Pitching and the Persistence of Excellence

April 7, 2018
                         Pitching and the Persistence of Excellence
 

              This is driven by the opening days of the 2018 baseball season, in which we have seen that Chris Sale is still a really good pitcher, and David Price also looked like a really good pitcher at least in his first two starts, while other very good pitchers have looked like they perhaps are not as good as they used to be.  There is some magic about it. . .about these opening days, but also about the fact that a player was really good last year and still is this year.  It’s hard to explain, but part of the legerdemain of baseball is the fear that a player who was really good last year will not be good this year.  A baseball fan is omnianxious.  He is afraid that the Felix Hernandez of 2015 will become the Felix Hernandez of 2016, that the Cliff Lee of 2013 will become the Cliff Lee of 2014, that the Roy Halladay of 2011 will become the Roy Halladay of 2012.  It is certain to happen sometime.  This pandemic fear that we have, that our gold will turn to lead, makes us watch the opening days of the baseball season with the hopeful dread of a banker discovering that his vault has been left open overnight.

              The purpose of this research is to examine the factors that go into sustaining excellence as a pitcher.  Of course, these issues have been examined a thousand times before in various studies and we may not learn anything new, but there is an answer to that, too.   So long as we lack perfect understanding, we must continue to do research. 

              The exact two questions we are asking are:

              1)  If a pitcher is excellent in one season, what are the odds that he will be excellent again in the next season, and

              2)  How does this vary with various inputs?

              But before we get to those, we have to start with questions like "What are we going to define as an excellent pitcher?"

              I’m going to define an "excellent" pitcher for purposes of this study in terms of Earned Runs Allowed, and specifically, I am going to say that a pitcher is excellent if:

              1)  His ERA is at least 10% better than the league average, and

              2)  He allows at least 7 runs fewer than a league-average pitcher would allow.

              The "7 runs" criteria has two purposes.   First, it excludes pitchers who pitch 3 innings and don’t allow a run, or pitchers who pitch 10 innings with a 1.80 ERA, from being included on the "excellent pitcher" list.  Second, it gives me a "slide indicator" that I can position to get the percentages I want.   I wanted to define about 20% of pitchers as excellent, and I wanted to define "10% better than the league" as "excellent".   By adding this second figure, the "runs better" figure, I can cut it off wherever I need to, to get both 20% of the pitchers and pitchers with ERAs 10% better than the league.   It happens that 7 runs better is the serendipitous point.   By these two cutoffs, 19.7% of all pitchers in the study are identified as "excellent".  

              The 20% standard is generous enough that almost every pitcher who is around for a few years will sometimes be identified as "excellent".   20% is not a Hall of Fame standard; it’s a "good pitcher that year" standard.   Don Aase, always at the top of the file, has three seasons in his 14-year career in which he is identified as "excellent".  

              It also happens that these two measurements tend often to be about the same, for the very best pitchers.   I used three figures to define the list:  M1, M2 and M3.  

              M1 is the pitcher’s ERA, divided by the league ERA, subtracted from 1.00, multiplied by 100.

              M2 is the number of earned runs a league-average pitcher would have allowed in the number of innings pitched by this pitcher, minus the number of runs that this pitcher did allow. 

              M3 is the minimum of M1 and M2. 

              My study group was "all pitchers who recorded at least one out in a season from 1920 to 2015"—eliminating pitchers who got no one out, because division by zero is messy and creates problems, and they’re not relevant to the study although they could be marginally relevant to the percentages, starting at 1920 because the game was too different before 1920 to be relevant, not including 2017 because we need to back it off a year so that we have a "look ahead year" in the data, and not including 2016 because I haven’t yet updated this Excel file I am using to include the 2017 data, so the last year I can use is 2016.   I wish I could park-adjust the data, but I don’t have the ability to do that in this file.

              Anyway, when we sort pitchers by M3, we find that the best pitchers usually have nearly identical M1 and M2 figures.  The #1 pitcher by M3 is Pedro Martinez, year 2000, whose figures are 65, 77, 65, which would not qualify as "nearly identical".  But the #2 pitcher is Bob Gibson, 1968 (62, 63, 62), #3 is Greg Maddux, 1994 (63, 60, 60), #4 is Greg Maddux, 1995 (61, 59, 59), #7 is Zack Greinke, 2015 (58, 56, 56), and #8 is Roger Clemens, 2005 (56, 55, 55).  

              An interesting feature here, right away, is that most of the very top seasons are recent. . .in your memory if you are over 50.   These are the top 25 seasons ranked by M3:

             

First

Last

Team

Lg

Year

Pedro

Martinez

Boston Red Sox

AL

2000

Bob

Gibson

St. Louis Cardinals

NL

1968

Greg

Maddux

Atlanta Braves

NL

1994

Greg

Maddux

Atlanta Braves

NL

1995

Dwight

Gooden

New York Mets

NL

1985

Pedro

Martinez

Boston Red Sox

AL

1999

Zack

Greinke

Dodgers

NL

2015

Roger

Clemens

Houston Astros

NL

2005

Kevin

Brown

Florida Marlins

NL

1996

Roger

Clemens

Toronto Blue Jays

AL

1997

Pedro

Martinez

Montreal Expos

NL

1997

Dean

Chance

Los Angeles Angels

AL

1964

Jake

Arrieta

Cubs

NL

2015

Ron

Guidry

New York Yankees

AL

1978

Lefty

Grove

Philadelphia Athletics

AL

1931

Sandy

Koufax

Los Angeles Dodgers

NL

1966

Dolf

Luque

Cincinnati Reds

NL

1923

Zack

Greinke

Royals

AL

2009

Warren

Spahn

Milwaukee Braves

NL

1953

Roger

Clemens

Boston Red Sox

AL

1990

Carl

Hubbell

New York Giants

NL

1933

Randy

Johnson

Seattle Mariners

AL

1997

Clayton

Kershaw

Dodgers

NL

2013

Lefty

Gomez

New York Yankees

AL

1937

Tom

Seaver

New York Mets

NL

1971

 

              A lot of recent seasons in there.   This doesn’t always happen.  There are a million ways to rank pitchers, but many times when I draw up a list, it seems to favor players from years ago.   For some reason the opposite happens at the top of this list. 

