Winnable Games
On Monday, May 17, the Boston Redsox led the New York Yankees 9-7 entering the ninth inning. The historical record tells us that a team has more than a 90% chance of winning that game. The Redsox lost, turning a winnable game into an actual lost game.
Let’s count a winnable game as any game where the team has at least an 81.21% chance of winning at any point during the game prior to entering the ninth inning. We see this happened 2430 times last year. Not coincidentally, there were 2430 games played in 2009. Basically, if you have more winnable games than actual wins, then this means you lost a few more games than expected. You can call it bad timing, or non-clutch, or whatever term you like.
In 2009, the Mets had 77 winnable games, but ended the season with only 70 wins. They also had 89 lose-able games (chance of winning as low as 18.79% or worse), and ended up with 92 losses. They led the league in not winning as many games as they should have. On the other end of the spectrum were the clutch-filled New York Yankees. The Yanks had 97 winnable games, compared to their actual 103, and 69 lose-able games, but ended up losing only 59. That’s an 8 game improvement.
The 2009 Rays had an interesting season: they had 19 games that were considered both winnable and losable. For example, on August 30, 2009:
In the bottom of the 8th inning, up by two, with two outs, they were in a winnable position: 86.4% chance of winning. After Polanco hits a three-run HR two batters later, they had a 83.9% chance of losing. The Rays led the league in most winnable/lose-able games. The Chicago Cubs were on the other end, with only four games that were both winnable and lose-able.
Finally, the Arizona Diamondbacks provided their fans with the most thrills. They had 99 lose-able games, but ended up with the win 17 times to lead the league. The Pirates were only able to win 5 of their lose-able games.
As for 2010, the Diamondbacks have taken a reversal, with 22 winnable and 19 lose-able games, compared to an actual 16-23 record. The Tigers on the other hand have 18 winnable and 24 lose-able games, compared to an actual 22-16 record. And the Redsox have been involved in seven games where they had the chance to both win and lose the game, to lead the league.
Is this article title a potshot @ Randy Winn? One might indeed say that the game and the Yankees’ prospects at victory were “Randy Winned” into oblivion, and, in so far as any noun can be ‘done unto’ by a transitive verb (glass that you break is breakable, a baby that you can kiss is kissable, a dog that you loved was lovable),… you might say that game was “Winnable.” That which is “Winned” is eminently “Winnable.”
Why 81.21%?
Threshhold selected to make this true:
Ah, thanks. I didn’t catch that and had the same question.
I was also wondering where that percentage came from, but that’s not true. A couple of minutes worth of playing with retrosheet data turns up this game (and given that my search was from the beginning of the season, and how close this is to the beginning, I’m sure there were others)
http://www.fangraphs.com/wins.aspx?date=2009-04-09&team=White%20Sox&dh=0&season=2009
Hmm… decided to do another minute’s worth of searching:
http://www.fangraphs.com/wins.aspx?date=2009-04-18&team=Blue%20Jays&dh=0&season=2009
http://www.fangraphs.com/wins.aspx?date=2009-04-18&team=Giants&dh=0&season=2009
Let me try to clarify that… not-well-stated objection. Your statement says that there were 2430 games, and that you chose that percentage so that every game would be winnable. My point was merely that they weren’t all winnable (by this definition), even if you did come up with 2430 times where a game was winnable by one team. (In a back-and-forth game, would your search return that game more than once if one team reached “winnability” more than once?)
Right, good point.
Since there are 2430 games, and since some games are not winnable by either team (win% between 20% and 80% heading into the 9th), then that means that some of the other games are winnable by both teams.
Is the objection that I have some games as winnable by two teams? If it helps, I can just as well separate the 2430 winnable games into say 2000 games winnable by one team, and 215 games winnable by two teams.
Sorry, replying out of thread; it won’t let me reply to your reply for some reason.
Well, the 81.21% seemed pretty arbitrary, and saying it was chosen to make winnable games equal to played games… Given that that doesn’t mean that every game was winnable (by someone), I just am not sure what the point of setting them equal is, either. I can’t think of a better spot for where to set the threshold; I guess I’m just wondering if there’s something about using that spot as a threshold that I’m missing. (Plus, the phrasing in the article seemed to me to imply that every game was winnable at some point, by that definition, when that’s obviously not the case.)
Sum of all winnable games was 2430. Sum of all wins was 2430. If the Angels have 102 winnable games and 97 actual wins, and the Yanks have 98 winnable games and 103 actual wins (or whatever the numbers were), then that to me provides a quick and easy comparison point.
