Pankin - Walking Bonds - SABR presentation (August 17, 2003)
Warning: 30 page PDF file.
For those interested, my take was published here and there.
--posted by TangoTiger at 07:23 PM EDT
Posted 3:37 p.m.,
August 19, 2003
(#1) -
trantor
PDF is a pain, but worth going through.
Seems the more "dodgerized" the Giants lineup gets, the more incentive to walk Bonds.
OTOH, perhaps a coincidencce, but earlier this year, teams pitched to Bonds, and he hit poorly - it was only after a couple games with HR's that the walk pattern started, and his OBP starting rising.
Is this a chicken/egg situation where the more you decide to walk Bonds, the fiercer hitter he is when he does get a pitch to hit, or was he merely mortal early this year, and got hot about the same time as he was being pitched around.
thanks for posting this - I ejnjoyed it both this time, and your previous contribution!
Posted 3:47 p.m.,
August 21, 2003
(#2) -
aaron
August 20th, 2003. Aurilia on 3rd, Grissom at 1st. 1 out, bottom of the 9th, tied at 1 versus the Braves. Without a second thought, Bobby Cox walks Bonds to face Alfonso (can't remember who was pitching).
Tango, you seemed to indicate DO NOT WALK in your tables, but there was no hesitation on the part of Cox (or probably any other manager in that situation).
Interestingly, Alfonso (like Bonds) is a very tough strikeout. At any rate, he singled up the middle for the victory.
Would you really prefer to face Bonds with 1st and 3rd 1 out rather than a rather average these days Alfonso with the bases loaded? Another interesting point is that on the Grissom single, Aurilia gambled and went for third beating a bad throw by one of the outfielders. Seems like a bad time to gamble knowing that you are forcing Cox to walk Bonds no matter your success or failure.
Posted 4:12 p.m.,
August 21, 2003
(#3) -
tangotiger
(homepage)
Aurilia on 3rd, Grissom at 1st. 1 out, bottom of the 9th, tied at 1
Let's look at some win probabilities, assuming that you've got an average opponent, and you yourself are also average. (I'm guessing Smoltz was there, but I suppose we have a great pitcher for the Giants as well? Maybe not.)
Anwyay, bottom 9th, tied, 1 out and:
men on 1b/3b: .829
bases loaded: .835
uhmmm, I said "don't walk?"... let me see. According to the link above, I'm saying don't walk any time you have a runner at 1b or 3b with 1 out. Kinda strange, so let's look into it some more.
If Bonds gets a hit, game over. If Bonds gets a regular walk, he barely gains anything (.006 wins). If he gets an out, the Giants win prob goes down to .643.
So, a hit adds +.171 wins, an out drops -.186 wins, and a walk adds .006 wins.
You know, that really doesn't make any sense. The win prob cannot be .829, it must be much lower. In this case, it's very easy to figure out.
Win prob(when hit wins game) means:
freq(H) * (1 - winprob) = freq(out) * (winprob - .643)
(assuming that the walk is almost irrelevant)
That sets our winprob at .762.
This is really strange, I must have programmed something seriously wrong.
Thanks for pointing this out to me...
Posted 6:44 p.m.,
August 21, 2003
(#4) -
aaron
Actually, Gryboski was brought in for the ninth with the score at 1-1. Apparently the Braves are unfamiliar with LI *grin*. I guess with such a big division lead it is not so crucial to bring in Smoltz, but of course, one would expect the Braves to be gunning for what may be their biggest obstacle to the world series...
The day before?
4-3 Braves up to start the bottom of the 8th: Cox brings in Hodges who gives up a run before finishing the inning.
4-4 to start the bottom of the 9th: Cox leaves in Hodges who finishes the inning without giving up a run.
4-4 to start the bottom of the 10th: Cox brings in King. Bonds up first hits a homerun to win it.
As best I can tell, Smoltz had not worked since Aug 14, giving him 5 days of rest before the game I first mentioned. And as far as I know, he is not injured in any way. Seems a shame not to use him in either of these games.
Where was Smoltz??