              So this brings up an immediate question:  Is there some sort of "recency bias" here?  I didn’t intend for there to be a bias for one era over another; that’s why I used ERA relative to the league norm; it’s supposed to make things constant over time.

              And it appears that it does, generally; for some reason there are just a lot of recent seasons at the top.  Part of it is that I am now old, so what seems recent to me is actually more than half of baseball history.    Anyway, looking at the data by decades, the percentage of pitchers who are identified as "excellent" never goes lower than 18.5% (current decade) and never goes higher than 21.0% (1970s).   This chart gives the percentage of pitchers in each decade who are identified as excellent, and the percentage of those pitchers who are identified as excellent in one season who will repeat on the list the next season:

From

To

 

Ex Pct

Rpt %

1920

1929

 

19.1%

43.7%

1930

1939

 

19.6%

39.3%

1940

1949

 

19.8%

36.8%

1950

1959

 

18.7%

40.5%

1960

1969

 

18.8%

40.9%

1970

1979

 

21.0%

39.0%

1980

1989

 

20.2%

36.8%

1990

1999

 

20.0%

38.6%

2000

2009

 

20.5%

39.8%

2010

2015

 

18.5%

40.8%

 

              Over time, 39.4% of pitchers who are identified as excellent in one season will also be identified as excellent in the following season.   There are 6,861 pitcher/seasons in the study which are identified as "excellent", which is 19.7% of all pitchers in the study.  Of those 6,861 pitchers, 2,703 were again identified as excellent in the following season.  Both of these figures—19.7%, and 39.4%--are essentially constant over time, or at least stable. The "repeat percentage" dropped in the 1940s for an obvious reason, World War II.  If you would like to argue with me about what constitutes an "excellent" pitcher, feel free; I will just mark you down as a moron who didn’t understand the purpose of the research, and make a mental note to ignore anything you say in the future.

              We thus can draw our first conclusion:   A pitcher’s chance of being excellent in one season is more than twice as great if he was excellent in the previous season as they it is if he was not excellent in the previous season.   39.4% is exactly twice 19.7%, not "more than twice", but my statement is true nonetheless, for a reason that you can figure out if you work on it for a moment.

              In the 1920s, 82% of pitchers who were identified as "excellent" were starting pitchers, 5% were relievers, and 13% were in a mixed role, meaning that they had between 10% and 60% of their appearances as starts, and 40% to 90% in the bullpen.   Over time, these percentages have changed enormously, so that since 1990 there are more excellent relievers than excellent starting pitchers—and pitchers who are excellent in a mixed role have virtually disappeared:

 

From

To

% Mx

% Rel

% St

1920

1929

13.2%

5.3%

81.5%

1930

1939

15.1%

7.1%

77.8%

1940

1949

10.8%

13.0%

76.2%

1950

1959

11.5%

22.6%

65.8%

1960

1969

7.0%

35.2%

57.9%

1970

1979

5.8%

38.3%

55.8%

1980

1989

5.1%

47.4%

47.4%

1990

1999

3.0%

53.8%

43.1%

2000

2009

2.5%

58.3%

39.2%

2010

2015

2.4%

57.0%

40.6%

 

              So in modern baseball 41% of excellent pitchers are starters, 57% are relievers, and 2% make a transition during the season from one role to the other.   Another thing that has changed is the number of pitchers per team:

From

To

Pitchers

Teams

Pit/Team

1920

1929

394

160

2.46

1930

1939

397

160

2.48

1940

1949

454

160

2.84

1950

1959

442

160

2.76

1960

1969

560

200

2.80

1970

1979

736

248

2.97

1980

1989

816

260

3.14

1990

1999

1027

278

3.69

2000

2009

1282

300

4.27

2010

2015

753

180

4.18

 

              Modern teams use many, many more pitchers in a season than was true a hundred years ago.  Since the percentage of all pitchers who are excellent is the same, this means that many more pitchers per team are defined as "excellent" for the season. 

              Now to get to our serious question:  How does the "repeat percentage" vary with various inputs?   Is it different for starters than relievers, for example?

              A little.  The "repeat percentage" is 42% for starting pitchers, 37% for relievers, and 30% for pitchers in a mixed role. 

              I had expected that the "mixed role" percentage would be low, since those are mainly. . .well, pitchers who are not settled into a role, not fully proven.  Before 1975, a huge percentage of league ERA champions were these mixed-role pitchers, who moved in the season from one role to the other.   In 1957 the American League leader in ERA was Bobby Shantz, in 1959 Hoyt Wilhelm, in 1960 Frank Baumann, in 1962 Hank Aguirre, in 1970 Diego Segui, in 1972 Luis Tiant, all mixed-role pitchers.  The next year their ERAs went up, in every case, by a run or something very close to a run.  What happened was, a young pitcher or a pitcher who had been fighting injuries would start the year in the bullpen, and make 25 to 35 relief appearances with an ERA of 1.00 or 1.25.   About June 1 to July 1 the team would need starting pitching, so they would move him into the starting rotation, and he would continue to pitch moderately well making 15 to 20 starts, and winding up the season with a 2.50 ERA—and just skimming over the standard of 154 or 162 innings to qualify for the ERA title.   But he’s not really an outstanding starter, to begin with, and also, he’s not properly trained to be a starting pitcher; he’s doing something that is outside his conditioning program.   By the end of the year, very often he’s just holding on, and the next year he is not the same guy. 

              We don’t do that in modern baseball; we don’t rush guys from the bullpen into the starting rotation in mid-season.   But anyway, by 1972 I recognized the syndrome, so I expected the "repeat percentage" for mixed-role pitchers to be low.   In checking the facts, I see that in 1959 Hoyt Wilhelm was not technically a mixed-role pitcher, he was a starter.  If you point out things like this to me, I will also mark you down as a moron; I don’t care.  I’m addressing the general question, not the question of whether Hoyt Wilhelm was a starter or a mixed-role pitcher in 1959.  

