This bit of research actually began with a comment in the "Hey, Bill" section posted by a reader just before Felix Hernandez won the Cy Young Award, despite his lackluster record of being credited with wins and blamed for losses.   The post suggested that the "switch" in Cy Young voting patterns was not really a "switch", but the culmination of a long-term trend that could be seen as early as 1987, when Nolan Ryan drew some support in the NL Cy Young voting (12 points, tied for 5^th) despite a won-lost record of 8-16.    Ryan had lost two-thirds of "his" decisions, but had led the league in strikeouts (270) and ERA (2.76).  As far back as then, the reader suggested, voters were beginning to pay attention to something other than wins.

            When I saw that post—which unfortunately I cannot find—I thought, "Well, I should be able to measure that."  I should be able to measure, in any vote, how much weight is given to each category of performance, thus document how those things change over time.

            I came up with a method to do this, somewhat imperfectly, and I’ll explain the method in a minute.   But first, a brief history of the Cy Young vote, just to make sure that everybody is on the same page.

 

I.  A Brief History

 

            The Cy Young Award began in 1956, with the first award going to Don Newcombe, who had gone 27-7 for the Brooklyn Dodgers.   From 1956 to 1966 only one award was given each year, to the best pitcher in baseball, not the best pitcher in the league.   It was done that way because Ford Frick, who was then the Commissioner, felt strongly that it should be that way.   Frick thought that if you gave a Cy Young Award in each league, then sportswriters would start thinking that there was an award for pitchers and an award for position players, the MVP Award, and they would no longer vote for pitchers for the MVP Award.   Which, basically, turned out to be true.

            There have been 99 Cy Young Awards given—one a year 1956 to 1966 (11), and two a year 1967 to 2010 (88).

            In the first years of the vote, there was a very simple voting system in which only one sportswriter covering each franchise voted, and he voted for only one pitcher.    The "modern" system of Cy Young voting came into existence in two stages.   In 1967—shortly after the retirement of Ford Frick—the BBWAA began giving two Cy Young Awards each year, one for each league.   With that change they doubled the number of voters to two per team, but continued to use the very simple one man, one vote ballot.

            In the American League Cy Young vote in 1969, this system (predictably) malfunctioned; it wasn’t predictable that it would happen in the American League in 1969, but it was foreseeable that it would happen in some league before too long.  With only 24 voters voting, it was very easy to get a tie in the voting, with no way to resolve the tie.   In 1969 Denny McLain (24-9, 2.80 ERA) got 10 votes, and Mike Cuellar (23-11, 2.38 ERA) got 10.   The other four votes went to other pitchers.   Each pitcher got a half of a Cy Young Award; as I understand it Denny McLain got the right half and Mike Cuellar got the left half, but that’s not important right now.

            The MVP vote had always (back to 1911) used a "ranked" ballot, in which each writer listed several choices for the award.  This works much better.   It works better simply because it collects much more information, and you have a better chance of getting the right answer when you have more information.  After 1969 the BBWAA switched to a "ranked" ballot.   (Of course, as most of you probably know, beginning in 2010 the vote switched from a three-man to a four-man ballot.   That’s better still, but not hugely significant, at least yet.)

            There is a third relevant change here, but it’s hard to explain.   From 1963 to 1968, most of the Cy Young votes are obvious.   Sandy Koufax won the award unanimously in 1963, 1965 and 1966.  In 1968 Denny McLain was unanimous in the American League, and Bob Gibson was unanimous in the National.    In 1967 the votes weren’t unanimous, quite, but the winner in each league won 18 of the 20 votes, and in 1964 the winner won 17 out of 20.

            We still occasionally have votes like that, of course, but it is much more the exception than the rule—and has been since 1969.  But as a consequence of these three factors—the unified, two-league award, the simple ballot, and the domination of a few pitchers—we really have very little information about the thinking of Cy Young voters before 1970.   For each vote now we can do a pretty decent job of diagnosing what was considered important, but before 1970 we just don’t have much to work with.

 

 

II.  Analytical Method

 

I set up a system in which, if the player who was first in the league in a category also won the Cy Young Award, the player who was second in the category was second in the voting, the player who was third was third, and on down the line, then that would be considered to be a 100% agreement between the category and the Cy Young voting.

