Buy The Book
If you are going to compare to a baseline, you CANNOT compare to zero. You have to compare to the actual hitter’s true talent level. It’s not the average hitter that bunts, but a below average hitter. So, we EXPECT to see a negative win value relative to the average hitter, not only on bunts, but on non-bunts too.
Glove-slap: Mike.
Brian Burke does a good job for football.
***
He also signs off with some MGLism:
If you gave the Saints coaches the choice between receiving the punt and letting the Falcons roll the dice on 4th and inches, they’d take the punt every time and twice on Sunday. That tells you something, doesn’t it?
If you are looking for a quick equation:
chance of scoring
= 0.086
* (base + 0.5)
* (3 - outs)
This gives you:
Base Outs ChanceScoring
1 0 0.39
1 1 0.26
1 2 0.13
2 0 0.65
2 1 0.43
2 2 0.22
3 0 0.90
3 1 0.60
3 2 0.30
Close enough to reality for something quick that you need.
If someone wants to come up with something more robust, then play around with those arbitrary coefficients, and post your results!
If you want the actual data.
Colin posted a run expectancy chart for the 9th slot, when the pitcher is the batter and when a pinch hitter is the batter. One thing to be careful about is that the underlying contexts won’t be similar. For example, if you have a pinch hitter at bat, this likely means that the opposing pitcher may be a pretty good reliever. So, it’s an interesting chart, but I’d like to see a more apples-to-apples comparison.
You can also see a related chart here, which I think was 1956-2007, but I’m not sure (I posted the file June 2008).
http://tangotiger.net/retrosheet/reports/re_bo.htm
BASE OUT RE_PIT RE_PH DIFF
___ 0 0.50 0.49 0.00
___ 1 0.21 0.25 0.04
___ 2 0.05 0.09 0.04
1__ 0 0.79 0.84 0.05
1__ 1 0.38 0.50 0.12
1__ 2 0.12 0.19 0.07
_2_ 0 1.02 1.13 0.11
_2_ 1 0.56 0.64 0.08
_2_ 2 0.18 0.28 0.09
12_ 0 1.46 1.40 -0.06
12_ 1 0.69 0.87 0.18
12_ 2 0.24 0.43 0.19
__3 0 1.34 1.39 0.06
__3 1 0.75 0.93 0.18
__3 2 0.22 0.35 0.14
1_3 0 1.57 1.74 0.17
1_3 1 0.86 1.10 0.24
1_3 2 0.26 0.44 0.18
_23 0 1.77 1.96 0.19
_23 1 1.10 1.34 0.24
_23 2 0.28 0.57 0.29
123 0 2.04 2.32 0.28
123 1 1.20 1.51 0.32
123 2 0.42 0.79 0.37
Partial mail:
I’d like to comment on a shortcoming that WPA has as a story stat. Or to use your analogy, I maintain that there are cases where the screwdriver screws the wrong screw. I still love the stat though.
WPA does a great job when the story is batter vs. pitcher. As this is the main story line in most games, WPA gets it right most of the time. However, whenever the story relates to fielding, WPA misses the boat. What bothers me is not so much that it doesn’t credit great/terrible fielding, but apportions both the credit and the blame to the wrong players.
My best example is one of my most favorite games, Yankees vs. Mets on June 12th 2009. With the score in favor of the Mets 8-7, bottom of the 9th, men on 1st & 2nd, 2 outs, A-Rod hit a pop up off K-Rod to Luis Castillo. He dropped the ball and the Yankees won. A-Rod was awarded 0.818 WPA points (I think perhaps the most in his career for a single play!) and K Rod “awarded” -0.818. Neither of them deserved either the credit or the blame.
Similarly, in cases where a 1st baseman makes a tremendous play and turns an unassisted double play by doubling off the runner, the pitcher receives all the credit, the batter all the blame, while the fielder receives nothing.
My response:
There’s nothing in WPA that prevents the awarding of the play to the fielders.
I will point you to the very first time I did WPA, where I go out of my way to talk about Moises Alou, and Gonzalez, and really, all the fielders. NEVER do I say it was ONLY Mark Prior.
http://www.hardballtimes.com/main/article/crucial-situations
That Fangraphs or BR.com does not have the fielding data to do the split is not a problem with WPA. It’s a problem with them, or rather, the lack of data that describes the fielders in detail.
I’ll also point out that WPA only described who was INVOLVED. It does NOT try to establish who was RESPONSIBLE. In the above example, ARod and KRod were involved. Now, it would be nice if Fangraphs and BR.com captures the fielders as well. But capturing fielders is really not so clean. The best example there is that the 1B is simply the end point of the putout, and he is basically “just there”. Was he “involved”? You can make the case that he was involved. You can also make the case that he was incidental.
