Friday, January 03, 2014
IBB Bonds
?An interesting article that missed one important item, best captured here:
Looking at the all-time IBB leaders, their average WPA per IBB cluster around that number: .0107 for Hank Aaron, .0100 for Willie McCovey, .0102 for Manny Ramirez, .0106 for Ken Griffey, .0101 for Prince Fielder, .0105 for Albert Pujols, .0111 for Miguel Cabrera. Vladimir Guerrero is the far outlier, at .0120 per walk. Bonds, at his peak, nearly doubled that.
WPA assumes "average" future conditions, and average present conditions. A great hitter, like Aaron, Junior, Pujols, etc, are worth, on average, about +.07 runs per PA. But in IBB situations (runner on base, 1B open), it's probably more like +.10 runs per PA, which is +.01 wins per PA. And, lookie there, the WPA of their IBB was +.01 wins per PA. This is why we call IBB win-neutral: it's not that their win value is 0, but that their win value relative to the other options (pitching to a great batter) is win-neutral.
So, when the author states that Bonds doubled their numbers at his peak, well, I think that's fine. After all, at his peak, 2001-2004, generated +.17 runs per PA in all situations, which means that in IBB situations, he would have otherwise generated +.23 runs per PA, or +.023 wins per PA. That he was issued IBB below this level, on average, likely shows that many, most, of his IBB were good IBB.
But what about that bases loaded one? I talked about it here.
I have not RTFA yet, but Tango’s “IBB Bonds” chart is something that I reference all the time (at least the concept) in regards to exactly what he is talking about. It is very important. I use it all the time in my analyses of various in-game decisions and strategies.
Even though a model, be it a simulation, or a theoretical one, can never know the exact imputs, and there is always some chance that someone else (like a coach or manager) has extra information that would affect the results of that model, a good one will be able to tell you whether a certain strategy, decision, or condition is true or false even if the imputs are not exact or correct and even if there is information that the model is not privy to but someone else is or might be.
That is a very important point. Very few things bug me more (until I write about something else which bugs me 😉) than when a reader or critic says something like, “Well you (your model) don’t have all the relevant information, so your conclusion (said decision was right or wrong) can’t possibly be correct,” when it is obvious from the magnitude of the results or I have clearly stated that, “Regardless of any extra information or the quality of the data used, the conclusions will be the same!”
And it is not really an either/or thing like in Tango’s Bonds IBB chart. He just does that for convenience and sometimes it works out that way because the conditions we are looking at are discreet and limited rather than continuous. In other words, he may have just split up his analysis into something like, “If I think my model’s answer is correct 80% of the time, I’ll call it a sure walk or don’t walk, and everything else I’ll call “flip a coin.” Or whatever split he wants. Or, as I said, it could be that because there are a limited and discreet number of possible IBB situations, that some of them happen to be “I am 95+% sure that this is the right decision and the others are, “I am only 55 or 60% sure.” Or it could be that the the gain or loss even when you are sure is so small that you group it like 99% sure and I will call it a “walk or don’t walk” and the rest I’ll call it a “coin flip.”
In any case, the concept is an important one.
Now I’ll go RTFA.