            The last time this actually happened was in the National League voting in 1988.   These were the National League leaders in Wins in 1988:

 

1.  Orel Hershiser	23
Danny Jackson	23
3.  David Cone	20

 

            And this was the Cy Young vote:

 

1.  Orel Hershiser	120 points
2.  Danny Jackson	54 points
3.  David Cone	42 points

 

            These were the only three players mentioned in the voting; all 24 voters listed those three players, and only those three.  There is no clear disagreement between the "wins" category and the Cy Young voting, so we measure it as 100% agreement.

            Zero percent agreement, on the other hand, would be. . .well, saves in 2009.   In 2009 no pitcher who earned even one save was mentioned in the Cy Young voting in either league.   In other words, saves—based on the 2009 voting—appear to have NO impact on the vote.  That’s zero percent.

            This is the actual system, the details.   There are four stages in each measurement:

 

            1)  Define the field of competitors,

            2)  Calculate the potential score if the votes and the category line up perfectly,

            3)  "Score" the match for the category, and

            4)  Divide the score by the potential.

 

            The field of competitors I defined as "all pitchers eligible for the award who have a Season Score of at least 100."    There typically are about 40 to 50 pitchers in a league who will have a Season Score of 100 or higher.   This was defined by a guess, but as it turned out, there was only one pitcher in history—Lee Smith in the strike-shortened 1994 season—who has been mentioned in the Cy Young voting with a Season Score of less than 100.    He got one point in the voting.

            To calculate the potential score, line up the vote-getters like this:

 

            This is the data for the American League vote in 2010.   Multiply the top vote-getters’ total by 25, the #2 man by 24, #3 by 23, etc.   Like this:

 

            If the other category lines up perfectly with the Cy Young vote—which hasn’t happened since 1988--the total will be 11,148.

            Next, we line up the "category" totals for the category being tested.   Let’s say Innings Pitched.   We line up the Innings Pitched for the American League pitchers in 2010 who are considered Cy Young candidates.   It makes this:

 

            We give 25 points to the top man, 24 to the #2 man, etc.   Which makes this:

 

            And we multiply that by the pitcher’s points in the Cy Young vote:

 

 

            This total (9,020) we divide by the league potential (11,148), making .809.   In the American League in 2010 there was a 80.9% agreement between Innings Pitched and Cy Young votes.

            That’s a pretty high figure, but look at the data.   The pitcher who led the league in Innings Pitched also won the award.    The pitcher who was second in innings (Sabathia) was third in the Cy Young voting.   They line up reasonably well.

 

            I have to explain just two little wrinkles in the system. . .well below my norm (17.2 wrinkles and/or exceptions requiring explanation per report.)    The two little wrinkles are:

            1)  When two players tie in a category, the player who did better in the Cy Young voting is always listed first.   In other words, since Jered Weaver and Justin Verlander each pitched 224 1/3 innings, we list Weaver first because he got more points in the Cy Young voting, and

            2)  If a player has ZERO in the category (zero saves, or zero complete games or zero shutouts), he can’t be listed among the top 25 in the league, even if there are not 24 players in the league who did have at least one of those.   In some leagues there aren’t 25 pitchers who threw a shutout, or 25 pitchers who recorded a save, but we don’t give "points" to a pitcher based on zeroes.

 

III. The Study

 

            I tested the importance of 23 statistics in each of the 99 Cy Young contests.   Some of these are "raw" categories like wins, ERA and strikeouts, and some of them are "manufactured" categories created by putting data together.  Eight of them were raw statistical categories:

 

            1)  Wins

            2)  Saves

            3)  ERA

            4)  Innings Pitched

            5)  Complete Games

            6)  Shutouts

            7)  Strikeouts

            8)  Walks

 

            Four were simple, first-level statistical extrapolations:

 

            1)  Strikeouts Per Nine Innings

            2)  Walks Per Nine Innings

            3)  Strikeout to Walk Ratio

            4)  Home Runs Allowed Per Nine Innings

           

 

            And the other eleven were more complicated statistical extrapolations:

 

            1)  WL1           Won-Lost record (stated as 5 * Wins – 3 * Losses)

            2)  WL2           Won-Lost record (stated as 5* Wins – 2 * Losses)

            3)  WLS1         (1) on this list plus Saves

            4)  WLS2         (1) on this list plus 2 * Saves

            5)  WLS3         (2) on this list plus Saves

            6)  RSA6         The number of Earned Runs Saved by the pitcher as compared to a pitcher

                          &nb​sp;                     with a 6.00 ERA.

