272 stat fields tested across four positions. 41 replicated. This is the record of the other 231, including our own discarded ideas.
Which fantasy football stats do not work?
Published fantasy research is overwhelmingly a record of what worked. The failures are rarely written down, which makes the same dead ideas circulate indefinitely.
272 stat fields were tested across four positions and 41 replicated — an 84.9% failure rate. Among the casualties: player line-beating persistence across 58,890 graded lines came in at −0.055, not zero but negative, meaning the market over-corrects toward recent line-beaters and makes them slightly worse bets going forward.
Most measured quantities in football describe what happened rather than what a player controls. A stat has to be both attributable to the player and stable in the conditions that produce it to survive a replication bar, and most are one or neither.
Findings you cannot check against their failures are not findings.
Every field is swept season by season and required to hold its sign in every season tested. Fields that predict strongly in some seasons and reverse in others are reported as failures, not as conditional findings.
Figures on this page come from the GWTTKB methodology report, where the full testing record is published.
A failure to replicate is not proof that a metric is meaningless — it means the metric does not predict fantasy scoring forward under this test. Several of these are perfectly good descriptive measures.