The difference between correlation and causation, spurious correlation, and confounding variables.

What this dot point is asking

AQA wants you to explain why correlation does not prove causation, recognise spurious correlation, and identify confounding variables that could explain an apparent relationship. This is as much a reasoning skill as a calculation: examiners want you to challenge an unjustified "X causes Y" claim drawn from a scatter diagram.

Why correlation is not causation

When two variables are correlated, there are several possible explanations, and "the first causes the second" is only one of them. The second could cause the first; a third variable could cause both; or the correlation could be a coincidence. A classic example is that ice cream sales and drowning incidents are correlated, but neither causes the other; both rise in hot weather. Because you cannot tell which explanation is true from the scatter diagram alone, you cannot claim cause from correlation.

Spurious correlation

Large data sets can throw up striking but meaningless correlations purely by chance: with enough variables, some pairs will appear strongly correlated even though they are wholly unconnected. This is why a strong correlation alone, however striking, is not evidence of a real relationship, and why statisticians look for a plausible mechanism before taking a correlation seriously.

Confounding variables

In the ice cream and drowning example, temperature is the confounding variable: hot days raise both ice cream sales and the number of people swimming (and so drownings). The two original variables appear linked only because a hidden third factor drives both. To establish genuine causation you need a controlled experiment that holds confounding variables constant and changes only the explanatory variable, which is why observational correlations can suggest, but never prove, cause.

This topic ties together with the rest of the module and with experimental design. When you fit a line of best fit and make a prediction, you are using the correlation, but you must not therefore claim the explanatory variable causes the response. And it connects to the work on controlling variables and bias: a controlled experiment earns the right to talk about cause precisely because it holds the confounders constant, which an observational scatter diagram cannot do. The exam phrasing to watch for is any sentence of the form "this shows that X causes Y" drawn from a correlation alone, which you should always challenge by naming a plausible confounder and stating that an experiment would be needed to establish cause.