APBRmetrics

[Mike G]

Suppose a basketball "game" was sudden death: whoever scores first is the winner.

Let's say team A is comprised of players that are 10% "better" than those on team B. Team A will beat team B about 52.5% of the time.

If a "game" is to 10 points, the odds change. Now team A's W-L % against team B is going to be about .600.

Without changing anything regarding the makeup of the teams, or their relative abilities, but simply by allowing a game to go some 100 possessions (each), the likelihood of victory by Team A rises to about 80%.

In an NBA game, a team that averages 18 pts per position is overmatched by one that averages 20. Yet the players, on average, are within 10% in ability. And in real life, there are below-average players on even the best teams. When we divvy up win-shares, or player-wins, teams that are 90% as good as average have only 40% of the average number of wins.

I think, for purposes of predicting a team's win %, we should have something like player potential-wins. This would be the player's share of wins on an average team. A team with a total of 41 ppw among their players could in fact be expected to win 41. If that total is 51, expected team wins would be higher than 51. How much higher? Maybe that requires the formula with the 14th-powers.