One of the blogs I love (and link to on the sidebar) is Climbing the Ladder because all he ever does is look at random statistics that 99% of soccer fans wouldn't care about.
His latest post was about trying to predict what lead to success in the home-and-home playoff series that have existed since 2003. Here were the results
Note: I adjusted percentages to not count ties
Fewer games (18-7-7 72.0%)
Goal difference (21-11-0 65.6%)
Less experienced coach (19-12-1 61.3%)
Top goalscorer (17-11-4 60.7%)
Momentum (19-13-0 59.4%)
Higher seed (19-13-0 59.4%)
Goals for (18-13-1 58.1%)
Home record (18-14-0 56.3%)
Older team (18-14-0 56.3%)
Goals against (16-13-3 55.2%)
Won season series (12-10-10 54.5%)
Away record (17-15-0 53.1%)
Younger coach (17-15-0 53.1%)
2 year record (16-16-0 50.0%)
3 year record (14-18-0 43.8%)
Coach w/ better all time record (13-19-0 40.6%)
This is obviously super interesting to me, so I ask him for the data because I'd like to look at how the Elo Ratings do. He's a badass so he replies and it turns out Elo prediction went 21-11, matching goal differential at 65.6% success.
Below are the results of each series with the Elo difference and games-played difference for the winner of each series (winner is on the left):
2007 HOU vs DAL (92.41) (+2)
2006 HOU vs CHV (56.99) (+2)
2008 CLB vs KC (56.84) (-1)
2006 DC vs NY (53.85) (+1)
2003 CHI vs DC (53.84) (0)
2005 NE vs NY (43.24) (0)
2009 CHI vs NE (42.04) (+1)
2007 NE vs NY (40.12) (+3)
2008 CHI vs NE (34.88) (-7)
2003 NE vs NY (34.07) (-1)
2010 LA vs SEA (27.33) (-8)
2006 NE vs CHI (23.31) (0)
2003 SJ vs LA (20.61) (-4)
2003 KC vs COL (16.81) (-1)
2004 LA vs COL (12.96) (0)
2005 COL vs DAL (12.82) (-4)
2009 LA vs CHV (12.48) (-3)
2004 DC vs NY (12.30) (0)
2010 COL vs CLB (8.94) (-10)
2009 HOU vs SEA (8.71) (+5)
2004 KC vs SJ (4.91) (-1)
2008 RSL vs CHV (-30.67) (-4)
2010 DAL vs RSL (-39.07) (-7)
2010 SJ vs NY (-41.22) (-3)
2006 COL vs DAL (-51.11) (0)
2004 NE vs CLB (-61.74) (-1)
2009 RSL vs CLB (-68.02) (-6)
2007 KC vs CHV (-70.75) (0)
2007 CHI vs DC (-77.37) (-10)
2005 CHI vs DC (-81.97) (-4)
2005 LA vs SJ (-99.84) (+2)
2008 NY vs HOU (-105.34) (-14)
The really cool thing to me is that only once in 11 series (2005 LA) did a team with more games played and a lower Elo rating win the series.
So does this mean anything? Well, it's not quite statistically significant (although the fewer games one is pretty damn close to 95% confidence and is over 90% confidence) but I think that some of the results would be trends worth paying attention to next season if you happen to find yourself in a position to wager a little money.
Showing posts with label mls elo ratings. Show all posts
Showing posts with label mls elo ratings. Show all posts
Wednesday, November 10, 2010
Friday, August 13, 2010
An Introduction to Elo Ratings
One of the features of this site that I hope to build out even more are a set of proprietary Elo Ratings for Major League Soccer teams. The most up to date ratings can always be found by clicking that tab up above here that says MLS Elo Ratings. You may be familiar with the idea from the national team Elo ratings at www.eloratings.net or if you were ever a giant nerd who got really into chess. If you want some more information on how I developed it you can read this paper, otherwise just check out the 5 most common questions I get on it.
