Tuesday, December 6, 2011

Introducing a New Pitching Stat: ppERA (Part 1 of 5) - Motivation

(This is Part 1 of a five part series.  Part 2 will be an explanation of ppERA.  Part 3 will feature examples.  Part 4 will discuss the benefits of ppERA.  And Part 5 will consider objections and offer replies.)

The flaws with traditional ways of measuring a pitcher’s performance (such as Wins, ERA, Saves, etc.) have been exposed through decades of sabermetric analysis.  In the place of these stats, sabermetricians offer stats such as the following:

  • FIP (Fielding Independent Pitching):  Focuses on that which a pitcher has most control of – home runs, strikeouts, and walks.  This removes that which a pitcher largely lacks control of – whether batted balls fall for hits or not
  • xFIP:  FIP with an adjustment to stabilize HR/FB rate, which studies show is not something that seems to be within a pitcher’s control
  • tERA (True ERA):  Basically, tERA calculates the runs a pitcher is expected to give up over the number of outs he is expected to get.  This is done by figuring out the run and out expectancy for each PA ending event (K, BB, HBP, HR, Line Drive, Outfield Fly Ball, Groundball, Infield Fly Ball).

There are, of course, many more advanced stats that attempt to evaluate pitcher performance.  Each stat, however, shares an important common characteristic – each one calculates only actions that end plate appearances.  Anything that happens before the final pitch of a plate appearance is ignored.

This is a strange result.

Justin Verlander threw 3,941 pitches last year.  2,485 of those pitches did NOT end a plate appearance.  That’s 63.1% of the pitches he threw.  Even for a stat like tERA, which accounts for all PA-ending events, 63.1% of the pitches Verlander threw were irrelevant for the evaluation of his performance.  And for a stat like FIP, even less of his pitches (8.48%) mattered.

Intuitively, it seems that, on average, a groundball in an 0-2 count will be more weakly hit than a groundball in a 2-0 count (and thus results in a hit less often).  If we ignore everything that happens before the groundball, we have no way of accounting for this.

Furthermore, sabermetricians correctly attempt to remove irrelevant context.  Just as a stat like ERA incorrectly rewards/punishes pitchers for Strand Rate and team defense, every advanced stat rewards/punishes pitchers for a strike or a ball happening to occur when there were already two strikes or three balls, respectively. 

In other words, sabermetricians view each pitcher-hitter confrontation as an isolated event, correctly ignoring irrelevant context such as whether there are men on base.  It is my contention that this ignoring should be extended even further.  Each PITCH should be treated as an isolated confrontation between pitcher and hitter, and count should be disregarded.  (I understand that this is a controversial claim.  I will consider objections and offer replies in Part 5 of this series).

This is my motivation for the introduction of Pitch-by-Pitch ERA or ppERA, which is really just a variation of tERA.  ppERA, however, will count every pitch that leaves a pitcher’s hand. 

It will not be my contention that ppERA should be the only stat used for evaluating pitching performance.  Rather, I believe it should accompany the other (many) stats one considers when evaluating pitchers.

No comments:

Post a Comment