Learn & Understand

Correlation, Causation, Confounders, and Simpson's Paradox

The companion calculator computes the correlation coefficient, a measure of how strongly two variables move together in a linear relationship. It is one of the most used statistics in data science, and one of the most misused, because a strong correlation is routinely mistaken for evidence that one variable causes the other. It is not. Correlation can arise from coincidence, from a hidden third variable, or even reverse when data is grouped differently. Understanding why correlation is not causation, and the traps of confounders and Simpson's paradox, is essential to not being fooled by a compelling number.

Correlation Is Not Causation

The single most important lesson in interpreting correlation is that it does not imply causation. Two variables can be strongly correlated for reasons that have nothing to do with one causing the other. They might both be driven by a third factor; the causation might run the opposite way from what you assume; or the correlation might be pure coincidence, especially when many variables are examined and some line up by chance. A correlation coefficient measures association, the tendency to move together, and association alone cannot distinguish among these very different underlying stories. Treating a correlation as proof of cause is one of the most common and consequential errors in data analysis.

Confounders: The Hidden Third Variable

The most important source of misleading correlations is the confounder, a lurking variable that influences both of the correlated variables and creates an association between them that is not causal.

The classic pattern: ice cream sales correlate with drowning incidents, not because ice cream causes drowning, but because hot weather, the confounder, increases both. Countless spurious correlations work this way, with an unmeasured common cause manufacturing the association. This is why establishing causation requires controlling for confounders, through careful study design or controlled experiments, rather than trusting a raw correlation from observational data.

Simpson's Paradox: When the Trend Reverses

An even more unsettling trap is Simpson's paradox, in which a correlation or trend present in aggregated data reverses when the data is broken into subgroups. A treatment can appear worse overall yet be better within every subgroup; a relationship can be positive in the whole dataset and negative in each part of it. This happens when a confounding grouping variable is distributed unevenly across the groups, so the aggregate mixes together populations that behave differently. Simpson's paradox is a dramatic demonstration that a correlation computed on pooled data can be not just misleading but exactly backwards, and that how you group the data can flip the conclusion. It is a powerful reason to examine subgroups, not just the overall number.

Pearson's Own Assumptions

Beyond the causation traps, the correlation coefficient itself has a specific limitation to remember: it measures only linear relationships. Two variables can be tightly related in a curved, U-shaped, or cyclical pattern and still produce a correlation near zero, because the relationship is not a straight line. So a low correlation does not mean no relationship, only no linear one, and plotting the data is essential to see patterns the coefficient misses. For relationships that are monotonic but not linear, rank-based correlation measures capture the association where the standard coefficient fails.

Interpreting Correlation With Discipline

Use the calculator's correlation coefficient as a measure of linear association, and interpret it with discipline: a strong correlation is not evidence of causation, confounding variables routinely manufacture spurious links, Simpson's paradox can reverse a trend when data is grouped differently, and the coefficient sees only linear relationships. Always plot the data and consider what else might be driving the association. The calculation gives the association; understanding causation, confounders, and Simpson's paradox is what keeps that association from fooling you.