sabermetrics and absolutes

[JLN]

I want to start this post by apologizing for the length of it but I wanted to put something together to get my mind off potential Zambrano and Harden injuries.

 

I love reading this board and I have learned a lot from it but recently I’ve felt the need to comment on some of the sabermetric talk because it has gone overboard in terms of absolutes (e.g., “bunting is bad” and “clutch hitting doesn’t exist”). My claims for this are basically based on Bill James’ article, “Underestimating the Fog” and the game theory chapter from The Book. While I obviously am not nearly as smart as James or Tango, Lichtman, and Dolphin I want to bring out a couple of the main points.

 

In James’ article he basically is arguing that the methods used to reach some of the common conclusions in sabermetric research (including things like pitcher wins, cERA, streakiness, and lineup protection) are flawed so our conclusions may be flawed as well. James’ argument is that the randomness of data (the fog) might be much more dense than we give it credit for so the inability to find results may not mean the effects aren’t actually there but instead that researchers have yet to find appropriate methods to find those effects.

 

The game theory chapter is probably more relevant to the issues I’m talking about. I’m not going into the details about game theory but I’m going to use the example of the sacrifice bunt (because it sparked my interest from the 9/2 game thread). They basically argue that although run expectancy and win expectancy might sometimes dictate that a sacrifice bunt would be a poor choice it is a good idea to sometimes bunt anyway. By bunting occasionally, even sometimes when the defense is playing in, you can force the opposing manager to play for the bunt and gain the benefit of the increased likelihood of getting a hit at some later point. (This also means is that in cases where bunting or not bunting results in a similar run or win expectancy managers are actually bunting right around the proper amount.) Anyway, I can’t explain it any better than the authors but I think that example explains it pretty well.

 

So I would argue that we can’t make definite statements based on some of the sabermetric research because our level of confidence based on the research should not be that high. Also, based on game theory, as long as the opponent isn’t so horribly managed that they do the same exact thing every time in every situation, questionable managerial decisions (like bunting) could sometimes be beneficial even when win expectancy disagrees.

 

I’m not very smart so I hope anyone that is familiar with this stuff will correct all of my errors.

[Hairyducked Idiot]

One of the great flaws of sabermetrics, it's sort of an offshot of Heisenberg uncertainty.

 

Almost all of the data we study was collected in the context of player and teams who *weren't* trying to maximize sabermetric principles. Exactly what you were talking about with the bunt thing.

 

The run expectancy charts were created when everyone was sac-bunting X amount. If we drop sac-bunting to Y-amount, we change the situations and need new charts.

[jersey cubs fan]

Sac bunting is stupid. Bunt for a hit all you want, as long as you succeed.

[Warren Brusstar]

Sac bunting is stupid.

 

Not always.

[FergieJ31]

That is probably my favorite Bill James article; it gets referenced and discussed every now and then around here. I've mentioned it in a few of my posts over the years.

 

IMO, the number one misuse of baseball statistics is the old "correlation is not causation". But a close second is this one: Absence of evidence is not evidence to the contrary. The fact that sabermetricians have failed to turn up evidence that clutch hitters exist doesn't imply clutch hitters do not exist. Underestimating the Fog describes this very well.