Printer-friendly

Contact Author

May 19, 2005

Postseason Odds

The History

by Clay Davenport

"You show the Orioles with a 60% chance of winning the division, and the Yankees not even at 10%. How can this be?"

Yes, I've gotten multiple e-mails just like that so far this season, people wondering how I can justify such long odds in the Orioles' favor so early in the season. You can look up the methodology in articles found here and here from last season. Even though we didn't start running the postseason odds reports until August last year, I haven't changed the program in any way: The settings are determined solely by the way a team has played so far this season, without any knowledge whatsoever of which players are on the team. It doesn't know or care that Yankee pitchers are all a run or more behind their recently established levels of performance, and it doesn't know if Brian Roberts will really hit for a .390 EqA over the entire season. It only knows what the team has done to date.

Given that it has no idea whether a team can be described as overperforming or underperforming, you have to wonder just how good a job it can do with the simulations. To answer that, I wrote a version of the program that would calculate the adjusted standings and expected post-season odds for any date in history, with hearty thanks once more directed to the incomparable work of Retrosheet. The differences between the historical runs and the current season are:

• That I "only" ran the simulation 10,000 times per season, instead of the one miiiilllliiiooonnn times the current season gets.

• For the vast majority of seasons, I did not have access to team batting and pitching statistics through such-and-such date, which meant that I could not calculate whether the teams were scoring or allowing more runs than expected. I was forced to assume that what I saw in terms of runs scored and allowed was real. I still adjusted for the team's Pythagorean projection and for the quality of the schedule already faced, so we've still got two out of three standings adjustments working for us, and that ain't bad.

For timing, I had several options, but chose to go with the simple "days since Opening Day". As of Monday, May 16, we were 44 days removed from the first game of the season, back on April 3; I ran the adjusted standings and playoff odds using 45 days. In 2005, teams had played about 36 games apiece in those first 44 days; historically, teams have averaged about 35 games, with lows of zero (mid-season replacement teams in some 19th century leagues) to a high of 47 for several teams in 1944--wartime restrictions led a certain compression of the schedule.