Posted 10:58 p.m.,
August 21, 2003
(#5) -
Tangotiger
This is really strange, and I'm going to post my thoughts on the matter tomorrow.
The win probs that I have listed are correct, and I have one "sure-fire" way of doing them, and I have a second "fail-safe" way to verify them. I can't verify them for this 9th inning scenario.
As soon as aaron gave me the situation (man on 1b/3b, 1 out, tied, bottom of 9th), this was such an easy "walk now" situation (since the runner being on 1b or 2b almost virtually doesn't matter), that it really stunned me that I said "don't walk".
Anyway, I'll give more details tomorrow, and maybe one of the clever Primates can point out my flaw.
Posted 1:19 a.m.,
August 22, 2003
(#6) -
aaron
One more time for Cox...
August 22. Smoltz hasn't worked for over a week now.
Bottom of the ninth, tied at 3. Cox brings in Cunnane to face the 9,1, and 2 batters. 3 up, 3 down
Bottom of the tenth, tied at 3. Cox brings in Hodges to face Grissom and then Bonds. With 1 out, Bonds hits it out to win the game.
Oh where, oh where was Smoltz??
Posted 10:09 a.m.,
August 22, 2003
(#7) -
tangotiger
Let's figure out how to calculate run expectancy (RE). Given the following:
- safe play occurs 33% of the time
- RE AFTER a safe play is 1.10 runs
- RE AFTER an out play is 0.30 runs
What is the RE BEFORE this PA?
RE = .333 x 1.10 + .667 x .30 = .5667
So, we can also say:
- LWTS value of safe play = 1.10 - .5667 = +.533
- LWTS value of out play = .30 - .5667 = -.2667
And
.333 x .533 = .667 x .2667
With me?
Ok, now let's try to do this with Win Expectancy.
What's the chance of scoring at least one run (and thus winning the game), with a man on 1b, 3b, 1 out, bottom 9th, tie game, assuming lg average opponents?
You have a 66% chance of scoring your run IN THIS INNING, and winning the game. The other 34% chances you get into Extra Innings, and you have a 50/50 shot at winning.
So,
WE (bottom 9th, tied, man on 1b/3b, 1 out) = .66 + .34x.50 = .830
Doing the same thing
WE (bottom 9th, tied, man on 1b/3b, 2 outs) = .28 + .72x.50 = .640
Ok, so those are our known true WE.
Now, doing the same as we started with RE:
- 26% chance of hit or RBOE = WE of 1.00
- 9% chance of walk = WE of .840
- 65% chance of out = WE of .640 (maybe less cause of DP)
So,
WE = .26 x 1.00 + .09 x .840 + .65 x .640 = .752
But, we expected .830
Can someone figure out where I'm going wrong?
Posted 10:12 a.m.,
August 22, 2003
(#8) -
tangotiger
After I posted that, it finally hit me: sac flies! The WE can actually increase substantially, following an out, and my Bonds program looks like it did not consider the SF properly.
Posted 6:38 p.m.,
August 22, 2003
(#9) -
aaron
Tango
Not sure if anyone else is reading this but...
I had assumed you were including runs on sacrifices and ground outs of course as I alluded to with my statement about Alfonso (along with Bonds) being tough to strike out. Perhaps in other game situations these events are negligible, but in the late innings of close games they are crucial!
Hope that solves the problem.
Posted 6:57 p.m.,
August 22, 2003
(#10) -
Tangotiger
Yup, the win prob tables I published did consider the SF and grounders scoring the runners, etc.
But, my walk/don't walk Bonds did not balance that properly. Kinda embarrassing really. It's easy enough to fix, maybe 1 or 2 hours of work. Not sure when I'll do it though, but I'll target the 1st game of the playoffs.
Posted 7:43 p.m.,
August 23, 2003
(#11) -
Danny
Tango,
I bit off topic, but if you get a chance I'd love to see the LI of relievers this year. Comparing Smoltz/Gagne/Foulke would be very interesting. Thanks.
Posted 10:56 p.m.,
August 23, 2003
(#12) -
Tangotiger
Sorry, but I need the pbp event files to do that. I'll only be able to look at this in the off-season.