Reading anchor correlation studies without fooling yourself

Reading anchor correlation studies without fooling yourself

Every year a fresh correlation study reports that top-ranking pages have 'more diverse anchors' or 'higher branded share,' and the finding gets repackaged as a directive. The methodology rarely supports the directive, and the reverse-causation problem is usually fatal.

The core defect: ranking position is the independent variable in these studies, but it is treated as the outcome. Pages rank well, then accumulate diverse, branded anchors because they are visible and citable. The diversity is a consequence of ranking, not necessarily a cause of it.

— Correlation coefficients in these studies are typically weak, often well under 0.2.
— A weak coefficient on observational data with plausible reverse causation supports almost no actionable claim.

On one hand, the consistency across years and providers is mildly reassuring — the same direction keeps appearing. On the other, shared methodology means shared bias; ten studies repeating the same flaw are not ten independent confirmations.

Limitation: none of these studies can randomize anchor profiles, so by construction they cannot establish causation, only association.

Open question: what would a single well-designed natural experiment — two near-identical pages, anchors as the only manipulated variable — be worth against the entire shelf of correlation reports we currently cite?