Correlation vs Causation

Correlation means two variables move together; causation means one variable is actually responsible for changing the other. Mistaking the first for the second is one of the most consequential errors in marketing analysis.

What Correlation vs Causation Means in Marketing

When two things happen at the same time or move in the same direction, it is tempting to assume one causes the other. Your email sends go up; so does revenue. Your SEO improves; so does your conversion rate. Your paid social spend increases; so does direct traffic. In each case, the correlation might exist without any causal link.

Correlation is a statistical relationship: two variables moving together, positively (both up) or negatively (one up, one down). Causation is a mechanistic claim: changing one variable directly produces a change in the other. The correlation might be caused by a third variable (a confounder), by coincidence, or by the reverse relationship you didn’t consider.

Marketing data is full of correlations that don’t indicate causation. Seasonal patterns mean many metrics rise and fall together regardless of what campaigns are running. Overall market growth lifts multiple channels simultaneously. An improving product increases retention, which improves every downstream metric that depends on retention. When you observe a correlation between your marketing activity and an outcome, you have a hypothesis, not a conclusion.

This matters most when you’re making budget decisions. Attribution models show correlations between ad exposures and conversions. The model cannot tell you whether the ad caused the conversion or whether the user would have converted anyway.

How the Correlation-Causation Distinction Works

Correlation. Measured statistically as a coefficient between -1 and +1. A correlation of +0.8 means the two variables tend to move together strongly. A correlation of -0.4 means they have a moderate negative relationship. A correlation of 0 means no discernible relationship. Correlation makes no claim about direction of cause.

Confounding variables. The most common reason a correlation is not causal is a third variable driving both. Brand awareness and customer acquisition might both increase in January: not because awareness causes acquisition, but because January is a high-purchase intent period when both naturally rise. The confounder is seasonality.

Establishing causation. The gold standard is a randomised controlled experiment: assign units (users, regions, customers) randomly to treatment and control groups. Any difference between them is attributable to the treatment because randomisation controls for confounders. In marketing, A/B testing, geo experiments and incrementality tests are the practical tools for establishing causation.

Retrospective analysis. When experiments aren’t possible, techniques like difference-in-differences, regression discontinuity and instrumental variables attempt to infer causation from observational data. These are used in media mix modelling and econometric attribution. They produce better estimates than raw correlation but still carry assumptions that can be wrong.

Correlation vs Causation Example

An analytics report shows that users who receive more than three emails per month have a higher average order value than those who receive fewer. It would be easy to conclude that sending more email causes higher purchase value. But the correlation is almost certainly running the other way: high-value customers are more engaged and receive more emails because they open them; the engagement is a reflection of customer quality, not a product of email frequency. Increasing email frequency to low-engagement customers based on this correlation would likely decrease AOV from those customers rather than improve it.

Why Correlation vs Causation Matters for Marketers

Marketing attribution is fundamentally a causal question: did this campaign cause this sale? Attribution models answer it with correlations dressed up as causal inference. The correlation-causation confusion leads to over-investing in channels that look good in attribution models because they are associated with conversions, rather than channels that actually drive them. Holding this distinction clearly in mind when reading performance reports is one of the most practically useful analytical skills a marketer can develop.