15.  Roger Craig, 1963

                Roger Craig had a famous Tough Luck season in 1963, pitching for the expansion Mets, when he was 5-22 with a 3.78 ERA (5-21 as a starter, one loss in relief.)   In the section above he shows with a Good Games Record of 18 – 11.

                What struck me is that no other approach to evaluating that season that I am aware of would show Craig as deserving of a record as good as 18-11.  His ERA was higher than the league norm, and yes, it was a hitter’s park, but Lee Sinins’ Complete Baseball Encyclopedia still shows him as ten runs worse than league average, park adjusted (-10), and as deserving of a won-lost record of 12-15.

                This Good Games Record is vulnerable to the flaws created when you draw a straight line through the data.   A game is either on one side or the other; a "6" is the same as a "10" for this purpose, a "0" is the same as a "4".    Details can get lost.   I’m not claiming it’s a perfect method. 

                Still, I have more confidence in this analysis than I have in any analysis based simply on season totals.   Season totals miss things, too; getting roughed up a few times looks exactly the same in the post-season totals as pitching consistently poorly.   Pitching against a bad team looks the same as pitching against the ’61 Yankees.  

                The question I would pose here is, could the traditional methods be systematically over-rating the pitchers on good teams, and under-rating the pitchers on bad teams, because the traditional methods ignore the issue of the quality of competition faced?    Almost all lines of sabermetric analysis, I believe, build on the assumption that the quality of competition you face evens out over the course of the season.

                But, of course, it doesn’t even out; I started to say that it doesn’t necessarily even out, but a more accurate statement would be that it necessarily doesn’t even out.  Suppose that you are playing in an 8-team league, with a .620 team, a .585 team, a .550 team, a .515 team, a .485 team, a .450 team, a .415 team, and a .380 team.  Suppose that a starting pitcher makes 5 starts against each of the other 7 teams.   The pitcher on the .620 team (the 100-win team) not only has the advantage of having a better team behind him—which we understand and adjust for—but also has the advantage that he is in effect competing in a .483 league, .483 being the average winning percentage of his opponents.    The pitcher on the bad team—the 100-loss team—is in effect competing in a .517 league.  It’s not an immense difference, but it is not a trivial difference, either.   It’s a factor.

                Of course, this difference gets smaller if there are more teams in the league—but it gets larger if you are playing for a worse team.   The 1963 Mets were not a .380 team, the bottom feeders of my theoretical example; they were a .315 team.  The average quality of competition that Craig would have faced in 1963, assuming he faced each of the other nine teams an equal number of times, would have been .521.

                Has Craig’s season been systematically undervalued by other analysts, and by myself up to now, because we have failed to systematically adjust for this bias?   I would suspect that the answer is "yes".   I can’t say that for certain, because I don’t really understand all of the analytical approaches that everybody else uses; I would be a better person if I did understand them all, but I don’t.   It just seems to me that there could be something there that everybody has been missing up to now.

16.  Rotation Patterns

                In one of the charts above I showed you the number of pitchers starting on two days’ rest or less, by year.   While I was doing that, I decided to count the number of pitchers starting on 3 days rest, 4 days rest, etc.   In the chart below "2 days" means 2 days rest or less, and "7 days" means seven days’ rest or more.   Some of the data from the 1950s will be screwy because we are missing games.   If we’re missing a couple of starts then a pitcher who started on three days’ rest might show up as starting on 17 days rests, so. ..be careful with the data from the 1950s.   The percentages on the right are the percentage of 3-to-6 day totals, ignoring the off-rotation numbers.   This has the effect of making the missing data irrelevant to the percentages, since the starts after "Missing" starts would always be off-rotation. 

                Note that the percentage of pitchers starting on 3 days rest increased from 29% to 44% in five years, 1961 to 1966.  Up to 1960 managers used kind of loosy-goosy rotations, playing left/right matchups and moving pitchers from the rotation to the bullpen to get the matchup they wanted.  Between 1959 and 1966 the belief that pitchers would be more effective if used in a regular rotation swept the game, resulting in much more stable rotations.

                We can see in this chart that the four-man rotation remained the norm until 1973, and that the shift toward the five-man rotation really began in 1974.   Some teams were using five starters in some seasons before 1974, but before 1974 it was not a trend.   The great shift from the four-man to the five-man rotation happened between 1973 and 1984.

                We can also see, however, that there continues to be some shift in the patterns even in the last ten years.   In 2003 there were 51 starters working on three days rest; in 2013 the number was 5.  In 2002 there were twice as many starters working on four days rest as on five days rest, 62% to 31%; by 2012 that ratio was only 54-39.   As recently as 1995, only 22% of major league starters were working on five days’ rest.   Within a very few years, it appears likely that that number will go over 40%.

17.  Why the Steroid Era Creates Dominance

(Apart from the fact that Pitchers Also Used Steroids)

                Two of the most dominant starting pitchers of all time, Randy Johnson and Pedro Martinez, did their best work in the very heart of the steroid era.  We will hear not infrequently that what these pitchers did was especially remarkable because it was done in the steroid era, when run-scoring levels were at historic highs.

                Well, yes, but. . ..In a certain sense a high run context acts as a magnifying glass on the statistics, increasing the separation between the great and the ordinary.