            7)  RSA7         Earned Runs Saved by the pitcher as compared with a 7.00 ERA.

            8)  RSALg       Earned Runs Saved by the pitcher as compared to the league ERA

            9)  Leveraged RSAL

                                 ​;   The Same as (8) on this list, but multiplied by (IP + 2* Saves) / IP

            10)  SOBB       (2 * SO) – (3 * BB)

            11)  Season Score        A standard method that I use to rank pitcher seasons

 

           

IV.  The Results of the Study

 

            We can first dismiss from the study some of these eleven statistical extrapolations in the last group; they were things that I thought might work, but didn’t.   I had initially thought that WL1 (Wins times 5 minus losses times 3) was about the "right" weight of losses versus wins, but it clearly isn’t.   The most instructive example is the 1962 Cy Young vote, in which the three leading contenders were Don Drysdale (25-9), Jack Sanford (24-7), and Bob Purkey (23-5).   If you line those three up by WL1, they go Purkey (100), Sanford (99), Drysdale (98).    If you line them up by WL2 (wins times 5 minus losses times 2) they go Drysdale (107), Sanford (106), Purkey (105).   In fact, the voting went Drysdale-Sanford-Purkey, so WL2 has a value for that vote of .986, whereas WL1 is at .935.    There’s not a big difference between them, but WL2 is a better predictor than WL1, so we can just get rid of WL1.

            We’ll get rid of WL1, WLS1, WLS2, and RSA7, as those turned out to be just less-accurate versions of other "compiled" stats.

 

            Warning you now that I am about to give you a misleading statistic.

            Among the basic statistics, Wins have been, as you might guess, by far the best predictor of Cy Young voting performance—and remain so today.   Over the entire history of the award, wins are 88.9% predictive of Cy Young voting performance.

            Focusing just on the last ten awards, wins are 88.5% predictive.

            There have been recent awards, yes, in which Wins did not seem to carry much weight.   In the 2010 Cy Young vote, Wins come in at just .716, and in the National League in 2006, the award won by Brandon Webb with Trevor Hoffman second in the voting, they come in at just .665.   But overall, wins are not meaningfully less predictive of Cy Young voting performance in recent history than they have been overall.

            The reason this is true is that the major "disruption" to the pattern of the biggest winners winning the Cy Young Awards is actually not the recent trend to consider other starting pitchers.   The biggest disruption to that pattern was the era in which relievers won Cy Young Awards.

            From the beginning of the Award in 1956 to 1973—an era of 26 awards--no reliever won the Cy Young Award, and, in general, relievers were rarely mentioned in Cy Young voting.  In this era, there was a 96.2% agreement between Wins and Cy Young Votes.

            From 1974 to 1992—an era of 38 awards--eight relievers won the Cy Young Award:

            1974 NL  Mike Marshall

            1977 AL  Sparky Lyle

            1979 NL  Bruce Sutter

            1981 AL  Rollie Fingers

            1984 AL  Willie Hernandez

            1987 NL  Steve Bedrosian

            1989 NL  Mark Davis

            1992 AL  Dennis Eckersley

 

            And many, many other relievers were prominently mentioned in Cy Young voting.   In general, some reliever would be a serious Cy Young candidate as often as not.   These relievers did not have high Win totals, and, in this era, there was only an 83.6% agreement between Wins and Cy Young voting performance.

            Since 1993—an era of 36 awards--only one reliever has won a Cy Young award (Eric Gagne in 2003), and relievers are mentioned in Cy Young voting less and less often.   Since 1993, Wins are 89.4% predictive of Cy Young voting performance.

