Correlation Is Not Causation

Ten pages rank on page one for your keyword. Measure them and you learn what they have in common. You do not learn why Google chose them, and for the work in front of you that turns out not to matter.

What a correlation gives you is a shortlist of things the winners share. What turns that shortlist into a decision is everything that happens after the coefficient, and that is the part the phrase in the title tends to talk people out of.

What Ten Pages Have In Common

The ten pages on page one are not a sample of the web. They are the ten Google chose to put in front of people who typed that phrase, which makes them the only group worth measuring.

So a correlation here is a description of that group. Pages near the top load faster, or carry more images, or run longer. It is a fact about the pack, not a law about search.

That is enough to act on, as long as the next two steps happen.

What The Validation Throws Out

Clear outliers come out first, using Tukey's method. One page that is wildly unlike the other nine will otherwise set the shape of the whole correlation.

Then Cook's Distance checks what survived. A correlation that only exists because one remaining page is unusual gets invalidated rather than reported. A single odd page can manufacture a 0.7 by itself, and reporting that number would be worse than reporting nothing.

One more thing to read correctly: position one is the goal, not position ten. A positive correlation means the smaller values sit in the better positions, which reads backwards until you have seen it once.

Variance Decides The Bucket

Strength alone does not decide what to do. What separates a requirement from an opportunity is how much the ranking pages differ from each other.

When every page on page one satisfies a factor, it is table stakes. Matching it earns you nothing except the right to compete, and missing it is disqualifying. That is CRITICAL.

When the ranking pages vary widely and the measurement still points at the factor mattering, there is room to move. That is OPPORTUNITY, and it is where the work goes.

The other two are honest about their own limits. GAMBLE is a relationship the evidence is too thin to confirm, so test it rather than commit to it. AVOID has no measured relationship with rank, and effort spent there buys nothing.

What A P-Value Does Not Prove

A p-value says how often noise alone would throw up a number this large at this sample size. It does not prove a pattern is real, and it says nothing whatever about cause.

With ten pages it is rarely decisive, because ten is the whole population of page one rather than a sample you can grow. That is not a flaw in the method, it is the size of the thing being measured, and it is why the bucket carries more weight than the coefficient. The longer version of that argument works through a real table.

Anyone who stops at "correlation is not causation" is right about the statistics and finished thinking one step too early.

The Last Step Is Yours

The analysis produces a sentence like: the ranking pages carry more images than yours.

That means three different things depending on what your page is. On a how-to it means steps need screenshots, because the searcher is following along. On a product listing it means the item itself is under-shown. On a sales page it probably means proof, not decoration, and more stock photography will make it worse.

Nothing in the measurement knows which of those you are. You do, and that is the step you are being handed rather than sold.

That Is What Intent Means

Google picked those ten knowing what people who type that phrase are trying to do. So the things the ten have in common are a description of that want, written in measurements.

That is search intent stated concretely, rather than sorted into informational, navigational, commercial and transactional. Two keywords that look interchangeable can produce different profiles, and when they do, they have different intents no matter how similar the words are.

Which is why the same advice is wrong twice: once because it came from an article rather than a measurement, and once because it came from a different keyword than yours. Outranking the pages above you starts from the ten you are actually behind.