Master Melvin

A few thoughts about finding the appropriate value for Mel Ott

            1) The cornerstone of sabermetrics, the first observation upon which all the rest relies, is that the purpose of the game is to win, and that therefore each player’s value *is* the number of games that he has won for his team. 

            2)  WAR assumes that value is not in the number of games won for the team, but in the number of games won for the team above the number that could have been won by an easily available alternate had the player not been there.   That’s OK; I’m not arguing with that.  I would point out that the calculation of WAR can never be MORE accurate than the calculation of wins, since it relies upon a prior calculation of wins, and must always be less accurate, because it relies upon (and always will rely upon) substituting arbitrary values for unknowns.  

            3)  There is an additional caveat with the cornerstone observation, which is that some wins are more valuable than other wins.  If two teams meet late in the season and one of them goes into the game tied for first and the other is 40 games out of first and long since eliminated, then the game is more important to one team than it is to the other, and thus a win by one team has more value than a win by the other team. 

            From a logical standpoint, point (3) is as important as point (2); however, as there is a more-or-less agreed-upon protocol for handling point (2), but no agreement as to how to handle point (3), we thus make adjustments for point (2) while ignoring point (3).  Relevant to this discussion, however, is neither point (2) nor point (3), but point (1); I merely wanted to acknowledge that point (1) is not an absolute truth. 

            4) Point (1) is that Wins are the foundation of analysis.   Point (4) is that runs are NOT the foundation of analysis.   Runs are the pathway toward wins.  

            5)  WINS are not park-dependent.  The park does not win games.  In each park in each game, excepting ties, there is one win and one loss.  In a balanced schedule, each team plays (essentially) the same number of home and road games.  We don’t park-adjust WINS; we park-adjust RUNS. 

            If you park-adjust WINS, you wind up adjusting wins out of existence.  You can’t do that.  Wins are the be-all and end-all of analysis.   You can’t accidentally adjust wins out of existence in the course of your analysis. 

            6)  The Polo Grounds, where Ott played his home games his entire career, was a home-run hitter’s park.  Ott hit 511 home runs in his career—323 in his home park, 188 on the road.  I think it is the most pronounced home/road home run split of any player in history with a large number of home runs.

            7)  Because it was a HOME RUN park, people assume that it was a HITTER’S park.  I myself believed that to be true, early in my career, before we developed the robust information sources that we have now.  But it actually wasn’t a hitter’s park, at all; it was actually a pitcher’s park.   This chart gives the park Home Run Factors and park Run Factors for the years 1926 to 1947, when Ott was there.  As most of you know, a factor of 100 means that the park was average in this respect.  200 means that twice as many home runs per game were hit in that park as in the team’s road games.

            If we overlay that data with Ott’s Plate Appearances for each season, we can calculate his career park effects.  The Home Run effect of the park, relevant to Ott and weighted by his plate appearances per year, is 216 for home runs, but 97 for runs.  

            8)  Two conclusions from that:  a) Ott’s home run totals for his career were LESS effected by the park than his team overall, and b) the runs that Ott created were, on average over his career, MORE valuable than if they had been created in a neutral park. 

            9)  A modern reader will not understand how it is possible for a park to have a home run factor of 275, but a park run factor of 89.   What people may not get is that at time (a) home runs were less common as an element of offense than they are now, and (b) in some parks there were almost no home runs at all, but many runs still scored.   In 1931 Cincinnati hit only six home runs all year in their home park, and Boston only 16.  No team other than New York hit more than 51 home runs in their home park. 

            10)  What is relevant to the question of Mel Ott’s value is not what he would have done or might have done in some other park or at some other time, but what he DID do in the time and place where he played.  

11)  We establish that value—his offensive value--by the ratio of his runs created to the wins of the team, using either actual wins—which I prefer—or expected wins, which is used by (I believe) both of the WAR methods. 

12)  When we do those things, Ott has 528 Win Shares, 107.8 WAR by baseball reference, 110.5 WAR by Fangraphs.  Jimmie Foxx has 435 Win Shares, 96.6 WAR by Baseball Reference, 101.8 WAR by Fangraphs.    These numbers are not inflated in any way, shape or form by park effects; rather, they adjust for park effects, thus removing the advantage that Foxx had from playing in a better hitter’s park. 

