The Pirates are making the playoffs, barring an unprecedented disaster. Pittsburgh’s postseason odds are at 99.6 percent, meaning there is a 1-in-250-chance the Bucs are shut out of the playoffs.
However, posting an 8-9 September record that includes being swept in St. Louis has pushed the Pirates from being favorites in the NL Central division (12.4 percent odds).
There is a 78-percent chance the NL Wild Card game will be a Pirates-Reds matchup. Pittsburgh and Cincinnati play six of their final nine games against each other, and it will be little more than a battle for Wild Card home field unless one of the two teams dominates the other over those six games.
Okay, more numbers. Let’s look at the NL Central standings.
- St. Louis: 89-63
- Pittsburgh: 87-65 (2.0 games back)
- Cincinnati: 87-66 (2.5 games back)
And the Wild Card standings.
- Pittsburgh: 87-65
- Cincinnati: 87-66
- Washington: 81-71 (5.5 GB)
- Arizona: 77-74 (9.0 GB)
Yeah, the Arizona Diamondbacks are still not mathematically eliminated. Let’s just get to the individual projections.
- Baseball Prospectus — 99.7% playoff odds, 10.7% division odds
- Clay Davenport — 99.5% playoff odds, 12.4% division odds
- FanGraphs — 99.6% playoff odds, 14.2% division odds
If you want to, you can also check out Baseball Prospectus’ projections in nifty graph form on MLB.com.
There is not much more to add about this set of projections. The most likely scenario is the Pirates finish about 92-70 and play the Cincinnati Reds in a Wild Card Coin Flip Game that would include 13 hit-by-pitches (estimated). Unless the St. Louis Cardinals decide to start losing, the last two Pirates-Reds series will only decide if they both have to fly back to Pittsburgh for one more game.
One more point, though. The odds and projection saw this coming.
The Pirates had baseball’s best record on July 1, the season’s halfway point. But all the systems recognized that Pittsburgh’s true talent level was that of a .500 team, projecting the Bucs to go about 41-40 in the second half and finish with 92 wins.
That might have sounded silly at a time when the team was steamrolling its opponents, but projection systems are adept at recognizing when a team is playing over its head over a small sample. So far, those projections have been right-on. The Pirates are 36-35 since July began and appear headed directly at 92 wins.
Are these odds perfect? Of course not, and they’re always being adjusted and fine-tuned. However, fans should recognize the projections give unbiased, impartial forecasts that have a higher success rate than the ol’ eye test.
The future is scary. Bring a friend and a calculator.
Agreed that the projection systems since Julyish seem to have pegged a likely outcome for the Pirates season. I can’t help but find them dubious, though – not on the basis that they’re math-heavy systems, but on the basis that they are not _rigorous_ in calculating probability.
If we think about each game as an outcome, we only have a trial of 162 to determine what the trend of a team is. The Law of Large Numbers sort of chuckles at you bemusedly if you mention your N is 162.
“But Philip! Clay Davenport runs simulations thousands of times! The Law of Large Numbers surely applies there.”
Yes, but his underlying assumptions are bunk and hooey, because they have to be. You can’t accurately simulate a 162-game season – not enough computing power on the planet. Look at their April 2 projections: http://web.archive.org/web/20130402064155/http://www.baseballprospectus.com/odds/
79-82. Naaaaht great.
I think you begged the question there. Additionally, what is your definition of accuracy? Fangraph offers a coin flip mode for comparison.
Confused about what assumption I’m making – care to point it out?
Davenport’s using PECOTA to run his team simulations – errors in individual player performance are going to be multiplied when you draw it out to teams. It’s an okay way to make guesses about how players will perform given past history and comparables (looks more accurate than most) but rigorous probability theory it ain’t. I can guarantee you that they are not simulating every pitch of every at bat of a season and using the inclusion-exclusion principle to sum probabilities. It’s nearly impossible to do for even simple models. Really hard to do for baseball games, which thanks to things like foul balls can have potentially infinite sequences.
Wikipedia espouses that their avg error rate across teams hasn’t been terrible, though it’s only up to 2008:
“2003 5.91 wins; 2004 7.71 wins; 2005 5.14 wins; 2006 4.94 wins; 2007 4.31 wins. Silver conjectures that the improvement has come in part from taking defense into account in the forecasts beginning in 2005. In 2008 the average error was 8.5 wins.”
Obviously there’s baseball left to play but if you take a gander roughly half are off 10+ wins. I definitely want to see how it stacks up at the end of the year.
As for preferred accuracy: feels like anything over 5% error is getting iffy to me, and the avg error rate feels less important than large deviations for some teams. There are quite a few of those this year.
I apologize; I did not understand your criticism of the projection systems, and I set up a straw man.
I read your assumption to be that the projection systems for players are terrible/flawed thus any playoff odds derived from such are terrible/flawed. I think the ZiPS, CHONE, PETCO, systems are fairly accurate at player level, (not in the statistical sense, perhaps not inaccurate is a better phrasing.) I do not think they are bunk given that my priors for comparison were macroeconomic forecasting and ex-player conjecture.
You are correct, if anyone thinks these playoff odds are rigorous probability calculations they are misrepresenting them. Rereading those final paragraphs I now see your concern.