              Probably the percentage for relievers is lower than the percentage for starters because there are more pitchers there who are closer to the margins of the definition.  Remember the "M3" definition—the minimum of the two numbers that define excellence.  Later we will study this, but there is no doubt that pitchers whose M3 figure is 30 are going to be more likely to repeat as excellent pitchers than pitchers whose M3 figure is 8.  Because starting pitchers pitch more innings, they have larger M3 figures, thus more "safe" pitchers who are not near the margins of the definition. 

              In the 1970s, when I started doing baseball research, it was an article of faith among baseball writers that relievers were much less consistent than starters.   Older writers would wax grand and eloquent about the phenomenal consistency of Rollie Fingers, as a reliever, when most relievers were good one year and not the next.   There were two reasons that this was generally believed.  One was that, in the 1950s, when these sportswriters were born in the business, most relievers were guys with second-line stuff, junkballers, and they may well have been somewhat less consistent.  Guys like Jim Konstanty and Luis Arroyo came out of nowhere in their 30s, had monster seasons, and returned the next season to the mists of mediocrity.  The other reason was that if you don’t study things, you form all kinds of beliefs that turn out, with research, just to be wrong.  These guys didn’t study anything, so they believed a lot of nonsense.  Even if you go back to the relievers of the 1930s, 1940s, if you study it with modern methods and adjust for the instability of the data in smaller samples, you’ll find that there’s not a lot of difference between the consistency of starters and that of relievers.  It’s actually always been pretty much the same.  Also, I know there is one of you out there who wants to argue with me about Rollie Fingers’ phenomenal consistency.  Feel free to make your argument; I’ll make a note.

              While we’re here. . . it’s not really what we’re here for, but I suppose I should give you a list of the pitchers who have the most seasons which are considered excellent by our present definition.  Roger Clemens leads the list, with 19 seasons, followed by Tommy John and Mariano Rivera, 17, Warren Spahn with 16, Nolan Ryan, Tom Seaver, Hoyt Wilhelm, John Smoltz, Bob Feller and Whitey Ford, 15, Bert Blyleven, Tom Glavine, Curt Schilling and Gaylord Perry with 14, and Don Sutton, John Franco, Billy Wagner, Lefty Grove and Pedro Martinez with 13.   Mostly Hall of Famers. 

              Now to the issue of age. . . .64% of teenagers who had excellent seasons were able to repeat as excellent pitchers the next year, or 9 out of 14. (You are probably wondering who the teenagers were who qualified as "excellent" pitchers.   The list is Bob Feller in 1936, 1937 and 1938, Ralph Branca in 1945, Don Drysdale in 1956, Wally Bunker and Billy McCool in 1964, Larry Dierker in 1966, Gary Nolan in 1967, Don Gullett and Bert Blyleven in1970, Fernando Valenzuela in 1980, which was the year before Fernandomania, Dwight Gooden in 1984, and Felix Hernandez in 2005.)

              This chart summarizes the percentage of excellent pitchers at each age who are able to repeat at the next age:

 

 Age

Repeat 

Pct 

<19

14

9

64%

20

33

19

58%

21

104

37

36%

22

189

80

42%

23

322

119

37%

24

444

178

40%

25

574

248

43%

26

690

255

37%

27

626

245

39%

28

600

225

38%

29

558

243

44%

30

537

209

39%

31

451

182

40%

32

386

149

39%

33

315

132

42%

34

251

79

31%

35

186

79

42%

36

173

65

38%

37

119

44

37%

38

101

41

41%

39

70

31

44%

40

51

13

25%

41

26

7

27%

42+

41

14

34%

 

              The numbers go up and down a little, but there is no obvious pattern.   This conclusion is re-inforced when we group the numbers at age 25 with those at age 24 and 26:

 Age

 #

Repeat 

Pct 

 #

Repeat 

Pct 

<19

14

9

64%

47

28

60%

20

33

19

58%

151

65

43%

21

104

37

36%

326

136

42%

22

189

80

42%

615

236

38%

23

322

119

37%

955

377

39%

24

444

178

40%

1340

545

41%

25

574

248

43%

1708

681

40%

26

690

255

37%

1890

748

40%

27

626

245

39%

1916

725

38%

28

600

225

38%

1784

713

40%

29

558

243

44%

1695

677

40%

30

537

209

39%

1546

634

41%

31

451

182

40%

1374

540

39%

32

386

149

39%

1152

463

40%

33

315

132

42%

952

360

38%

34

251

79

31%

752

290

39%

35

186

79

42%

610

223

37%

36

173

65

38%

478

188

39%

37

119

44

37%

393

150

38%

38

101

41

41%

290

116

40%

39

70

31

44%

222

85

38%

40

51

13

25%

147

51

35%

41

26

7

27%

118

34

29%

42+

41

14

34%

67

21

31%

 

              The percentage of excellent pitchers repeating as excellent is the same at age 38 as it is at age 25.   Age is not an obvious variable in a pitcher’s chance of repeating a successful season.

              Well, then. . .what DOES age mean, for a pitcher?   It’s not that easy to say.   There is a narrative established in my head, so I see it the way I see it.   It’s basically a game of attrition.  If you have 1,000 successful pitchers at age 32, 30% of them will get hurt at age 33, leaving 700, 30% will get hurt at age 34, leaving 490, etc., until they’re all gone.  But IF a pitcher is still successful, it’s pretty much irrelevant to a third party whether the pitcher is 33 or 40—not absolutely, but pretty much.

              What changes is his ability to come back.  A 26-year-old pitcher gets hurt, he may have time to re-build his career.  A 33-year-old has less time to come back, and a 40-year-old is out of time.

              Let’s turn our attention now to the strikeout rate.   The strikeout rate can be looked at either in absolute terms, strikeouts per nine innings, or relative to the league.   We’ll start with strikeouts in absolute terms.   I sorted the 6,861 pitchers according to their strikeouts per nine innings; the highest strikeout rate in the group was by Aroldis Chapman in 2014 (17.67 strikeouts per nine innings) and the lowest was by Ernie Wingard in 1924 (0.95 strikeouts per nine innings).   Then I divided the pitchers into ten deciles of 686 pitchers each, except that the fifth group from the top had 687 pitchers.  

