Bar arguments

The Agitator quotes a not too bad George Will baseball article, and uses it to suggest that Greg Maddux might be the best pitcher the game has ever seen, especially taking into account the era he pitched in. While the interesting take away from that article is the anecdotal evidence of Maddux pitching to the score (the degree to which this actually takes place is fairly important in player evaluations, but very difficult to ascertain), I'd like to address the claim about Maddux's historical stature.

Bill James: ...my friend Dan Okrent, when he was editing Sports Illustrated, ran a cover story on Maddux, entitled "The Best Pitcher Sincer...?" The thesis of the article was that Maddux was the greatest pitcher, certainly the greatest right-handed pitcher, in many years.
I might have been a part of the article, except that I didn't buy the premise. Greg Maddux, as great a pitcher as he is, has never been the equal of Roger Clemens. Maddux has had, throughout his career, quite exceptional Earned Run Avrages, compared to the league norms. The best of these are:

1994 1.56(Maddux) against 4.21(League)              .371
1995 1.63(Maddux) against 4.18(League)              .390
1997 2.20(Maddux) against 4.20(League)              .524
1998 2.22(Maddux) against 4.23(League)              .525

You have to go back many years to find ERAs as low as this, relative to league, and this is the basis of the argument for Maddux as the greatest pitcher of his generation. Maddux's best relative ERAs are better than Clemens':

1997 2.05(Clemens) against 4.56(League)              .450
1990 1.93(Clemens) against 3.91(League)              .494
1998 2.65(Clemens) against 4.65(League)              .570
1986 2.48(Clemens) against 4.18(League)              .593

Maddux is ahead by about .07 times the league norm, which is a difference of seven or eight runs for a pitcher pitching 250 innings.
But while relative ERAs certainly are a valid indicator of pitching excellence, they are not a definite measure of a pitcher's contribution to his team. There are at least three other things which might be considered, which are:

1. Park Effects.
2. Innings Pitched.
3. The contribution of the defense to the prevention of runs.

Both Maddux and Clemens have pitched most of their careers in hitter's parks, so the adjustment for park effects is minor, but it favors Clemens by 1%.
The larger question is, is pitching at this level for 200 innings the same as pitching at this level for 250 innings?
Of course it is not. Maddux is hardly a wuss; he has led his league in innings pitched five times, and has pitched 250 innings or more in a season four times.
Still, in his busiest seasons Maddux has pitched 268, 267, and 263 innings. Clemens has pitched 281, 271, 264, and 264. In his four best relative-ERA seasons, Maddux pitched a total of 896 innings. (It would have been more except for a strike, but that's spilled milk.) In his four, Clemens pitched a total of 981.
So who do you want: a pitcher with a 1.90 ERA in 224 innings, or a pitcher with a 2.20 ERA in 245 innings?
It's close--maybe too close to call. If the league ERA is 4.20, the former pitcher is saving 57 runs relative to the league average; the latter, 54 runs. But if you compare the pitcher not to the average but to the replacement level, you get a different answer: 102 runs a year for "Maddux," 103 for "Clemens" (assuming a replacement level of 6.00 runs/9 innings).
The other issue, the contribution of the fielders behind a pitcher to his ERA, is a trickier one. To extricate a pitcher's ERA from the fielding behind him is virtually impossible, and I certainly could not claim to have done so in a logically compelling manner. Nonetheless,
(a) my statistical method shows that the defenses behind Maddux have contributed more to his success than the defenses behind Clemens,
(b) I don't really think there is any doubt that this is true.
The Atlanta Braves' pitching of the last ten years has been great--but their defense has helped. Clemens has not had those kind of defenses behind him.
Maddux has not only exceptional ERAs, but also exceptional won-lost records. However, since Clemens career winning percentage is better than Maddux's (.647 to .640, through 2000), it is hard to make that into an argument for Maddux. Clemens has a better winning percentage although he has pitched, overall, for teams which were not quite as good.
Clemens' best seasons, as I see them, are 1997 (32 Win Shares), 1986 (29), 1987 (28) and 1990 (also 28). Maddux's best seasons are 1995 (30 Win Shares), 1992 (27), 1994 (26), and 1997 (also 26). I'm not suggesting it is cut and dried or that Maddux is not a worthy candidate. But in my opinion, Roger Clemens is the greatest pitcher of this generation.

If you're still reading after all that, me again. All of the above was written to reflect events through the 2000 Baseball season, but may have only taken Win Shares into account through 1999. Since then, the two of them have had the following results:
MadduxERA+IPWin SharesClemensERA+IPWin Shares

So far this year, Clemens is way ahead of Maddux. It's still a tough call, but it doesn't look to me like Maddux had done enough to pull ahead of Clemens from 2000-02, and now Clemens looks to be playing significantly better. Maddux can't be the best pitcher of all time, because he's not the best pitcher currently playing.