Game State Matrix (October 13, 2003)
Empirical data on how often a team wins at the end of each inning. This work is similar to Phil Birnbaum's, except Phil's data covers many more years and all the base/out states as well. However, the presentation in this link is rather interesting.
--posted by TangoTiger at 06:08 PM EDT
Posted 6:16 p.m.,
October 13, 2003
(#1) -
studes
(homepage)
Cool graph. I'd like to know what software he used to create it.
However, there are enough weird things going on in this graph to make me question some of the data points. Little "splotches" of red or light blue seem to crop up in backgrounds of yellow or fuschia (really, what are some of those colors?), when I would think the distribution of each probability would follow a fairly normal pattern.
Posted 6:21 p.m.,
October 13, 2003
(#2) -
tangotiger
It was based on empirical data, and even though it's 2000 games, it's still ONLY 2000 games. Believe me, you need far far more than that, which is why a simulator or a math model would be better, IF those models can match reality in terms of better pitchers coming into the game in close games, etc.
If you want studes, take Phil Birnbaum's data from several weeks ago, and reconstruct this graph with Phil's data.
As for the software used, it's mentioned elsewhere on Rhoid's site.
Posted 6:28 p.m.,
October 13, 2003
(#3) -
tangotiger
(homepage)
Studes,
The software was developed by the person in charge of the baseball Rhoids' site. You can check out the above link.
Posted 8:48 p.m.,
October 14, 2003
(#4) -
David Smyth
This is a much-discussed topic, but the problem I have with this kind of stuff is that what is an independent situation or opportunity for an individual is not the same as such for a team. So, for example, if a reliever has a 1.8 leverage index, that's fine to evaluate his win probability added----but, "who created" that 1.8 boost, and who should deserve credit for it? Should it be the reliever 100%? Should it be the reliver and the manager? Should it be the reliever, the manager, and the starters in his games who got it to that point?
Yes, runs at certain times have more importance, but the attribution of those runs is not as straightforward as some seem to be portraying...
Posted 9:14 p.m.,
October 14, 2003
(#5) -
Patriot
I agree with that. One thing that has always bothered me about BA w/ RISP and the like...somebody on that team did something to get into scoring position. You create more runs by doing better w/ RISP then with nobody on base, but if you perform poorly with nobody on, you're denying your teammates opportunities.
Posted 12:47 a.m.,
October 15, 2003
(#6) -
Tangotiger
If it's the bottom of the 9th, tie game, bases empty 0 outs, say that gives the home team a .65 chance of winning. If he hits a triple, say that gives the team a chance of winning of .95. So, that triple with bases empty add .30 wins. The next batter, if he drives in that runner will add only .05 wins.
To answer the question being put forth: it's all taken care of. The win probability is based on assuming a typical random distribution of FUTURE events to establish the CURRENT win probability.
So, to start the 9th, we *expect* 30% of the time the team to score. And the other 70% of the time? Well, it goes into extra innings, and you get (an almost) 50-50 chance of winning. .30 + .70/2 = .65
You get the triple with 0 outs, and you have a nearly 90% chance of scoring. Why? Because you put yourself on 3B, and you *expect* the next 3 batters to perform at league average, and therefore, the batter-runner will reap the large majority of the share of getting on base.
On the other hand, if it was 2 outs, and the batter walked, he'd add very little value to his team winning. The batter that drives him in would get the lion's share of the credit.
Posted 1:01 p.m.,
October 15, 2003
(#7) -
studes
(homepage)
I personally like the leveraged index for pitchers very much, because we know that different innings have varying levels of importance, and this seems like a "straightforward" (at least, conceptually) way to quantify that. But we do need to be clear that it is a value stat only (just like ARP) and does reflect managerial choices (as well as other factors, such as whether the team was involved in a lot of close games compared to the league average).
I also like it for batters, but only in a system that assigns run creation to batters, such as Tango's example.
I hadn't thought of this, but can we use leveraged indices to quantify the effectiveness of a manager?
BTW, Tango, thanks for the links.