THE BOOK--Playing The Percentages In Baseball

"Ackbar”.  I love it, brings back old primer memories.

“He’s running into a trap!”

The BE point really changes with the profile of the batter (and pitcher of course), specifically whether he is a singles or power hitter and his walk rate (a steal has no effect is the batter is going to walk, right?), and not with his “overall” quality.

Also, I would think that you almost MUST use different WE’s for lefty and righty batters, if nothing else, since a lefty batter has a larger advantage with that runner being held on first.

Game theory for basestealing only involves NOT stealing sometimes when your chance of stealing is above the BE point (otherwise the defense would just pitch out every time).  I can’t really think of any other game theory involved in basestealing.

I have a great project if anyone wants to take it on.  If not, I’ll do it myself one of these days.  Here it is:

For every basestealing opportunity, we can figure out the BE point (even approximately) based on the WE of the game. No problem there.  We can also figure out the average CS rate of the average basestealer in any given situation, based on the catcher’s and pitcher’s historical CS rate (regressed and context-adjusted hopefully).  We can even adjust for park (artificial turf is much easier to steal on).

Now we can go through the retrosheet PBP database for any time period, and do two things:  For every single basestealing opp, one, see how much extra WE we could pick up if the runner stole if his personal (against that pitcher and catcher) SB rate is higher than the BE.  Two, see if he attempted a steal or not.  If he did and his personal rate were higher than the BE rate, we give him and his team (manager) no credit or debit.  If his personal rate is lower than the BE point (he should not have been stealing), then we dock him credit equal to how much WE he costs his team by attempting the steal (easy to estimate).  If he does not go and he should, then he gets docked.  If he does not go and he shouldn’t, then he gets no credit or debit.

After all is said and done, we can ascertain several things:  One, how much each player gains or costs his team in WE based on the frquency of his SB attempts (and our estimate of his true SB talent of course).  Same thing for teams and managers.  Two, how much a typical team costs themselves in runs by not using an optimal SB attempt strategy.

A few problems that would have to be addressed:  Game theory.  Even if he “should” go, a player cannot go every time, otherwise the defense would pitch out every time they think the player “should” go (assuming they could figure that out).  Sometimes a player wants to attempt a steal, but the batter ends up putting the ball in play or ending the inning before he attempts that steal.  Again, because of game theory considerations, the runner cannot attempt the steal on the first pitch every time even if he should go.  Even if he did, the batter would sometimes put the ball in play on the first pitch and you certainly don’t want your batter taking a pitch just to let he runner steal (as opposed to CW, right?).

Speaking of the runner going and the batter putting the ball in play, this is a “hidden” benefit of the steal which I have never seen addressed, which may subsntantially reduce the BE point of most if not all steal attempts.  Unless the advanatge of the being on the move is cancelled by the distraction to the batter on that pitch - I don’t know.

Finally, I thought that Black Hawk and Rally were the same person.  I wish that people would use their real names or at least one AKA and stick to it.

I thought that Black Hawk and Rally were the same person

Angels fans, eh? They all look the same!

I think Rally is Chone, but that both of these guys are Angels fans?  And Shredder is in the mix too I think.  I agree… just one darn handle.  I’ve been Tango (or Tangotiger) since the beginning of my online days (yeesh, over 10 years already).

Yup, I’m also Chone.

I was Rally Monkey for awhile on BTF, then Anaheim Rally Monkey of Maryland, shortened to AROM.  At least I try and keep Rally in there, sorry for the confusion.

I understand the game theory aspect of it, but in this case Aybar is just too terrible of a percentage basestealer for him to run in that situation.

From a purely subjective point, I’m happy that Aybar got thrown out against the Indians.  Go Tribe!

On a couple of points that MGL made: We can model a runner’s chances of “making it” based on a number of factors using logit regression (and entering in batter, pitcher, catcher, and runner characteristics as well as game-state information).  Get a few years worth of data together and it could probably produce a pretty good equation.  I’m also confused why you wouldn’t credit the manager based on sending the runner when the percentages said it would be a good idea to do so.  If we think that a runner has an 80% of making it and BE is 70%, then isn’t he costing his team by not going?  One other is issue is the “hidden” steal attempt where a runner goes, but the batter fouls the pitch off.

The idea of whether the runner going actually opens a hole for the hitter is something that interests me.  I’m putting the finishing touches on my third (and last) article on throwing to first.  Preliminarily I’ve found a “hole” effect, but I need to do a bit more work to see how big it is.

