Clemens '86 vs. Martinez 2000

Dipping into my archives today...

There's a thread in the Red Sox newsgroup today about comparing various pitching seasons for dominance. Just over three years ago, I did a fairly extensive piece on two of them, and posted it on the Providence Journal's web board. It eventually ran (in a significantly shortened manner) in the Providence Journal on the following Sunday. Below is the entire piece, in all its unexpurgated glory...

---

Nothing in this piece should be construed as criticism of Roger Clemens, Art Martone, or Bill James. But Art's Sunday Beyond the Box Score in which he used Win Shares to assert that Clemens has been a better Red Sox pitcher than Martinez, and, in particular, that Clemens' 1986 season was the equal of Martinez' 2000, made me suspicious. Because I don't think that's correct. So I looked at it here in...


Beyond Beyond the Box Score

I think that a strong case can be made that Roger Clemens has contributed more to the Boston Red Sox, in terms of pitching performance, than any other pitcher in franchise history. A case can be made for Cy Young, but, speaking just in career terms, those are the only two that really belong in the discussion.

But I'd like to address the contention that Clemens has been the "best pitcher" team history. In particular, I want to address the 1986 and 2000 seasons, and examine the performances, and determine whether the Win Shares assessment of the two season as being equivalent is really merited.

To start with, it's important to remember what Win Shares is, and what it is not. Win Shares is a method of dividing the contributions to a team's final record among all of the players on the team, so as to apportion credit accurately for the team's record. It is not, directly, an assessment of player performance, merely an attempt to assess player contribution. This sounds like semantics, and to some extent it is, but it's a legitimate point and it needs to be addressed.

I'm not going to go into a full discussion of the Win Share system for a couple of reasons. First, it's long and complicated, and not easily summarized. Second, I'm not certain that I understand all of it. But I do understand that the entire starting point is that there are three Win Shares for every team win, and they're apportioned among the players in such a manner as to measure each players contribution, so there is a certain amount of teammate dependence in the system.

Taken at it's extreme, for example, Barry Bonds could have a season better than his 2001 season, and if the Giants pitching staff were so bad that they lost every game, he'd get no Win Shares, despite having the best offensive season in baseball history. It is in many ways similar to the everlasting argument of whether a player on a last place team can legitimately win an MVP award. "We came in last with you, we could've come in last without you." If, on the other hand, a team was constructed which won 162 games, there are only 486 Win Shares to go around, even if every player has the best season in history. Those are obviously extreme and unrealistic scenarios, but they illustrate some of the limits in the system.

In part, the Win Shares for a starting pitcher depends on teammate performance, as the formula includes Wins and Losses. I believe that, on a per season basis, W/L record is too teammate dependent, and can skew assessments of pitching performance. When offenses don't score, pitchers don't get wins. Period. And while, over the course of a career you'd think that there's some evening out, and that winning percentage has decent correlation with pitching performance, it's not a good first order evaluation of it.

To apply that to a Win Share scenario, consider two pitchers who throw identical lines, 7 innings, 2 ER, same number of hits and walks - identical. Then consider that each pitcher's team score 3 runs in the 8th inning and wins 3-2. In a system of evaluation in which player production is being measured, these two performance would measure identically. But pitcher A is going to get more Win Shares than pitcher B. How is that possible? Because pitcher B was pitching at home, and the reliever who pitched the top of the 8th gets the win, and pitcher A was on the road, and still the pitcher of record when his team scored in the top of the 8th.

Now, before analyzing the seasons, I'd like to emphasize that nothing which follows should be construed as disparaging either of these pitchers in any way, or either of these seasons. There is no question, it's not debatable, that both of these are outstanding seasons from outstanding pitchers. Each won the Cy Young award and each deserved it. Anyway, let's look at the specifics of the two pitcher seasons, Clemens '86 and Martinez '00, and look at strictly pitching performance. Here are their pitching lines:

Clemens 1986 vs. Martinez 2000

	Year	Ag	Tm	Lg	W	L	G	GS	CG	SHO	SV	IP	H	ER	HR	BB	SO	ERA	*lgERA	*ERA+
Clemens	1986	23	BOS	AL	24	4	33	33	10	1	0	254	179	70	21	67	238	2.48	4.19	169
Martinez	2000	28	BOS	AL	18	6	29	29	7	4	0	217	128	42	17	32	284	1.74	4.97	285

The first thing that many people would see in these lines is the W/L record. Martinez's 18-6 is obviously a very good record. Clemens' 24-4 is a record for the ages. His .8571 winning percentage is one of the top 30 in ML history. His 24 wins is one of only 94 times in the post-dead ball era that a pitcher's won more than 23.

