No announcement yet.

A new way to look at Pitching Runs?

  • Filter
  • Time
  • Show
Clear All
new posts

  • A new way to look at Pitching Runs?

    I've been looking at Pitching Runs and trying to find a different way to express the value of a pitcher.

    What I've come up with is a method that I believe could better evaluate a pitcher's performance in the context of the opposition he faces; for, as we know, not all schedules are created equal. The idea is simple enough but the implementation is a bit tedious.

    Over a given season, a team scores a bunch of runs. From that total, we can estimate their average runs per inning by dividing their total runs scored by innings played (more on why it's an estimate below).

    team runs per inning = total(runs scored) / total(innings played)
    So, the estimated runs per inning is the number of runs one can expect a given team to score in a given inning.

    On the other side, we have a given pitcher who pitches some number of innings in a game against some team and gives of some number of runs. We can say that if that were an average pitcher (not necessarily the same as a "league-average" pitcher), he would give up an expected number of runs that can be calculated using the opposing teams runs per inning:

    expected runs allowed = opposing team's runs per inning * innings pitched
    Now with our pitcher's expected runs allowed and his actual runs allowed we can calculated how many runs he saved his team as compared to an "average" pitcher:

    runs saved = runs allowed - expected runs allowed
    A summation (not an average) of a pitcher's runs saved over the course of a season should indicate how many total runs he saved his team. This seems like a useful metric to describe a pitcher's performance as it implicitly factors in the run production of the pitcher's opposition.

    There are, of course, a few pitfalls. As mentioned above, the runs per inning value is an estimate. This is only because I haven't yet found a reliable way of calculating a team's innings played. For a team that plays 162 games, loses every home game, and never plays an extra inning game, 1458 innings are played. Of course, we know this isn't the case: teams don't always bat in the ninth, when they do they don't always complete the inning (due walk-off hits), and teams fairly often play extra inning games; not to mention games that are called early due to weather. Luckily, things balance out for the most part, but rarely does a team play exactly 1458 innings in a full season. The effect on the final runs saved stat is minimal, but probably not negligible.

    Also, it's a bit more tedious to compute than most stats. This is due to the nature of the stat: a line of 8 IP, 2 R against the Royals should not be equivalent to the same line against the Red Sox. Thus, the numbers are considered on a per-game basis.

    So, is this worth the code used to calculate it? Maybe
    It it some ground-breaking revolutionary answer to Life, the Universe, and Everything? That's doubtful
    Is it a pretty neat indicator of a pitcher's performance in the context of the particular lineups he faces? Yep
    Has it been done before? Maybe
    Is there still a lot of work to be done? Absolutely

    Finally, I'll post the numbers I have so far, calculated through games played yesterday. I've only included starters who have pitched in 17+ innings (roughly the cutoff for ERA leaders at this point in the season), which leaves me with 108 pitchers. For the sake of brevity, I'll only include the top and bottom 15. And, as you can see, pitchers are identified by their BR ID, this is only because I've been to lazy to write code to switch these for the pitcher's name, and I figure the BR IDs are decipherable to most everyone here.

            Pitcher         PRV             /ip             starts
    1       sheetbe01       -10.645         -0.38           4
    2       peavyja01       -9.719          -0.324          4
    3       haranaa01       -9.476          -0.271          5
    4       greinza01       -9.04           -0.312          4
    5       leecl02         -8.609          -0.38           3
    6       hernafe02       -7.504          -0.262          4
    7       saundjo01       -7.275          -0.248          4
    8       hamelco01       -6.927          -0.239          4
    9       lohseky01       -6.792          -0.279          4
    10      millwke01       -6.405          -0.2            5
    11      carmofa01       -6.22           -0.271          4
    12      blackni01       -6.169          -0.244          4
    13      westbja01       -6.017          -0.203          4
    14      evelada01       -5.816          -0.246          4
    15      linceti01       -5.784          -0.304          3
            Pitcher         PRV             /ip             starts
    94      arroybr01       4.986           0.234           4
    95      burnea.01       5.105           0.239           4
    96      garlajo01       5.254           0.216           4
    97      bondeje01       5.269           0.233           4
    98      mechegi01       5.469           0.231           4
    99      jennija01       5.953           0.293           4
    100     roberna01       6.496           0.3             4
    101     cainma01        6.722           0.339           4
    102     moseldu01       7.547           0.384           4
    103     oswalro01       8.058           0.351           4
    104     gorzeto01       8.383           0.484           4
    105     lillyte01       9.768           0.524           4
    106     verlaju01       11.287          0.464           4
    107     millean01       14.073          0.797           4
    108     sabatc.01       17.714          0.985           4
    So, thoughts? Opinions? Critiques? Revelations that I'm not, in fact, re-inventing the wheel and that this has been done before?
    Last edited by Debaser; 04-21-2008, 04:56 PM. Reason: woops
    Chicks dig the long ball.

  • #2
    Quality of lineups faced is definitely an important variable to consider. It *tends* to even out in the long run, but not always. Baseball Prospects has a report on it (and quality of pitchers that hitters face):

    Beyond The Boxscore (still with some lime)


    • #3
      Those numbers are definitely interesting, but I'm not sure that they would be of any direct use here. I'm really just concerned with the endgame (runs) and not so much the factors that contribute to run production. Thanks for the advice, though.
      Chicks dig the long ball.


      • #4
        You presume that the Yanks runs per game scored would be what that pitcher would have faced, when in fact you don't know if ARod is in the lineup or not. That's why BP's data shows the OBP and SLG numbers of the guys that the pitcher actually faced. You can easily compare this to the Yanks actual overall OBP and SLG to see whether the pitcher faced a standard Yanks lineup or not.
        Author of THE BOOK -- Playing The Percentages In Baseball


        • #5
          One way to find out if this is a new way or not would be to join SABR and then submit your study as a report to them. The odds of this being a genuinely new method of value rather than a re-invention of the wheel are not high.

          The key thing is to find out whether this really tells us any thing of real value.
          Buck O'Neil: The Monarch of Baseball


          • #6
            There's a question of horses and carts here too.

            The quality of a 3R/7IP performance of a pitcher is relative to the line-up he's facing. But, as is an output of 3R/7 innings for an offense depending on the pitcher they are facing.

            I understand that on one level that is exactly what we are trying to prove, but semantically, the used of the term "expected runs" seems problematic. As, one would not be expected to score at your average rate against one of the game's best.

            Also, and not to complicate this further, but another thing about a line-up and a pitcher's efficacy there against that could belie the objective quality of each respective party is the handedness distribution. A team who is overall a league average offense but is loaded with lefties may be a tougher task for (all or some particular) righties than an overall better team with a more even handedness distribution.

            This possibly even segues into Tango's previous point. Lefties were routinely sat against Randy Johnson. Duke Snider hardly was ever in the line-up when a lefty was on the mound, etc.

            That's the Catch-22 of trying to more precisely define these things than the standard (even simple relative) metrics do. Once you address one esoteric consideration, it begins to beg a host of others...
            Last edited by digglahhh; 04-22-2008, 09:14 AM.

            In the avy: AZ - Doe or Die


            Ad Widget