Behind the Scoreboard: How can you tell if a playoff team will win?

2009-10-08T11:58:40-04:00

2009-10-05T12:53:56-04:00

2009-10-06T17:58:06-04:00

MrKid72

Eureka 576346

9 I think that looking at the whole season to try to predict the post-season is a big problem because: 1. Teams switch from a 5-man to a 4 (or sometimes 3) man rotation in the postseason; 2. There are personnel differences between the regular season and the post season due to injuries, trades, etc. A 25-man roster becomes effectively about an 18-man roster in the post-season. Somebody needs to figure out how to compare the top 18 (9 position players, 9 pitchers) on each team, and scrape off the stats from the bottom of the barrel players who won't be appearing in the game unless they are more than 8 runs ahead or behind. 7144087

2009-10-06T12:59:39-04:00

Big Dawg 69

Greenville 63737

8 Dodgers win and Torre moons the Steinbrenners from the mound--or not. 7137880

2009-10-06T12:22:32-04:00

NL East Fan

Trenton 683695

7 Hey Eyepatch, you need to get that crystal ball calabrated. If it was working right, it would show the Yankees losing in five to the Phillies. 7137022

2009-10-06T05:36:19-04:00

http://img.fannation.com/upload/user_profile/image/634/226/thumb/616px-DarthNihl.jpg 769 UT

THEeyepatch

Salt Lake City 634866

6 You know what..... I will just use my crystal ball to predict..... hold on...... seeing something.............. wait a second.............. YANKEES WIN WORLD SERIES!!!! 7130694

2009-10-06T02:06:38-04:00

meyds90

Florissant 883528

5 How did you account for the fact that your two independent variables, a team's record and home field advantage, are correlated? The team with the higher record in the playoffs is by rule given home field advantage, is it not? 7130679

2009-10-06T01:27:21-04:00

prettyrickymyers

wyoming 512459

4 way to make something simple sound really complex 7130573

2009-10-05T23:55:50-04:00

RCR

Aurora 532319

3 Sa-weet article! Thanks for the insight. I am looking foward to the refinements in the playoff model. 7130365

2009-10-05T21:14:03-04:00

clarkandaddison

Chicago 883428

2 great job. looking forward to your future analysis. 7128994

2009-10-05T20:24:55-04:00

submariner

No City 835813

1 doo doo del toro 7128725

Sky Andrecheck writes for the <a href="http://www.hardballtimes.com/">Hardball Times</a>, an awesome site that you should be reading if you aren't already. He'll be a regular contributor to this blog throughout the postseason. October is here, and while 162 games have given us an idea of who the top teams are in the regular season, the playoffs are something else, widely and accurately seen as a crapshoot in which even the best teams need a lot of luck to win. (Since the beginning of the wild-card era, the team with the best record has gone on to win the World Series just twice in 14 years -- Red Sox in '07 and Yanks in '98.) So should we throw the regular-season stats out the window when it comes to the postseason? Or is there something in teams' regular-season play that can help us predict postseason success after all? You've heard the cliches: Good pitching beats good hitting, "small ball" teams usually do well, teams that live and die by the home run usually die by it in the playoffs and home-field advantage takes on vastly added importance in October. Do any of these theories have merit, or are they baseball's equivalents to old wives' tales? I examined all postseason games stretching back to 1969 and looked at a variety of team statistics: The regular-season winning percentages of the two teams, the teams' Pythagorean records, the strength of each team's pitching staff (including runs allowed, strikeouts and walks), and the strength of each team's offense (including runs scored, home runs, walks, strikeouts and stolen bases). Putting these statistics into a model, we can determine which factors are important in winning postseason games. (For those statistically minded, it's a <a href="http://en.wikipedia.org/wiki/Logistic_regression">logistic regression model</a>, designed to model binary outcomes -- in this case whether the team in question won or lost.) This allows us to predict outcomes based on a variety of different characteristics of each team. As it turns out, very few of the commonly held theories are true. Think that teams who steal a lot of bases have an advantage in the playoffs? When controlling for other factors, the team with the speed advantage did not perform any better in the playoffs. How about the "good pitching beats good hitting" mantra? Again, when taking into account the records of the two teams, teams that thrived via pitching and defense fared no better than the slugging teams. Think that power staffs that dominate with strikeouts are ripe for postseason success? The team with the power pitching advantage gained no special edge. And, when accounting for other factors, the team that won via the long ball fared no better or worse than its punch-and-judy counterparts. In fact, none of the team characteristics had a <a href="http://en.wikipedia.org/wiki/P-value" target="_blank">statistical p-value</a> that was close to statistically significant. How about a team's <a href="http://en.wikipedia.org/wiki/Pythagorean_expectation">Pythagorean record</a>? Supposedly a more consistent indicator of team performance than vanilla run differential, is it actually an indicator of postseason success? Regular records and Pythagorean records are highly correlated, so it's difficult to separate the two, but the model clearly preferred looking at a team's regular record. The model suggests giving a team's regular record about 80 percent of the weight, and giving only 20 percent of the weight to the Pythagorean record. However, the Pythagorean record is not statistically significant in the model, meaning that you'll probably have nearly as good of prognosticating ability simply using a team's regular-season record alone. So what does make a difference? Sadly (though unsurprisingly), there's no magic bullet. Simply put, teams that win in the regular season have a better shot at winning in the postseason. After debunking the myths, the model says that just two main factors matter: The winning percentage of the two teams and home-field advantage. When controlling for the records of the two teams, home field advantage boosts a team's probability of victory by about 6.2 percent, which is slightly higher than the regular-season home field advantage, which boosts the home team by about 4 percent. Therefore, in a playoff game between two evenly matched teams, the home team will have about a 56 percent chance to win. Additionally, each game ahead in the standings boosts a team's chances of winning by about 0.7 percent. Therefore, in a playoff game between a 100-win team and a 90-win team, we would expect the 100-win team to win about 57 percent of the time. So, what's the quick and dirty way of determining the probability of winning a postseason game? You can use this simplified formula: Chance of Team A Winning = 44% + (Team A Wins - Team B Wins)*0.7% (+ 12% if Team A is home). A graph below shows the teams' probability of winning based on the two factors. <img src="http://img.fannation.com/upload/si_blog_post_images/43841/andrecheckprob2.gif" alt="Andrecheckprob2" width="500" height="315" /> Perhaps more interesting is a graph showing the probability of winning an entire five-game series, not just one game. In this case, the home-field advantage is not quite as important, but the differences between the two teams are magnified. <img src="http://img.fannation.com/upload/si_blog_post_images/43861/andrecheckprob1.gif" alt="Andrecheckprob1" width="500" height="276" /> So, what does the blind model think of this year's postseason? The probabilities are as follows: NYY 67 percent chance of victory vs. DET/MIN LAA 55 percent chance of victory vs. BOS LAD 58 percent chance of victory vs. STL PHI 54 percent chance of victory vs. COL Of course, this model doesn't explain it all. As we'll be talking about throughout the postseason, many other factors can sway these probabilities. Personnel changes such as midseason pickups and injuries can sway things, as can a top-heavy starting rotation and bullpen. And then, of course, there is the question of how particular teams match up with each other. Still, the blind model can be a good starting point in predicting the fate of each postseason series. We'll get more refined in the days to come. Sky Andrecheck writes for <a href="http://www.hardballtimes.com/">HardballTimes.com</a> and is an author in the <a href="http://www.actasports.com/detail.html?&id=080">Hardball Times 2010 Baseball Annual</a>. He also writes regularly for <a href="http://baseballanalysts.com/">BaseballAnalysts.com</a>. He is a statistician by trade.