For people that look mainly at W/L record, the conclusion from Win Shares that these two seasons are equivalent might come as a surprise. As stated above, I'm not of that number, and I'm shocked that Win Shares show these seasons as equivalent. Frankly, looked at as pitching performances, the Martinez season seems to be so superior as to defy comparison. What would I look at to come to that conclusion?

The first thing is that number on the far right, ERA+. Adjusted ERA+ is a figure which expresses a pitchers ERA in terms of the league and park context in which he compiles it. Basically, you adjust the pitcher's ERA based on how easy it is to score runs in his home park (it is much easier to score in Coors field, for example, than in Yankee stadium.) Then, you divide the League ERA by the pitchers ERA and multiply by 100. A pitcher who is absolutely league average would compile an ERA+ of 100. Clemens' 169 says that he was 69% better than a league average pitcher. That's an outstanding figure, and led the AL in 1986. But Martinez' 285 is ... well, it's difficult to express how good that is. It's the best single-season ERA+ in Major League history since the AL was started in 1901. The only season higher in any Major League was Tim Keefe's 294 in 1880.

So, at the pitcher's primary function, preventing runs from scoring, Martinez's season was clearly better than Clemens'. So what are the mitigating circumstances which could bring Clemens up to match Martinez? Maybe Martinez was helped by his defense a lot more than Clemens was. But if so, it's not obvious from the available data. In fact, as he put fewer baserunners on via walk (32 vs. 67) and struck out more batters (284 vs. 238), it seems as though Clemens' ERA was more defense-dependent than Martinez'.

Another thing, and one which does, in fact, work in Clemens favor, is that he pitched more innings. Better than average innings pitched have value, and, all other things being equal, the pitcher who maintains a better than league average ERA over more innings is more valuable than one who does it over fewer innings. Clemens threw 37 more innings in 1986 than Martinez did in 2000 (254 vs. 217) and made 4 more starts (33 vs. 29) and those factors do, in fact, help make up for the difference in run prevention. But with the difference in run prevention that exists, is it enough? I don't think it's even close.

Let's look at what the Red Sox would have to do in those 37 innings to get Martinez' 2000 performance to match Clemens' 1986 performance. They'd need a pitcher to make 4 starts, pitch 37 innings, and allow 51 hits, 35 BB, 4 HR and 28 ER. That works out to a 6.81 ERA and 2.32 WHIP for those additional innings. And Pedro plus replacement would still have 46 more strikeouts, even if the replacement pitcher struck out 0.

And none of that takes into account the fact that 1986 was a more pitcher-friendly (or, more appropriately, less pitcher-hostile) environment than 2000. Martinez' didn't just have a lower ERA+, he had a lower actual ERA, despite the fact that the league ERA was nearly 20% higher. An ERA that was 30% lower in an environment where run scoring was 20% easier. There's just no comparison on the run prevention.

So, if Martinez was so much better, why was his W/L record so much worse? Did he choke? Was he unable to win the close games? I believe that the answer is clearly no to those last two questions. It's a fairly straightforward issue. Clemens' 24-4 mark represents excellent pitching backed up by excellent run support. Martinez' 18-6 record represents otherworldly pitching and lots of no-support.

In examining the performances of each pitcher, we find the following cumulative stats on the average performances in wins and losses:

Clemens 1986 vs. Martinez 2000

	IP	H	HR	R	ER	K
Clemens wins	8.13	5.33	0.63	2.21	2.04	7.71
Clemens losses	6.5	5.5	0.75	3.5	2.75	5
Martinez wins	7.63	3.72	0.28	0.78	0.72	10.28
Martinez losses	8	5	1	2.17	2.17	10

We can see that, as expected, each pitcher pitched better in his wins than his losses. What might not be expected is that Martinez actually pitched more innings per game in his losses than his wins. But it's actually an amazing line. Pedro Martinez lost six games in 2000. His average line in those games was 2.17 runs allowed in 8 innings.