And Sean, pick an alias already!

Pizza Cutter (whoever that is), I am already docking a player/manager when he does not go and he should, so if you give him “credit” when he goes and should go, you are double counting.

That being said, I have not put a whole lot of thought into the “accounting system.” My original thought was to have an accounting system that represented runs above or below the optimal strategy, which would mean only to credit or dock when the player/manager dis what he was NOT supposed to do.  You could also use an accounting system that came up with a number that was above or below what the average team does I suppose which would require different rules for crediting.

I have looked into the hidden benefits of having the runner(s) in motion and they are substantial.  Retrosheet has what looks to be pretty complete data for 2004 and 2006 for runners in motion.  For those two years there were 3540 runners in motion when there was not a full count with 2 outs.  The increase in run value added over runners not in motion was .0443 runs per play or 155 extra runs.  This brings the break even point for steals to about 59%.  Figured with WPA the break even point is about 60.6%.

Wow, that is a lot!  Can you give us the breakdown of what happens when a player is in motion and the ball is in play, such as s,d,t,hr,outs, dp, etc.?  I assume that some of these are hit and runs which would mean that the batter was forced to swing at any pitcher which would reduce the positive outcome of the ball in play not counting the runner being in motion.  IOW, if you get .0043 extra runs per play on all “in motions” the breakdown between hit and runs and straight steals would likely be “more than that” for straight steals and “less than that” for hit and runs, unless of course, the batter’s attempt to hit the ball in the hole during a hit and run (if that is even substnatially possible) more than makes up for the fact that he has to swing at a lot of bad pitchers.  Is there any way to try and distinguish bewtween a hit and run and straight steal?  For one thing all the hit and runs will be on a hitter’s count, and never with 2 strikes, of course the outcome of a ball in play on a straight steal with 2 strikes is also not going to be that good simply because the batters has 2 strikes.

There would be a number of different possibilities of why runners are in motion.  There are hit and runs were the batter is required to swing, run and hits where the batter can chose to swing, runner running on his own which would be functionally equivalent to a declared run and hit, straight steals where the batter swings to miss to protect the runner, and straight steals where the batter is not required to swing.  Almost impossible to separate between these choices. 

I haven’t broken down all the results, as I was concentrating on the ones that I was presuming would provide most of the extra runs.  Triples and home runs when the runners are going would not provide any extra runs unless you found that they occured more often when the runners were going, and then you would have to show that that result was not a sample size abberation.  But a single when the runners are going results in the runner on 1st ending up on third or scoring 702 times on 931 singles.  Of 1113 ground outs only 120 resulted in double plays while about 43% of the ground outs in the same base out situations when the runners aren’t going result in double plays.  These are the two big gainers.  Line drive outs cost some runs, of course, due to double plays plus a few more on fly ball outs.  As you noted some more runs are lost due to a lower BABIP because of being forced to swing at whatever the pitcher threw.

I have expressed my opposition to evaluating strategic decisions using run expectancy or WPA many times.  The reality of strategic decisions is that they are based on more factors than can be included in any analysis short of a full blown game simulation.  The same is true when looking at runners moving.  There are many factors that I have not tried to include in this initial look.  All I really wanted to show was how far off traditional calculations of break even points on stealing probably are.

You certainly have to look at home runs and triples (and everything else including K’s) if you want to see the effect of the runner going regardless of whether it was a hit and run or straight steal!  Otherwise you will overvalue the runner being motion since I assume that triples and HR rates will go down when sometimes the batter is forced to swing at bad pitches (in hit and runs) and sometimes he swings to “protect the runner” (although I don’t even know what that means - “protect the runner").

BTW, there really is not such thing as a run and hit anymore.  There really are only two situations where the runner is in motion.  One is the manager-called hit and run where the runner runs (often times he does not get a good jump and often times it is not a particularly fast or good baseruner) and the other is when the runner attempts a steal (either on his own or “suggested” by the manager).  When the runner steals on his own, the batter can do anything he wants.  In our research for The Book, the data suggested that a veteran batter pretty much ignores the runner when he is a basestealer while the young batter either easily gets distracted by a basestealer on second or he takes a lot of pitches in order to “let” the runner steal.  Of course taking pitches that you might otherwise swing at is a stupid strategy unless the runner has like a 90% success rate or greater I would think.  If the runner has a success rate around or a little higher than the BE point, then if the batter deliberately takes some pitches that he might otherwise swing at, overall the runner/batter is going to reduce their WE by attempting a steal.