              Of the top 686 pitchers in strikeout rate, 376 repeated as successful pitchers in the following season.   Of the bottom 686 pitchers, only 223 repeated as successful pitchers in the following season:

 

 

Highest Strikeout Group

55%

2nd Highest

47%

 

43%

 

42%

 

37%

 

36%

 

34%

 

33%

2nd Lowest

33%

Lowest Strikeout Group

33%

             

              Obviously the strikeout rate influences the "repeat success" rate, as we would expect that it would based on hundreds of previous studies.   But what if, instead, we sorted the pitchers by strikeout rate relative to the league strikeout rate?  Would that improve the sorting power of this factor?

              It does, yes:

Highest Strikeout Group

50%

2nd Highest

50%

 

49%

 

44%

 

40%

 

39%

 

36%

 

33%

2nd Lowest

30%

Lowest Strikeout Group

25%

 

              We sort the data by deciles and then measure the standard deviation of the decile measurements.  If the measurement becomes more effective, the top groups move further away from the bottom ones, and the standard deviation of the decile measurments increases.   It does increase, and markedly, from .074 to .089.   I refer to this, in the rest of this article, as "74 to 89". 

              If strikeouts sort the pitchers by the likelihood of future success, what about walks?   This chart sorts the pitchers in the same way based on raw walk rates:

 

Lowest Walk Rate

48%

2nd Lowest

45%

 

43%

 

45%

 

39%

 

38%

 

36%

 

34%

2nd Highest

37%

Highest Walk Rate

31%

 

              Here the standard deviation of the decile measurements is .054 (54).  The walk rate has significant effectiveness as an indicator of the ability to remain successful, but less power than the strikeout rate.  The wildest pitcher who qualified as successful in a season was Tommy Byrne in 1949, who walked 179 men in 196 innings, but posted a 3.72 ERA.  The low-walk guy was Carlos Silva in 2005, who walked only 9 in 188 innings. Let’s do walks per nine innings compared to the league norm:

 

Lowest Walk Rate

49%

2nd Lowest

44%

 

44%

 

44%

 

40%

 

39%

 

34%

 

35%

2nd Highest

36%

Highest Walk Rate

30%

 

              Here again the effectiveness of the measurement increases when we adjust for the league norm, although the increase is not as large; our standard deviation here is 57.  Probably adjusting for the league walk rate has less impact than adjusting for the league strikeout rate because there is much less historic variation in walk rates than there is in strikeouts.   Let’s do strikeout/walk ratio, raw:

 

Best Strikeout/Walk Ratio

57%

2nd Best

49%

 

46%

 

41%

 

38%

 

37%

 

34%

 

35%

2nd Worst

31%

Worst Strikeout/Walk

26%

 

              This is the most effective predictor of the ability to remain effective that we have found, with a standard deviation as discussed above of 91. 

              Now I have two directions I can go in.  I have a method that I developed probably ten years ago of comparing the strikeout to walk ratio to the league norm, called the strike zone winning percentage, and I’ll look at that in a moment.   But Tom Tango argues that strikeout to walk ratio is not the optimal evaluator of strikeouts and walks, for this reason.   Suppose that two pitchers pitch 200 innings each.  One pitcher strikes out 100 batters and walks 25, a 4-to-1 ratio.   The other pitcher strikes out 150 batters and walks 75, a 2-to-1 ratio.   Are these pitchers the same in their impact, or different?

              Tom argues that these pitchers have different strikeout to walk ratios, but essentially the same impact on the runs allowed column, since, he says, the benefit of a strikeout is more or less equal to the cost of a walk.   He thus argues that what matters is the strikeout to walk margin.  

              So that sub-divides, too; we could study strikeout to walk margin, raw, which we will do later, or strikeout to walk margin per inning, which we will do now.   Strikeout to walk margin, per inning:

Best K-W Margin

58%

2nd Best

50%

 

42%

 

42%

 

40%

 

32%

 

33%

 

37%

2nd Worst

34%

Worst K-W Margin

27%

 

Ted Wingfield in 1925 had a better-than-league ERA although he had 30 strikeouts and 92 walks, a margin of 2.91 more walks than strikeouts per nine innings.   Anyway, this does in fact score higher on our test than the strikeout to walk ratio, 92 to 91 (.092 to .091, standard deviation of the deciles.)   The best 10% of these pitchers, in their strikeout margin per inning, are 58% likely to remain excellent in the following season, which is the best percentage we have yet seen.

              Strike zone winning percentage is the pitcher’s strikeouts, multiplied by the league walks (or walks per game, it doesn’t matter), divided by the same plus the league strikeouts multiplied by the pitcher’s walks.  In other words, using P-SO for pitcher strikeouts and L-SO for League strikeouts:

              K Zone Winning Pct  = (P-SO * L-BB) / [(P-SO * L-BB) + (L-SO * P-BB)]

              The only pitcher in the data who has a K Zone winning percentage of .900 is Dennis Eckersley, who was at .916 in 1989—and again in 1990, the same percentage.  The best by a starting pitcher is .849, by Bret Saberhagen in 1994.    Previous studies have shown that the strike zone winning percentage correlates very well with the pitcher’s actual winning percentage.

              Anyway, the strike zone winning percentage—which is the strikeout/walk ratio, but league-adjusted—blows everything else we have seen so far out of the water in terms of its effectiveness at predicting continued excellence:

Best K Zone Win Pct

59%

2nd Best

54%

 

48%

 

47%

 

41%

 

35%

 

33%

 

33%

2nd Worst

27%

Worst K Zone Win Pct

20%

              That’s a score of 122—by far the best we have seen.  We have the highest retention percentage on one end, the lowest on the other.

              Well.  . .let’s see.  If the "Score" of Strikeout to Walk Ratio is 91, Strikeouts Minus Walks per inning is 92, and league-adjusted strikeout to walk ratio is 122, then how about league-adjusted Strikeouts Minus Walks per 9 innings?  Do I need to write out the formula, or is it self-‘splanatory?   Better safe than sorry. . . .

              [(P-SO -– P-W) /  P-IP ]  /   [(L-SO -– L-W) / L-IP]

              Well (skipping chart), that scores at just 92. . .it turns out that there is an obvious problem with that.   There are some leagues in which the difference between the strikeout rate and the walk rate is only .03 to .05 per nine innings.   A pitcher with an advantage of 1.00 strikeouts above walks per game, in such a league, comes out at 20 or 30.  Ewell Blackwell in 1947 comes out at 107.69.   This causes that formulation to lose its predictive power.

