The main things that a pitcher really gives up are home runs and walks.   The rest of it depends on the defense, right?   The pitcher gives up some runs—when he gives up a home run and when he walks people—but for the rest, it’s up to the defense.

            Yeah, I know about strikeouts and hit batsmen; we’ll get there.   I’m just saying. . .real simple idea.  What if we ranked pitchers by how many walks and home runs they give up?  How well would that match with more sophisticated performance measures?

            Pretty well, it turns out. . .you can be the judge of that, but I was kind of astonished at how well it did.  I ranked pitchers by this simple metric:

            Walks

            + Hit Batsmen

            + 4 Times Home Runs

            Per 9 innings

            I call it Herbie, for HR BB. This is SUCH a simple concept that somebody must have done this before, but. .it’s new to me.   When I was thinking about this, walking home from the park, I assumed that it probably wouldn’t work because it would either discriminate against pitchers like Robin Roberts, Catfish Hunter, Ferguson Jenkins, Eric Milton and Bartolo Colon, who give up lots of home runs but win by not walking people, or it would discriminate against pitchers like Whitey Ford, Dave Stewart, Fernando Valenzuela and Kerry Wood, who give up some walks but stay away from the long ball. 

            And it does discriminate, but not too badly.  Robin Roberts led the National League in Herbie in ’51 and  ’52, was close his other good years, while Whitey Ford led the League in Herbie in ’64, and was third in the AL in ’56 and ’58.   Actually, it does real well. . .obviously there are limits to what you’re going to get out of a formula like this, but it seems like the pitchers Herbie identifies as the best in the league are the best in the league fairly often.  

            I compared the Herbie leader to the ERA leaders for the years 1930-1959—a total of 60 competitions.   In those years there were ten pitchers who led their league in both ERA and Herbie (Dazzy Vance, 1930, Carl Hubbell, 1933, 1934 and 1936, Cy Blanton, 1935, Spud Chandler, 1943, Harry Brecheen, 1948, Warren Spahn, 1953, Mike Garcia, 1954 and Bob Friend, 1955.)    There were ten more seasons in which the league leader in ERA was second in Herbie, and (oddly enough) ten more in which he was third.   Altogether, the league leader in ERA finished 1-3 in Herbie in 30 of the 60 leagues, finished 1-5 in ERA in 37 of the 60 leagues, and finished 1-10 in Herbie in 53 of the 60 leagues.   Many times the pitcher who led the league in Herbie seems like a more reasonable candidate for the league’s best pitcher than the actual ERA leader.  In 1959, for example, Hoyt Wilhelm was the ERA leader in the American League (15-11, 2.19), whereas Camilo Pascual was the Herbie leader (17-10, 2.64).  I think most people would look at those seasons and conclude that Pascual was actually the better pitcher.  In 1951 the National League ERA leader was Chuck Nichols (11-8, 2.88), but the Herbie leader was Robin Robers (21-15, 3.03), and in 1952, combining those two, the NL ERA leader was Wilhelm (15-3, 2.43), but the Herbie leader was again Roberts (28-7, 2.59). 

            As the leagues have gotten larger, of course, the likelihood of the Herbie leader and the ERA leader being the same person has gotten less, but the relationship doesn’t seem any less close.   I looked at all pitchers pitching 180 innings from 1950 to 2006.  The four best Herbie scores in that group were by Greg Maddux in 1994, 1995, 1996 and 1997.   (That last sentence is astonishing, but it’s an astonishing statement about Greg Maddux, rather than about Herbie, and I’m writing about Herbie, so I don’t have anywhere to go with that.)  The worst was by Jose Lima in 2000 (7-16, 6.65 ERA, 48 home runs allowed in 196 innings.)

            The next question I asked was whether Herbie could be translated into ERA.  I was trying to keep it simple.   The simple formula for translating Herbie into ERA is:

            (Herbie – 1) / 2 + 1 = ERA

            In other words, a Herbie score of 8 equals an ERA of 4.50:

            (8 – 1) / 2 + 1 = 4.50

            And a Herbie score of 7 equals an ERA of 4.00.   The best Herbie score of 1950-2006 was 2.36, by Maddux in 1994.  A Herbie of 2.36 equals an ERA of 1.68.  His actual ERA was 1.56.   The worst Herbie score of that era was 12.01+, by Lima.  A Herbie of 12.01+ is equivalent to an ERA of 6.51.  His actual ERA was 6.65.

            If this translation always worked this well, Herbie would become an important sabermetric tool.  Unfortunately it doesn’t.   It works very well for groups.   A Herbie score of 6 translates to an ERA of 3.50.   If you take all pitchers with 180 innings, 1950-2006, there are 205 pitchers with Herbie Scores of 5.90 to 6.10.   Their average Herbie score is 6.00, and their average ERA is 3.51—almost exactly what the translation would predict.  The translation works well for many, many pitchers—for example, Bret Saberhagen, 1989, had a Herbie ERA of 2.16, and an actual ERA of 2.16.   Fernando Valenzuela in 1986 had a Herbie ERA of 3.14, and an actual ERA of 3.14.   Don Cardwell, 1960, and Cliff Chambers, 1951, both had Herbie ERAs of 4.38, and actual ERAs of 4.38.   Kevin Gross, 1995, had a Herbie ERA of 5.52, and an actual ERA of 5.54. 