If someone wants to create a process for WPA that also includes the fielders, then go ahead. You may get to the point that it ends up hiding the story you are trying to show.
Studes calls WPA the story stat. A win expectancy chart (and its associated leverage index) gives you a numeric representation of how you feel watching the game. Imagine that, how you feel, quantified.
Well, Poz unearths a gem of a game with WPA:
Well, Art Shamsky had the greatest WPA ever for a single game. His performance is the very peak of what man can do to win a baseball game. He homered to give his team the lead in the eighth, homered to tie it in the 10th, homered to tie it again in the 11th, there is not much more a baseball player can do. And so what happened? Art Shamsky’s Reds lost to the Pirates that day.
His WPA was +1.50. Remember, with every team starting at 0.50, this means the winning team, as a team, will be +0.50, and the losing team will be -0.50. With Shamsky at +1.50 as the losing team, this means that the rest of the team was -2.00!
Poz asks:
WPA—Win Probability Added—is one of the most interesting statistics out there, and to be honest I do not see why it has not become more popular among mainstream baseball fans. Maybe it needs a better presentation, a better name, a public relations person because WPA, it seems to me, speaks so clearly to what so many baseball fans love about baseball: The winning plays.
And he’s right. WPA was first presented back with the Mills brothers 40 years ago. The first time I talked about WPA was the day after the Cubs/Marlins game, easily the game where you can feel that something huge was happening after every play. And when I posted it on my blog, the readers immediately got it. (also see above chart)
Why do some people give it a bad rap? Because they take the stat out of its comfort zone, the story stat, and lump it in with other stats, where they then try to dissect it, expose its limitations, and then decide that because of those limitations, it has no value whatsoever! It’s like looking at OBP, seeing its limitations (BB = HR), and then deciding it’s crap. Imagine that. The person doesn’t know how to use a screwdriver, and then complains that it can’t hammer in a nail. “Why do we need a screwdriver if I have a nail?” Well, how about I have a screw, and I need that Philips?
I believe the best way to sell WPA, WE, and LI is to hammer home those Cubs/Marlins/Game6 types of games. Because those games can be explained, to a certain degree, numerically. And once you can do that, once you have a process that you can categorize emotional games on a numeric scale, it then becomes possible to find games like Art Shamsky’s.
That’s why we quantify things: to find even more emotional games.
How sure do you need to be to go for it? This reader does the work for one scenario.
Using a different Win Expectancy chart, and trying a different scenario, here’s what I got. This one is a pure sac fly, so the batter made the second out, and you have to decide whether to send the runner or not.
If it’s the bottom of the 9th: if down by 1, you have a .15 chance of winning. If you score, you tie it up, and now you have a .55 chance of winning. If you are thrown out, you are down to 0. So, when you’ve got a +.40 potential gain against a -.15 potential loss, you only need to be safe 27% of the time. If it’s the top of the 9th: down by 1, you have a .10 chance of winning. If you tie it up, you are at .38. Thrown out means 0. In this case, it’s a gain of +.28 against a loss of -.10. That’s 26% of the time.
What if it’s a single (and the batter takes second on the throw): in the bottom of the 9th, it’s .18 chance of winning with runners on the corners, .60 if you score and draw the throw, and 0 if you are thrown out. That’s +.42 gain against -.18 loss, or 30% chance of needing to be safe. In the top of the 9th, it’s .15 winning with runners on the corners, .46 if you score and draw the throw, and 0 if you are thrown out. That’s +.31 against -.15, or 33% chance of needing to be safe.
Basically, go for it if you have at least a 1 in 3 chance of scoring. Sit tight otherwise.
This reader is lamenting the work of Darren Oliver, even though:
Darren Oliver, a key piece in last year’s run to the playoffs and still-quality reliever in the present whose political capital is running out. I say that he is still a quality pitcher in the present not to purposefully antagonize anyone, but because it is the truth: he’s allowing around two-thirds of a home run per nine innings, he’s still maintaining a strikeout-to-walk ratio in the vicinity of 4:1 (though the strikeouts have dropped off this year, from 9.5 K/9 down to 7.9 K/9), his BABIP has plummeted from .310 to .254, and his left-on-base percentage is still sitting in the 76-77 percent range. Hell, his ERA has even fallen by nearly half a run relative to last year. Why, then, are we so perturbed by his mere presence in close-and-late situations?
