Baby Winexp on Games 1 and 2 (with Toy Surprise inside)
by Apricot on Oct 5, 2006 4:37 PM PDT
Email
PrintFirst as a special reward to people who read these Winexp diaries...
I am pumped up to announce that I have gotten a killfile script developed for the Firefox browser. It allows you to hide all comments by people that annoy you on AN (and in fact any SB Nation blog). Every person commenting now has a kill (or unkill) link attached. So I could hide all of Sal's posts for instance with a click and unblock later. This only hides them from you, not anyone else. There are certain people in the game threads that I will be adding immediately to my killfile. Freedom... freedom!
In short, install Firefox 1.5, install Greasemonkey, and then install this Greasemonkey script written by Daniel Martin, customized very very quickly at my request. You might have to restart your browser.
(All I know is that it is working great for me. Actually, I use Safari, but I'm going to start using Firefox for AN specifically for the killfile.)
Okay, on to the games. Baby W has already made her commentary on Game 1, so go there if you missed it.
Game Two
In some ways Baby W has very little to tell us about the game. (What is this? Search AN for "winexp" for background.) Only by watching the game can you see that the Kotsay homer was at least partly if not fully due to Torii, and the Kotsay double play was fully due to Morneau. We could manually change the credit, but I prefer to leave human judgment out of it as much as possible. In the same way that Kotsay officially gets a HR and GIDP, you know what went into that judgment and to take it with a grain of salt... e.g. an outfielder never gets an error without touching the ball and a GIDP could be a weak grounder or a blast to the wrong glove. Reyes gets the 2 ER, Kotsay gets the HR. All stats have their limits and knowing where they are is important.
Having said that, Baby W says the heroes of the game are Kotsay, Duke and Morneau. Most interesting to me is how harsh Baby W is to Loaiza. He pitched an excellent game, I thought. But in her mind, pitching with a 2 run lead starting in the bottom of the 6th, you should win 3 out of 4 games (~75%). To let them tie it right away drops your chances to about 40%, which is a huge swing. So... no cookie for Loaiza, whom she deems approximately neutral.
Second interesting thing to me is the icy calm she feels in situations that terrify me (and AN). For instance, Huston in the bottom of the 9th, 2 out, up 3, runners on 1st and 2nd, tying run at bat. How often do you think the home team pulls out the win? Enough to be having a heart attack? (My answer: yes.) Well according to history, the home team wins 39 times in 1000 (~4%). The mathematical model commonly used for winexp says the home team wins 43 times in 1000. In fact, go back to right after Kotsay's homer. Up two, top of the 7th, 2 outs, your chances of willing are around 78% - 80%. And yet it didn't feel in the bag at all.
It would be interesting to isolate the historical data for playoff teams, but I don't think it would change much. Both sides are going to get better by selection, so the odds will probably stay similar.
TOP PLAYS
2006-10-04. Top 7.2. R: 0. B:1. P: Dennys Reyes. Mark Kotsay in the park home run to center, Kendall scored (0.776, +0.329)
2006-10-04. Bot 6.0. R:-1. B:. P: Esteban Loaiza. Justin Morneau homered to right (0.585, +0.176)
2006-10-04. Bot 6.0. R:-2. B:. P: Esteban Loaiza. Michael Cuddyer homered to left (0.409, +0.136)
2006-10-04. Top 5.0. R: 0. B:2. P: Boof Bonser. Marco Scutaro doubled to right, Swisher scored (0.693, +0.122)
2006-10-04. Top 5.0. R: 0. B:. P: Boof Bonser. Nick Swisher doubled to right (0.571, +0.091)
2006-10-04. Top 3.1. R: 0. B:1. P: Boof Bonser. Mark Kotsay grounded into double play first to shortstop, Kendall out at second (0.425, -0.061)
2006-10-04. Top 5.1. R:+1. B:3. P: Boof Bonser. Jason Kendall singled to left, Scutaro scored (0.741, +0.060)
Name WXC Events
MIN Justin_Morneau +0.125 4
MIN Michael_Cuddyer +0.085 4
MIN Pat_Neshek* +0.043 3
MIN Juan_Rincon* +0.011 5
MIN Boof_Bonser* -0.002 25
MIN Joe_Nathan* -0.012 4
MIN Jason_Tyner -0.023 5
MIN Luis_Castillo -0.033 5
MIN Torii_Hunter -0.051 4
MIN Nick_Punto -0.069 5
MIN Jason_Bartlett -0.086 5
MIN Joe_Mauer -0.093 4
MIN Rondell_White -0.115 4
MIN Dennys_Reyes* -0.320 2
Name WXC Events
OAK Mark_Kotsay +0.228 5
OAK J_Duchscherer* +0.185 6
OAK Mark_Ellis +0.080 4
OAK Kiko_Calero* +0.075 5
OAK Frank_Thomas +0.069 4
OAK Nick_Swisher +0.067 5
OAK Huston_Street* +0.047 6
OAK Marco_Scutaro +0.038 4
OAK Jay_Payton -0.008 4
OAK Jason_Kendall -0.030 5
OAK Esteban_Loaiza* -0.047 23
OAK Milton_Bradley -0.066 4
OAK Eric_Chavez -0.098 4
34 comments
|
0 recs |
Do you like this story?