            I began by saying that Wins, over the entire history of the award, are 88.9% predictive of Cy Young votes, whereas in the last five years (10 awards) they are 88.5% predictive.   But this is misleading, because the predictive significance of wins, for starting pitchers, was much higher than that for much of the award’s history.

 

            This is the data for each of the eight "basic" categories studied:

 

 

 

 

            Saves, even in the era when some relievers won the Cy Young Award, were never remotely equal to wins in importance in the Cy Young voting.   In the American League in 1984, not only did a reliever (Willie Hernandez) win the award, but another reliever (Dan Quisenberry) finished second in the voting.   In that race, Saves measure as 63.5% predictive of Cy Young votes.    In the National League in 1974, when Mike Marshall won and three other relievers were mentioned in the voting, they measure at .555, and in three other awards won by relievers (1981 AL, 1989 NL and 2003 NL) they measure at greater than 50%.  Wins, on the other hand, measure as more than 50% predictive of voting in 95 of the 99 elections.

            There is a barrier to "raw ERA" measuring as highly predictive, which is that there are normally a half-dozen "non serious" candidates who have extremely low ERAs, but no chance to win the Cy Young Award.   In 2009, for example, the three best ERAs in the National League were by Hong-Chih Kuo (1.20), Billy Wagner (1.43) and Mike Adams (1.76), none of whom was even mentioned in the Cy Young voting (although I don’t quite understand why Wagner wasn’t.)   In the American League the best ERAs were by Andrew Bailey (1.47) and Chris Perez (1.71), neither of whom was mentioned.   There have always been pitchers like this; in 1956 Marv Grissom had an ERA of 1.56.   These pitchers cause the measured relationship between ERA and Cy Young votes to be fairly low, although (as you can see) ERA has become more significant over time.

            Just explaining. .."Walks", above, are measured in the same way as if walking people was a positive thing.    As you can see, Cy Young winners throughout history have not infrequently had high or very high walk totals, simply because more walks come with more innings pitched, and innings pitched are the second-best predictor of Cy Young performance, among these eight basic stats.    In 1958 Bob Turley won the Cy Young award despite leading the major leagues with 128 walks.  In 1959 Early Wynn did the same thing, leading the majors with 119 walks, but winning the Cy Young Award, while Sam Jones led the National League in walks, but was second in the Cy Young voting, causing the measured link between walks and Cy Young votes for 1959 to be .942.   This still happens, but it happens less, and I’ll discuss this more in a moment. 

 

            Among the four "simple calculation" categories, the most predictive has been the strikeout to walk ratio, but there has been a striking increase, over time, in the importance of home runs allowed.

 

 

            HRF is "Home Run Frequency".  In the early history of the award, Home Run frequency was not at all predictive of Cy Young success.   In 1956 Don Newcombe won the first award despite allowing 33 home runs, second-highest total in the majors.    In 1968 Denny McLain won 31 games and was unanimously voted the American League Cy Young Award despite leading the majors in home runs allowed, also with 31.    Home Run frequency was less than 10% predictive of Cy Young votes in 9 of the first 26 Awards.

            Over time, although raw home run totals really have not increased all that much, home run frequency has become much more predictive of Cy Young success.   In 1956 there were 1,239 major league games and 2,294 home runs, or 185 home runs per 100 games.  In 2010 there were 2,430 games and 4,613 home runs, or 190 home runs per 100 games.   Again, more on this in a moment.

            Looking now at the data for the "formula" predictors.   There were 11 formulas tested, of which five were dismissed earlier in this report.   The data for the other six is as follows:

 

 

 

 

            I was looking at the question of how well the won-lost record, unaided, predicts Cy Young voting.  I started with 5W-3L, and realized that 5W-2L would work better, but. . .that doesn’t work great, either.    The measured predictive value of wins, unaided, was .889; of the won-lost record, .883.

            Wins only measure as more predictive than the won-lost record because of the way we treat ties.   Ties, in this system, are always interpreted as favorable to the match.   There are more ties in "wins" than in "won-lost record", so that causes wins to measure higher than the won-lost record.    But even allowing for that. ..losses don’t seem to mean much.   I was surprised by this; this, to me, was one of the largest surprises in the data, that LOSSES don’t seem to have very much meaning in Cy Young voting.   22-4; 22-12.   It’s all about the same.

