For fifty years, lunar scientists divided into two camps. One group argued the Moon had a powerful magnetic field for a billion years — as strong as Earth's, sustained by a long-lived dynamo. The other argued the field was always weak or absent. Both had evidence. Both had Apollo rocks. Neither could explain the other's data.
Nichols, Wade, and Stephenson (Oxford, Nature Geoscience 2026) resolved the contradiction by finding the variable no one had controlled for: titanium content.
Every Apollo sample recording a strong magnetic field contained more than 6 wt.% titanium. Every sample below that threshold recorded a weak field. The correlation is stark. The mechanism: episodic melting of titanium-rich material at the lunar core-mantle boundary generated brief, intense magnetic episodes — possibly as strong as Earth's field, but lasting no more than 5,000 years, possibly as short as decades.
The Moon had a magnetic field. It was almost always weak. The exceptions were titanium-driven blips.
The fifty-year debate was an artifact of where Apollo landed. The astronauts touched down in mare — the smooth, dark basaltic plains that were safe for landing. Mare basalts are titanium-rich. The rocks they collected happened to be the ones that recorded the rare events. A random sampling of the lunar surface would have shown overwhelmingly weak-field evidence with occasional anomalies. Instead, the landing sites selected for titanium-rich terrain, and the researchers inherited a dataset biased toward the exception.
This is the purest form of sampling bias: the measurement apparatus — where you land your spacecraft — determined which history you recovered. The data was real. The rocks genuinely recorded a strong field. The field genuinely existed. But the question was never “did the Moon have a strong field?” It was “how often?” And the answer — almost never — was invisible because the sampling was non-random in exactly the dimension that mattered.
The same structure appears wherever access determines evidence. Medical knowledge is biased toward diseases that bring patients to hospitals. Species ecology is biased toward organisms large enough to count. Fossil records are biased toward environments that preserve bone. In each case, what gets measured is what's accessible, not what's typical.