And BTW, here is my opinion on the “we can’t model everything that happens in a real life game” thing:  Some things we can model more than others - for example, the IBB.  Other things are a little more complicated to model, but with some effort can be done reasonably well - like the sac bunt.  I don’t think there are any things that we can’t model substantially. However, once we recognize how much of a particular situation we can model, we can still make better strategy decisions and here is why:  If we think we can model something very close to what happends in “real life” than we can be confident that what the computer tells us to do is the right thing.  (Sure we may make a few mistakes in the grey area, but by definition those won’t cost us too much).  If we think we can’t model something that well, we can still come up with rules of thunmb which will be accurate as long as we make sure that we are in the “safe zone.” IOW, using our models that are not that precise, we simply make conservative recommendations.  For example, even though it is difficult to precisely model sac bunt strategy, we can safely say that if we think that it is close either way (whether to attempt a bunt or not), if the batter is a bad bunter and slow, a bunt is mostly likely a bad idea.  If the batter is a good bunt and fast, you cannot go wrong with a sac bunt attempt.  And then again, if you mix up your strategy you gain even more from game theory considerations.

"protect the runner”

haven’t heard this for a while, but used to be the case that a batter might swing and miss deliberately on the theory that it would distract the catcher.

All events including HRs, 3Bs, and Ks were included in the numbers I gave you in post #8.  I just didn’t break them out separately as I did for taking extra bases on a single, or double plays on a ground out.

Finally, I thought that Black Hawk and Rally were the same person.  I wish that people would use their real names or at least one AKA and stick to it.

Yeah, that happens a lot.

I use the same handle everywhere.  Rally is sometimes Chone or AROM.

I don’t use my real name because my job is not “troll around sabermetric websites all day”.

I don’t use my real name because my job is not “troll around sabermetric websites all day”.

That’s a great way to answer.  I’ve been asked many times to reveal my identity, as if it’s even important to the other person.  I would simply decline citing privacy reasons, and still I get heckled about it.  There was even some SABR executive on the baseball-databank group who insisted that if I did not reveal myself, that he wouldn’t correspond with me!  Out of the thousands of electronic correspondances I’ve had in my internet years, he’s pretty much the last guy I’d reveal myself to.  Suffice to say, I’m not joining SABR any time soon.

Anyway, we all have a right to privacy.  But for continuity’s sake, use the same alias/identity in your circle of subjects.  So, any sports postings I make always carries Tango-something.

Some of us use aliases for professional reasons.  I suppose if you dug around enough, you could find my real name, but my day job is as a clinical psychologist.  We get more than the average amount of stalkers, so I need to protect myself.  (I actually had a bit of an argument with the folks at MVN for them to allow me to use Pizza Cutter as my display name.) Besides, Pizza Cutter is so much cooler than my real name.

Sure, I didn’t mean to suggest that anyone use their real name if they don’t want to.  Only that they somehow make it easier for people to discern who they are.  Not that they have to do that either.  After all, this is a free country.  People can do whatever they want as long as they don’t overtly hurt anyone else - you know, gamble on the internet, deposit or withdraw more than $10,000 in the bank, go to the doctor and not have to give their SS#, etc.  Wait a minute, I must have the U.S. mixed up with some other country…

MGL, now that you’ve learned we’re two different people, the easiest way to differentiate Rally and myself is to remember that Rally is the smart one.

BTW, there really is not such thing as a run and hit anymore.

Interestingly, the Aybar play that started this discussion was a run-and-hit; Kendrick chose not to swing.  (Which may raise another question:  is there a point at which a hitter should swing at a bad pitch in order to protect a bad basestealer [by either putting the ball in play at a likely disadvantage or fouling it off for a strike]?)

Of course, what the Angels post facto are describing as a “run-and-hit” may well fit under:

When the runner steals on his own, the batter can do anything he wants.

And if so, as Kendrick is a young batter, it’s certainly possible (though not definite) that:

the young batter ... takes a lot of pitches in order to “let” the runner steal.

came into play.

I suspect the hit-and-run is particularly difficult to model as a large part of its success will depend on what kind of pitch the batter is forced to swing at.  We just don’t have the data ("Pitcher A will throw a changeup on the outside corner 65% of the time to RHB on a 2-1 count with no outs and Batter B has a .397 BABIP on such pitches” or something like that) to individualize different combinations of pitchers, batters, and baserunners.