1102

The Moderator

New York 7 79803 Sky Andrecheck writes for the <a href="http://www.hardballtimes.com/">Hardball Times</a>, an awesome site that you should be reading if you aren't already. He'll be a regular contributor to this blog throughout the postseason. October is here, and while 162 games have given us an idea of who the top teams are in the regular season, the playoffs are something else, widely and accurately seen as a crapshoot in which even the best teams need a lot of luck to win. (Since the beginning of the wild-card era, the team with the best record has gone on to win the World Series just twice in 14 years -- Red Sox in '07 and Yanks in '98.) So should we throw the regular-season stats out the window when it comes to the postseason? Or is there something in teams' regular-season play that can help us predict postseason success after all? You've heard the cliches: Good pitching beats good hitting, "small ball" teams usually do well, teams that live and die by the home run usually die by it in the playoffs and home-field advantage takes on vastly added importance in October. Do any of these theories have merit, or are they baseball's equivalents to old wives' tales? I examined all postseason games stretching back to 1969 and looked at a variety of team statistics: The regular-season winning percentages of the two teams, the teams' Pythagorean records, the strength of each team's pitching staff (including runs allowed, strikeouts and walks), and the strength of each team's offense (including runs scored, home runs, walks, strikeouts and stolen bases). Putting these statistics into a model, we can determine which factors are important in winning postseason games. (For those statistically minded, it's a <a href="http://en.wikipedia.org/wiki/Logistic_regression">logistic regression model</a>, designed to model binary outcomes -- in this case whether the team in question won or lost.) This allows us to predict outcomes based on a variety of different characteristics of each team. As it turns out, very few of the commonly held theories are true. Think that teams who steal a lot of bases have an advantage in the playoffs? When controlling for other factors, the team with the speed advantage did not perform any better in the playoffs. How about the "good pitching beats good hitting" mantra? Again, when taking into account the records of the two teams, teams that thrived via pitching and defense fared no better than the slugging teams. Think that power staffs that dominate with strikeouts are ripe for postseason success? The team with the power pitching advantage gained no special edge. And, when accounting for other factors, the team that won via the long ball fared no better or worse than its punch-and-judy counterparts. In fact, none of the team characteristics had a <a href="http://en.wikipedia.org/wiki/P-value" target="_blank">statistical p-value</a> that was close to statistically significant. How about a team's <a href="http://en.wikipedia.org/wiki/Pythagorean_expectation">Pythagorean record</a>? Supposedly a more consistent indicator of team performance than vanilla run differential, is it actually an indicator of postseason success? Regular records and Pythagorean records are highly correlated, so it's difficult to separate the two, but the model clearly preferred looking at a team's regular record. The model suggests giving a team's regular record about 80 percent of the weight, and giving only 20 percent of the weight to the Pythagorean record. However, the Pythagorean record is not statistically significant in the model, meaning that you'll probably have nearly as good of prognosticating ability simply using a team's regular-season record alone. So what does make a difference?