1. So what's the deal?
The Elo Ratings are based on a chess ranking system centered around a base value of 1500. A team's ranking goes up or down after each game depending on the quality of their opponent, the game's result, and the goal differential.
Anything over 1600 is up there with the best teams in MLS history. Anything below 1400 is super awful.
2. What about home field advantage, season-to-season carryover, expansion teams?
Based on historic MLS results home field advantage is worth about 85 points on a 1500 point scale, so we just add 85 points to the home team right before running the calculations.
When a new season starts, a team's rating moves halfway back to the 1500 midpoint. So if the Columbus Crew end the 2010 season with a 1540 rating they will start 2011 with a 1520 rating.
Expansion teams come in at a 1500 rating. It's not the best method by any means since most of these teams tend to be pretty poor, but it starts to correct itself over the course of a season.
3. Is it accurate?
Yes and no. It takes around 15-20 games for the ratings to normalize to the new team strength, which means at least half the season. Even then, if one team makes a big improvement during the season, like bringing in Thierry Henry and Rafa Marquez during the summer transfer window, it will take longer for the ratings to normalize than it will for a fan to notice who is playing well.
4. But my team is so much better than you say!?
Actually, even though I admit it's not perfect, odds are you are wrong here. Go back and look at your team's schedule. Look at how many of your wins were at home, or how many are against bad teams, or the overall goal differential. Beating the patsies by a goal will not help your Elo Rating very much at all.
5. Do you do other leagues?
Kinda. I've done the legwork of putting together the ratings for Premier League teams, but I haven't watched the changes for a season yet, so I'm not sure how good they are right now. It took a lot of work to get the MLS Elo Ratings developed since I used every results from the league's first season in 1996 and I want to make sure the other leagues are equally good before I unleash them on the masses.
I do answer a couple questions a month from people looking to create their own ratings though, so maybe you'll see some more pop up from other people.
Anyway, I hope that gives you a good background on how these work. If you have any questions just send me an e-mail or post in the comments and I'll get back to you as soon as I can.
1. So what's the deal?
The Elo Ratings are based on a chess ranking system centered around a base value of 1500. A team's ranking goes up or down after each game depending on the quality of their opponent, the game's result, and the goal differential.
Anything over 1600 is up there with the best teams in MLS history. Anything below 1400 is super awful.
2. What about home field advantage, season-to-season carryover, expansion teams?
Based on historic MLS results home field advantage is worth about 85 points on a 1500 point scale, so we just add 85 points to the home team right before running the calculations.
When a new season starts, a team's rating moves halfway back to the 1500 midpoint. So if the Columbus Crew end the 2010 season with a 1540 rating they will start 2011 with a 1520 rating.
Expansion teams come in at a 1500 rating. It's not the best method by any means since most of these teams tend to be pretty poor, but it starts to correct itself over the course of a season.
3. Is it accurate?
Yes and no. It takes around 15-20 games for the ratings to normalize to the new team strength, which means at least half the season. Even then, if one team makes a big improvement during the season, like bringing in Thierry Henry and Rafa Marquez during the summer transfer window, it will take longer for the ratings to normalize than it will for a fan to notice who is playing well.
4. But my team is so much better than you say!?
Actually, even though I admit it's not perfect, odds are you are wrong here. Go back and look at your team's schedule. Look at how many of your wins were at home, or how many are against bad teams, or the overall goal differential. Beating the patsies by a goal will not help your Elo Rating very much at all.
5. Do you do other leagues?
Kinda. I've done the legwork of putting together the ratings for Premier League teams, but I haven't watched the changes for a season yet, so I'm not sure how good they are right now. It took a lot of work to get the MLS Elo Ratings developed since I used every results from the league's first season in 1996 and I want to make sure the other leagues are equally good before I unleash them on the masses.
I do answer a couple questions a month from people looking to create their own ratings though, so maybe you'll see some more pop up from other people.
Anyway, I hope that gives you a good background on how these work. If you have any questions just send me an e-mail or post in the comments and I'll get back to you as soon as I can.
Subscribe to:
Posts (Atom)