If on the other hand I selected a winnable game as say the odds being greater than say 75% and that only one team reached that level, then maybe I get a total of say 1800 winnable games (for one team). That puts the average at 60 for each team.
Then what? I’m adding extra work then if say the Mariners had 62 winnable games by this definition and 85 actual wins. Should I just add 21 games to each team to baseline them to the 81 win average? Should I multiply by 81/60?
There’s plenty of arbitrary choices to be made. I don’t know what any arbitrary choice is better than the other, but I have to make a choice at some point, so I can paint the story.
If you are suggesting that this arbitrary choice doesn’t paint a good story, that’s fine. If, let’s say, it’s a big surprise that the Mets left their fans with a real downer or the Yanks really were clutch, then your objections would have more merit.
Otherwise, we can sit here and try to tease out the best definition, and still end up win the same place (which is what I think we’d get anyway). So, I don’t know that it’s a big deal, considering that the baseline that I’ve chosen requires no further adjustments. 2430=2430.
Thanks, that makes perfect sense now.
Interesting analysis. Can you post the numbers for all teams on your website? Is there any easy way to compute this? Might be a starting point to evaluate ‘managerial strategy’, though I am guessing it is mostly luck. Thanks
I think this is largely due to luck, but I still really like looking at records with this type of WPA filter. It might not show true talent level, but the teams that win alot of losable games are fun to watch and those that lose alot of winnable games (like the Mets as of resent) are unbearable.
I would be very interest in any recurring traits found in the two extremes here. I sense that teams which win alot of losable games are good at getting on base, see alot of pitches and are generally good hitting teams, where as the teams that lose alot of winnable games have pitchers in the bullpen would put a lot of men on and suffer from high slugging against them. I guess the samples in even a full season aren’t large enough to really draw strong conclusions from, but it is still an interesting idea.
It will be a combination of timely hitting and quality relief pitching.
It’s not the heart of the article, but the “not coincidentally” caught my eye.
2430 showing up in both places does in fact seem like a coincidence to me. It’s indeed unsurprising that the number of “winnable” games is close to the total games played, but exactly…? I’m not adept at searching last year, but here’s an example of a game that wasn’t “winnable” for either team (4/17/08 Rockies vs Padres), while some are winnable for both teams, as you noted.
If you’re double counting some games, the total could be higher or lower than 2430. If not, the total would likely be lower. Either way, surprising.
Am I missing something?
So now of course you point out where the percentage choice came from… 🙂
Yeah as a mets fan I’d be interested in seeing this stat posted for teams too. I feel like we’ve had a lot of winnable/losable games that have come down to a few bad decisions.
Is WPA adjusted by team and stadium, or is it totally context neutral? It seems like a team with a potent lineup and a lousy pitching staff (Brewers, Diamondbacks, etc) will win from behind and lose from ahead more often because there are more runs overall.
It’s probably not a huge difference because no teams score enough runs to make a 2-3 run lead trivial. More of an “I wonder…”
I can post a handy linkable file. If you think this is something worthwhile, say like the “average LI” for a game, and “average win expectancy” for a game (those two help you determine the closeness of a game), maybe you can petition David to add “winnable” and “lose-able” games, and make it searchable.
Speaking of which, a page that has a game log for each team would seem in order. For example, there should be a “GAME” tab next to Batters/Pitchers/Fielders here
Once you do that, we can see all 162 games for each team, see their team totals as batters and pitchers for WPA and the like. And, you can include “winnable” and “losable” games, or, simply, list the actual numbers I used, which is the max/min win expectancy before the 9th inning.
David has a huge pile I’m sure to do.
I feel a correlation between losing winnable games and Meltdowns is out there…. See 2010 Mariners bullpen.
Yeah, I was surprised noone had mentioned the M’s until that last comment. Do they lead the league in the disparity between winnable games and actual wins in 2010?
DBacks worst, Mariners 2nd worst, so far. Rangers actually third-worst if you can believe it.
Tigers, Astros(!) and Rays top 3.
Wow, you wanna talk about Winnable Games…check out the Braves/Reds today. Reds were over 98% for almost the majority of the game, and were even up to 99.9% in the 9th inning. Lost it.
This needs it’s own category.
I wonder whether I might suggest a different term, one that is, I recognize, no more precise on deeper analysis but that might be clearer for the casual reader, viz., “presumptive win”, an appellative that I have used to describe games of the sort of which you write (using a less sophisticated, purely conjectural bar of 80 per cent), a suggestion that you should, of course, feel free to ignore.
On an unrelated note, might I trouble you to check, if you’ve runs the numbers for all of the teams, whether I’ve the record for the Brewers (1 lose-able win, 5 winnable losses) correct? Thanks.