Comments
what about a final regular season winexp diary?
[kill];[hide comment]
heartattackworthiness
What winexp doesn't capture here is the range of possible outcomes of a play. The reason you were so tense about that plate appearance was that the result could have been a huge drop in win probability, from nearly 1.0 to well below 0.5.
Consider the case of the starting pitcher getting the first batter out. That result in a small increase in winexp. The worst that could happen, giving up a leadoff home run, results in a small decrease. So you are not that excited by the first out, because the worst possible outcome wasn't that bad anyway.
In the case of the closer getting the last out with the tying run at the plate, the winexp only goes up a little bit, but you know it COULD have gone down a lot.
When Street got the final out you weren't excited because of what he did, but because of what he didn't do.
So how might we measure the "heartattackworthiness" of an event? Simple - by measuring the difference between the worst possible outcome and the best possible. Then you could rank the most heartattackworthy. I think you'd find find Street involved in a lot of them.
by matthias on Oct 5, 2006 5:20 PM PDT reply actions
Good point
I think Hunter's play
In the same way, the Dodgers yesterday probably made the biggest running gaffe on the bases this season, as well.
Quick: Most significant baserunning error in 2003. Eric Byrnes. Close second: Miguel Tejada.
[hide comment]
Wait
i didn't really block it
Yeah, I figured it out below
Go A's. Go Tigers.
question
Oh, now I get it
Byrnes play sidenote: Worst I've ever sworn in front of my kids--no close second. It's legend in our house.
what is this weird blank comment?
Clicking on kill
Unfortunatly, kill files lead to . .
I want a kill file that removes all comments about kill files.
I used to think I wanted a killfile
Kinda off topic
I like websites that have that added little feature. For instance, yahoo has the "Y!", and Google has the boxed in G, and MSNBC has the peacock, etc..
it's called a favicon
Blez
by sf drift king on Oct 6, 2006 9:38 AM PDT up reply actions
More thoughts on closer winexp
To see this imagine a simplified baseball world in which closers alway pitch in exactly one situation: bottom of 9th, one run lead, one runner on, two outs. And the outcome is always either an out to save the game, or a home run to lose it. And imagine the league average is to get the save 90% of the time and blow it 10% of the time. Then winexp will show a .9 chance of winning this situation, so a save will be worth +.1, and a loss will be -.9. If Street matches the league average, he'll have one -.9 for ever nine +.1s, and end the season with zero.
And if a mop-up guy - call him Sweetbeck - only pitches the ninth inning of games with a more than five run lead, and has average performance, he will make virtually no difference to the outcomes and also end up with a zero.
So... does this mean Street has contributed no more to his team than Sweetbeck? No. He has produced league average performance in extremely critical situtations where the average is set by the best players in the game (Rivera, Nathan etc.) who routinely play in that situation.
So whatever winexp contribution is measuring here, it's not measuring value to the team. It's measuring something more like VORP, where you have to think very hard about what "replacement player" would mean.
