Strong and Weak MVP Classes

              Somebody observed recently that the American League this year had a strong crop of MVP candidates and the National League a weaker crop.  I didn’t really think that this was true, and having studied the issue I know now that it is not true, but also, we got into a discussion here recently about Roger Maris and the 1960 American League MVP Award.  My point was that while Maris may (or may not) have deserved the 1960 AL MVP Award, it was an exceptionally weak group of MVP candidates.   Maris hit .283 with 39 homers, 112 RBI, not really an MVP season.    If we take all players in baseball history who have hit 38 to 40 homers, driven in 107 to 117 runs and hit between .271 and .295, there are 13 players.   None of the others won an MVP Award; none of them was a serious MVP candidate.   Edwin Encarnacion in 2015 hit .277 with 39 homers, 111 RBI; he finished 12^th in the MVP voting.  Roy Sievers in 1958 hit .295 with 39 homers, 108 RBI, finished sixth in the MVP voting.    Jesse Barfield in 1986 hit .289 with 40 homers, 108 RBI, finished sixth in the MVP voting.   A season like that normally finishes about sixth to tenth in the MVP voting, but it happened that in 1960 there wasn’t really anybody else, so he won the Award.

              In the course of that discussion I realized that many of the readers couldn’t deal with that point.  They would argue that Maris in 1960 deserved the MVP Award, not because this was actually the issue we were discussing, but because that was the argument that they understood.   The other argument they didn’t understand and didn’t have any tools to deal with.   So then I was thinking. . .well, OK, how do we deal with that?

              On Saturday night (11-11-2017) I lay in bed thinking about this problem, and when I woke up Sunday morning I saw clearly how to deal with the problem.   Does that ever happen to you?  It happens to me about twice a year, I think, where I go to bed worrying about some problem and when I wake up I understand the problem and have the sense that I have understood the problem since the middle of the night.   The subconscious is a funny thing.   The rational mind is a servant; the subconscious is a wild animal.  It does whatever the hell it wants to do.   Occasionally, when you turn off your rational mind, the complications drop out and the solution to the problem becomes obvious.

              Anywho, this was the solution that was obvious to me on Sunday morning.   We can use Win Shares to measure the strength of an MVP class.   My first thought was that we could use Win Shares minus 25, squared, but that doesn’t quite work; that approach would conclude that whenever you have one superstar having his best season, that that’s the best MVP "class" of the decade.   That’s not exactly what I was going for.   I settled instead on this formula:

              Win Shares minus 25 if Win Shares are greater than 25, plus

              Win Shares minus 30 if Win Shares are greater than 30, plus

              Win Shares minus 35 if Win Shares are greater than 35, but

              Never greater than a total of 30. 

              So if a player has zero to 25 Win Shares, he counts as zero toward the strength of the MVP class.   Players with 25 or fewer Win Shares are not generally MVP candidates, although a few players have won MVP Awards with 25 or fewer Win Shares, but it isn’t a common thing.   You become an MVP candidate when you get above 25 Win Shares.  This transition leads to this spectrum of Win Shares vs. MVP candidate Weight Points (MVPCWP):

              It often happens that the player who has the most Win Shares in a league also wins the MVP Award.   It sometimes happens that he does not, too, and I am not decrying or criticizing that.   In the American League this year the player with the most Win Shares (Jose Altuve) probably WILL win the MVP Award, whereas in the National League the two players who tied for having the most Win Shares (Charlie Blackmon and Joey Votto) probably will not win the MVP Award, which will likely go to Giancarlo Stanton.    Votto and Blackmon have 33 Win Shares each; Stanton has 29, the same as Aaron Judge.   The point is not that whoever has the most Win Shares should win the Award; the point is that Win Shares track MVP voting performance well enough for Win Shares to be used as a surrogate for a player’s strength as a MVP candidate, divorced from the issue of whether there are other stronger candidates in that particular season.  

              With that explanation, a few points:

              1)  The class of American League MVP candidates this season is actually the weakest since 2009 and the weakest in either league since 2009, although the 2016 National League class actually is almost the same.   The class measures at 38 MVP Candidate Weight Points, whereas the National League in 2016 was at 39.  

              2)  The reason the AL measures as having a weak MVP class, in a sense, is Mike Trout’s injury.   Trout normally registers as a very strong MVP candidate, but since he missed 48 games with an injury he earned "only" 29 Win Shares this year—the same as Aaron Judge and Giancarlo Stanton—which makes him an MVP candidate, but not a STRONG MVP candidate; just another guy that the voters in their infinite wisdom might happen to take a liking to.   That sounds derisive but I don’t mean it that way; I realize that the MVP voters do in fact have more insight into the value of the players than do my mathematical formulas which lead to Win Shares. 

