Identifying a Team’s Best Pitcher

                After I published the article entitled Openers and Aces (May 9), there were posted requests to explain how I had decided what who was each team’s best starting pitcher.  That decision was based on the Season Score, which is to say that it was based on a combination of Wins, Losses, Innings Pitched, ERA, Un-Earned Runs Allowed, Strikeouts and Walks.  It seems like an obvious choice for this purpose.   The weakness of the Season Score is that it doesn’t adjust for the team or for the league, but since neither of those is an issue in this case—comparing teammates—it seems right.

                The first poster speculated that I must have decided who the best pitcher on each team was based on Wins and Losses.   I don’t know what I could possibly have done that would cause anyone to think I would identify the best pitcher on a team by Wins and Losses, but it tends to confirm my suspicion that I must have been a very wicked person in a previous life.    Debating the issue of who should be identified as the best pitcher on a team, the Pirst Foster complains about Steve Trachsel with the 2006 Mets, which seems like a non-instructive example, since there is no method by which Trachsel would show up as the #1 starting pitcher on the 2006 Mets.

                A Pecond Soster brings up the equally useless and non-instructive example of Tom Sturdivant and the 1957 Yankees—non-instructive, because Sturdivant would appear to be the #1 starter on that team by any method you could come up with; I don’t see how you would avoid it.    An instructive example would be one in which one pitcher was a team’s #1 starter by one method, while another was the #1 starter by some other method.   The 2013 White Sox or Red Sox.   On the Whites, Jose Quintana pitched 200 innings and had the best won-lost record among the team’s starters (9-7).   Chris Sale was 11-14, but led the team in ERA, strikeout/walk ratio and innings.  On the Reds, Jon Lester led the team in innings pitched and won-lost record (15-8); John Lackey was 10-13 but had a better ERA and a better strikeout/walk ratio.   Season Score likes Sale more than Quintana, Lester more than Lackey.

                I got to wondering about the general issue, and. . .has there ever been a team on which one pitcher led the team in ERA, a different pitcher had the best won-lost record, a third pitcher the best strikeout/walk ratio, while a fourth pitcher led them in innings pitched?

                I would have bet that that had never happened.   In fact, it has happened 24 times since 1900. . ..very surprised.  On the famous 1971 Baltimore Orioles team which had four twenty-game winners, for example, Dave McNally had the best won-lost record (21-5), Jim Palmer the best ERA (2.68), Pat Dobson the highest total of strikeouts minus walks (187-63), and Mike Cuellar pitched the most innings (292).

`               The White Sox did this in both 2006 and 2008.   In 2006 their #1 pitcher by won-lost record was Jon Garland (18-7), by ERA Jose Contreras (4.27), by strikeouts minus walks Javier Vazquez (184-56), while Freddy Garcia led the team in innings pitched, 216.   Remarkably enough, that team’s other starting pitcher, Mark Buehrle, had about as good an argument to be considered the staff ace as any of the other four, although he didn’t lead the team in anything.  His stats that year were a little bit down, but. . .he’s Mark Buehrle.   At least you can count on him to do what he does.   In 2008 the best won-lost record was by Gavin Floyd (17-8), the best ERA John Danks (3.32), the best strikeouts minus walks Vazquez again (200-61), and the most innings pitched Buehrle (218.2).

                It’s not only good teams that have done this; the 1963 Kansas City A’s were led in won-lost record by Diego Segui (9-6), in ERA by Moe Drabowsky (3.05), in strikeouts minus walks by Orlando Pena (128-53), and in innings pitched by Dave Wickersham (238).   Drabowsky had the best ERA on the team but the worst won-lost log (7-13).   All four guys really were half-decent pitchers who had decent careers, but they were overpowered by pitching for what was otherwise a horrible team.    The 1980 Mets (67-95) were led in won-lost record by Mark (Don’t Call Me Erma) Bomback (10-8), in ERA by Pat Zachry (3.01), in strikeouts minus walks by Pete Falcone (109-58), and in innings pitched by Ray Burris (170.1).   Fifth Starter:  Craig Swan.   Falcone and Swan; it’s a two-bird rotation.

