That's the primary reason I haven't written about pitching. Getting an additional reliable starting pitcher is the biggest remaining issue facing the Twins this offseason. And part of the reason I haven't written about that issue is that the rest of the remarkable Twins blogosphere(3) has it covered. As does, of course, the TwinsCentric Offseason GM Handbook.

But the bigger reason is that I have trouble being too opinionated about pitching. And that's because pitching is a craps shoot. My gut reaction is that the more passionate you are about predicting a pitcher's performance next year, the stupider you are. And I say this because I've been pretty passionate about predicting pitching in the past, and I've ended up feeling pretty stupid. Because all we really know is that pitching is a craps shoot.

And this is something you can show statistically, and I'll do so in my typical half-as ... er ... back-of-the-napkin way. Let's take a look at starting pitchers from 2008 and starting pitchers from 2009 and see which of their stats from 2008 could've best predicted their ERA in 2009.(4)

We'll do this by deriving something called a "correlation coefficient." Ooh, I lost a lot of you with those last two words, didn't I? Let's try again.

We'll do this by giving each statistic a rating between 0 and 100. Zero means the 2008 statistic had almost no relation at all to the player's ERA in 2009 - you couldn't have predicted it at all. 100 means it would have been in lock step with the ERA in 2009, so you could have predicted the ERA exactly. (5)

Here are the results:

2008 Stat | Rating |

FIP | 46 |

RAR | 45 |

SO | 45 |

K/9 | 40 |

ERA | 38 |

RE24 | 36 |

WPA | 36 |

HR/9 | 34 |

Strikes | 32 |

LOB% | 31 |

Pitches | 31 |

IP | 30 |

CG | 28 |

Balls | 26 |

K/BB | 25 |

HR/FB | 25 |

AVG | 24 |

GS | 22 |

W | 21 |

BB | 20 |

WHIP | 20 |

LD% | 20 |

HR | 17 |

H | 15 |

BABIP | 15 |

L | 2 |

BB/9 | 2 |

GB% | 0 |

GB/FB | 0 |

I could go on for awhile about this rating, and I still might, but let's talk about the top of the table a bit first.

The best predictor is FIP, which is a fairly new pitching metric that stands for Fielding Independent Pitching. It measure a pitcher's effectiveness with something that looks like ERA, but is based only on plays that do not involve fielders: home runs allowed, strikeouts, hit batters, walks, and, more recently, fly ball percentage, ground ball percentage, and (to a lesser extent) line drive percentage. It's worth a longer entry later.(6)

Anyway, FIP edges out a close race over RAR (Runs Above Replacement) and ... Strikeouts? Notice that it isn't strikeout rate. That's next on the list at 40. It's just plain strikeouts. I guess it's not shocking that for starting pitchers the total number of strikeouts is a better indicator of next year's ERA than a pitcher's strikeout rate, but I didn't anticipate that.

Anyway, fell free to fill up the comments sections with questions/observations about the values in this list. I would love to dissect these values more in a later entry.

But back to the craps shoot. What is most striking to me is that the highest number is just 46. Why? Because of this....

Let's say you were taking a look at a list of all batters last year but your list of stats was missing the RBI column. You wanted to figure out about how many RBI a player had based on the other stats they had. Could you come close using home runs? How about hits?

You probably could. HR would have a rating of 93 in predicting the number of RBI. And hits would have a rating of 94.

But you wouldn't expect to be able to predict it very well with triples, right? We don't associate guys who hit triples with guys who put up big RBI numbers. The Twins leaders in triples this year were Denard Span, Michael Cuddyer and Carlos Gomez. Needless to say they didn't drive in the most RBI.

Well, you would be right about that, too. Triples are a crappy indicator. They only have a rating of 51 in predicting the number of RBI. But that is five points higher than the best indicator we have for next year's ERA.

Which makes our indicator a little worse than crappy. Because pitching is a craps shoot.

(1) There's lots of reasons to write. Sometimes it's because you're inspired. Sometimes it's because you have a passion about something. Sometimes it's to stretch.

And sometimes it's just because you have a blockage. That's today. I don't know exactly where this will go, but I'm pretty sure it's been blocking all other writing for the last few days.

(2) And, by the way, that's how I'm going with "craps shoot." As opposed to "crapshoot" or, god forbid, "crap chute." I'm pretty sure the term is supposed to refer to throwing dice in a game of craps. If it is something closer to the latter spelling, please, don't tell me.

(3) And while we're interrupting this entry half a dozen times before it finds its legs, let's talk about that remarkable Twins blogosphere. Does everyone know about MNGameDay.com? Where you can find a feed of the latest entries from something like 50 Minnesota Twins blogs? No? That's because I don't think I've mentioned it here in something like two years. Because I'm an idiot.

(4) Some specifics are in order for those geeks playing along at home. I'm limiting the study to those pitchers with more than 15 starts in 2008 and in 2009. That gets me (by my computations) 94 pitchers, which isn't a huge sample size, but not half bad either.

(5) Again, for the geeks: I'm computing a correlation coefficient between the arrays, taking the absolute value of it (since I'm just as interested in negative correlations) and multiplying it by 100.

(6) Especially because Carl Pavano's FIP last year was just 4.00, compared to his ERA of 5.10.

## 5 comments:

Maybe I have misread your post, but if I am interpreting things correctly, you are making a false comparison. My impression is that you are comparing the ability for a pitching statistic to predict a pitcher's era in the future to the ability of a batting statistic to predict a batter's RBI total in the present. A player very rarely has a lot of hits or a lot of home runs without getting a fair number of RBIs. However, it is relatively common for a batter to have a 'breakout' season one year and come back terrible in the next.

John -

I think all that demonstrates is that ERA is volatile from year to year. That is not a great surprise since it is pretty sensitive to even minor changes in performance.

The components of FIPS, walks, home runs and strikeouts, are a lot less volatile. When one of those goes south, it usually has a significant cause like an injury.

A guy who depends on "FIPS" to be effective is not going to pitch very many innings or keep his job in the rotation once he loses it. So most of them won't show up in your sample.

The Mathematician is right. For a true comparison, you would have to calculate the correlation between a player's 2008 hits, say, and his 2009 RBI. That would likely be quite a bit lower than 94 but still plenty higher than 46. So the point is the same, just not quite as sharp.

Great to see that list of correlations -- I've never seen it laid out like that. Very cool. But I'm curious now -- what is the correlation between 2008 home runs or hits and 2009 rbi's?

I guess all I see is that predicting ERA is a crapshoot, but ERA isn't the best indication of pitching anyway. Just as your example gives samples of predicting RBI, OPS would be a more useful hitting stat to predict.

Post a Comment