              We can solve that problem in one of two ways.   First, we could switch to SUBTRACTING the league average from the pitcher’s data, rather than dividing by it. . ..I’ll try that first:

Largest Relative Margin

61%

2nd Largest

54%

 

48%

 

45%

 

40%

 

34%

 

34%

 

31%

2nd Lowest

30%

Worst Relative Margin

18%

 

              Oooh. . .that scores at 127.   Nice. 

              The other option is to divide the pitcher’s strikeout margin by the league strikeout margin, PLUS X.   In other words, before we divide the pitcher’s strikeouts minus walks per nine innings by the league data, we add something to the league data so that it can’t be near zero.  

              What is X?   What is the number that you add to the league data to maximize the result?   It is easy to experiment.   It turns out that the maximum result is achieved by adding 1.20.   1.20  works better than 1.10 or 1.30; 1.20 works better than 1.19 or 1.21 This is the data for that:

Best Relative K Margin

63%

2nd Best

54%

 

49%

 

43%

 

39%

 

34%

 

30%

 

32%

2nd Worst

27%

Worst Relative K Margin

24%

 

              That scores at 126. . . basically the same as subtracting the league number, but not better than it.   I promised earlier to check the impact of just using raw strikeouts minus walks, which would be +311 for Sandy Koufax in 1965 (382 and 71) and -42 for Dick Fowler in 1949 (43 strikeouts, 115 walks.)  This is the data for that sort:

Best Strikeout Margin (Raw)

62%

2nd Best

50%

 

47%

 

45%

 

38%

 

37%

 

33%

 

33%

2nd Worst

26%

Worst Strikeout Margin (Raw)

24%

 

              That scores at 117, which is a very good number.  Let’s check the impact of the Home Run Rate. Raw Home Run rate:

 

Lowest HR allowed rate

37%

2nd Lowest

43%

 

45%

 

40%

 

40%

 

40%

 

41%

 

39%

2nd Highest

37%

Most HR Allowed

33%

 

              Nothing there so far. . .that scores at 33.  Let’s move on to Home Run Rate compared to league norm.   Slick Castleman in 1937 was very effective despite a home runs allowed rate that was six times the league average:

 

Lowest HR allowed rate

39%

2nd Lowest

42%

 

43%

 

40%

 

41%

 

38%

 

42%

 

39%

2nd Highest

36%

Most HR Allowed

35%

 

              Well, that’s even worse.  There doesn’t appear to be anything there that helps the prediction.   I’m moving on. 

              Let’s look at excellence retention by innings pitched.   The left-hand column below is the average innings pitched by the pitchers in the group:

 

282

Most Innings Pitched

59%

240

2nd Most

50%

216

 

41%

192

 

36%

157

 

33%

118

 

35%

93

 

35%

78

 

38%

67

2nd Fewest

40%

47

Fewest Innings Pitched

26%

             

              That scores at 93, which means that it actually has considerable value as a predictor of the ability to sustain excellence, although not as much as the strikeout/walk data.  59% of the pitchers who were excellent in 255.1 innings or more were able to repeat as excellent the next season, whereas only 26% of those who were excellent in less than 58 innings were able to repeat. 

              OK, and here’s what I expect to be the biggest separator.  Margin of excellence.   Remember the M1, M2, M3 figures, from the start of this article 13 pages ago?

 

              M1 is the pitcher’s ERA, divided by the league ERA, subtracted from 1.00, multiplied by 100.

                  M2 is the number of earned runs a league-average pitcher would have allowed in the number of innings pitched by this pitcher, minus the number of runs that this pitcher did allow. 

                  M3 is the minimum of M1 and M2. 

 

              One would expect pitchers who are HIGHLY excellent to be more likely to retain excellence than pitchers who just scraped by the qualifying standard, right?   The minimum M3 figure to be described as "excellent" is 7.   In the chart below the left-hand figure is the average M3 of the group:

34

Highest M3 figure

62%

24

2nd Highest

53%

20

 

43%

17

 

41%

14

 

40%

13

 

35%

11

 

36%

10

 

28%

9

2nd Lowest

31%

7

Lowest M3 Figure

25%

 

              That scores at 113.  It turns out that the M3 figure is a good indicator for the retention of excellence—but not the best we have seen.  I thought it would be the best indicator we had seen, but it isn’t.   Also, it turns out that the M1 figure (which is simply the ERA relative to the league norm) is a very poor indicator of the ability to retain excellence (41), whereas the M2 figure is a very good indicator (118), so that when you mix M1 and M2 together into M3, it’s actually a poorer indicator than M2 by itself.  This is M2:

38

Highest M2 figure

65%

25

2nd Highest

51%

20

 

45%

17

 

40%

15

 

39%

13

 

35%

11

 

36%

10

 

28%

9

2nd Lowest

31%

7

Lowest M2 Figure

25%

 

              Before we move into the wrapup stages here, another thing I should look at is consecutive seasons of excellence.   Obviously, it would seem, a pitcher coming off of five consecutive seasons of excellence must be a better bet to repeat as excellent than a pitcher coming off of one good year, right?   So let’s look at that. . .

              Mariano Rivera had 16 consecutive seasons of excellence, the most ever.   In the chart below, "10" means "10 or more":

Consecutive

Count

Repeat

Pct

10

36

21

58%

9

23

15

65%

8

34

23

68%

7

49

35

71%

6

80

51

64%

5

138

84

61%

4

283

140

49%

3

568

293

52%

2

1417

592

42%

1

4233

1449

34%

 

              Pitchers coming off their first (consecutive) season of excellence are 34% likely to repeat as excellent the next year.   Although the data is irregular because the numbers get to be small, this increases generally until a pitcher having his 7th consecutive excellent season is 71% likely to have an eighth. 

              OK, so the best indicators I have found are the M2 figure and pitcher’s strikeouts minus walks margin, minus the league margin.   The final question, then, is how can we combine these two into one number which is more effective than either one of them by itself?

