The results are in for the 1999 season. The method: Our method for predicting the 1999 baseball season is based on the Markov Chain approach developed in the research paper A Markov Chain Approach to Baseball, published in Operations Research in 1997. The method takes a lineup for each team and defense/pitching factor for each team to compute an expected run distribution for any head to head team match up. That is, for each team we compute how often each team's lineup should zero runs against the other team, how often one run and so forth. From this information, we can compute the probability that each team will win a given game against the other. By considering how many games each team plays with each other team and in whose ballpark (there is a home team advantage factor based on league) and considering lineups with or without designated hitter or pitchers batting and considering how often each player plays and how each team's bench performs, we add the probabilities for each team to arrive at the number of games each team can expect to win in the season. Mark Kallus and Jerrold Volk of West Orange helped to determine the likely players and lineups for each team. Revisions of the method for the 1999 season: 1. The player data was taken from the last 4 years of a player's performance but weighted more heavily based on last year's performance. (If the player played only last season, the 1998 data was taken). 2. Pitchers had been weighted as all starters pitching the same amount and accounting for 2/3 of the innings pitched. Now all pitchers are weighted based on innings pitched. Previously, we considered only hits and walks against a pitcher, whereas for 1999 we considered also how often the pitcher gives up home runs. (A team's defensive rating based on this is divided by the league average to get a scale of how good the team's pitching/defense is.) 3. Last year we assumed each starter played 75 percent of games (when there was a DH) and 70 percent (when there was no DH). This year we considered in how many games a fielder played and weighted how much bench to include in that player's lineup position accordingly. All expected starters are given a minimum of 50 percent of the lineup position. Bench players were averaged based on their number of plate appearances. How we did last year: The method was used to predict the standings last year as well. Our predictions compared favorably with the publications used to obtain data on who was on each team, etc. (Sports Illustrated, Baseball Illustrated and The Star-Ledger). Last season, all did poorly at predicting who would make the playoffs. Of the 8 playoff teams, Sports Illustrated picked 3, Baseball Illustrated picked 3, the Star-Ledger picked 4, and our picked 4-1/2 (for calling it a toss-up between Baltimore and Boston). In picking the order of finish of teams in each division, Sports Illustrated and Baseball Illustrated had 47 percent greater error and The Star Ledger 39 percent greater error than our results. In predicting the number of games each team would win, the Star-Ledger had an error 27 percent larger (8.5 games/team) than ours (6.7 games/team) (SI and BI did not predict wins). So in all cases, our Markov Chain approach beat the experts!!!!!! The World Series: For the World Series, the model gave the Yankees a 74 percent chance beating the Padres. Other questions the method can answer: One of the aspects of the method is that we can quantify what should happen to the number of wins a team can expect when a trade is made or if a given player is injured. (e.g., the expected number of games for Rafael Palmiero was reduced due to his injury). We can also go back to see "what would have been" if certain changes were made. For example, we can aswer the following interesting questions: 1. Should opposing pitchers have walked Mark McGwire every time he was up last season? Answer: No. The Cardinals could have expected to win about 9 extra games had the opposition used this strategy. 2. Who has the most valuable player (MVP) in each league? Answers: Mark McGwire and Albert Belle (pitchers were not considered). Here, by MVP, we mean which single player in each league would have taken a team of average players and increased their number of wins by the greatest amount. (McGwire would have taken an average team from 81 wins to over 89 second was Bonds to almost 87. Belle's team wins 87 1/2 to Garciaparra's 86). It is interesting that no Yankee came close. The Yankees just have alot a very good players, they each contribute a few wins and it just adds up. The 1999 Standings follow below. Prof. Bukiet's 1999 Major League Standings October 4, 1999 -- Major League Baseball Standings Predictions posted March 11, 1999 National League East American League East Team Won Lost Pct. GB Team Won Lost Pct. GB Braves 103 59 .636 - YANKEES 108 54 .667 - METS 91 71 .562 12 Orioles 89 73 .549 19 Phillies 71 91 .438 32 Red Sox 88 74 .543 20 Expos 67 95 .414 36 Blue Jays 85 77 .525 23 Marlins 52 110 .321 51 Tampa Bay 69 93 .426 39 National League Central American League Central Team Won Lost Pct. GB Team Won Lost Pct. GB Astros 96 66 .593 - Indians 94 68 .580 - Cards 94 68 .580 2 Tigers 74 88 .457 20 Cubs 84 78 .519 12 White Sox 70 92 .432 24 Reds 83 79 .512 13 Royals 64 98 .395 30 Brewers 70 92 .432 26 Twins 50 112 .309 44 Pirates 70 92 .432 26 National League West American League West Team Won Lost Pct. GB Team Won Lost Pct. GB Dodgers 92 70 .568 - Mariners 95 67 .586 - Giants 86 76 .531 6 Texas 89 73 .549 6 Diamondbacks 83 79 .512 9 Angels 81 81 .500 14 Rockies 81 81 .500 11 Athletics 72 90 .444 23 Padres 79 83 .488 13