Causal vs Correlational Relationships

A correlation is a pattern where two variables tend to change together. Correlations have a direction. In a positive correlation, both variables rise together, the more hours you study, the higher your test score tends to be. In a negative correlation, one variable rises as the other falls, the more time spent gaming, the lower the test score might tend to be. If there is no clear pattern, there is no correlation.

Correlations also have a strength. A strong correlation has points that sit close to a clear line, so the pattern is easy to see. A weak correlation has points that are scattered loosely, so the pattern is faint. As you saw with scatter plots in Lesson 13, the shape of the cloud of points tells you both the direction and the strength of a correlation at a glance.

Causation means that changing one variable directly produces a change in the other. If you turn a dimmer switch up, the room gets brighter, the switch causes the brightness. Causation is a far stronger claim than correlation. A correlation just says "these two move together"; causation says "this one makes that one happen".

This is why the famous rule matters so much: correlation does not prove causation. Two variables can move together for three different reasons. First, one really might cause the other. Second, a hidden confounding variable might be driving both. Third, it could be pure coincidence, a fluke that vanishes with more data. When you see only a correlation, you cannot yet tell which of the three is true, so you must not jump to "cause".

A confounding variable (sometimes called a lurking variable) is a hidden third factor that drives two other variables at once. It is the single most common reason a correlation looks like a cause when it is not. Return to the hook: ice-cream sales and drownings are correlated, but neither causes the other. Hot weather is the confounding variable. Heat drives ice-cream sales up, and the same heat sends more people swimming, where more drownings sadly occur. Remove the heat from the picture and the link between ice cream and drowning disappears.

To catch a confounding variable, always ask: "Could some third thing be causing both of these at the same time?" If the answer is yes, the correlation may be entirely explained by that hidden factor, and the obvious causal story collapses.

If a correlation alone cannot show cause, how do scientists investigate a causal claim? They build a case from several lines of evidence. A well-designed controlled experiment can provide strong evidence: change one variable while keeping other relevant factors as stable as possible, then measure the outcome. Researchers also consider a plausible mechanism, a way one thing could produce the other, and consistency across repeated studies. Every study has assumptions and limitations, so conclusions should reflect the full body of evidence.

Two more clues help. A dose-response pattern, more of the cause giving more of the effect, points toward a real link. And the time order must be right, the cause has to come before the effect. When all of these line up together, scientists become confident the relationship is causal, not just correlational.

A pattern in a tiny sample may be due to chance. A larger, well-collected sample can make an estimate more precise and help scientists assess how much random variation could explain a pattern. But more data does not remove bias or confounding, and a very large study can detect a small effect that may have little practical importance. Scientists consider how the data were collected as well as how many measurements they have.

Statistics help scientists compare data with a specified chance-based model. A result is called statistically significant when the data would be unusual under that model, assuming the test’s conditions are met. This does not give the probability that a finding is true, prove an effect exists, measure its size or importance, or show that one variable caused another. Interpret results alongside effect size, uncertainty, study design, possible bias and confounding, and other evidence.