So, it seems the typical fan is not happy with Oliver, even though on the surface, he’s got all the right numbers. Apparently, the reason is that when it comes time to the context of the game:
Last year, Oliver logged 22 “shutdowns” against just nine “meltdowns,” which wasn’t exactly within the realm of the elite, but was still rather acceptable all things considered. The year before that, it was 23 shutdowns against 10 meltdowns. And one more year before that, it was 19 shutdowns against only nine meltdowns.
But this year? This year, it’s just 13 shutdowns against 12 meltdowns, with that latter figure being tied for the third-most meltdowns by any given relief pitcher in the majors this season, and easily beating out the seven meltdowns incurred by Neftali Feliz and Arthur Rhodes. What had always been a ratio of greater than two-to-one is now a ratio right around one-to-one, or, more specifically, 1.08, which ranks 112th out of 135 qualifying relievers this season. That’s troubling, to say the least.
So, we see that WPA is describing what has happened in a numeric sense, and gives some evidence for why fans are feeling the way they apparently are.
And, to conclude:
That said, win probability and its derivatives aren’t predictive in nature, so I don’t think you can simply go out and declare that Oliver will continue pitching poorly in higher-leverage game situations—hell, he could go out and begin turning that back around tomorrow, and it wouldn’t surprise me too much. It’s certainly worth keeping an eye on, though.
An excellent post all-around.
This is the run expectancy matrix for 1969-1992, which had 0.477 runs per inning.
1B 2B 3B 0 outs 1 outs 2 outs
__ __ __ 0.477 0.252 0.094
1B __ __ 0.853 0.504 0.216
__ 2B __ 1.102 0.678 0.325
1B 2B __ 1.476 0.902 0.435
__ __ 3B 1.340 0.943 0.373
1B __ 3B 1.715 1.149 0.484
__ 2B 3B 1.967 1.380 0.594
1B 2B 3B 2.343 1.545 0.752
This is the RE matrix from BPro, for current 2011 (0.479 runs per inning):
1B 2B 3B 0 outs 1 outs 2 outs
__ __ __ 0.479 0.261 0.094
1B __ __ 0.853 0.519 0.215
__ 2B __ 1.057 0.671 0.311
1B 2B __ 1.431 0.929 0.432
__ __ 3B 1.311 0.928 0.344
1B __ 3B 1.686 1.186 0.465
__ 2B 3B 1.889 1.338 0.561
1B 2B 3B 2.264 1.596 0.683
The numbers in the two charts are quite similar, with a few gaps. Since my chart is based on games played from 1969-1992, and BPro’s chart is based on the five weeks of 2011, we can see why my chart would be expected to be more stable.
I’m not surprised that there is such strong agreement, since the run expectancy matrix is run-environment driven. It’s not like they’d both agree that you start at 0.48 runs per inning, but then be wildly off in some other base/out state.
In any case, I’ve been pleading for baseball to return to the baseball of my youth, and that’s what we are getting. Get ready for the return of small ball! Speedsters, split screens of pitchers/runners, tons of useless pickoffs (that part should be addressed if small ball is the reality), jack rabbits in the outfield. And hopefully the end of highlights that are just run-of-the-mill home runs.
Lots of math, so be forewarned. However, please jump in, because this is my kind of fun.
Michael does a super job of analyzing the situation here.
The situation is this: in the top of the last inning, the score is tied, there are no outs, and the runners are on 1B and 2B. The batter has a 1-2 count with the ball in the air. Do you intentionally drop the ball, keeping the batter at the plate at 1-2, and runners at 1B and 2B. Or, do you catch the ball, let the runners go to 2B and 3B, and a new batter at the plate?
This is exactly what win expectancy is designed to answer. Michael starts with the basic win expectancy data, and then looks to try to make adjustments.
The end result is that the fielder made a defensible and gutsy call to not catch the ball.
Matt. I did not know this: (lgRPG^(1-z)) * 2, where z is between 0.27 and 0.29. Can you show me how this is derived, because I was doing it the brute force way. Sh!t, had I know it was this easy, I would never have proposed my method.
Thanks Brian:
And for your mobile needs:
http://live.advancednflstats.com/WPmobileLive.php.
Brian asks:
Don’t miss Andrew Foland’s 5-part tour de force on how to incorporate pre-game estimates of win probability with the in-game estimates. Awesome job, and I really appreciate the time Andrew has put in. This is something I’ve been asked to do for a while. I think it’s a great idea, and I developed a very similar method to do it.
However, I would be reluctant to make it the ‘official’ WP or WPA model. If the Patriots have a game against a weaker opponent pegged as a 70/30 match-up, and Tom Brady plays lights-out, should he only get 0.30 WPA instead of 0.50 WPA? Should he be penalized for being favored pre-game?