Allowing that there may still be other considerations such a peak value impact, defense and pennant race impact, Ott appears to be more valuable than Jimmie Foxx.

            13)  I’m in the middle of a long study comparing player’s performance in MVP voting to their "actual" value.   While I am not finished with the study, and Ott is not the MOST undervalued player I have found so far, he under-performed in MVP voting to a really remarkable degree, I think for two reasons:  (a) that he walked a lot, which was not appropriately valued by sportswriters in that era, and (b) that people misjudged the park effects then as they do now. 

            14)  Regarding point (3) above, I can suggest two options.  First, one could adjust each plate appearance by its "pennant impact"; in other words, figure what the team’s chances of winning the pennant are before the plate appearance and after the plate appearance, and sum up the pennant impact of each plate appearance. 

            There are, however, three HUGE problems with that approach.  First, in order to estimate the pennant chances before and after every plate appearance, you have to make dozens of assumptions about what other teams are going to do, and other issues.  Do that for 600 plate appearances, and you have hundreds of thousands of little unstated assumptions which are buried inside your calculation.  If those little unstated assumptions are wrong by just a tiny bit, the end product can be wrong by a lot.

            The second problem if you figure the increases and decreases in the chance of winning the pennant (or a game) is that it is a zero-sum calculation, in which the average player is at zero.   The average player does not have a value of zero.  The average player has a value which we could represent as the number of games he has won for his team, or as the number he has won above replacement level, but you can’t represent it as zero.  Thus, after figuring the pennant gains and losses, you have to twist the data somehow to adjust for that, which introduces another potential source of error.

            The third problem is that a single hit or a few hits can have an immense impact on a team’s chance of winning the pennant.   If you value the players on the 1978 Yankees based on each play’s impact on the team’s chance of winning the pennant, you almost certainly will conclude that Bucky Dent was more valuable to the team than Ron Guidry was.  (Guidry went 25-3 with a 1.74 ERA; Bucky Dent hit .243 with 5 homers, 40 RBI, but hit a three-run homer in the seventh inning of the deciding game.) This conclusion may be valid in a certain narrow sense, but the problem with it is that without Ron Guidry’s season, Bucky Dent would never have been in position to have any impact on the pennant race. 

            Figuring value by pennant race impact disconnects the player’s VALUE from his TALENT to a probably unacceptable extent.  Bucky Dent’s value in 1978 was much greater than the expected outcome of his talent.  That’s actually the issue with making adjustments based on actual team wins, rather than expected team wins.   If a team should win 85 games based on their runs scored and runs allowed but actually wins 95 games, do you evaluate them as an 85-win team or a 95-win team?  I would prefer to evaluate them as a 95-win team, but mine is a minority position.  It’s the same issue with studying pennant impact.

            15)  Because of these problems, incorporating pennant impact in value assessments in that way is probably impossible.   The other alternative is this, or something like this: 

            Suppose that, in addition to the actual games won by the team, we added:

o   Two games if they made Post-Season play,

o   Four games if they won the league championship, and

o   Six games if they won the World Championship. 

If you do that, then you give 2-3% advantage in value to a player on a team that wins 88 games and does win their division, as compared to a team that wins 88 games and does not win their division.

I don’t know that that’s unreasonable; I’m not saying it should be done, but I don’t know that it would be unreasonable.   In essence, the old MVP voters, before modern analysis, gave about a 10% advantage to a player on the league championship team.   That’s too much.  If you give that much of an advantage based on the team winning the pennant, that undermines the rest of the value analysis.   It almost guarantees that a player on the first-place team will win the MVP Award, unless it is an unusual year.   It makes a .270 hitter with 15 homers look like a better player than a .270 hitter with 15 homers on a second-place team. 

16)  It is a principle of modern analysis that a player should be evaluated on the basis of what HE has done, not on the basis of what his teammates have done.   That is a completely valid principle, of course.

But it’s not the only valid principle. 

If you say that the point of the competition is to win the pennant, therefore winning the pennant has to be factored into each player’s value, I don’t have a problem with that.   If you say that an 88-win season in which you win the World Series has to be evaluated differently than an 88-win season in which you finish fourth, I don’t have a problem with that.   I think that valid analysis can be done either way, as long as you don’t push it so hard that the "special value" of winning the pennant obliterates the "ordinary value" of playing well and helping your team to win games.

I will open this up to comments by BJOL subscribers tomorrow or Monday.