Notice: Trying to get property 'display_name' of non-object in /var/www/html/wp-content/plugins/wordpress-seo/src/generators/schema/article.php on line 52
keyboard_arrow_uptop

When I wrote my five-part series on home-field advantage in 2009, I noticed that it had been steady at about 54 percent for over half a century. It was 53.9 percent in the 1950s, 54.0 percent in the ‘60s, 53.8 percent in the ‘70s, 54.1 percent in the ’80s, 53.5 percent in the ‘90s, and 54.2 percent in the 2000s. However, in the last three years, we have seen home teams win 55.5 percent of the 7,288 games played, a very statistically significant difference. Does this suggest that a large change has actually taken place, or is it just a coincidence? If a change has taken place, what is causing it?

On one hand, there was only about a 1 percent chance that the league-wide home-field advantage of the last three years would be so far from the historical 53.9 percent of the last 60 years by sheer randomness. On the other hand, I clearly have cherry picked the last three years—I don’t plan on writing an article every year examining the previous three and declaring that it was 54 percent again—and perhaps the odds of getting any three-year period with such an extreme number may not be so small. However, if we expand the period to 2007-10, we still see a statistically significant difference in home-field advantage at the 2 percent level, and if we expand the period to 2006-10, we also see a statistically significant difference at the 2 percent level. Even the 54.68 percent home-field advantage from 2003-10 is statistically significantly different at the 5 percent level than the 53.97 percent home-field advantage of the last 60 years. The point is that I could have cherry picked a lot of starting and stopping points and still seen something so extreme.

The below table lists the league-wide home-field advantage since 1900:

Year

HFA

Year

HFA

Year

HFA

Year

HFA

1900

58.1%

1930

57.1%

1960

54.8%

1990

53.7%

1901

56.2%

1931

58.2%

1961

55.1%

1991

53.8%

1902

57.8%

1932

55.4%

1962

53.6%

1992

55.2%

1903

56.3%

1933

55.8%

1963

55.2%

1993

53.8%

1904

54.0%

1934

54.8%

1964

52.6%

1994

51.7%

1905

55.3%

1935

54.7%

1965

53.6%

1995

53.2%

1906

54.2%

1936

55.2%

1966

53.4%

1996

54.1%

1907

54.4%

1937

54.3%

1967

56.3%

1997

53.5%

1908

53.9%

1938

53.8%

1968

51.1%

1998

53.8%

1909

53.8%

1939

53.7%

1969

54.9%

1999

52.1%

1910

56.0%

1940

52.9%

1970

54.0%

2000

54.0%

1911

52.7%

1941

53.9%

1971

52.0%

2001

52.4%

1912

52.2%

1942

54.6%

1972

52.9%

2002

54.2%

1913

51.0%

1943

55.2%

1973

53.0%

2003

55.0%

1914

55.2%

1944

54.9%

1974

53.4%

2004

53.5%

1915

55.4%

1945

56.7%

1975

54.0%

2005

53.7%

1916

55.7%

1946

55.2%

1976

52.4%

2006

54.6%

1917

50.6%

1947

54.0%

1977

54.4%

2007

54.2%

1918

56.5%

1948

50.9%

1978

57.3%

2008

55.6%

1919

55.1%

1949

56.0%

1979

54.0%

2009

54.9%

1920

53.3%

1950

54.7%

1980

54.2%

2010

55.9%

1921

54.3%

1951

52.6%

1981

52.1%

 

 

1922

55.2%

1952

55.0%

1982

53.8%

 

 

1923

50.9%

1953

52.6%

1983

54.2%

 

 

1924

53.7%

1954

53.1%

1984

52.9%

 

 

1925

56.0%

1955

56.2%

1985

55.0%

 

 

1926

56.5%

1956

53.6%

1986

54.7%

 

 

1927

56.1%

1957

52.5%

1987

54.8%

 

 

1928

52.1%

1958

54.9%

1988

53.8%

 

 

1929

54.5%

1959

54.1%

1989

55.0%

 

 

You can start to notice how different 2008-10 look compared to other years from this table, but since the year-to-year fluctuations are very large even within one year, let’s look at the table again in three-year bursts. This will give a more clear sense of how abnormal the last three years are.