                If all baseball games were 1 to 0—that is, if every game ended when the first run was scored—then the distinctions between the best pitchers and the average would be minimized.  Sandy Koufax wouldn’t go 26-8; he would go 21-13.   Pedro Martinez wouldn’t go 23-4; he would 18-9.

                How do we know that?   If every game was 1-0, then the team’s winning percentage would be the same as their run ratio.    We know what those run ratios were.   In games started by Sandy Koufax in 1966, the Dodgers outscored their opponents, 165-104.   In games started by Pedro Martinez in 1999, the Red Sox outscored their opponents, 159 to 85.

                The more runs are scored in a game, the fewer games it takes for the better team—or the better pitcher—to assert dominance.   I am sure that there are a lot of you who know this and think that I am wasting your time in pointing it out, but a lot of people don’t get it.  In a moment we’ll list the best Good Game Percentages in the data.   A disproportionate number of these will be from the steroid era.

                The system does not favor pitchers from the steroid era, but, to a small extent, it does favor a great pitcher from the steroid era, as opposed to a great pitcher from another era.   In this sense, Pedro has an advantage over Gibson or Koufax. 

18.  Good Game Percentage Leaders

                Three seasons by Randy, three by Pedro, two each by Gibson, Carlton and Maddux, one each by Clemens, Koufax, Gooden, Guidry, Kershaw, Tiant, Mike Scott and Dave McNally.

19.  Performance Norms At Each Level

                In this study, pitchers who have a "10" performance have an overall winning percentage of .959, and an ERA of 0.47.  Pitchers who have a "0" performance have a winning percentage of .003 and an ERA of 18.52.

                Plus, remember that the competition faced by pitchers in Group 6 was tougher than the competition faced by pitchers in Group 5, etc.  The most relevant point here is that each step upward has more or less the same impact on the winning percentage.   The increase from Group 0 to Group 1 is only 35 points, whereas the increase from "2" to "3" is 132 points, but that gap mostly disappears if we focus on the winning percentage of the pitcher’s team, rather than the winning percentage of the pitcher:

                Because this is true, it would seem to me that it is an entirely reasonable way to determine a pitcher’s value by simply adding up the levels in his games.   The more steps up the ladder he has taken, the more games his team probably has won.

                I would suggest that there are four ways, growing out of this study, to determine who "should" win the Cy Young Award or who should have won the Cy Young Award:

                1)  To look at the Good Game Percentage,

                2)  To add up the levels,

                3)  To look at how far above average the pitcher is (5.00), or

                4)  To look at how far he is above replacement level.

                I can’t see that any one of those methods is clearly better than the others.   Therefore, I would argue that if the Cy Young Award winner ranks as the best pitcher in the league in any of these approaches, that it’s a reasonable selection and should be treated as an accurate selection.   If the Cy Young Award winner doesn’t come out on top by any of these measures, then it is up to somebody else to justify the voting. 

                However, I will note that the Replacement Level theory is not actually applicable to a game-by-game analysis.

20.  Leader Lists

                Most Level-10 performances in a season:

                Most Level-9 and Level-10 performances in a season:

                The older pitchers have a big advantage here because they were working in a four-man rotation.   Bunning, 1967, is the only pitcher on any of these first lists who didn’t win the Cy Young Award.   Most Level-8, 9 and 10 performances in a season:

                Most Level-7, 8, 9 and 10 performances in a season

                Mickey Lolich, 1971. . .second pitcher we have seen who didn’t win the Cy Young Award. 

                Most games started in a season without a Level-7 performance or better:  Steve Blass, 1973, 18.

                Most Level-10 performances in a career:

                Every pitcher with 100 Level-10 performances in his career is in the Hall of Fame except for Clemens, Pedro and Luis Tiant.   Tiant had exactly 100.    Most Level-9 or Level-10 performances in a career:

                Most Level-8, Level-9 or Level-10 performances in a career (reminding you that we don’t have complete data for the Warren Spahn/Robin Roberts/Whitey Ford generation:

                Most Level-7 and above performances in a career:

                Highest Average Performance Levels in a season, 20 or more starts:

                Highest totals in a season (10 points for "10", 9 points for a "9", etc.):

                Furthest above .500 (5.000) in a season:

                Furthest below average in a season:

                Furthest above replacement level in a season:

                Furthest below replacement level:

Highest average performance levels in a career, 100 or more starts:

                Lowest average performance level in a career, 100 or more starts:

                Highest Good Game Percentages, 100 or more starts:

21.  Mike Garcia, 1954

                Mike Garcia in 1954 was 19-8 with a league-leading 2.64 ERA—but never beat a team with a .500 or better record.  Other than Garcia, the most wins by any pitcher in my data who never beat a .500 or better team was 13; Garcia was at 19.

                Garcia’s team, the 1954 Indians, was so good that they kept most of the rest of the league under .500.   Other than the Indians, there were only two teams in the league with winning records—the Yankees (103-51) and the White Sox (94-60).   Garcia didn’t beat either of those teams, and nobody else in the league was better than 69-85.   So Garcia won 19 games, but never beat an opponent with a won-lost record better than 69-85.