Seeing the World Pitch by Pitch

Back in February of 2007, I had an informal exchange with John Beamer over J.J. Putz over the merits of Putz's 2006 improvement and whether it was repeatable (scroll down to No. 5). Unsurprisingly, most projection systems did not think so, but I surmised that was because they failed to grasp the driving forces of J.J.'s transformation, the splitter he picked up from Eddie Guardado which completely changed his profile as a pitcher. In literally the blink of an eye J.J. went from just another hard-throwing fastball guy with average control and no secondary out pitch, to a bona fide two plus-pitch pitcher.

While statistical research has done wonders into breaking down the hitting side of the game, including the aging and development curves that apply well to a broad number of hitters, the pitching (and individual defense) side are still lacking. Most notably, the aging and development curves on pitchers (and defenders) is all over the map. Pitchers do not follow a standard curve of development as hitters do. Instead, it's more of a series of plateaus, something that resembles a step-wise function for those of you out there keen with math.

ERA is still the measurement de jour used to evaluate pitchers. However, most of us have come to realize that ERA is not a very good way to quantify a pitcher's contributions. For one, it is incredibly team-dependent, in that a pitcher's ERA is largely influenced by the defense behind the pitcher and we all now generally agree that the pitcher bears little control on a batted ball once it is put into play (the DIPS theory).

For that reason, and several others, we moved onto to other measurements, eventually settling on the three true outcomes: strikeouts, walks and home runs. Those three categories served us pretty well in evaluating pitchers and formed the basis of the DICE, FIP and dERA statistics. Why? Well, they correlated much much better on a year-to-year basis than ERA, suggesting that they are less influenced by random noise and better reflect a pitcher's innate ability.

However, I hope that we can find something better, or at least more complete, than this and so I turned to mining Retrosheet and MLB's Gameday for pitch-by-pitch data. One of the things that we can do right away with this data is analyze how the outcomes are reached. For example, a pitcher's strikeout ratio is highly influenced by his percentage of swinging strikes generated (R^2 around 0.62) while a pitcher's percentage of called strikes alone bears almost no statistical influence on the pitcher's strikeout ratio (R^2 around 0.03).

 little bit of difference here  

First off we'll cover the average pitcher broken down in two ways (SP vs RP and LH vs RH) across the years involved (2003-7) to provide a typical pitcher profile. Afterwards, we'll look to see if there are any interesting trends from 2003 to present.

I break up the data in two ways, by pitch and by plate appearance and house it all in one table, what I coined the IPORT (for "Individual Pitch and Outcome Report Table") back in 2007. For every pitch thrown there are one of five possible results:

-batter takes for a ball
-batter takes for a strike
-batter swings and misses
-batter swings and fouls it off
-batter swings and keeps it fair

Of those pitches that fall into that last category, balls in play (incl. home runs for now), there are four possible categories for the type:

-line drive
-infield pop

Individual Pitches

For every (read: 99.99%) plate appearance, it will end in one of six ways:

-batter strikes out
-batter draws a walk
-batter is intentionally walked
-batter is hit
-batter hits a home run
-batter puts the ball in play (obviously not incl. home runs now)
-catcher interferes (we toss these out b/c the pitcher isn't involved)

Outcome Reports

What's fascinating to me is how little variation there is across all five years. With more years at our disposal we would be able to ferret out some interesting investigations and that is also one of my goals is to extend the coverage of this back to 1989 when Retrosheet first introduced pitch-by-pitch data. The reason I don't right now is because I have concerns over how they marked their batted ball types pre-2003.

Percentage of PAs resulting in a strikeout by RH SP
15.4, 15.5, 15.2, 15.4, 15.6

Percentage of PAs resulting in an unintentional walk by RH RP
7.9, 7.8, 7.7, 7.6, 7.8

Percentage of pitches resulting in a groundball by LH SP
9.1, 8.8, 9.0, 9.0, 8.7

Percentage of pitches resulting in a swinging strike by LH RP
9.4, 9.4, 9.6, 9.6, 9.5

Oh, you wanted actual interesting trends? Oh, hmm... well K/uBB looks like it might be increasing at a snail's rate across all four categories. BABIP might be inching upwards. There's just not much there. The state of the underlying game has been incredibly static for the last five years.

Ball% vs CalledStr%
If the batter chooses not to swing a pitch, how often is the pitch called a ball compared to called a strike? 68.89% are called a ball.

Foul% vs (LD%+GB%+FB%+IF%)
If the batter swings and makes contact, how often is the pitch fouled off compared to kept fair (incl. home runs)? 47.07% are hit foul.

The end goal of all of this will be to identify, out of the myriad of trackable statistics, the key indicators that best indicate future performance. As mentioned, swinging strike percentage is a terrific indicator of future strikeout rates and can help us in determining if a pitcher's raw ability to miss bats is improving or not. That way, we can identify real gains in talent, like Putz's, from lucky stretches like Jarrod Washburn's higher-than-normal strikeout rate in the middle of 2006.

This is all going to take a long while to sift through and run models on, the details of which will interest almost none of you, so for now I wanted get some good and/or interesting referential data in your hands. Future posts dealing with this data will link back to this original writeup and will also feature "IPORT" in the subject line.