              3)  The comparative weakness of this year’s AL MVP class is also illustrated by this point, that Jose Altuve’s season in 2017 was nearly identical to his performance in 2016 and calculates as having one less Win Share than in 2016, but in 2017 he probably will win the MVP Award, whereas in 2016 he finished third in the voting. 

              4)   Aaron Judge actually ranks not second in the American League in Win Shares, but tied for third, even with Trout but one Win Share behind Eric Hosmer.   I was surprised by that, so let’s track it back and see if we can figure it out.

              Eric Hosmer created 116 runs while making 434 outs; Judge created 131 runs while making 411 outs, so Judge appears to be ahead on that level.   Yankee Stadium, however, is more of a hitter’s park than Kauffman Stadium (Royals), so a run in Kansas City has more of an impact on the bottom line (wins) than one run in Yan Key Stadium.   The park indexes are 93 and 102.   When you adjust for the fact that the teams play only half of their games at home those numbers become 97 and 101, so a run for Hosmer has 4% more impact than a run for Judge, and when you adjust for some other stuff this becomes actually only a 2% edge for Hosmer, actually not that big a deal.

              A much bigger factor in the calculation is this:   that the Yankees were inefficient in the use of their runs, whereas the Royals were highly efficient.   This is counter-intuitive, because the Yankers won 91 games and the Royals only 80, but the Yanghees SHOULD have won 102 games, based on the number of runs they scored and allowed, but actually won only 91 games because they made inefficient use of their runs, whereas the Royals SHOULD have won only 71 games based on their runs scored and allowed, but actually won 80 because they made extremely efficient use of their runs.   Thus, a run by a Royals player INCREASES in value when you translate runs into wins, whereas a run by a Yanquese player DECREASES in value when you translate runs into wins.   Thus, Hosmer moves ahead of Judge in terms of WIN impact, although Judge is ahead of Hosmer in terms of RUN impact. 

              What we are dealing with is the difference between usual and specific outcomes.   If you compare a player who did what Judge did and a player who did what Hosmer did, the player who did what Judge did would USUALLY win more games for his team than the player who did what Hosmer did.   But in this specific case, Hosmer probably had very slightly more positive impact on the win column than Judge did, as best I am able to measure that. 

              5)  This is the strength of each MVP class in American League history (since the BBWAA award was introduced in 1931):

                                              AMERICAN LEAGUE

              And this is the strength of each MVP class in the National League:

                                          NATIONAL LEAGUE

              6)  In 1931, the first year of the BBWAA official vote. . . .(the BBWAA had actually voted on MVP Awards in 1930, but the organization had not voted to endorse the process and did not hold a banquet and hand out trophies, so the results aren’t usually listed anywhere that you can find.)   In 1931, the first year of the BBWAA vote, the American League had an exceptionally strong class of MVP candidates, while the National League had an exceptionally weak class of candidates.   The 1931 American League class was the strongest class of candidates in history until 1963, whereas the 1931 National League class was the weakest of all time until 2008, except for the strike seasons of 1981 and 1994.  

              This is one of those things that I sort-of knew anyway, although I didn’t ACTUALLY know it until I developed a method to study the issue.   The American League ERA that year was about a half a run higher, but the Win Shares method adjusts for the run context of each player, which completely takes that issue off the table; that has NOTHING to do with the measured strength of the MVP class.   But the American League had Lefty Grove (31-4 with an ERA less than half of the league norm), Lou Gehrig (184 RBI), Al Simmons (.390 with 22 homers, 128 RBI), Babe Ruth (46 homers, 163 RBI, .373), and Earl Averill (.333 with 32 homers, 143 RBI, 140 runs scored), whereas the NL MVP Award was won by Frankie Frisch (.311 with 4 homers, 82 RBI, 96 runs scored).   Frisch is one of the weakest MVP picks ever.   In part this may be because he had assets which are not measured in the statistics; in part it may be that he was a poor selection.   But in large part, it simply reflects the fact that nobody in the National League had an MVP season.   No pitcher in the National League won 20 games, whereas one pitcher in the American League won 30, and five pitchers won 20.   No hitter in the National League hit more than 31 homers, which is unusual in a league with a 3.86 ERA, and no hitter hit .350 in the National League for the first time since 1919, and the only time between 1919 and 1938. 

              7)  The American League generally had stronger classes of MVP candidates than the National League from 1931 through 1948.   In 1949, with the emergence of the black superstars, the National League took over as the league with the stronger MVP classes.  

              In the 1950s and early 1960s, the American League has an unmistakable lack of stars.  This shows up in many different ways, of which this is one.  The National League has Mays and Aaron and Banks and Frank Robinson and Duke Snider and Eddie Mathews and Gibson and Koufax; the American League has Mantle, and really nobody else who is on that same level.   This leads to the National League winning the All Star game pretty much every year, and also to their having a stronger class of MVP candidates every year except 1956 and 1961.