Year

HFA

Year

HFA

Year

HFA

Year

HFA

1900-02

57.2%

1930-32

56.9%

1960-62

54.5%

1990-92

54.2%

1901-03

56.8%

1931-33

56.5%

1961-63

54.6%

1991-93

54.2%

1902-04

56.0%

1932-34

55.3%

1962-64

53.8%

1992-94

53.7%

1903-05

55.2%

1933-35

55.1%

1963-65

53.8%

1993-95

53.0%

1904-06

54.5%

1934-36

54.9%

1964-66

53.2%

1994-96

53.1%

1905-07

54.7%

1935-37

54.7%

1965-67

54.4%

1995-97

53.6%

1906-08

54.2%

1936-38

54.4%

1966-68

53.6%

1996-98

53.8%

1907-09

54.0%

1937-39

53.9%

1967-69

54.1%

1997-99

53.1%

1908-10

54.6%

1938-40

53.5%

1968-70

53.5%

1998-2000

53.3%

1909-11

54.2%

1939-41

53.5%

1969-71

53.6%

1999-2001

52.8%

1910-12

53.6%

1940-42

53.8%

1970-72

53.0%

2000-02

53.6%

1911-13

52.0%

1941-43

54.6%

1971-73

52.6%

2001-03

53.9%

1912-14

52.8%

1942-44

54.9%

1972-74

53.1%

2002-04

54.2%

1913-15

53.9%

1943-45

55.6%

1973-75

53.5%

2003-05

54.1%

1914-16

55.5%

1944-46

55.6%

1974-76

53.3%

2004-06

54.0%

1915-17

53.9%

1945-47

55.3%

1975-77

53.6%

2005-07

54.2%

1916-18

54.1%

1946-48

53.4%

1976-78

54.8%

2006-08

54.8%

1917-19

53.9%

1947-49

53.6%

1977-79

55.2%

2007-09

54.9%

1918-20

54.8%

1948-50

53.9%

1978-80

55.2%

2008-10

55.5%

1919-21

54.2%

1949-51

54.4%

1979-81

53.6%

 

 

1920-22

54.2%

1950-52

54.1%

1980-82

53.5%

 

 

1921-23

53.5%

1951-53

53.4%

1981-83

53.5%

 

 

1922-24

53.3%

1952-54

53.6%

1982-84

53.6%

 

 

1923-25

53.6%

1953-55

54.0%

1983-85

54.0%

 

 

1924-26

55.4%

1954-56

54.3%

1984-86

54.2%

 

 

1925-27

56.2%

1955-57

54.1%

1985-87

54.9%

 

 

1926-28

54.9%

1956-58

53.7%

1986-88

54.4%

 

 

1927-29

54.2%

1957-59

53.8%

1987-89

54.5%

 

 

1928-30

54.6%

1958-60

54.6%

1988-90

54.2%

 

 

1929-31

56.6%

1959-61

54.7%

1989-91

54.1%

 

 

The important takeaway is that even though we see some pretty large fluctuations prior to World War II, there is a very steady home-field advantage in recent years that has barely moved until recently when it flirted with jumping in 2006 and then really took a big stride forward in 2008, and no three-year period since 1944-46 has had a home-field advantage this large, so the results seem to suggest something may be happening.

In last year’s articles, I also discovered that the magnitude of home-field advantage seemed to be very similar across teams, with year-to-year fluctuations in teams with big or small home-field advantages disappearing as quickly as they came (though the Rockies seemed to be one team that repeatedly had larger home-field advantages than other teams). However, if there has been a sudden league-wide change in home-field advantage, it may not have affected all teams equally, so I gathered the home-field advantage for all 30 teams over the 2008-10 period. But keep in mind that the margin of error for any one team is about +/- 9 percent, and that it is likely that one or two teams will be beyond that. In other words, one of these teams is more than 9 percent above or below their true home-field advantage capabilities.

