from Amazon
Monday, April 07, 2008
IBB Redux
Pretty much everything is a redux these days.
I was watching the Tigers, White Sox game tonight and with the Tigers down a couple of runs in the 6th, I think, they issued an IBB to AJ what’s his name to load the bases with 1 out. In The Book:
We suggest that it is generally not a good idea to load the bases with an IBB, even with 1 out. Obviously the decision is a complex one. To take a page out of the crazy BTF thread about the Justice post, I’ll say again that there is no human being on earth (unless there is some savant that can do it) that can figure out if that was the correct move or not, let alone a baseball manager. If the correct answer were a matter of life and death, he (Leyland) would have to hire a number cruncher to figure it out, and even then, hope that the number cruncher got it right, given all the variables that go into the equation. Of course, if the person offering the “life and death test” (let’s say GOD, because only he knows the 100% correct answer) was at least a little bit generous, he would offer Leyland and the number cruncher some “slop.” If the decision were “don’t walk him”, but by a hair, and the number cruncher says “walk him, but it is close,” then Leyland lives.
If the number cruncher says, “Walk him, and it ain’t even close,” then Leyland bites the dust. If God tells us that the correct answer is “walk him and it’s not close,” and the number cruncher says, “Don’t walk him,” the Tigers’ skipper gets axed (literally). Get it?
Obviously a guess (or whatever you want to call what a manager does when he makes these decisions) is going to be “right” a good portion of the time, but that is obviously not what I am talking about. I am talking about whether the manager, in any close situation (I am assuming that this and similar situations are close) like this, can KNOW that he is reducing the other team’s WE by walking the batter (whether he knows what that means or not). How would he do/know that? NO close and complex decisions can be figured out by intuition or experience. That just doesn’t make any sense.
Anyway, I thought I would look at the IBB again, with runners on 2nd and 3rd and 1 out. Since after the IBB, lots of things affect the RE and WE, we can take a shortcut and include all of those things automatically by looking at actually how many runs were scored after such an IBB. I did that for 2003-2007. I looked at all IBB’s that loaded the bases with 1 out, with a non-pitcher coming up and any inning less than the 9th.
In 5 years, there were 1007 such IBB’s, or around 6.7 per team per year, which is a lot, I think. 1.652 runs were scored after the IBB, on the average. One standard deviation is 2.14 runs, which amounts to .067 runs per instance for N=1007. Even though the distribution is not normal, I don’t think, we can say that the 1.652 is around 1.52 to 1.79 with a 95% confidence interval. That’s the last time I’ll talk about sample error. From now on, I’ll just assume that the 1.652 runs is the “real” number after an IBB in that situation.
Now, all we have to do is to compare that to how many runs we think would score if we let all those batters hit away with runner on 2nd and 3rd. What I will do is use those batters’ collective 3 year batting stats to determine their “real” batting line, and then I will tweak those numbers based on how ALL batters change their stats with runners on 2nd and 3rd with 1 out, when they are allowed to hit of course. I may or may not adjust for the pitchers on the mound when the IBB’s occurred (depending on how lazy or not I am). I have no reason to think that the pitchers on the mound are not a group of average pitchers. I am not including 9th and later innings, so closers tend not to be on the mound.
In general, when the IBB was issued, it tended to be in the later innings, but there were plenty of mid-inning IBB’s, and not a lot of early ones. Also, the score tended to be tied, or the issuing team either up by 1 or 2 runs, or simply behind in the game, as in tonight’s Tigers game.
Of course, you want to look at WE and not RE, but for now, I am going to compare the observed (empirical) RE, which was 1.652 runs with what we would expect if those exact batters would be allowed to swing away. I will first infer their collective stat lines and then I will use a Marov model to figure out how many runs would score to the end of the inning, given 1 out, and runners on 2ndt and 3rd, and then compare that to the 1.652 runs that DID score.
Any guesses as to the result? Any suggestions for changing the model?
As Dial would say, this has already been done before, I am sure, so my disclaimer here and in everything else I do from now on, is that I automatically give full citation to everyone else’s work before me, and I fully admit that I am probably just “tweaking” the work that has already been done. If you have not read the BTW thread, I apologize for being obtuse with that comment.
BTW, I constantly hear criticism about these kinds of models and analyses, along the lines of, “You don’t know all the variables, so we can’t know if your numbers are correct.”
I’ve said this a million times before and I’ll say it again. What we (I) do is to create a model that is pretty good, and then come up with an answer. If the answer is clear cut, regardless of those other variables, then we don’t care about them. If it is not clear cut, then we can say/do one of two things: One, try and include more variables in the model and come up with an answer that has more certainty, or simply say, “It is too close to call, either go with my answer (knowing that the certainty is not great) or go ahead and include those other variables “by the seat of your pants” and then make your decision.
What happens is that most of the time we include the variables that have the most effect on the model, such that if we find a clear cut answer, we don’t care much about the other variables. If that is not the case, that either we come up with an answer that is NOT clear cut, or there are variables that are or may be important, but for some reason we cannot model accurately (which is rare), then so be it.
Remember that good sabermetrics is NOT the search for an answer, per se. It is the search for the truth. Sometimes the truth is, “We don’t know.” Often, it is, “We think it is this, but we are only 80% (or 90%, or 55%) certain.