The following table shows a little more analysis, including a couple of categories that I call "Cheap Wins" and "Bad Losses". "Cheap wins" are games in which a pitcher gets a win despite giving up more than 4 ER, and is bailed out by his teammates. "Bad losses" are games in which a pitcher gets a loss despite allowing fewer than 4 runs.

Clemens 1986 vs. Martinez 2000

	ERA Wins	ERA Losses	ERA+ Wins	ERA+ losses	Cheap wins	Bad losses
Clemens	2.262	3.808	184.83	109.778	3	2
Martinez	0.852	2.438	576.33	201.436	0	6

It's obvious that the Win/Loss record discrepancy is a result of teammate performance, not individual performance. Clemens had 3 cheap wins, Martinez had 0. Clemens only had 2 bad losses while all 6 of Martinez' losses were games in which he allowed fewer than 4 runs.

But the truly staggering number from that table is Martinez' ERA+ in games in which he lost. To try to put this in perspective, there have been only 32 >200 ERA+ seasons in Major League history. Martinez had an ERA+ of 201 in his losses!

How about the Wins? Did Roger have more Wins because he did a better job keeping the opponent off the scoreboard and then "pitched to the score" afterwards? To determine that, I looked again at the individual Wins and determined how many times each pitcher allowed the opponent to take the lead ("Gave Lead"), how many times he had the lead and let the opponent tie it back up ("Lost Lead") and how many times he did each ("Both").

Clemens 1986 vs. Martinez 2000

	Gave Lead	Lost Lead	Both	Total
Clemens	7	2	3	12
Martinez	1	1	1	3

Again, we see that Clemens Win/Loss superiority over Martinez is almost entirely a result of not his, but his teammate's performances. In 50% of wins, 12 of 24, he let the opposition either score the winning run or score the tying run after his team had the lead. Martinez, on the other hand, only had 17% of his wins (3 of 18) where his offensive teammates "bailed him out." If we're examing pitching performance, there's just no justification for rewarding Clemens relative to Martinez for his W/L record. It's a product of his run support, and Martinez' teammates did not support him the same way. And it makes a difference. Using the Short Form Win Shares (which do NOT use Wins and Losses), Martinez still scores 28. Clemens scores 26.


So, what are the best seasons in Red Sox history? The first thing that I did to examine this was to look at all pitching seasons in which a Red Sox pitcher prevented at least 20 runs, of which there were 85. To generate runs prevented, we look at how many runs a pitcher would have allowed in the innings he pitched if he'd pitched to a league average ERA, and then subtract the actual number of ERs he allowed. This favors pitchers who a) pitch well and b) pitch a lot of innings. And, in fact, if you sort strictly by runs prevented, Cy Young tops the list, though Martinez 2000 is second.

Once I'd identified the 85 20+ runs prevented seasons, I adjusted for league context. With a 200 ERA+, a pitcher will generate 2 runs prevented per 9 innings in a league with an average ERA of 4, and only 1.5 in a league with an average ERA of 3. So we divide the runs prevented by the expected runs allowed to express it as a percentage.

Finally, we want to reward pitchers for throwing lots of innings, but not penalize pitchers who pitch in high-offense eras with different team construction (more starters, more relievers) and lively baseballs. We will assume, for the purposes of this comparison, that it's unreasonable to expect anyone to pitch more innings than the max in the league that year. We'll therefore add to the earned runs allowed by each pitcher the earned runs that would be allowed by the pitcher plus the earned runs that would be allowed by a replacement pitcher making up the rest of the innings to the league max. And then, we'll re-calculate an adjusted ERA+, using each pitcher's ER, plus the replacement ER bring the IP total up to the league max. In the event that the pitcher led the league in IP, obviously, there will be no adjustment ER added.