              M2 operates on a scale of 7 to 77, with a standard deviation of 9.49.   The Strikeout Margin operates on a scale of -5.1 to +10, with a standard deviation of 1.78.  If we simply add the two together, then, the M2 figure will dominate, and we won’t make any progress over M2 by itself. 

              Since the strikeout margin is the better predictor of the two (127 to 118), we want the strikeout margin to have more impact on the combination than the M2 figure.   If we multiply the Strikeout Margin number by six before adding it to the M2, that would make a standard deviation ratio of 10.7 to 9.5, about the right ratio.   So let’s try that. . . a newly created figure, M2 plus six times the pitchers strikeout margin minus the league norm:

Best Manufactured Number

71%

2nd Best

55%

 

45%

 

48%

 

40%

 

34%

 

31%

 

26%

2nd Worst

27%

Worst Manufactured Number

18%

 

              The predictive power of this manufactured number by the method I have been using is 158—much higher than any previous number we have seen.

              So a pitcher who has a manufactured number in the top 10% of all excellent pitchers has a 71% chance to repeat as excellent, while a pitcher who has a manufactured number in the bottom 10% has only an 18% chance to repeat as excellent—which is to say, less than a randomly selected pitcher.  A randomly selected pitcher has a 20% chance to be excellent this season; a pitcher who was excellent last year but who did not have a large margin of excellence (M2) and did not have a good strikeout to walk margin has only an 18% chance to repeat as excellent. 

              The final question I addressed was whether I could improve the predictive power of this manufactured number by factoring in the number of consecutive excellent seasons the pitcher has had.   I was not able to do so.   All of my attempts to improve this number by factoring in the number of consecutive excellent seasons made the predictor less powerful, rather than more, and so I abandoned the effort.  

              OK, then; our "Excellence Repeat Predictor" is:

              The pitchers strikeouts minus his walks, per nine innings,

              Minus the league average of the same,

              Times 6,

              Plus the number of runs the pitcher has saved compared to a league-average pitcher, as measured by ERA.  

              In all of baseball history, the ten pitchers most likely to repeat as excellent were:

             

First

Last

Year

Pedro

Martinez

2000

Pedro

Martinez

1999

Randy

Johnson

2001

Randy

Johnson

1999

Randy

Johnson

2002

Randy

Johnson

2000

Pedro

Martinez

2002

Pedro

Martinez

2003

Lefty

Grove

1931

Mariano

Rivera

2008

 

              All ten of them did in fact repeat as excellent, except for Randy Johnson in 2002.  In tomorrow’s article I will review the excellent pitchers of 2017, and assign "likely repeat scores" to each one. 

 

Tomorrow’s Article

              I was going to publish these two articles on Monday and Tuesday of next week, but it occurred to me that some people might have fantasy drafts over the weekend and might be angry with me if I held off publishing this.  

              Suppose that we use the method above to

              1)  Identify everyone who qualifies as an excellent pitcher in 2017, and

              2)  Estimate, based on the method explained here, that pitcher’s probability of repeating as excellent again in 2018.

 

              We have to extrapolate the method just a little bit, but it’s pretty straightforward.   I think there were 137 "excellent" pitchers in 2017.   These are detailed below, with an estimate for each pitcher of his likelihood of repeating excellence in 2018.   Pitchers are listed by the team that they were with at the end of the 2017 season.

              I am sure I must have included here some pitcher who has since had Tommy John surgery, and consequently has no chance at all to be excellent in 2018.   My apologies. 

 

 

ANGELS

Name

IP

ER

SO

BB

ERA

Estimate

Bridwell,Parker

121.0

49

73

30

3.64

16%

Petit,Yusmeiro

91.3

28

101

18

2.76

58%

Parker,Blake

67.3

19

86

16

2.54

64%

             

ASTROS

Name

IP

ER

SO

BB

ERA

Estimate

Verlander,Justin

206.0

77

219

72

3.36

50%

Morton,Charlie

146.7

59

163

50

3.62

38%

Keuchel,Dallas

145.7

47

125

47

2.90

35%

Peacock,Brad

132.0

44

161

57

3.00

53%

Devenski,Chris

80.7

24

100

26

2.68

57%

Giles,Ken

62.7

16

83

21

2.30

61%

             

ATHLETICS

Name

IP

ER

SO

BB

ERA

Estimate

Blackburn,Paul

58.7

21

22

16

3.22

15%

             

BLUE JAYS

Name

IP

ER

SO

BB

ERA

Estimate

Stroman,Marcus

201.0

69

164

62

3.09

41%

Happ,J.A.

145.3

57

142

46

3.53

31%

Leone,Dominic

70.3

20

81

23

2.56

45%

Osuna,Roberto

64.0

24

83

9

3.38

63%

             

BRAVES

Name

IP

ER

SO

BB

ERA

Estimate

Ramirez,Jose

62.0

22

56

29

3.19

17%

Freeman,Sam

60.0

17

59

27

2.55

23%

Vizcaino,Arodys

57.3

18

64

21

2.83

33%

             

BREWERS

Name

IP

ER

SO

BB

ERA

Estimate

Davies,Zach

191.3

83

124

55

3.90

17%

Nelson,Jimmy

175.3

68

199

48

3.49

54%

Anderson,Chase

141.3

43

133

41

2.74

48%

Suter,Brent

81.7

31

64

22

3.42

20%

Swarzak,Anthony

77.3

20

91

22

2.33

58%

Knebel,Corey

76.0

15

126

40

1.78

73%

Hughes,Jared

59.7

20

48

24

3.02

17%

Hader,Josh

47.7

11

68

22

2.08

56%

             

CARDINALS

Name

IP

ER

SO

BB

ERA

Estimate

Martinez,Carlos

205.0

83

217

71

3.64

39%

Lynn,Lance

186.3

71

153

78

3.43

23%

Nicasio,Juan

72.3

21

72

20

2.61

37%

Lyons,Tyler

54.0

17

68

20

2.83

43%

Brebbia,John

51.7

14

51

11

2.44

38%

Tuivailala,Samuel

42.3

12

34

11

2.55

21%

             

