Printer-friendly

October 4, 2007

Lies, Damned Lies

The Greatest Pennant Race Comebacks

by Nate Silver

I’ve had kind of a lucky year. The PECOTA projection I made in the offseason that gained the most notoriety is that the White Sox would finish 72-90; that turned out to be their actual finish. After that, the next most controversial projection was that Dustin Pedroia was going to have a very good year; now he looks like a shoo-in for the Rookie of the Year Award. And in a July article for Sports Illustrated, we noted that the Secret Sauce predicted that the Red Sox would meet the Cubs in the World Series, an outcome that now looks entirely possible (though incrementally less so after the Cubs’ loss last night).

That doesn’t mean we’ve gotten everything right. PECOTA missed high on a lot of rookie hitters. We thought the Devil Rays would somehow win 78 or 79 ballgames. We thought big breakouts were coming for Jeremy Bonderman and Daniel Cabrera. Overall, it was a pretty normal year for PECOTA; our philosophy if we get six hits for every five misses, we’ve done pretty well. But those bad predictions didn’t generate a lot of press hits, and the good ones did.

So it was kind of fortuitous that I wrote an article last week about the greatest pennant race collapses in baseball history. At the time that I wrote that piece, the Mets still had a 96 percent shot at making the playoffs. By the time we published it, they were down to 85 percent, but still looked relatively safe. Then a terrible thing happened, and the Mets joined the ranks of the rich and infamous. The topic got a lot of coverage; I did an update for SI, and Clay did another take over here. As you might guess, I’m pretty tired of this subject by now, but I got so many reader requests to do the opposite of this topic—the greatest pennant race comebacks in baseball history—that I’m going to aim to please.

As it happens, most of the greatest comebacks conceived of strictly as the longest odds overcome toward the middle of the season rather than toward the end. This is the result of a sort of statistical quirk—if we’re looking at those teams that had the longest odds against coming back, what we want are the closest numbers to zero that aren’t exactly zero; basically, those numbers that are between 0.001 and one percent. In the middle of the season, a team cannot really be “mathematically” eliminated; they could always win their last 72 games in a row or whatever and make the playoffs. You might have to run 100 simulations before they made the playoffs, or 10,000, or one million, or 100 million.

So the middle of the season is when it’s easiest to achieve these small, but nonzero numbers. At the end of the season, on the other hand, the function is a bit less continuous. With four games left to play, you either have a somewhat tangible chance to make the playoffs, or you are at exactly zero and no combination of wins and losses will get you in. Maybe you wake up in the morning with a 5 percent chance to make the postseason—low but not that low—and if you win it goes up to 10 percent, and if you lose it goes to exactly zero.