For the replacement pitcher, we'll set the ERA at 1.5 times the league average. Why? Because we've got to use something, and we assume that some of the innings not pitched trickle down to below average pitchers. 1.5 is an easy number to use (James uses 1.52 in roughly the same role during the Win Shares calculation.)

So, using that method, here's the league-adjusted, IP-adjusted list of the 20 best run prevention seasons in Red Sox history.

Best Run Prevention - Red Sox history

	Year	Lg ERA	ERA	IP	ER	ExR	RPr	ExIP	RpERA	ExRn	TotR	AdjERA	AdjERA+
Martinez, Pedro	2000	4.92	1.74	217	42	119	77	21	7.48	17	59	2.24	220
Young, Cy	1901	3.66	1.62	371.33	67	151	84	11	5.57	7	74	1.73	211
Martinez, Pedro	1999	4.87	2.07	213.33	49	115	66	18	7.4	15	64	2.49	196
Young, Cy	1902	3.57	2.15	384.67	92	153	61		5.43		92	2.15	166
Clemens, Roger	1991	4.1	2.62	271.33	79	124	45		6.24		79	2.62	157
Clemens, Roger	1990	3.92	1.93	228.33	49	99	50	39	5.95	26	75	2.51	156
Clemens, Roger	1986	4.19	2.48	254	70	118	48	18	6.36	12	82	2.73	153
Lowe, Derek	2002	4.47	2.58	219.67	63	109	46	20	6.79	15	78	2.93	153
Clemens, Roger	1994	4.81	2.85	170.67	54	91	37	13	7.31	10	64	3.16	152
Parnell, Mel	1949	4.2	2.77	295.33	91	138	47		6.38		91	2.77	151
Clemens, Roger	1992	3.95	2.41	246.67	66	108	42	19	6	13	79	2.67	148
Martinez, Pedro	2002	4.47	2.26	199.33	50	99	49	40	6.79	30	80	3.01	148
Clemens, Roger	1987	4.47	2.97	281.67	93	140	47	4	6.8	3	96	3.02	148
Martinez, Pedro	1998	4.66	2.89	233.67	75	121	46	18	7.08	14	89	3.18	146
Wood, Joe	1912	3.33	1.91	344	73	127	54	49	5.07	28	101	2.3	145
Young, Cy	1903	2.96	2.08	341.67	79	112	33		4.5		79	2.08	142
Grove, Lefty	1937	4.62	3.02	262	88	135	47	16	7.03	13	101	3.26	142
Hughson, Tex	1942	3.66	2.59	281	81	114	33		5.56		81	2.59	141
Grove, Lefty	1936	5.04	2.81	253.33	79	142	63	48	7.66	41	120	3.57	141
Wakefield, Tim	1995	4.72	2.95	195.33	64	102	38	26	7.18	21	85	3.45	137

OK, that's busy, so let's walk through it. Name, Year, League ERA, ERA, IP and ER are fairly self-explanatory. ExR is Expected Runs, the number of runs that an average pitcher would have allowed in that many IP. RPr is Runs Prevented (Expected Runs - Earned Runs). ExIP is the number of IP less than the league leader. RpERA is the Replacement Pitcher's assumed ERA (1.5 * the League ERA). ExRn is Extra Runs, the number of runs allowed in the Extra Innings at the replacement ERA. TotR is the actual Earned Runs allowed plus the Extra Runs. AdjERA is the pitchers ERA recalculated using the ExIP and ExRn. And finally, AdjERA+ is the ERA+ after the same adjustment. (Where there are gaps in the table, the pitcher led the league in IP, so no adjustments were necessary.)

There have been 16 Major League seasons in which a pitcher compiled an ERA+ of 220 or more. Martinez managed a 220 in 2000 even after adding 21 IP at a 7.41 ERA.


Conclusion:

While Roger Clemens can fight it out with Cy Young over who has had the most productive pitching career in a Boston Red Sox uniform, I believe that Pedro Martinez is the "best" pitcher ever to pitch for the Sox. And his 2000 season was significantly better than Clemens' 1986 season, and, indeed, any other season by any pitcher wearing a Red Sox uniform.