THE BOOK--Playing The Percentages In Baseball

Chances of scoring one run, 1st and 2nd with 2 outs: .106
Chances of scoring one run, man on 2nd with 2 outs: .147

So Pedroia intentionally walking decreased the Red Sox chances of scoring by .041, but increased their chances of winning by .004.

Strange.

What’s the chance of scoring *at least* one run in the first case?  Is it more than .106, or is that what you meant?

I think the numbers in #1 are from the run frequency table and are representative of exactly 1 run.  I think the corresponding of “at least one run” are .231 (1-0.769) for 1st and 2nd and .223 for 2nd only (1-0.777).  I used http://www.tangotiger.net/RE9902score.html

Greg - Chance of scoring at least 1 run baseout state 202 is .238 2005-2009 from my empirically based RE table.  For baseout state 1202 it is .245.

You definitely need the “at least 1 run”:
http://www.tangotiger.net/RE9902score.html

My bad, I read the table wrong.

Reprinting from Fangraphs:

http://www.fangraphs.com/blogs/index.php/those-are-the-little-things-that-i-dont-think-you-can-see-in-the-box-score-ever

The win expectancy chart that I linked to is based only on the inning, score, base, out.

There is NOTHING to stop anyone from creating a win expectancy chart that includes the identity of the batters, fielders, pitchers, players in the dugout, last time a reliever was used, manager, park, the count, the time of day, the climate, and anything else you think can affect the win expectancy of a game. All you need to do is put in the time and effort to create such a system.

Win Expectancy, as a framework, is perfect. Win Expectancy, as an implementation, is limited to whatever parameters you use.

***

All Retrosheet years (I think this was 1956-2008 when I did it):
http://www.tangotiger.net/retrosheet/reports/re.htm

All Retrosheet years, win expectancy:
http://www.tangotiger.net/wins.html

The parenthetical promise was done many years back:
http://www.tangotiger.net/RE9902score.html

It’s also in The Book, which you can read for free from Amazon’s Look Inside.

Why don’t more runners do it? The problem is that if Borbon doesn’t catch the ball, then Scutaro probably scores if he is standing near second base, while he doesn’t if he is tagging up.  In analyzing the decision to tag up or not, you have to include the Win Probabilities associated with an outfield error or the runner misjudging whether the outfielder will catch it.

Right, Ken.  The numbers we’ve been batting around presuppose that the runner knows the batted ball is a flyout, but of course he can only presume that as it flies through the air, with differing degrees of certainty based on teh situation.

If we had a lot more data, we could construct a Win Expectancy table based on “runner sees a hard hit ball at approximate angle X fly over the shortstop’s head towards left-center field”, and then look at all the possibilities (ball caught + no advance, ball caught + advance to 2nd via tag-up, ball not caught + advance to 2nd, ball not caught + advance to 3rd, ball not caught + advance to home).

One of those things for the to-do list once the complete BaseballFx data system is complete and available

I think I’ve learned enough from you, Tango, to recognize that the advantages of that more complex win expectancy chart are likely to be outweighed by the disadvantages, mainly in terms of small sample size.  Bill James’s archetypal “smart” ballplayer (i.e., Joe Morgan) does instinctively work all that into the calculation, but if he starts by learning how it works in gross for thousands of games and millions of plays, he can make his adjustments from a more reasonable base beyond his own instinct.  It doesn’t take a lot of brains to know you’re not going to run on Ichiro if you have Greg Luzinski on second base, but most plays are nowhere near those endpoints.  You don’t need the data on how likely some specific middling runner and some specific middling arm will succeed, in part because it doesn’t exist and in part because it will be misleading because the sample size is too small.  You just use the basic data for nearly all cases and know enough about your own capabilities and the scouting reports on the opposing players to adjust for the more extreme cases.  That, I suspect, is a more precise and less of a luddite position than what Lowell actually said, but it’s not too far from his position.