Interesting read. Assuming the math was done correctly, are there any holes in the author’s analysis? Could any of his assumptions be debated qualitatively? Also, has anyone done something similar with different results? It would be cool if there was a way to incorporate all teams in all of the world’s team sports, I.e. Soccer, hockey, etc.
The math involved in the Elo Rating is probably beyond me, but how are the 1939 Yankees better than the 1998 Yankees?
1939 Yankees: 106-45, 70.2%
1998 Yankees: 114-48, 70.4%
With the postseason:
1939 Yankees: 110-45, 71%
1998 Yankees: 125-50, 71.4%
The three teams behind the 1939 team had win. % of .589, .565, .552
1998- .568, .549, .543
Looks like the 1939 team had slightly tougher competition.
1998 was an expansion year - there were two other teams over 100 wins that year (both in the NL) so there’s some argument that talent dilution made it a bit easier to win a whole lot of games. Maybe?
Does this analysis account for the different levels of variance in different sports? In baseball, if you put the best team in the Major Leagues up against the worst, the worst will typically still have something like a 1 in 3 chance of winning. Put the best football team up against the worst, though, and it’ll be a single-digit percentage.
It has to. Otherwise, the answer if blindingly simple; the 1972 Miami Dolphins finished the regular season 14-0, then went on to win the Super Bowl, becoming the only major league US sports franchise to finish a season undefeated.
From the article:
Though, I still find it a little interesting that the author would basically equate NFL and NBA games. An NFL team has only ~20 games in a season to rack up elo, whereas an NBA team has ~100. So one would think the NBA scores are a bit more exact.
WAG-The NBA only uses 5 players who play the whole time (more or less). The NFL uses 11 on offense and a different 11 on defense with a total 53 players to mix and match.
So in effect, there’s 4 games going on, each offense and defense is playing an individual game.
As long as the number of games is large enough, the number of games shouldn’t matter.
You mean, it should be an insignificant factor?
I’ve found that 538’s use of statistics in sports is nearly useless. They make lots of cool looking calculations, then make bold statements, but the predictive ability of anything that they do is extremely poor. It is good for making lists that cause lots of arguments, though, which is probably what they are looking for. More clicks on the controversial articles.
I don’t get it, either.
Digging into the ELO ratings is near impossible on a casual basis; it basically amounts to awarding or subtracting points from a team after every win or loss and the number of points is based on a bunch of things, so at the end of the 1939 season the Yankees have more points than the 1998 Yankees do.
If you look at the overall numbers, I would assume that two factors push the '98 yankees down a bit:
-
They were a little lucky to go 114-48; their runs scored and allowed were not quite that good. The 1939 Yankees actually were UNLUCKY to go 106-45.
-
1998 was an expansion year and one of the expansion teams was in the Yankees’ division.
Those are the math factors, anyway. The 1998 Yankees are an interesting team, too, in that they were just a really, really deep team. No one on the team put up a super season; the best player, Derek Jeter, had basically a Derek Jeter season. Nobody on the team won any of the 3 big awards, though I would have voted for Derek Jeter. Jeter and MAriano Rivera are the only players on that team certain to end of in the Hall of Fame (Tim Raines already is, but was not a huge part of the team’s success) ; the 1939 Yankees had Joe DiMaggio, who absolutely is an inner circle Hall of Famer, and several other Hall of Famers as well. The team just kinda LOOKS greater.
Of course we’re ignoring timeline. The 1939 Yankees were a pre-integration team in a time when no one threw a slider.
In the book The Crooked Pitch: An Account of the Curveball in American Baseball History, there are quotes from DiMaggio and Musial on the difficulty of hitting a slider along with some grumblings from old-timers that a slider was nothing more than the nickel curve of their day.
Yes, they looked at standard deviations above the league ELO mean for each team.
Not necessarily. The ELO system gives more points for beating a good team than a bad one, and (in 538’s method for the NBA, NFL and MLB) more points for winning by a lot. So going 14-0 by beating the worst teams in the league by one point each time will probably give a lower ELO than going 12-2 against the best teams, with twelve blowout wins and two one-point losses. And the '72 Dolphins had a notoriously weak schedule.
They equated NBA and NFL games, not seasons; so yes, after a full season, the NBA ratings are far far more exact than the NFL ones. An MLB season is in the middle; an individual MLB game is much less valuable, but there are a lot more of them. (see https://fivethirtyeight.com/datalab/what-the-diamondbacks-rough-start-says-about-mlb-and-the-nfl-and-nba/)
That means we can be fairly sure the Warriors are the best NBA team, and a little less confident the Cubs were the best MLB team last year, while the Patriots are more likely to have just gotten lucky [though I think the MLB playoffs are roughly about as random as the NFLs, and much less so than the NBA’s]
I’m not sure I understand how you’re drawing this conclusion from the nature of basketball and baseball. If basketball is more exact that the NFL for having 82 games plus playoffs instead of 16 plus playoffs, MLB is more exact that that, with 162 games plus playoffs.
(The Cubs were absolutely the best team in MLB last year. They were not as historically great as the Warriors were, but any system that does not state the Cubs were the best baseball team in 2016 is simply wrong.)
It’s not just about the number of games; it’s about how often the better team wins each individual game.
Consider if the NBA for some bizarre reason decided to shorten their games to 8 minutes instead of 48. Clearly, in the shorter game, getting a few lucky bounces on shots would matter a lot more relative to overall skill, and the Knicks would beat the Warriors far more often in 8 minute games than in 48 minute games.
Well, using the statistical measures 538 looked at (and short version, they look reasonable to me, though there are probably other ways to look), one full MLB game is about as random as 8 minutes of an NBA game. So, yes, there are more discrete games in an MLB season, but the overall level of luck still matters more than in an NBA season.
Basketball is more exact than football because it has more games IF each game has the same information content. Which 538 claims it does. Baseball has much less information per game.
I have problems with the 1944 St. Louis Cardinals, who won 105 games, but during the talent-depleted height of World War 2. Meanwhile, the 1942 Cardinals won 106 games against much tougher competition (Brooklyn won 104 games and finished just two games behind.)
It seems like the 1942 team was penalized because Brooklyn was tougher than anything the 1944 team faced.
The variance is also why you can get away with a single championship game for the Superbowl, but you need a whole 7-game series for the World Series. If you played just one game for the baseball final, you couldn’t be at all confident that the right team won.
I’m not sure what it says if basketball has the same per-game variance as football, but still has a championship series.
Having a wartime team on the list is preposterous. You have to consider timeline at some point, and the quality of MLB talent was really, really significantly impacted by the war. There’s a reason Mort Cooper was a good but not fabulous pitcher and suddenly became Greg Maddux during the war and then suddenly wasn’t anymore after the war. The 1944 Cardinals have a bunch of guys who curiously happened to have terrific years that year but went back to being ordinary when everyone came home in 1946.