Who is the most consistent starting pitcher in baseball?

                I’ve got several lists here, and I’ll start with the most predictable of them, which is the list based on "decent game percentage, 2011".    A "decent game" is any game in which the starting pitcher has a game score of at least 45. . ..figuring that there is no marked difference in a pitcher’s ability to win at 45 as opposed to 50, but if the starting pitcher goes under 45 the team rapidly loses any real chance to win the game.   The consistent-pitchers list by that standard is a pretty predictable one:

            Doug Fister is maybe a little bit of a surprise there, Ervin Santana, and perhaps the biggest surprise on that list is that Dontrelle Willis, despite an ugly record (1-6 with a 5.00 ERA) actually pitched decent in 11 out of 13 starts.  

                Another way to look at consistency is not "How often did he pitch well?", but "How consistent was he from start to start?"  

                By that standard, the most consistent starting pitcher in the majors in 2011 was Mat Latos of San Diego.

            With a standard deviation of Game Scores of only 10.9, Latos was not only the most consistent starting pitcher of 2011, but actually the most consistent in a season in the five years that I studied, 2007 to 2011.   

                The least consistent starting pitcher in the majors, by this standard (25 or more starts) was Jason Vargas of Seattle.

            And, again, Vargas was not merely the least consistent starting pitcher of 2011, but the least consistent in a season over the five years of the study. 

                Vargas’ inconsistency was probably an asset to him in a limited sense.    Vargas and Latos have several things in common. . .both young pitchers in their second year in the rotation, both of them pitching for West Coast teams losing 90+ games in extreme pitcher’s parks (Park Run Index of .82 for San Diego/Latos, .85 for Seattle/Vargas.)    Latos pitched better than Vargas in terms of strikeouts, ERA and average Game Score, but Vargas finished 10-13, whereas Latos was 9-14.   Of course I’m sure run support factored into that, but if you’re pitching for a bad team, you probably will win more games with a mix of exceptionally good games and exceptionally bad games than you will with a steady line of pretty-good games, I’m guessing.  

                Mat Latos in his ten best starts of 2011 pitched 72 innings with 71 strikeouts, 10 walks, a 1.75 ERA, but just a 5-3 won-lost record.  Vargas in his ten best starts was significantly better:  79 innings, 65 strikeouts, 10 walks and an ERA of 0.68.   He was 5-1 in those games. 

                Latos in his ten worst starts pitched 55.1 innings, struck out 50, walked 24, a 5.53 ERA, a won-lost record of 2-7.     Vargas in his ten worst starts was much, much worse:  44 innings, 78 hits allowed, a 12.07  ERA, 22 strikeouts, 25 walks, and a won-lost record of 0-9.   

                Take a second to look up Jason Hammel of Colorado—ERAs the last four seasons of 4.60, 4.33, 4.81 and 4.76.    Pretty consistent, right?

                In terms of game-to-game consistency, he’s actually been the most consistent pitcher in the major leagues over the five-year period:

            Minimum, 100 starts over the five seasons.  The surprise on that list is the realization that game-to-game consistency seems not at all connected to excellence.    Cliff Lee is actually one of the least consistent starting pitchers in the majors, in terms of pitching at the same level every time out:

            And, really just because I like doing this kind of stuff. . ..Jason Hammel in his 35 best starts over the years 2007-2011 has pitched 233 innings, 1.85 ERA, 18 wins, 4 losses, strikeout/walk data of 172-56.    Ricky Nolasco in his 35 best starts, on the other hand, has pitched 263 innings, has struck out 295, walked 38, posted a 1.47 ERA and has won 31 games and lost only 1.   

                But when she was bad, she was horrid.   Jason Hammel, in his 35 worst starts over the five seasons, has pitched 150 innings with a 10.08 ERA, 91 strikeouts, 72 walks, 270 hits allowed and a won-lost log of 3-21.    That may not sound good, but Ricky Nolasco, in his 35 worst starts, has 1 win, 26 losses, an ERA of 11.50, and has given up 305 hits in 157.1 innings, although his strikeout/walk ratio even in those games is still good (114 to 55).  

                The key question, of course, is whether we are measuring something real here, or whether, like batting averages with runners in scoring position, this is merely data that appears after the fact if you go looking for it.   I don’t know the answer to that.   I drew up lists of pitchers with very high and very low standard deviations of game scores over the years, and looked to see whether there was internal consistency in the lists.  There was some.    There were 19 pitchers who appeared on the same list twice, whereas there were only 12 pitchers who appeared once on each list.   On the other hand, there were two pitchers who appeared twice on one list and once on the other. . . .I didn’t know what to do with those guys, so I didn’t include them in the 19 and 12 count.

                If you look at the 25 pitchers who had the highest single-season standard deviations of Game Scores, they had 56 other seasons with 20 or more starts, in which they had lower-than-average standard deviations of Game Scores 25 times, higher-than-average 31 times.   If you look at the 25 pitchers who had the lowest single-season standard deviations of Game Scores, they had 47 other seasons with 20 or more starts, which included 24 seasons with lower-than-average standard deviations, and 23 with higher-than-average standard deviations.

                All of this suggests the possibility that we could be measuring something that really exists, or else that there could be some sort of consistent bias in the data, such as higher standard deviations in hitter’s parks or higher standard deviations for power pitchers or something.    I don’t know.    I know that Mat Latos was very consistent last year and that Jason Vargas was not, but whether that means anything or was just something that happened, I really don’t know. 

Percentages of Players Having A Better Next Season

            Suppose that you take all players who have 400 plate appearances in a season, and you ask "What percentage of those players will have a better season the next year?"  

                The generic answer, for all players at all ages, is "37%".    63% will have worse or at least slightly worse years the next season; 37% will have better or at least very slightly better.  For 21-year-old regulars, the number is 56%.   In the years 1900 through 2010 there were 210 twenty-one-year-old regulars in the major leagues, of whom 117 had better years as 22-year-olds than they had had as 21-year-olds.    That’s 56%.

                For 40-year-olds, it’s 7%.   In the years 1900 to 2010 there were 42 forty-year-old regulars in the major leagues, of whom 39 had worse years at age 41 than they did at age 40.    The three exceptions are Steve Finley, Stan Musial and Honus Wagner.

                This is the full data:

            You can sort of estimate the percentage at any age by the formula (22/(3* age – 24)); that’s for those of you who won’t remember where the data is, but will remember that formula.   That formula will tell you that, if you had a 50-year-old regular, the chance that he would be better at age 51 would be 17%.  

Presumably the number is less than 50% at all ages over 22 because of injuries, and also because of players who are taken out of the lineup for substandard performance.  If you take players who have 400 plate appearances in a season, 100% of them are healthy enough to have 400 plate appearances in a season.   Not all of them will be healthy again the next season, and not all of them will play well enough to hold their jobs. 

You can sort of control the second factor by eliminating from the study players who create less than 4.50 runs per 27 outs in the "base" year of the study.   But that, surprisingly, causes the numbers above to go down, rather than up; they go down across the board.   The number at age 27 drops from 39% to 33%, etc., so that’s not really helpful.