              8)  There are two factors acting on the quality of the pool of MVP candidates.   There is (a) the compression of talent over time, and (b) expansion.   The compression of talent over time means that the difference between the best players and the average players, over time, gets to be less and less.  This happens everywhere.   Software engineers.   The best software engineers in the Bill Gates/Steve Jobs generation made billions of dollars because the difference between the best software engineer and the 100^th best software engineer was enormous.   This is less likely to happen now, because the difference between the best software engineer and the 100^th best software engineer is nowhere near as large.   The same is true in baseball.

              The compression of talent constantly reduces the measured quality of the pool of MVP candidates over time, while expansion increases it.   The combined effect of these is that the 1930s and 1940s had stronger (and more obvious) pools of MVP candidates than the 1950s did, but that from the 1950s through today it is basically a draw.   With expansion and the addition of eight games to the schedule, the average measured strength of an MVP pool shot up between the 1950s and the 1960s, but otherwise has been essentially stable since 1950. 

              8)  Then the DH rule throws us a curve.  The DH rule changes the distribution of Win Shares.   You have 9 regulars instead of 8, but the average team wins the same number of games, thus has the same number of Win Shares; thus, the likelihood of a player getting 35 or 40 Win Shares is diminished.  

              When I draw up a list of the ten strongest leagues and the ten weakest leagues in terms of the quality of the MVP candidates, it turns out that all ten of the "strong" leagues are National League seasons, and 9 of the 10 weakest pools of MVP candidates are American League seasons.   There are two reasons for this, which are almost even in their impact on those lists.   One is that in the 1960s, when the totals shot up, the National League had many more stars than the American League, and the other is that the DH rule changes the distribution of Win Shares.  

              I could, of course, adjust this problem out of existence by putting in another step in the transition from Win Shares to MVPCWP (MVP Candidate Weight Points.)   For example, if I said 24 that Win Shares in the American League after the DH rule was equivalent to 25 in the National League, that would take care of the problem.   But the thing is, it’s a "real" phenomenon, as opposed to a statistical illusion.   It’s like the 1981 and 1994 strikes; I could adjust the effects of that strike out of existence, but what we would be adjusting out of existence is not a statistical illusion, but reality.   The reality is that, because of the strikes, very few players had seasons in 1981 or 1994 which would ordinarily make them MVP candidates.   That’s not an illusion; it’s a fact of life.   The same with the DH impact on Win Shares; it’s not an illusion; it’s a fact of life.  One generally should adjust for statistical illusions—that is, the park makes Nolan Arrenado look like he is a greater hitter than he is--but not for changes in the underlying reality.  

              What I decided to do instead was this:  that when I list the ten leagues with the best and worst pools of MVP candidates, I will limit each league to 7 positions on the ten-league list.   Then I have given you more complete data (above); you can make your own list if you don’t like mine. 

              9)   These are the 10 leagues with the weakest ever pools of MVP candidates:

              10)  And these are the ten leagues with the strongest pools of MVP candidates.   Remember that these numbers are adjusted for the run context:

              11)  Any total less than 50 represents a weak pool of MVP candidates; any number greater than 100 represents a strong pool of candidates.   The numbers this year were 38 and 53. . . thus, pretty weak pools of MVP candidates in both leagues.   Not to take anything away from Altuve; he’s a great player and had a great season.

              12)  Obviously, I made a number of completely arbitrary choices as to how to measure the size of an MVP pool.   All measurements are built on arbitrary choices.   An "inch" is an arbitrary unit of measurement; a foot is, a pound is, a ton is, a light/year is.   All units of measurement depend on arbitrary choices.  

              But no matter how you measure it, Yao Ming is taller than I am or you are.   The conclusion is not arbitrary because it is stated in arbitrary units.   And I would argue that these are not an arbitrary conclusions, either.   You could study this issue by some entirely different approach; you could use WAR rather than Win Shares, and you could scale the system in some different manner that you chose, and you could measure the top 10 candidates in each league rather than all players with more than 25 Win Shares, but no matter how you did it, you would reach conclusions very similar to mine.   No matter how you studied it, you would conclude that the 1931 American League had a very strong pool of MVP candidates, while the 1931 National League had a very weak pool of MVP candidates.   No matter how you studied it, most of the leagues above would be listed as having the strongest and the weakest pools of MVP candidates.   These are not arbitrary conclusions; they are essentially objective conclusions, reached by a process which involves a certain number of arbitrary choices. 

              I’ll open this up for comments by readers tomorrow.   Thanks.