I’ve talked about this in the past, and I guess Brian post-dates those discussions. No problem, as I presume lots of readers don’t go through my archives looking for stuff like that.
Ok, this is how it works. Suppose that Halladay is pitching, and the Phillies are at home. Let’s say that they have a .750 chance of winning. In order to go that, you need to “preallocate” a certain number of wins to Halladay, to the Phillies being at home, to Rollins, and Utley et al. Then, you let them accumulat their +.250 wins or -.750 wins depending on whether they win or lose.
And what happens after one million games? Well, Halladay’s in-game WPA will be exactly ZERO. The sum of his pre-allocated WPA will be exactly equal to his talent level. In short, given enough games, adding up a player’s in-game WPA is useless.
But:
1. That is deathly boring
2. We really don’t know what his true talent level is for each game
3. There aren’t enough games to get a zero for in-game WPA, and now the whole thing will look strange
So, rather than doing the extremely complicated method of preallocating wins on a player by player basis, simply start everyone with ZERO pre-allocated wins, and let them earn their in-game WPA without knowledge of their talent level or their team’s talent level. When you sum it up, you will obviously get zero pre-allocated wins, and the in-game WPA will be their entire impact.
If you are a professional bettor, well, you desperately need pre-allocated wins.
Check out the archives. I have a few threads on the topoc.
Peter does a good job at bringing light to WPA:
But the best thing about Win Probability is that it captures, like no other stat, how we viscerally experience sports. As fans, we carry in our guts a reflexive, fuzzy calculation of our team’s chances of winning a game—a nervous tension that explodes after triumphant plays and collapses after moments of agony. Win Probability expresses that emotion with mathematical precision.
...
“I introduced my first WPA story the day after the Bartman/Cubs collapse game in 2003,” says Tom Tango, co-author of The Book: Playing the Percentages in Baseball. “How can you capture that game with numbers? I think only WPA could do it. It’s a story stat.” In the eighth inning of that notorious playoff game against Florida, the Cubs’ Win Probability crashed from 95.6 percent to 1.8 percent, according to Tango’s calculations. But Steve Bartman’s interference caused just 3.1 percentage points of the plunge.
...
Is there anything Win Probability can’t do? Well, yes. If you’re trying to figure out whether Joey Votto is going to outhit Prince Fielder next year or how much the Reds should pay their MVP, this is not your stat. WPA looks back, not ahead; it measures accomplishment, not skill.But you know when you watch poker on TV, and the screen displays each player’s chances of winning as the cards come out? And how you don’t really need to understand anything about flops or implied odds to pick up the flow of a game—you can just figure out what’s happening as those little percentages change? Those numbers are Win Probabilities. And it’s only a matter of time before they start popping up during baseball or football broadcasts. Unobtrusive? Check. Easy to comprehend? Check. Revealing? Check. Best of all, unlike most Next Great Stats, it’s not just about brains. It’s also about the heart.
That’s run expectancy by the 24 base/out states, and Leverage Index by base/out. Sobchak does all the work for you:
http://www.chancesis.com/2010/12/19/run-expectancy-and-base-out-leverage-index/
My numbers are here:
http://www.insidethebook.com/ee/index.php/site/comments/leverage_index_by_base_out_states/
I didn’t double-check anything, but it looks ok. Note that I do things a bit differently because of the partial or potential-for-partial innings.
There’s a new graphic guy in town, and his name is Josh Maciel.
Click image to see bigger:
Feb 11 16:48
Reader Mail of the Day: Why do we need X years of fielding data? And what about outliers?
Feb 11 16:10
Clutch analogy
Feb 11 15:58
MGL: Today on Clubhouse Confidential
Feb 11 11:54
Who is Jeremy Lin?
Feb 11 10:29
Dwight Evans
Feb 11 02:12
Performance through the ages
Feb 10 23:01
For Your Soul
Feb 10 21:07
Hero of the month: Brittney Baxter
Feb 10 18:32
Moneyball at Villanova
Feb 10 17:00
Psst… wanna intern in Canada?
THREADS
February 11, 2012
Clutch analogy
February 11, 2012
Who is Jeremy Lin?
February 10, 2012
Jose Molina
February 10, 2012
Reader Mail of the Day: Why do we need X years of fielding data? And what about outliers?
February 10, 2012
Performance through the ages
February 10, 2012
Hero of the month: Brittney Baxter
February 10, 2012
Win expectancy charts used in football… in 1983!
February 10, 2012
Dwight Evans
February 09, 2012
Psst… wanna intern in Canada?
February 08, 2012
Moneyball at Villanova
Recent comments
Older comments
Page 1 of 320 pages 1 2 3 > Last »Complete Archive – By Category
Complete Archive – By Date