                These are the 24 teams which were led in the four categories by four different pitchers, using the term "Ace" to indicate the man my little formula decides is the team’s best pitcher:

                With regard to the 1966 Cleveland Indians. . .that team, which was third in the American League in ERA, may have had their best pitcher in the bullpen most of the season.   Luis Tiant opened the 1966 season throwing shutouts in his first three starts and threw a fourth shutout on May 24, but was sent to the bullpen after four bad starts in June, and didn’t return to the starting rotation until September 20.    He wound up making only 16 starts, thus failing to qualify for these charts, which require 20 starts—but led the American League in shutouts, with 5.

                The real bone of contention here is the appropriateness of including the pitcher’s won-lost record as one element by which we evaluate his performance.   I can’t tell you how surprised I am to find myself the designated defender of Won-Lost records, since I campaigned against over-reliance on them for the first 20 years of my career, and have probably, I would guess, done more than anyone else ever to discourage people from relying on pitcher won-lost records; maybe not, I don’t know, I’ll leave that up to you.    In any case, I’m not really a defender of won-lost records, but the people who don’t want to pay any attention to them at all are out of their minds, and somebody needs to tell them so.   I’ve explained why several times before, but I’ll take one more stab at it.

                The radical anti-won/lost advocate is making two mistakes.   First, he is focusing on flaws in the won-lost category, and ignoring the fact that all of the other categories that we use to evaluate pitchers have the same sort of flaws—sometimes worse, sometimes not as bad, but they all have the same kind of flaws.   And second, he is asserting that there is no information in the won-lost category which is not redundant of the other pitching categories such as innings, strikeouts, walks, and ERA.   That’s just dead wrong, because there is information in the won-lost record which is not redundant of the other pitching categories.  If a pitcher does anything well, such as showing composure in close games, winning games 4-3 and 8-7, that could be reflected in his won-lost record, even if it doesn’t show up anywhere else in his record.   It is wrong to assume that won-lost records reflect some such hidden skill, because in many cases they absolutely don’t; in many case they reflect Good Luck, and nothing else—but it is equally wrong to assume absolutely that no pitcher has a better won-lost record because he has pitched well in close games or because of some other hidden talent.    We can’t assume that absolutely because we don’t know that absolutely.

                But there’s something else here that I’ve never written about, so let me try to explain it.    Game Specific Run Effects.    Game Specific Run Effects include Temperature, Wind, the Home Plate Umpire, and various other unknown or difficult to quantify elements such as whether the grass is wet or dry, the humidity, the tendencies of the other umpires, and  . .well, who knows.

                At risk of confusing the issue, a brief diversion.   I went to a game in Kansas City about 1994, at which five or six baseballs were hit an enormous distance—five or six fair balls, and several foul balls.    I have never understood what happened; the wind was not howling out, the air was not warmer than any other days, but. ..you couldn’t avoid the feeling that there was something happening here that we didn’t understand.   Balls that ordinarily would go 400 feet, on that day, were going 430.    I remember sitting there, watching those long, long home runs, and thinking "something is going on here that none of us understands."   I have had that feeling at ballgames on other occasions—sometimes that the balls were just dying in the air, at other times that they were flying through it with un-accustomed ease.

                My off-the-topic question to you is, do you think there could be some fundamental atmospheric condition to which the human race is so far entirely blind, in the same way that we were blind to barometric pressure before 1643?    Our five senses are in some ways limited; there are things around us that we can neither see nor hear nor smell nor feel nor taste.   I’m not talking about ghosts or anything paranormal; I’m talking about the possibility of there being something entirely normal that we merely lack the ability to sense, even with the enhancements of scientific tools, because scientific tools, in the end, merely make our senses more acute.   Do you think there could be something of that nature, which affects the travel of a baseball in flight, but which we simply have not yet recognized and measured?    Or is that too speculative for you, too far out? 

