explain both?
- Does the causal direction make logical sense?
- Was data collected over time to check precedence?
- Are there experimental or quasi-experimental controls?
Frequently asked questions
Can a high R² mean X causes Y?
No. R² measures how much variance in Y is explained by X in a linear model. A high R² means good fit, not causal mechanism. Confounders can produce high R² without any direct causal link.
What is a spurious correlation?
A statistically significant association with no meaningful causal connection, often from coincidence, data mining, or hidden confounders. Classic examples include unrelated time series that trend together.
Does zero correlation mean no relationship?
No. Pearson r detects linear association only. Variables can have strong curved or threshold relationships with r near zero.
How do randomized trials help?
Random assignment balances confounders across groups, so outcome differences can be attributed to the treatment rather than pre-existing differences between groups.
Explore Statistics module · See our calculation methodology · Editorial policy