Tuesday’s Crazy Comebacks
After Brian Matusz struck out Willy Aybar in the 8th inning of last night’s Rays-Orioles game, Matusz was probably feeling pretty confident about leaving the game with a win. At the time, the Orioles led the Rays 3-0, and with only 5 outs remaining in the game, the Rays’ chances of winning were slim – 5.5%, to be exact.
Similarly, Brian Bannister had to feel good after recording the first out of the seventh inning. At the time, Bannister’s Royals led the Tigers 5-0 with only 8 outs remaining for Detroit. The Tigers’ chances at the time sat at a mere 2.9%. Even though Bannister allowed a run at the hands of Gerald Laird before exiting the game, the Royals’ win probability was still over 90% when he was replaced by Roman Colon.
Naturally, I wouldn’t be mentioning either of these situations if the improbable hadn’t occurred – neither Bannister or Matusz recorded a win, and in both cases their teams lost.
The Royals’ bullpen worked with remarkable efficiency to blow the lead. Colon gave up two doubles to only one out in the three batters he faced, allowing the Tigers to close the gap to 5-3. Dustin Hughes gave up a single and a walk to load the bases, setting the table for Juan Cruz. After a walk to Miguel Cabrera closed the gap to 5-4, Cruz finished the job by allowing a two run double to Carlos Guillen. By that point, the Tigers’ win probability had skyrocketed to 79.2%, a gain of 76.3% in merely 8 batters. The Tigers managed to hold the lead, and the game finished with a score of 6-5.
The collapse of the Orioles in the 8th was also quite rapid – Jim Johnson immediately recorded an out after replacing Matusz, but then combined with Will Ohman to allow 3 straight run scoring hits to Evan Longoria, Carlos Pena, and B.J. Upton. The hits combined for 4 runs, and put the Rays up by a score of 5-3, boosting the Rays win probability up to 85.3%. The Orioles hitters nearly reclaimed the game, however, as Luke Scott tied the game with a two run shot in the bottom half of the inning. Ohman and Cla Meredith combined for a quiet 9th inning, and a Nick Markakis single brought Baltimore back to a 72.4% chance to win, but the Orioles bullpen just wasn’t deep enough to handle the Rays in the 10th. Carlos Pena sealed the game for Tampa Bay with a 3 run shot off of Matt Albers. Rafael Soriano managed to save the game, despite a solo home run by Ty Wigginton, and the Rays won 8-6.
After dominating the majority of the game, both the Orioles and Royals were let down by their bullpens. The odds of both comebacks happening on the same night are a mere 0.16% chance – we would expect two comebacks of this magnitude to happen on the same night only about 0.25 times per season. Tuesday’s heroic comebacks (or unbelievable choke jobs, depending on your perspective) certainly provided us with some entertaining baseball, possibly on a level we won’t see again this season.
Jack Moore's work can be seen at VICE Sports and anywhere else you're willing to pay him to write. Buy his e-book.
Does win probability account for sub-par bullpens? I’d assume that a team would have a higher chance of coming back against Kansas City’s bullpen Vs. Oakland’s(from last year). It may only account for a couple % difference, but I would absolutely say that a team has a better chance of coming back from down 5 runs to win against a subpar bullpen, and probability should reflect that.
As I understand it, win probability does not take into account anything about the players on either team. It doesn’t care if you have Pujols at the plate or Willie Bloomquist, and it doesn’t care if your closer is Mariano Rivera or Rick “Wild Thing” Vaughn without his glasses. It just reflects the game state and the past history of outcomes of games in that state by all teams over multiple seasons.
Have there been any studies done on unlikely wins? Following the Tigers, I feel like more than half of their games are won by a team that at some point had a win probability less than 25%. Maybe our hitters just take a while to warm up and our bullpen is historically inconsistent?
Where did you get that last 0.25? I’m pretty sure it’s way off.
Let’s say every game gets to the point where one team has a 97% chance of winning. It goes from 50% at the start to 100% in one teams favor at the end, so at some point it has to pass 97%, right? This is not exact, because baseball happens in discrete steps (plate appearances, outs, steals, etc.), making the changes in win probability not continuous, but perhaps it’s close enough.
There are 2,430 games in a season, so that means that 2,430 * 0.03 = 73 comebacks of the type you describe will happen on average. I can’t recall off the top of my head the next step here, but assuming these are randomly distributed, two will occur on the same night several times a season.
I know this is nit-picking, but I feel like I’ve seen several claims like this thrown in at the end of FanGraphs articles recently.
You’re exactly right. The 0.25 comes from multiplying 162 times 0.16% — but that’s the number of times you’d expect two comebacks of that magnitude if there were only two games a day (or two games a day that reached that chance of victory).
Gratuitous mathery: Say there are 13 games a day that reach 97% chance of victory for one side before the end. (Games that are close late may jump straight from something below 97% to 100% — ironically, Royals-Tigers did that in the ninth inning — and teams don’t always play on the same day, so let’s use that as a guesstimate.) That gives 78 ways for exactly two of those games to involve comebacks — for each of those pairs there’s a (.03)^2 x (.97)^11 = 0.064% chance that those two games will end in a comeback and none of the other eleven will. Times 78, that’s a 5% chance that exactly two games will end in a comeback from a 3% WPA. So we should expect that to happen about eight times a season.
Just adding to this train of thought.
Assuming about 13 games a night on average have some point where one team has about a 5% chance of winning, that means that you would expect that at least one comeback will occur 50% of the time (.95^13). At least 2 comebacks will occur about 15 percent of the time. That’s about once a week. The chances mentioned above are the chances that those two particular games would feature comebacks.
But as Jack said, he wouldn’t be writing about it if it didn’t happen.
“I know this is nit-picking, but I feel like I’ve seen several claims like this thrown in at the end of FanGraphs articles recently.”
I’d like to second that. Probability is rather complicated, and it’s easy to think you’ve got it when you don’t.. The article is good without that extra bit.