Every game was priced from real bookmaker moneylines pulled from the historical odds archive, then de-vigged so the two sides add to 100%. That de-vigged number is the market's honest win probability. A factor only counts as an edge if games matching it beat that probability by more than chance would explain. Divisional status resolved from real franchise-division membership, handling relocations so a moved franchise keeps its correct division for each historical season. Divisional games compared directly against non-divisional games from the same seasons. Both sets scored the same way: real favourite win rate versus the summed de-vigged market probability.

edge lab · factor writeup · 6 min read

Divisional Games: Does Familiarity Really Compress the Margin?

Q: What the numbers said

Sample: 5,295 games; Window: 2006–2025; Signal: z < 1.17 Divisional favourites won at their implied rate. Non-divisional favourites won at their implied rate. The difference between the two gaps was not distinguishable from zero. Whatever familiarity does to a football game — and it plausibly does something to the score — it is already inside the number by the time you see it. Why this keeps getting repeated: Divisional games do produce more upsets in raw counts, because divisional games are more often close on paper to begin with. That is a property of the matchups, not a discount the market missed.

Q: What it means for your picks

Divisional status on its own is not a reason to avoid a team in a survivor pool. A 78% divisional favourite is still a 78% favourite. If you want a real reason to be nervous about a game, use the market's own number and the injury report, not the rivalry. Head-to-head history for any pairing is browsable in Explore, and every other theory we ran is listed in the Edge Lab.

The theory says division rivals know each other too well for favourites to cover. Across 5,295 games, divisional favourites won at exactly their implied rate.

"Throw the records out, it's a division game." It is the most repeated line in football and it has the ring of truth: these teams play twice a year, share a scouting file, and hate each other. Familiarity should compress margins and make favourites sweat. Across twenty seasons of real games, it doesn't.

What we tested

The hypothesis: in games between division rivals, the favourite underperforms the market's expectation, because familiarity narrows the real talent gap.

How we tested it

Every game was priced from real bookmaker moneylines pulled from the historical odds archive, then de-vigged so the two sides add to 100%. That de-vigged number is the market's honest win probability. A factor only counts as an edge if games matching it beat that probability by more than chance would explain.

Divisional status resolved from real franchise-division membership, handling relocations so a moved franchise keeps its correct division for each historical season.
Divisional games compared directly against non-divisional games from the same seasons.
Both sets scored the same way: real favourite win rate versus the summed de-vigged market probability.

What the numbers said

Sample: 5,295 games
Window: 2006–2025
Signal: z < 1.17

Divisional favourites won at their implied rate. Non-divisional favourites won at their implied rate. The difference between the two gaps was not distinguishable from zero. Whatever familiarity does to a football game — and it plausibly does something to the score — it is already inside the number by the time you see it.

Why this keeps getting repeated

Divisional games do produce more upsets in raw counts, because divisional games are more often close on paper to begin with. That is a property of the matchups, not a discount the market missed.

What it means for your picks

Divisional status on its own is not a reason to avoid a team in a survivor pool. A 78% divisional favourite is still a 78% favourite. If you want a real reason to be nervous about a game, use the market's own number and the injury report, not the rivalry.

Head-to-head history for any pairing is browsable in Explore, and every other theory we ran is listed in the Edge Lab.

More: back to the research register · methodology · guides