Still Searching for Clutch Pitchers

More than two decades ago, Pete Palmer contributed what I think is one of the best baseball statistical analysis efforts ever done. The results were published in The National Pastime in 1985, in an article entitled “Do Clutch Pitchers Exist?”

Palmer examined pitchers with at least 150 decisions between 1900 and 1983, accounting for how many runs each pitcher allowed, how many were scored on his behalf, and what his career won-lost record “should” have been based on that data. He was searching for “clutch” pitchers: men who won significantly more games than expected because of some unusual ability to pitch to the score and emerge victorious in the close games. With 23 years of additional data, and newly available research tools, now seems a good time to revisit this project.

The overwhelming majority of the time, a team’s won–lost record correlates to the number of runs it scores and the number it gives up. It follows that the same is true about pitchers: if a pitcher has a winning record, most likely it is because he allowed fewer runs than average (reflected in his ERA), or his team scored more runs than average, or both.

There is another factor involved in statistical results: luck, or what statisticians call random chance. For example, if you flip a coin 100 times, you’d expect to get heads about 50 times, but you might get a little more or a little fewer than 50 just by luck. In fact, based on the laws of random chance, there is a 68% chance you’d get within one standard deviation of that total (between 45 and 55 heads), and a 95% chance you’d get within two standard deviations (between 40 and 60 heads).1

What Palmer found is that most pitchers wound up with about as many wins as they should have, with variations within those rules of random chance. In other words, if you win more games than expected, you’re lucky, and if you win fewer, you’re unlucky. His conclusion: “Clutch pitchers do not exist.”

Palmer has updated and fine-tuned his research since then. For one thing, he essentially eliminated modern relievers, because their inclusion skewed the data. Many had much lower winning percentages than expected because of their usage patterns: entering almost exclusively with their teams ahead, they are more likely to suffer a loss than earn a win. Thus, Palmer’s current study includes only pitchers (501 in all) with at least 200 starts and 200 decisions between 1876 and 2006.

In the original study, Palmer used a complex method to estimate a pitcher’s run support, based on his innings pitched, his team’s offense, his own batting performance, and the Linear Weights formula. Thanks to Retrosheet, he is now able to use actual run-support figures (though the figures are not broken down to show runs scored while the pitcher is actually in the game).

Nevertheless, the results of the updated study are very similar to those of the original, and produce the same conclusions.

According to Palmer’s formulas, the number of runs needed to produce an extra win over the course of a season is equal to ten times the square root of the number of runs scored by inning by both teams. Using this theory, it is possible to project a pitcher’s won-lost record based on the number of runs scored and allowed. For example, Johnny Allen made 241 starts in his career, during which his teams scored 1,393 runs, an average of 5.78 per game. Since Allen pitched a total of 1,950.1 innings (the equivalent of 216.7 nine-inning games) in his career, we estimate that his teams scored 1,253 runs (216.7 times 5.78) on his behalf. Meanwhile, Allen allowed a total of 924 runs, an average of 4.26 per nine innings. He thus projects to have had 329 more runs scored on his behalf than he gave up.

To figure out Allen’s expected won-lost record, we need to determine the number of runs per win in his era. In this case, that number is ten times the square root of (5.78 plus 4.26 divided by nine), or 10.56 runs per win. We divide the 329 by 10.56, determining that Allen should have been 31.2 wins above .500. Since he had 217 decisions in his career, his projected wins are 31.2 plus half of 217, or 139.7. So Allen should have gone about 140–77 based on his runs scored–runs allowed patterns. In fact, his career record was 142–75.

Incidentally, Palmer has expanded his study to determine how many of a pitcher’s “extra” wins (wins over .500) can be attributed to his pitching, and how many to his offensive support. For example, Whitey Ford, an excellent pitcher on a great team, finished with a 236–106 record, or 65 games over .500 (171–171). Palmer finds that 38 of those wins were attributable to Ford’s pitching, 22 were courtesy of the Yankees’ bats, and the other five were due to luck.

In a sampling of 501 pitchers, we would expect to find about 160 (32%) who finished more than one standard deviation above or below projection, 25 (5%) who finished more than two, and one (0.25%) who finished more than three. The actual totals are 161, 16, and zero (with Red Ruffing just missing, at 2.98), respectively. Thus, the results are about what we would expect from random chance, and there is no evidence of clutch pitchers.

