THE BOOK--Playing The Percentages In Baseball

That’s a fantastic paper.  I was particularly pleased to discover a Poisson implementation in Excel.  Interestingly, I think, Ryder’s results suggest that scoring first actually leads to a larger probability of winning than does the analysis you refer to (unless I’m misinterpreting Ryder’s results, which is always possible).

Yes, Alan is looking at a team that has a 58% chance of winning to start the game (3.0 v 2.5 goals per game).

The 83% number for 15 min into a game is significantly higher than what would be expected in a purely Poisson distribution (without home ice considered). It should be about 69% based on the scoring rates of the recent regular season.

That there was no score for a quarter of the game suggests that the pace of the game is a defensive slug-fest. This would effectively reduce the rate parameter in the Poisson distribution, and make it less likely that a team can come back from a 1-goal deficit. (Either that or their sample size is too small.)

In case anyone is interested, I’ve been doing live in-game win probability graphs for the playoffs. It’s a slightly different model than Ryder’s, but it factors in penalty plays.

http://wp.advancednflstats.com/nhl

In regards to scoring first, whether it’s hockey, baseball, soccer, or maybe even football, doesn’t it just make sense that the team that scores first will obviously win more often? Announcers throw this out there like it’s a unique stat, or that the process of scoring first somehow is what makes a team more likely to win. Isn’t it probably true that they have it backwards? That is, the team more likely to win will probably score first. In other words, the better team scores more often than the worse team, and “scoring more often” means they’ll probably score first, too.

#3, Scoring first probably matters more in sports where there is less scoring.  I wonder what the winning %’s are for basketball.  It would be interesting to see the numbers across sports depending on game scoring.

Brian was marked for moderation and is now open.

Brian: 15 minutes LEFT, not into the game.

Oh. I should learn to read more carefully. In that case, the purely Poisson win probability is 80%, likely well within statistical error of the study’s 83%.