What Is a Correlation?

Correlation is not causation: two metrics can move together without one causing the other. Spotting a correlation is easy. Ruling out coincidence and confirming a cause is much harder, which is why the warning gets repeated so often.

A correlation itself is a real, measurable pattern in data. What’s speculative is the causal explanation people attach to it, and that interpretation is what can be misleading or inaccurate. This is true in any field, including SEO.

How correlation is measured

The usual measure is Pearson’s correlation coefficient, written as r. It describes the strength and direction of a linear relationship between two variables and runs from -1 to +1:

• +1 is a perfect positive correlation: both variables rise together in a straight-line pattern.
• 0 means no linear correlation. The variables can still be related in a curved, nonlinear pattern that r doesn’t detect.
• -1 is a perfect negative correlation: one variable rises as the other falls.

Two caveats apply. Pearson’s r is undefined when either variable never changes, because a constant has no variance to measure. And r only fits variables measured on a numeric scale; when data comes as ranks, like search positions, statisticians use Spearman’s rank-order correlation instead. Strength matters as much as direction: a coefficient near 0 supports almost no conclusion, so check the number itself before you accept a study’s headline claim.

Correlation vs. causation

Correlation means two things change together. Causation means a change in one produces the change in the other. When two measures track each other with no mechanism connecting them, the result is a spurious correlation. Tyler Vigen’s Spurious Correlations project collects thousands of real examples, including per-capita margarine consumption tracking the divorce rate in Maine.

Vigen builds these through what statisticians call data dredging: comparing 25,237 variables against each other, 636,906,169 correlation calculations in total, until random matches surface. As the project puts it, any sufficiently large dataset will yield strong correlations completely at random. Causation is established the opposite way: start with a hypothesis, change one variable in a controlled test, and hold everything else steady.

How SEO correlation studies work

An SEO correlation study measures which page features co-occur with higher rankings across a large sample of search results: backlink counts, anchor text patterns, content length, HTTPS, page speed. Backlinko’s “We Analyzed 11.8 Million Google Search Results” (updated April 14, 2025) is a well-known example of the format.

A study like that can show which traits top-ranked pages share. It can’t show that a trait causes the ranking. Pages that already rank attract more backlinks simply because more people see them (reverse causality), and a third factor such as brand size can drive both the links and the rankings. A correlation study can suggest a ranking factor, but it can’t confirm one.

What to do with a correlation study

Treat correlation studies as a source of hypotheses, never as instructions. Before you act on one:

1. Check whether Google actually documents the factor. Our ranking factor entry separates confirmed factors from ranking myths.
2. Validate the change on your own site with a controlled split test, changing one variable and measuring against an unchanged control group. Our A/B testing guide walks through the setup.

A correlation study tells you what’s worth testing first. Your own test tells you whether it actually works.