Team Stats

The module reads every game of the day in four stats: threes, rebounds, assists and turnovers. For each team it shows what it does on average in its recent games and what its opponent allows, and from those a projection for this game. Both teams together are set against the league's usual game, so you see at once whether more threes or rebounds than usual are on the way.

Under the projection is a ladder of lines: how often similar past games went over every line around it, the fair price of Over and Under and how often both teams' recent games ended over. Premium adds each team's own lines, the game in all four stats and the meetings.

A team's projection is the average of two numbers: what the team does in its recent games and what its opponent allows. Both are pulled a little toward the league average — a few games are few, and extremes even out — and the league's home advantage is added. Over more than 21,000 games this pairing missed less than the team's own form alone.

The rate over a line is read from the spread such counts have: it grows with the square root of the projection, as in a Poisson distribution. On past games the rates matched the results within a few percentage points at every level. Rows where the data provider sent obvious errors (hundreds of threes, or no turnovers in a whole league) are left out.

A fair price is a price without margin: 100 divided by the historical rate. If a bookmaker posts a line you see in the ladder and pays more than the fair price for it, its price is higher than the historical data implies. If it pays less, the margin eats the difference. It is a comparison with history, not a promise of a win.

Recent games know nothing of line-ups: an injured point guard changes the assists, a missing centre the rebounds. That is why lines close to the projection and games with both teams at full strength work best.