CUBS

Name

IP

ER

SO

BB

ERA

Estimate

Arrieta,Jake

168.3

66

163

55

3.53

32%

Hendricks,Kyle

139.7

47

123

40

3.03

36%

Montgomery,Mike

130.7

49

100

55

3.38

18%

Edwards Jr.,Carl

66.3

22

94

38

2.98

41%

Duensing,Brian

62.3

19

61

18

2.74

31%

Strop,Pedro

60.3

19

65

26

2.83

29%

Davis,Wade

58.7

15

79

28

2.30

48%

             

DIAMONDBACKS

Name

IP

ER

SO

BB

ERA

Estimate

Greinke,Zack

202.3

72

215

45

3.20

65%

Ray,Robbie

162.0

52

218

71

2.89

69%

Walker,Taijuan

157.3

61

146

61

3.49

28%

Godley,Zack

155.0

58

165

53

3.37

42%

Bradley,Archie

73.0

14

79

21

1.73

56%

Hernandez,David

55.0

19

52

9

3.11

32%

             

DODGERS

Name

IP

ER

SO

BB

ERA

Estimate

Darvish,Yu

186.7

80

209

58

3.86

39%

Kershaw,Clayton

175.0

45

202

30

2.31

77%

Wood,Alex

152.3

46

151

38

2.72

61%

Hill,Rich

135.7

50

166

49

3.32

52%

Ryu,Hyun-Jin

126.7

53

116

45

3.77

21%

Jansen,Kenley

68.3

10

109

7

1.32

80%

Watson,Tony

66.7

25

53

20

3.38

18%

Baez,Pedro

64.0

21

64

29

2.95

23%

Fields,Josh

57.0

18

60

15

2.84

35%

Avilan,Luis

46.0

15

52

22

2.93

26%

Morrow,Brandon

43.7

10

50

9

2.06

51%

             

GIANTS

Name

IP

ER

SO

BB

ERA

Estimate

Bumgarner,Madison

111.0

41

101

20

3.32

34%

Gearrin,Cory

68.0

15

64

35

1.99

24%

Strickland,Hunter

61.3

18

58

29

2.64

21%

             

INDIANS

Name

IP

ER

SO

BB

ERA

Estimate

Kluber,Corey

203.7

51

265

36

2.25

81%

Carrasco,Carlos

200.0

73

226

46

3.28

67%

Clevinger,Mike

121.7

42

137

60

3.11

33%

Shaw,Bryan

76.7

30

73

22

3.52

27%

Allen,Cody

67.3

22

92

21

2.94

60%

Miller,Andrew

62.7

10

95

21

1.44

74%

McAllister,Zach

62.0

18

66

21

2.61

34%

Otero,Dan

60.0

19

38

9

2.85

19%

Goody,Nick

54.7

17

72

20

2.80

49%

Olson,Tyler

20.0

0

18

6

0.00

26%

             

MARINERS

Name

IP

ER

SO

BB

ERA

Estimate

Leake,Mike

186.0

81

130

37

3.92

20%

Paxton,James

136.0

45

156

37

2.98

62%

Diaz,Edwin

66.0

24

89

32

3.27

40%

Vincent,Nick

64.7

23

50

13

3.20

22%

Zych,Tony

40.7

12

35

21

2.66

16%

             

MARLINS

Name

IP

ER

SO

BB

ERA

Estimate

Urena,Jose

169.7

72

113

64

3.82

16%

Barraclough,Kyle

66.0

22

76

38

3.00

24%

Steckenrider,Drew

34.7

9

54

18

2.34

55%

             

METS

Name

IP

ER

SO

BB

ERA

Estimate

deGrom,Jacob

201.3

79

239

59

3.53

59%

Blevins,Jerry

49.0

16

69

24

2.94

44%

             

NATIONALS

Name

IP

ER

SO

BB

ERA

Estimate

Gonzalez,Gio

201.0

66

188

79

2.96

47%

Scherzer,Max

200.7

56

268

55

2.51

78%

Strasburg,Stephen

175.3

49

204

47

2.52

75%

Kintzler,Brandon

71.3

24

39

16

3.03

16%

Albers,Matt

61.0

11

63

17

1.62

46%

Madson,Ryan

59.0

12

67

9

1.83

65%

Doolittle,Sean

51.3

16

62

10

2.81

54%

             

ORIOLES

Name

IP

ER

SO

BB

ERA

Estimate

Givens,Mychal

78.7

24

88

25

2.75

43%

Brach,Brad

68.0

24

70

26

3.18

27%

Bleier,Richard

63.3

14

26

13

1.99

16%

             

PADRES

Name

IP

ER

SO

BB

ERA

Estimate

Chacin,Jhoulys

180.3

78

153

72

3.89

18%

Stammen,Craig

80.3

28

74

28

3.14

25%

Hand,Brad

79.3

19

104

20

2.16

70%

             

PHILLIES

Name

IP

ER

SO

BB

ERA

Estimate

Nola,Aaron

168.0

66

184

49

3.54

46%

Neris,Hector

74.7

25

86

26

3.01

40%

Garcia,Luis

71.3

21

60

26

2.65

22%

Milner,Hoby

31.3

7

22

16

2.01

15%

             

PIRATES

Name

IP

ER

SO

BB

ERA

Estimate

Rivero,Felipe

75.3

14

88

20

1.67

66%

Schugel,A.J.

32.0

7

27

14

1.97

17%

             

RANGERS

Name

IP

ER

SO

BB

ERA

Estimate

Cashner,Andrew

166.7

63

86

64

3.40

15%

Claudio,Alex

82.7

23

56

15

2.50

27%

             

RAYS

Name

IP

ER

SO

BB

ERA

Estimate

Cobb,Alex

179.3

73

128

44

3.66

22%

Faria,Jake

86.7

33

84

31

3.43

25%

Colome,Alex

66.7

24

58

23

3.24

20%

Hunter,Tommy

58.7

17

64

14

2.61

44%

Cishek,Steve

44.7

10

41

14

2.01

28%

             

RED SOX

Name

IP

ER

SO

BB

ERA

Estimate

Sale,Chris

214.3

69

308

43

2.90

79%

Pomeranz,Drew

173.7

64

174

69

3.32

37%

Reed,Addison

76.0

24

76

15

2.84

42%

Price,David

74.7

28

76

24

3.38

29%

Kimbrel,Craig

69.0

11

126

14

1.43

82%

Kelly,Joe

58.0

18

52

27

2.79

18%

Maddox,Austin

17.3

1

14

2

0.52

29%

             