Here are some highlights of the new study:

• Of 501 qualifying pitchers, 100 (20%) came within one win of projection (rounding off to the nearest integer). Only four pitchers were more than 15 wins off projection.
• Of the 161 pitchers who were at least one standard deviation off projection, 102 were over projection and 59 were Of the 16 who were at least two standard deviations off, 14 were over projection but only two were under. The average pitcher among the 501 was one win over projection. This could be because those who are “lucky” in the win column are more likely to get 200 decisions.
• Two of the three luckiest pitchers were named Welch: Mickey (+21) and Bob (+17). The unluckiest, by far, was Red Ruffing (- 24). Table A shows the pitchers who exceeded projection by the greatest number of wins, while Table B shows those who came in under projection by the most. Table C shows the projected and actual records of some other pitchers of interest, including several commonly regarded as “clutch” pitchers.
• Several pitchers might have made the Hall of Fame, or at least become more serious candidates, had they only matched their projected records. They include Bert Blyleven (287–250 to 299–238; I think somehow he would have managed one more victory), Carl Mays (208–126 to 217–117), and Jim McCormick (265–214 to 280–199).
• On the other hand, Rube Marquard (201–177 to 195–183), Early Wynn (300–244 to 297–247), Happy Jack Chesbro (198–132 to 187–143), and Smiling Mickey Welch (307–210 to 286–231) might not be as Happy or Smiling anymore, on the outside of Cooperstown looking in.

 

Table A: The Luckiest: (Most Wins over Projection, 1876–2006)

Player	W	L	Proj W	Diff	StdDev
Mickey Welch	307	210	286.4	+20.6	+2.47
Greg Maddux	333	203	315.2	+17.8	+2.21
Bob Welch	211	146	194.2	+16.8	+2.52
Clark Griffith	237	146	221.8	+15.2	+2.17
Christy Mathewson	373	188	357.9	+15.1	+1.85
Roger Clemens	348	178	334.0	+14.0	+1.78
Harry Gumbert	143	113	129.0	+14.0	+2.43
Randy Johnson	280	247	266.4	+13.6	+1.97
Bill Hutchison	183	163	169.5	+13.5	+1.93
Ed Morris	171	122	157.9	+13.1	+2.12

 

Table B: The Unluckiest: (Most Wins under Projection, 1876–2006)

Player	W	L	Proj W	Diff	StdDev
Red Ruffing	273	225	297.3	-24.3	-2.98
Jim McCormick	265	214	279.7	-14.7	-1.90
Dizzy Trout	170	161	183.7	-13.7	-2.12
Bob Shawkey	195	150	208.4	-13.4	-1.95
Walter Johnson	417	279	430.1	-13.1	-1.39
Bert Blyleven	287	250	299.1	-12.1	-1.43
Murry Dickson	172	181	182.9	-10.9	-1.61
Ned Garver	129	157	139.9	-10.9	-1.80
Sid Fernandez	114	96	125.5	-10.5	-1.91
Bob Friend	197	230	207.5	-10.5	-1.34

 

Table C: Others of Interest (Through 2006)

Player	W	L	Proj W	Diff	StdDev
Grover Alexander	373	208	364.9	+8.1	+0.95
Bob Gibson	251	174	249.9	+1.1	+0.14
Sandy Koufax	165	87	160.8	+4.2	+0.73
Pedro Martinez	206	92	201.2	+4.8	+0.86
Nolan Ryan	324	292	318.5	+5.5	+0.67
Curt Schilling	207	138	210.4	-3.4	-0.45
Tom Seaver	311	205	305.0	+6.0	+0.74
John Smoltz	193	137	200.6	-7.6	-1.23
Warren Spahn	363	245	366.6	-3.6	-0.34
David Wells	230	148	222.6	+7.4	+1.02
Cy Young	511	316	511.9	+0.9	-0.10

 

In a clutch performance, BILL DEANE pitched his team to the Cooperstown Co-Ed Softball League Playoff Championship in 2006.

 

Notes

1. In this case—a binomial distribution—a standard deviation is the square root of P₃Q₃N, where P is the probability of success (50%), Q is the probability of failure (1 – P, or again 50%), and N is the number of tries (100). So the standard deviation here is 5. Finding the standard deviation (or sigma) for expected wins is a much more complex process, varying from pitcher to pitcher based on his number of decisions and his winning percentage. The average sigma in this group is 6.2 wins.