                OK, put that question aside, or post your answer if you prefer; back to the main topic.   Game Specific Run Effects vary at least with temperature, wind, and umpiring, and they vary as well with other game conditions which we more dimly understand.    Some days the game is 12 to 10, not simply because the pitching is bad and the hitting good, but because it is hot, the wind is blowing out, and the umpire won’t give a pitcher a corner; some days it is 1 to 0, not because the pitcher is great but because it is cold, the wind is pushing in and the umpire is calling everything a strike.    Game Specific Run Effects; GSRE for short.

                The question is, do GSRE even out for a pitcher, over the course of a season?   

                We don’t really know—but I don’t see how they possibly could, a pitcher making only 32 or 33 starts in a season.  Let us suppose that there are two pitchers, teammates, and that Pitcher A makes 8 starts under strong hitter-friendly GSRE, while Pitcher B, just by the luck of the draw, makes NO starts under strongly hitter-friendly GSRE.   Pitcher A is supported by 180 runs, has a 4.20 ERA and finished the season 16-11; Pitcher B is supported by 110 runs, has a 3.30 ERA and finishes the season 9-13.

                The radical anti-won/lost advocate will tell us that this difference in offensive support is merely luck, and consequently that it should be factored out of the pitcher’s evaluation—but the fact is that we do not KNOW that it is entirely luck.  The Radical Anti-Won/Lost Advocate is implicitly asserting that Game-Specific Run Effects even out over the course of the season, but the fact is that we do not know that they even out over the course of the season, and in my view it is enormously unlikely that they even out over the course of the season.  32, 33 starts. . .that just isn’t enough for anything to even out, generally speaking.

                I set up a model to estimate what the residue of Game-Specific Run Effects over the course of a season might reasonably be. . .I don’t know if this is interesting or not, but it was interesting to me so I’ll pass it along.   Let’s assume that an average team scores 4.50 runs per game, and let’s assume that that expectation is set by five factors:

                1)  0.50, standard for every game,

                2)  A random number between 0 and 2, reflecting the temperature at game time,

                3)  A random number between 0 and 2, reflecting the wind conditions,

                4)  A random number between 0 and 2, reflecting the umpiring, and

                5)  A random number between 0 and 2, reflecting the four factor, the sum of the other known and unknown elements which work to increase or suppress scoring.

                I use four random numbers to represent the game conditions, rather than one, because when you use four random numbers you get a bell-shaped curve, whereas when you use one you get a flat-line distribution.     The more random numbers you use, the tighter the distribution of the bell-shaped curve.

                Anyway, using these assumptions, the average Expected Runs Scored for a team in a game is 4.50, but the standard deviation is 1.15.    The actual standard deviation of runs scored for a team in a game is 3.07, so what we are implicitly saying here is that of all the variance in runs scored in a game, 14% is caused by Game Conditions, and 86% is caused by other factors (such as luck, the performance of the pitchers, and the performance of the hitters.)   That seems to me like a reasonable estimate.  

                Well, IF those conditions described above are an accurate model, then the standard deviation of Game Specific Run Effects, over the course of 32 or 33 starts, is .2013 runs per game; let’s call is .200.    In other words, there probably are Game Specific Run Effects embedded in a pitcher’s ERA on the level of .200 runs per game more or less than average.

                Did that make sense?   Let me try again.   I took the five-step assumptions outlined above, and simulated Game Conditions for hundreds of thousands of games.   Then I broke out groups of 32 and 33 starts (40% 33 starts, 60% 32 starts), and looked at the issue of whether or not Game-Specific Run Effects would even out over the course of 32 or 33 starts.  The standard deviation of Average Expected Runs for pitchers making 32 or 33 starts was .200.

                What that means—assuming that my model is reasonable—is that if a pitcher’s run context is estimated at 4.50 runs per game without any consideration of Game Specific Run Effects, then the "real" number would virtually never be 4.00 or 5.00, but that it would quite commonly be 4.30 or 4.70.

                And if Game-Specific Run Effects are real, then ONLY the won-lost record is going to automatically adjust for them; the ERA won’t, the strikeout to walk ratio won’t.   The won-lost record will, because the Game-Specific Run Effects should affect both teams in a game.  Effect, affect. . .I’ve never been able to figure that one out.  And don’t try to explain it to me.