            Wins, Losses, and Saves, however, can easily be combined into a simple predictive metric; 5W – 2L + Sv is 93.6% predictive of the Cy Young vote.   This, however, has become progressively less reliable over time, and it appears that this is still happening, that this is still becoming less reliable.

            RSA6 is Runs Saved versus a 6.00 ERA .  A pitcher who pitched 225 innings with a 6.00 ERA would allow 150 earned runs.   When Jason Schmidt in 2004 pitched 225 innings and allowed 80 earned runs, then, he was 70 runs better than a pitcher with a 6.00 ERA—70 RSA6.

            RSA6 is, essentially, a stand-in for ERA.   I understood before doing the study that ERA could not measure as highly predictive of Cy Young voting because of the problem with pitchers posting 1.50 ERAs in 60 meaningless innings.   What could predict Cy Young performance, however, is Runs Saved—that is, pitching a lot of innings with a low ERA.

            RSA6 is much more predictive than ERA (.855 vs. .605), but it still is not highly predictive—or anyway has not been until the last few years.   Over the history of the award there have been many pitchers with outstanding innings & ERA combinations who basically never showed up in Cy Young voting.

            In 1956 Early Wynn had the best innings & ERA combination, pitching more innings than Don Newcombe with a better ERA, and Wynn did go 20-9, so it’s not like he was a zero in that area, either.   But Wynn was not mentioned in the Cy Young voting.

            In 1958, when Bob Turley won the Cy Young Award with a 2.98 ERA, his teammate Whitey Ford had a 2.01 ERA, leading the majors in RSA6.  Ford was not mentioned in the Cy Young vote.

            In 1959 Warren Spahn led the majors in RSA6 and won 21 games as he always did, pitching more innings than the Cy Young Award winner (Wynn) with a better ERA, but was not mentioned in the Cy Young voting.

            In 1960 Don Drysdale led the majors in RSA6, pitching essentially as many innings as the Cy Young Award winner (Vern Law) with a better ERA, but was not mentioned in the Cy Young voting.

            In 1964 Drysdale again led the majors in RSA6, but again was not mentioned in the Cy Young voting.

            In 1969 Bob Gibson led the National League in RSA6 (314 innings with a 2.18 ERA), pitching more innings than the Cy Young winner (Tom Seaver) and with a better ERA, but was not mentioned in the Cy Young voting. 

            And look at the names of the pitchers who were slighted—Early Wynn, Whitey Ford, Warren Spahn, Don Drysdale, Bob Gibson.   These were Hall of Fame pitchers, the biggest names of their era—but in seasons in which they merely had fantastic combinations of innings pitched and ERA, but without great won-lost records, they were entirely ignored by the Cy Young voters.

            In recent years this does appear to have definitively changed.   Felix Hernandez, Zack Greinke and CC Sabathia have won American League Cy Young Awards in 2007, 2009 and 2010 when they did not lead the league in wins or have the league’s best won-lost records, but did lead the league in RSA6.

            In the first era (1956-1973, 26 awards) the Cy Young voting is 96.2% consistent with wins and 96.8% consistent with WLS3, but only 87.6% consistent with RSA6.   But in the last ten votes (2006-2010), the voting is 93.1% consistent with RSA6, as opposed to 88.5% consistent with Wins, 87.2% consistent with WLS3.   It does appear that we can conclude that pitching a lot of innings with a good ERA may now be more important, in Cy Young voting, than posting a good won-lost record.   An analysis of the data suggests that this change may have occurred as early as 1997, and certainly occurred no later than 2005.

            I would have thought that RSA6 would be a better predictor than RSAL (Runs Saved vs. League Average), and it is, but not by the margin that I would have anticipated.   RSA6 favors a pitcher who pitches more innings, thus favors a workhorse pitcher of the type that has usually done well in Cy Young voting.   Nonetheless, RSAL is essentially as good, as a predictor of Cy Young voting, as RSA6.