REDS

Name

IP

ER

SO

BB

ERA

Estimate

Castillo,Luis

89.3

31

98

32

3.12

35%

Iglesias,Raisel

76.0

21

92

27

2.49

52%

             

ROCKIES

Name

IP

ER

SO

BB

ERA

Estimate

Gray,Jon

110.3

45

112

30

3.67

30%

Rusin,Chris

85.0

25

71

19

2.65

31%

Neshek,Pat

62.3

11

69

6

1.59

68%

             

ROYALS

Name

IP

ER

SO

BB

ERA

Estimate

Duffy,Danny

146.3

62

130

41

3.81

25%

Minor,Mike

77.7

22

88

22

2.55

50%

Alexander,Scott

69.0

19

59

28

2.48

21%

Buchter,Ryan

65.3

21

65

26

2.89

26%

             

TIGERS

Name

IP

ER

SO

BB

ERA

Estimate

Fulmer,Michael

164.7

70

114

40

3.83

18%

Greene,Shane

67.7

20

73

34

2.66

28%

             

TWINS

Name

IP

ER

SO

BB

ERA

Estimate

Santana,Ervin

211.3

77

167

61

3.28

36%

Berrios,Jose

145.7

63

139

48

3.89

24%

Rogers,Taylor

55.7

19

49

21

3.07

19%

Busenitz,Alan

31.7

7

23

9

1.99

17%

             

WHITE SOX

Name

IP

ER

SO

BB

ERA

Estimate

Infante,Gregory

54.7

19

49

20

3.13

19%

Giolito,Lucas

45.3

12

34

12

2.38

19%

             

YANKEES

Name

IP

ER

SO

BB

ERA

Estimate

Severino,Luis

193.3

64

230

51

2.98

71%

Gray,Sonny

162.3

64

153

57

3.55

30%

Montgomery,Jordan

155.3

67

144

51

3.88

23%

Sabathia,CC

148.7

61

120

50

3.69

20%

Green,Chad

69.0

14

103

17

1.83

76%

Robertson,David

68.3

14

98

23

1.84

71%

Kahnle,Tommy

62.7

18

96

17

2.59

72%

Betances,Dellin

59.7

19

100

44

2.87

47%

Warren,Adam

57.3

15

54

15

2.35

32%

 

 

 

 

 
 

COMMENTS (28 Comments, most recent shown first)

Arrojo
MarisFan61, in the Bill James handbook, they do project pitchers' stats for the upcoming season and many of the pitchers listed are young with more minor league exposure than major league exposure.

I buy it every year - Bill used to say that he had nothing to do with the pitcher projections. I think last year he finally came around to endorsing them. But the methodology is never explained (nor should it be - it's proprietary)! For all we know, the minor league to major league projections for these guys incorporate some of the suggestions you've made (and maybe even the hitter projections as well - the last time he discussed the methodology for that I think was in the old Bill James Abstracts).

So perhaps he is chuckling at us, one step ahead...

If I ever get some time - maybe this summer - I'll go back and use Bill's method from this article and apply it to the projected K/BB from the handbook (Bill does caution that projecting summary stats - W, L, ERA is an "impossible task" but projecting Ks and BB's are more realistic (I'm paraphrasing), so the ERA required for this task may cause the whole thing to break down. (Not to mention he is using last year's K/BB rather than this year's projected K/BB).
12:03 PM Apr 10th
 
MarisFan61
P.S. to Arrojo -- Maybe to clarify it better:
Remember, my first reply to you was that while the commonly available data on hitting do quite well project hitters from the minors to the majors, I don't think it's capable of reliably predicting what you asked about on pitchers (i.e. K and BB rates in the majors) and I said a little about why not, and suggested some 'deeper' data that might do a better job on it.
It's just an opinion. :-)

I don't see that anyone else has followed up on it, either the incapability of commonly available data nor the possibly better capability of those things I mentioned.
4:27 PM Apr 9th
 
MarisFan61
(Arrojo: I didn't say it wouldn't apply for hitters. I simply didn't address it, because it's not what we're talking about here.
I could be a good separate subject, but I think not right here -- and I'd also add, while it might help on hitters too, and maybe people like Bill and others working for teams are already using such things (!) on both pitchers and hitters, the common stats already do a quite-good job on hitters.)
11:56 AM Apr 9th
 
Brian
" I will just mark you down as a moron who didn’t understand the purpose of the research, and make a mental note to ignore anything you say in the future. "

Before you ignore anything in the future, shouldn't there be a study on "moron repeat percentage"?
11:36 AM Apr 9th
 
Zeke**
Bill, when you say excellent relievers are less likely to repeat because they are "closer to the margins," is that just an effect of relievers' smaller sample size, or something about their skill distribution?

I know it's more likely for a mediocre pitcher to randomly exceed the study's standard of excellence by a few runs over 80 innings than over 180, but I wasn't sure if the 7 run limit or some other aspect of the study controlled for that...
9:05 AM Apr 9th
 
Arrojo
MFan61, yes I know you were only talking about the potential issues involved in projecting minor league pitchers' walk and strikeout rates into major league performance.

My point was, Bill James has already developed a method for projecting these minor league stats for hitters to the majors. If we already have a method for projecting hitters' minor league K and BB rates, why are the items you point out (record against the good hitters, swing and miss rates, etc) only applicable to pitchers' projections? That's another way of phrasing 'why can't we use the same method for projecting pitchers' K and BB rates from the minors'? You threw out some additional factors - if they are not needed for hitter projections (record against the good pitchers) why are they needed for pitcher projections?
8:06 AM Apr 9th
 
jemanji
Another super article, thank you.

I'm surprised that the chances for star pitchers to be "excellent" are as low as they are. A very interesting insight.


1:39 AM Apr 9th
 
MarisFan61
???

(I didn't at all address predicting hitters!
I didn't say anything, 'for' or 'against,' about those things predicting hitting. I simply wasn't addressing it, because this is about pitchers.)
8:20 PM Apr 8th
 <