SABR 38 Day Three: Stats, Stats, Stats
David Smith, the wonderful man who gave the world Retrosheet, just gave a nice presentation about the importance of strike one. It was the second part of a study he started at last year’s convention in St. Louis, and just as stuffed with interesting info and just as interesting.
As it turns out, it’s really bad to start a count 0-1 if you’re a hitter. Duh, right?
Well, it’s worse to start an at-bat with a swinging strike than a called strike, which is in turn worse than fouling a ball off. The very worst situation is if you’re a lefty, facing a lefty, and miss on your first swing, you’re in trouble. Statistically, you’re likely to end the at bat with a hit only 18.6% of the time — a .186 batting average.
It’s all a little dense and numbers heavy, but when you get down to the essence of the thing — that strike one is not only very important, but it’s telling just how it happens — it’s an interesting study.
It’s similar to Sal Baxamusa’s article on the “memory” of a count, published in December of 2006 by The Hardball Times, but much more in depth.
— — —
The esteemed Pete Palmer and Dick Cramer set out to investigate that great ephemeral statistic: clutch hitting. It’s been almost impossible to nail down, but their new study comes closest.
So, what conclusions did they reach? Here is their summary, in their words:
Game situation has nothing to do with average batting performance.
The variation in in career “clutchiness” among the 897 players with >3000 BFP’s between 1957 and 2007 seems random.
David Ortiz and Mark Grace are tied for roughly 80th and 100th among the 897 players, despite their renown as clutch hitters.
Interesting stuff, although the myth of “clutchiness” isn’t dead yet.
— — —
The next presentation is from Seattle Times columnist Jeff Angus. He set out to take a look at how Bill James’s Game Score has aged over the past 20 years.
It turns out that Game Scores have changed very little over that time, and are an excellent indicator of a starter’s quality — on the surface.
His conclusion that you can use Game Score to analyze pitchers may be a bit too much, but he’s effectively shown that Game Score is a useful, telling tool for the average fan to make a determination as to the quality of a pitcher or a start.
I’m concerned that he’s taking an average of an average, and then running it through a simplifying formula that reduces it to an integer, therefore creating a stat that’s just too basic, and not very telling. Still, there seems to be some grain of truth buried in Game Score, and there’s definitely more to be looked into on the subject.
I’d like to see tons more data fed into his model, on a start by start basis for every starter and every season, and see how Game Score fares across time.
Interesting stuff, though.
— — —
The last presentation of the convention is by Matt Souders. He’s trying to analyze the circumstances surround run scoring. He’s assembled a matrix that takes into account multiple factors from park factor to defense to pitching to team offense. It’s all very thorough.
The problem I see is that by feeding so much data into it, he’s going to make accurate predictions of runs scored — but is it a prediction, or just a return of the run scoring data he plugged into it in the first place?
This brings up a problem I’ve been seeing with some statistical analysis in baseball. Since there’s no true way to independently test these variables outside of the game (i.e., measuring batting average in a lab, defining defensive range on a practice facility), you might just be running in circles.
If you’re using one factor to measure another, there’s no ground level or absolute measure. It’s all relative to itself, and any variance isn’t really significant. You’ll predict exactly what you put in.
It’s like using an unmarked stick to measure the width of a stream. The stream may be exactly three sticks wide — or the stick is one-third of a stream long. There’s no absolute baseline to compare anything to.
My concern is that by taking into account virtually every factor in the game, you’re just measuring it against itself — and you’ll get ridiculously high correlation numbers like the .9994 that’s cited in this presentation.
Then again, I don’t have a degree in statistics, so I could be wrong. I’m just using common sense here. If you’re a stats professor, please let me know if I’m off-base on this.
— — —
Except for the upcoming trivia contest — in which I’m sure I have no chance of advancing beyond the preliminaries — the convention is fairly well wrapped up for this year. I’ll be back later with a complete wrap.