I had thought that "leveraging" the Runs Saved against League by considering Saves might increase the significance of Runs Saved against League as a predictor of Cy Young voting.   To my surprise, this proved to be totally untrue.   The study reports on one leveraging formula, but actually I checked several.   Anything you do to leverage the RSAL by considering saves will reduce, not increase, the connection between RSAL and Cy Young votes. 

            Now, perhaps the most interesting thing in the study is the increase in the significance of the strikeout and walk data.   Like ERA, a "raw" strikeout/walk ratio is not a good predictor of Cy Young performance because there will be pitchers with dominant performance in limited roles who don’t show up for the Cy Young ballot.   To cover this problem I created SOBB, which is 2 * Strikeouts minus 3 * Walks.   This favors a good strikeout/walk ratio, obviously, but it also pays attention to innings pitched.    50 strikeouts and 5 walks is a far better strikeout/walk ratio than 250 to 70, but 50-5 scores at 85 SOBB, whereas 250-70 scores at 290.

            Our data shows a steady growth in the significance of SOBB over time, from .669 in the first era to .807 in the third era.   But the question is, is this an increase in the perceived importance of the stat—like ERA versus wins—or is it an increase in the actual importance of the performance area.

            I believe that this results from an increase in the actual importance of the performance area.   There’s an intricate question here, and I’m not certain I have it right, but I believe that there’s a significant shift involved here.  In the history of baseball up until about 1980, 1990, strikeout pitchers were also walk pitchers.    From 1950 to 1959, these are the top single-season strikeout performances in baseball:

 

            But almost all of those pitchers also walked a large number of batters:

 

 

 

            Whereas the Cy Young pitchers of that era who had very good control—Vern Law in 1960, Don Newcombe in 1956, up to Randy Jones in 1976—had very few walks and very few strikeouts.   The strikeout pitchers were good but not great pitchers.   The winning percentage of the pitchers on the list above is just .578.

            In other words, the starting pitchers of that era could be divided into power pitchers and finesse pitchers.   It was not so much that one group was better than the other, but that there were good pitchers in each group.   This had always been true, from 1870 to 1980—that "power pitching" was a type of pitching, rather than a type of excellence. 

            But post-1980, this becomes dramatically less true.   In modern baseball everybody is a power pitcher, basically, and most of the best pitchers are those who get the most strikeouts.    Look at the top strikeout seasons of the last ten years:

 

 

            These pitchers, despite their high strikeout totals, issued only an average number of walks—and thus posted a winning percentage, as a group, of .688.

            As strikeouts have become more common, having a good strikeout/walk ratio has become more important—not perceived as more important, like ERA, but actually more important.   It’s a very large change in the game.   This change probably also explains the increased significance of Home Run Frequency as a Cy Young predictor.   From 1956 to 1980, Home Run frequency probably was not a good predictor of overall excellence for a pitcher because those pitchers who gave up the fewest home runs were also those who gave up the most walks.    But in modern baseball this is not true, or is (at least) much less noticeably true, which makes home run frequency more importance, since the home runs are not competing with walks.

            And—no surprise to me—the Season Score tracks with Cy Young voting at a very, very high level throughout the history of the award, much higher than any other stat that we checked.    There are only three Cy Young contests in history in which the Season Score does not match the Cy Young voting at a rate of at least 90% (1956, 1959, and the 1987 National League.)

            The Season Score tracks with Cy Young voting because that is what it is designed to do.   If I could figure out a way to make it track better with the Cy Young voting, I would change the formula so it tracks better.   It is not surprising that a formula designed to imitate Cy Young voting patterns matches well with Cy Young voting patterns.

            While this match may have slipped very slightly in recent years, due to the decreased importance of Wins and Losses, we are not yet in position to conclude that there has been a meaningful divergence of Season Scores and Cy Young voting performance.   Taking just the last ten votes, the Season Scores are 97.1% predictive of Cy Young voting, the same as their overall performance over the 99 votes. 

            The charts below summarize the significance of each statistic in each of the 99 Cy Young votes.  In the 2001 National League and the 2002 American League votes, the Season Scores shows as 100% predictive of Cy Young voting, but these are not actually "true 100s".   It’s actually 99.96% in the National League in 2001, and 99.97% in the American League in 2002:

 

 

​