More thought later if time permits...
by matthias on Oct 6, 2006 8:22 AM PDT reply actions
All true, but
by green star oakland on Oct 6, 2006 8:50 AM PDT up reply actions
yes, and that's what I'm doing...
by matthias on Oct 6, 2006 9:26 AM PDT up reply actions
what 0 WXC means is interesting
So I interpret 2*WXC as "games over .500" contributed. So having a WXC of 0.000 could be pretty reasonable if you've played a lot of games. If you haven't then it's not meaningful. The same way a 0.75 ERA is incredible if you've pitched all year and not so much after a couple of short appearances.
There is also a whole other issue about how many significant digits one should attach to WXC. I personally think it's accurate to about +/- .5, so a possible better way to report the results would be to report 2*WXC and round off, giving a "Games Over .500 Contributed" stat. However, baseball has trained me to love 3 sig digs (BA, ERA, etc), so I report WXC to 3 decimal places.
settings a baseline and including leverage
So the first step is to get the baseline right. I sit here typing comments on AN, and each comment contributes exactly zero to the A's probability of winning their games. My winexp contribution for the season currently is 0.00000000, which is better than many of the A's players. Yet the A's organization stubbornly refuses to pay me even the major league minimum for my league-average performance.
Why don't they pay me? Here's the thing: winexp says they start each game with a 50% chance of winning. But that's only true if they show up and play, and play decently. When the ump says "play ball!" they actually have to put some guys out on the field who are capable of playing the game at a professional level. It costs good money to find someone to take the first at-bat who will on average not reduce their chance of winning. If they sent me up to the plate, I would be a guaranteed strikeout. If they put nine guys like me in the lineup, they would have zero chance of winning and might as well forefit.
The winexp based value metric should produce zero for a player who always produces the worst possible result in every situation. Such a player is no better than me; a team made of such players would do no better playing than forefiting. So there's a start - getting zero right is always a great feature of any metric.
Figuring how much to add for non-pessimal outcomes is trickier. For each situation a player faces there's a worst possible outcome, a best possible outcome, and a range of outcomes in between including a mean outcome. We know to assign zero if the play achieves the worst outcome. What about better outcomes? Baby winexp currently assigns the same value to mean outcome for every situation. One approach would be to preserve that property and assign an arbitrary positive number (such as 1.0) to achieving the league average result in every situation.
But there's a problem with that rule in that it conflicts with the first. The average result can be arbitrarily close to the minimum result, so if you try to do a linear function from 0 at min to 1 at mean you could get a near infinite slope with a near divide by zero.
So the amount of credit you five for each play has to very with the range of possible outcomes. In other words, you have to include leverage. Here's rule 2: for zero-leverage situations, you must give zero credit. And generally, the more leverage, the more possible credit you should have (we'll have to formalize that).
More later. I have to squeeze some real work in before the game starts :-).
by matthias on Oct 6, 2006 12:34 PM PDT up reply actions
Ah, so much for work.
For any situation let Wmin be the worst possible change in winexp that can result. Let P(w): [-1,+1] => [0,1] be a function giving the probability of an a winexp change of w or better.
For each event, resulting in a winexp change of w, credit the player with a wvm contribution of (w-Wmin) * (1 - P(w)).
This function has all the properties I want as described above. A player who produces the worst possible outcome always gets zero credit; as the outcome improves the credit increases. And an outcome that was sure to happen no matter what gets no credit either; the more unlikely the (positive) result the more credit.
by matthias on Oct 6, 2006 12:56 PM PDT up reply actions
Absolutely
My train of though was off in the direction of the issues of interpretation, and how the anti-stats folks (certainly not intended to include you) often mis- or over-interpret numbers to show that they are flawed. Rather than "lies, damn lies and statistics" there are "liars, damn liars, and pundits".
by green star oakland on Oct 6, 2006 9:51 AM PDT up reply actions
Camino question
by green star oakland on Oct 6, 2006 8:50 AM PDT reply actions
I don't know
1 question..
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
Certain photos copyright © 2012 by Associated Press or Getty Images. Any commercial use or distribution without the express written consent of Associated Press and Getty Images is strictly prohibited.
