Correlation ≠ causation
About this app
Seven cases from storks to Nicolas Cage: guess the structure, then use partial correlation and an experiment to test what produces the correlation – third factor, direction, selection or chance.
Why correlation does not prove cause
The correlation coefficient r measures how closely two quantities vary together, on a scale from −1 to 1. It says nothing about cause and effect. The same scatter plot can come from five structures: X affects Y, Y affects X, a confounder Z affects both, a selection shows only part of the data, or chance. Two checks help: partial correlation, which removes Z, and an experiment in which X is assigned at random. Only a real effect X → Y survives the experiment.
What the app shows
The app walks through seven cases, including storks and births, ice cream sales and swimming accidents, red wine and lifespan, restaurants following Berkson’s paradox, and Nicolas Cage films. Two cases use real data, the others simulated data. On the left is the causal graph, on the right the scatter plot with a regression line. An r scale compares three views: observed, controlled for Z and experiment. A grey band marks where r falls in 95 % of samples when there is no relationship, alongside n and the p-value.
Things to try
Read the observation first and pick the structure you suspect behind it. Once the case is revealed, switch between “Observed”, “Test Z” and “Experiment” and see whether the relationship disappears or holds. In the simulated cases you set the strength of the effect and draw new samples. The “Types” panel summarises which structure vanishes under which check.
Frequently asked questions
- What is the difference between correlation and causation?
- Correlation means two quantities tend to rise or fall together. Causation means one quantity actually changes the other. A correlation can also arise from a confounder, a reversed direction of effect, sample selection or pure chance.
- What is a spurious correlation?
- A spurious correlation is a statistical relationship without a direct effect between the two quantities. A classic example is storks and births: larger countries have more storks and more births. Once country area is taken out, the relationship shrinks considerably.
- What is partial correlation?
- Partial correlation measures the relationship between X and Y after removing the influence of a third variable Z. It can be read as the correlation of the residuals after regressing X and Y on Z. If it falls close to zero, Z largely explains the observed relationship.
- How do I use the app?
- For each of the seven cases you guess the structure, reveal it and then test it with partial correlation and a simulated experiment. The causal graph, scatter plot and r scale show what happens to the relationship under each check.
Subject: Data & decisions
More from Data & decisions
- Anchoring effect
- Benford's law
- Berkson's paradox
- Cherry-picking: base period
- Cherry-picking: base period 2.0
- The three-cowboy duel (truel)
- The three-cowboy duel 2.0
- The three means
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps