Strategy and Metrics

Correlation Versus Causation

Also called Spurious correlation

The difference between two measurements moving together and one of them actually producing the other.

Quick facts: Correlation Versus Causation

Category
Strategy and Metrics
Also called
Spurious correlation
Level
Intermediate
Affects
Budget decisions, attribution conclusions, test design, reporting claims
Where to see it
GA4 exploration reports, ad platform attribution reports, geo split tests, spreadsheet comparisons
In this article4
  1. How correlation and causation differ
  2. Why the distinction matters
  3. Where it goes wrong
  4. How to act on it

How correlation and causation differ

Correlation means two measurements move together — when one rises, the other tends to rise as well. Causation means one of them is making the other happen. Correlation is easy to observe in any reporting tool. Causation cannot be observed at all; it has to be argued for, and usually tested.

There are three ordinary reasons why two lines move together without one causing the other. A third factor may be driving both: in Nepal, festival season lifts demand, which raises ad spend and sales at the same time without either causing the other. The direction may be reversed: rising sales often lead a business to increase its budget, so spend follows revenue rather than producing it. Or it may simply be coincidence, which happens more often than people expect when you compare enough lines against each other.

Why the distinction matters

Because budget decisions are made on it. Almost every conclusion drawn from an analytics report — this channel works, this page drives sales, this campaign lifted enquiries — is a causal claim built from correlational evidence. That is not automatically wrong, but it is an inference, and it should be labelled as one.

Attribution reporting makes this harder rather than easier. An attribution model distributes credit among the touchpoints a converting customer happened to pass through. It describes a path; it does not establish that removing a touchpoint would have cost you the sale. Branded search is the classic case: it sits near many conversions because people who already decided to buy from you look you up by name.

Where it goes wrong

The usual error is reasoning from sequence. Something was launched, a number went up, therefore the launch worked. Seasonality, a competitor pausing, press coverage, a public holiday and pure chance all produce the same shape, and none of them are visible in the report you are looking at.

The second is choosing the metric after seeing the data. If you examine enough figures, something will always have moved in a pleasing direction, and reporting that one as the result is not analysis. Decide what would count as success before the test starts.

The third is confidence from a small amount of data. A short period or a low volume of conversions produces swings that look like patterns, and the smaller ad auctions many businesses in Nepal work with make this more likely, not less.

How to act on it

When a decision is large enough to matter, test it rather than infer it. Hold a group or a region back from a campaign and compare, stagger a launch across markets, or turn something off deliberately and watch what happens — that is what a holdout test is for, and it is the difference between reported performance and incrementality.

When a test is not practical, write down the prediction first, note what else changed in the same period, and state your conclusion with the uncertainty attached. Building that habit into how you set up analytics and tracking is worth more than any dashboard, because it stops a coincidence from becoming a strategy.

Do and do not

Do

  • State the prediction before you look at the result
  • Test important claims with a holdout or a staggered launch
  • Note what else changed during the same period

Do not

  • Conclude cause from a change that merely followed in time
  • Pick the winning metric after seeing the data
  • Read attribution credit as proof of contribution

Questions people ask about this

Does this mean analytics reports cannot be trusted?

They can be trusted for what they measure, which is what happened and in what order. The caution applies to the next step, where a path through the data is described as the reason for the outcome. Use reports to notice things and to narrow down candidates, then confirm the important claims with a test before committing serious budget to them.

How can a small business test causation without a big budget?

Geography and timing are the cheapest tools available. Run a campaign in some cities and not others and compare the difference, or switch something off for a defined period and watch what happens to enquiries. Neither is perfect, but both beat inferring cause from a report, and both can be done at whatever budget you already have.

Is branded search traffic really caused by my ads?

Sometimes partly, and it is very hard to tell from a report. People search for a business by name after hearing about it somewhere, so branded search sits close to conversions regardless of what created the awareness. The honest way to find out is to change branded activity deliberately for a period and observe whether total enquiries move, not just where credit lands.

Related terms

Found this useful?

Share it, or ask an AI to summarise it

Back to the glossary

Knowing the term is the easy part

Applying it to your own site and budget is the work. Book a call and I will tell you what actually applies to you.