            But there are also many pitchers whose actual ERA is not near their Herbie ERA, and these discrepancies are neither random, meaningless, nor transient.   Take, for example, those 205 pitchers who have Herbie scores between 5.90 and 6.10.  They include:

Steve Carlton, 1969     Herbie Score:  5.98      Herbie ERA:  3.49        Actual ERA:  2.17

Tom Seaver, 1969        Herbie Score:  6.09      Herbie ERA:  3.55        Actual ERA:  2.21

Joe Horlen, 1968          Herbie Score:   5.95     Herbie ERA:  3.47        Actual ERA:  2.37

Camilo Pascual, 1963 Herbie Score:     6.10     Herbie ERA:  3.55        Actual ERA:  2.47

Randy Johnson, 2001  Herbie Score:  5.95       Herbie ERA:  3.47        Actual ERA:  2.49

John Burkett, 1998       Herbie Score:  6.00      Herbie ERA:  3.50        Actual ERA:  5.68

Mark Clark, 1998        Herbie Score:  6.07      Herbie ERA:  3.53        Actual ERA:  4.84

Joe Blanton, 2006        Herbie Score:  6.07      Herbie ERA:  3.53        Actual ERA:  4.82

Bart Johnson, 1976      Herbie Score:  6.10      Herbie ERA:  3.55        Actual ERA:  4.73

Allan Anderson, 1990 Herbie Score:  5.92        Herbie ERA:  3.46        Actual ERA:  4.53

            John Burkett’s 1998 season is, in a sense, a fluke.   The discrepancy there (2.18 runs) is easily the largest discrepancy between a pitcher’s Herbie ERA and his actual ERA among the 3,449 pitchers in the study.  Nobody else was over 1.92. 

            But there are meaningless flukes and there are meaningful flukes.  This is a meaningful fluke.   Maybe Carlton had some good luck in 1969, but the difference between Carlton and Burkett is not luck, but is, rather, the many facets of performance that we have not encompassed within this measurement.   Burkett’s actual ERA was always worse than his Herbie ERA, every year he pitched 180 innings except one.  Steve Carlton’s actual ERA was better than his Herbie ERA every year from 1969 through 1981, although not usually a run better.  

            Obviously, the main thing we haven’t considered yet is strikeouts.  Most of the pitchers who had better actual ERAs than Herbie ERAs were power pitchers; most of the pitchers on the other end were on the other end.    Couldn’t we adjust Herbie for strikeouts, and come up with a stat that is a better measure of the pitcher’s true value?

            Well, yes, we could.  But we’re crossing a line there.   In fact, I’m going to draw a line here to indicate that we’re crossing a line.  I’ll see you on the other side of the line.

      =========================================================

            There are two kinds of invented stats in sabermetrics.   There are serious analytic methods, and there are little tools that are fun to play around with but don’t really mean much, like Secondary Average, Game Scores, Season Scores and Frustration Scores for baseball games.  

            Herbie is just a fun thing to play around with.   You can figure it in 20 seconds; you can convert it into an ERA in another 20 seconds.  “How do you most accurately predict a pitcher’s ERA, based on his strikeouts, walks and home runs allowed?” is a serious sabermetric question.   If we were going to take it to that level, we’d have to do a whole bunch of studies and tests and stuff, review the formulas that already exist for that sort of thing and take seriously the issue of whether the new one is better than the old ones.   I’m not really up for that. 

            Well, OK, here’s how you can do that, sort of.   Replace Herbie with Herbie-2:

            Innings Pitched Times 3, minus strikeouts

            Divided by 3

            Plus Walks

            Plus Hit Batsmen

            Plus 4 times Home Runs

            Times 9

            Divided by Innings Pitched

            Now the best Herbie-2 score since 1950 is no longer Maddux, it’s Pedro Martinez in 1999.  

            You can convert Herbie-2 into ERA by this formula:

            (H2 – 5) * .424

            That formula is more accurate, for predicting ERA, than the Herbie ERA.  It’s way off for Martinez in 1999 (1.30 vs. 2.07), but on average, it’s more accurate.

            But not dramatically more average.   Herbie ERA has an average error, for pitchers pitching 180 innings or more, of 0.47.   This formula has an average error of 0.41.   I’m sure we could improve the formula more by messing around with the weight of a home run vs. a walk, but that’s real work, and I don’t foresee a big payoff to it.

            You can never be sure what is a serious sabermetric tool and what is just something to mess around with, because many times the things you think you’re messing around with turn out to be extremely useful when you’re seriously studying something.  But we have two Herbies here, one which is very simple but not tremendously accurate at predicting ERA, and one which is a lot more complicated but not tremendously accurate at predicting ERA, either.   For my purposes, I’ll stick with the simple one.