There are many ways, statisticians and mathematicians have come up with, to determine outcomes and strengths of players or teams. One way, in Chess, is finding a players rating by a rating system. At Chesspark, we use a variation of the Glicko rating system. I have always wondered if this rating system would work well with team sports. So, being that I love baseball, I did some research (google search) on baseball and chess ratings. What follows, is a my first experiments in using glicko atings to determine game outcomes and finding the strength of a team.
First, I could not find any information on baseball and Glicko. I did find some on ELO and baseball. There is a site that does ELO ratings on soccer (football) teams as well. ELO and Glicko are rating systems used in Chess or other games involving two players. Team sports do not necessarily use these methods to determine the outcome of the next game. Baseball has a well known method called the Pythagorean expectation. It uses what the teams have done in previous games to determine the outcome. It uses runs scored and runs allowed. You can use this equation in other sports too. I would like to compare ratings versus the Pythagorean expectation and also investigate combining the two.I gathered data from http://www.retrosheet.org. Wrote a python script to parse the data, and compute the ratings. I will release code in upcoming blog posts. Using the script and data we can calculate the final ratings of teams at the end of the season. The results are interesting.
Glicko Rating results for the Major League Baseball 2007 Season:
====================== teams by rating ========================== team rating rd total wins - losses New York Yankees 1580.0 28.465171 162 94 - 68 Boston Red Sox 1571.0 29.260524 162 96 - 66 Cleveland Indians 1564.0 28.649712 162 96 - 66 Los Angeles Angels of Anaheim 1562.0 28.888047 162 94 - 68 Arizona Diamondbacks 1537.0 28.440955 162 90 - 72 Colorado Rockies 1534.0 28.165939 163 90 - 73 Detroit Tigers 1533.0 28.574597 162 88 - 74 Seattle Mariners 1532.0 28.210795 162 88 - 74 Toronto Blue Jays 1529.0 28.210438 162 83 - 79 Philadelphia Phillies 1529.0 28.447520 162 89 - 73 San Diego Padres 1523.0 28.317698 163 89 - 74 New York Mets 1510.0 28.604631 162 88 - 74 Los Angeles Dodgers 1505.0 28.615645 162 82 - 80 Minnesota Twins 1495.0 28.312780 162 79 - 83 Atlanta Braves 1494.0 28.618743 162 84 - 78 Chicago Cubs 1494.0 28.402241 162 85 - 77 Oakland Athletics 1488.0 28.395109 162 76 - 86 Texas Rangers 1487.0 28.586974 162 75 - 87 Milwaukee Brewers 1474.0 28.445591 162 83 - 79 St. Louis Cardinals 1464.0 28.453097 162 78 - 84 Washington Nationals 1462.0 28.858994 162 73 - 89 San Francisco Giants 1462.0 28.315478 162 71 - 91 Chicago White Sox 1461.0 28.461948 162 72 - 90 Kansas City Royals 1461.0 28.856424 162 69 - 93 Baltimore Orioles 1460.0 28.501870 162 69 - 93 Tampa Bay Rays 1450.0 28.742325 162 66 - 96 Houston Astros 1448.0 28.453125 162 73 - 89 Cincinnati Reds 1442.0 28.818955 162 72 - 90 Florida Marlins 1438.0 28.412856 162 71 - 91 Pittsburgh Pirates 1424.0 28.681253 162 68 - 94
NOTE: This is a great example on how close in skill professional baseball teams are.
Now, lets take a game from the 2007 season and determine the outcome via glicko and the same percentage via the Pythagorean expectation. I will use Atlanta Braves games since they are my favourite team. Lets take the 42nd game Atlanta played. This game was versus the Boston Red Sox who was a high rated team all year.
NOTE: This determination is very simplistic and missing some steps, a more in-depth example may follow in other blog posts.
Boston had a record of 29 - 12 (Winning percentage 70.7%)
Atlanta had a record of 25 - 17 (WP 59.52%)
2007-05-19 ATL 1529.0 vs. BOS 1675.0
2007-05-19 ATL 52.562% vs. BOS 72.554%
By rating, Boston should win, they are the strongest. The chance for Atlanta to win is about 47.7%. The outcome of the game was Boston 13 and Atlanta 3. Boston crushed them. Before this, Atlanta split a series with Washington, a very weak team.
By winning percentage, we see that Boston's real percentage is lower than what their percentage should be and we see that Atlanta's real percentage is higher than the determined one. Using that we see that Boston is favoured to win unless luck comes into play.
Both methods show Boston being the favourite, but 10 runs seems too much. :)
Determining the actual chance that Boston will win is a bit more complicated than the above example, with both methods. It is easy to see that using each way shows that Boston is stronger and should win the game, and they did. You can also use other statistical methods to determine how many runs both teams could score and use that to help with determining the outcome. I want to find if I could use ratings to help or replace some of these.I believe ratings is a great indicator of how strong the team is right now (or at the moment) and is a simpler way when looking at strength of schedule and other factors in finding out who will win a series or finish first in the standings.
In posts to follow, I hope to start predicting the outcome of current games, and developing a system to make this easier for me and others. I plan on using XMPP, Pubsub and BOSH. I also want to investigate finding a pitcher or batter's rating using glicko. After that, who knows.
The baseball information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at "www.retrosheet.org".