Team

Stadium

Year Built

HFA 2008-10

Pirates

PNC Park

2001

21.3%

Tigers

Comerica Park

2000

18.7%

Rockies

Coors Field

1995

17.7%

Twins

Target Field

2010

16.8%

Rays

Tropicana Field

1998

16.0%

Red Sox

Fenway Park

1912

15.2%

Mets

Citi Field

2009

14.0%

Diamondbacks

Chase Field

1998

12.8%

Astros

Minute Maid Park

2000

12.6%

Nationals

Nationals Park

2008

12.1%

Blue Jays

Rogers Centre

1989

12.0%

Mariners

Safeco Field

1999

11.9%

Braves

Turner Field

1996

11.9%

Athletics

Oakland-Alameda County Coliseum

1966

11.8%

Orioles

Oriole Park at Camden Yards

1992

11.7%

White Sox

U.S. Cellular Field

1991

11.3%

Dodgers

Dodger Stadium

1962

11.1%

Yankees

Yankee Stadium

2009

11.1%

Cardinals

Busch Stadium

2006

10.3%

Giants

AT&T Park

2000

9.9%

Rangers

Rangers Ballpark in Arlington

1994

9.1%

Reds

Great American Ball Park

2003

8.6%

Indians

Progressive Field

1994

8.6%

Cubs

Wrigley Field

1914

7.0%

Padres

PETCO Park

2004

6.6%

Royals

Kauffman Stadium

1973

4.5%

Brewers

Miller Park

2001

4.5%

Phillies

Citizens Bank Park

2004

3.5%

Angels

Angel Stadium of Anaheim

1966

2.9%

Marlins

Sun Life Stadium

1993

2.7%

Home-field advantage for the average team has historically been about 8.0 percent (54%-46%), but 23 of 30 teams have larger home-field advantages than that in the last three years.

Although determining individual home-field advantages is a fool’s errand in most circumstances, there has been research showing that domed stadiums may show small home-field advantage trends than others. Additionally, I found that one particularly pronounced source of home-field advantage emerged in the number of extra-base hits that go for triples instead of doubles. I believe that this was because outfielders are better at playing the bounces in their own stadium, so road teams do not hit as many triples.

One of the major trends in ballparks came in 1992 when the Orioles built Camden Yards. This quirky stadium became a template for many other new “mallpark” stadiums built since then, each with their own idiosyncrasies. Since we know that quirks can cause extra home-field advantage, perhaps teams are beginning to exploit this more in recent years. While this may be a feeble theory, it is worth noting that the correlation between having a stadium built since Camden Yards was built has a .19 correlation with home-field advantage over the last three years. The older stadiums such as Wrigley Field, Angel Stadium, and Kauffman Stadium all are home to teams among the bottom six in home-field advantage, while the only  one in the top 10 is that decidedly quirky Fenway Park. It is difficult to say for sure that this is the cause, but if this is more than a fluke, my best guess is that teams are learning to take advantage of their quirky stadiums.

Even with this possible evidence, I am still skeptical that a change has taken place. While it is certainly possible that home-field advantage means more than it used to in baseball, I would still expect it to sit at about 54 percent in 2011, simply because there is such a historical precedent for flukes in league wide home-field advantage balancing out.

 What do you think? Is this a real change? If so, what else could be causing it? 

Thank you for reading

This is a free article. If you enjoyed it, consider subscribing to Baseball Prospectus. Subscriptions support ongoing public baseball research and analysis in an increasingly proprietary environment.

Subscribe now
You need to be logged in to comment. Login or Subscribe
joelefkowitz
12/03
On the one hand, if you remove Fenway from the equation (which certainly qualifies as "quirky", which is what we're looking for in stadiums built since Camden) the correlation jumps to 0.26, but on the other hand there are only 7 data points that qualify as "Not built since Camden" which might be too small to mean anything. For example, Having the word "Stadium" in the name of your stadium has 6 qualifying data points and a strong negative correlation of 0.42, but I'm pretty sure the fact is irrelevant.
joelefkowitz
12/03
Also FWIW, the three sets of data starting at 1976 and ending at 1978 show a similar variance as 2006-2008 (heck, they're even right next to each other in the table), but jumps back down to a more expected level in 1979, so at the very least, this kind of shift (for this amount of time anyway) isn't unprecedented.
swartzm
12/03
Yeah 1978-80 at 55.2% was as close as it got, but no three year period had 55.5%, so it's slightly unprecedented. And it is significantly different from the 1950-2010 average at the 99% confidence level. It's just that it still could be a coincidence, and I think that's still the most likely explanation. I think after 2011 and 2012, we'll have a better sense of this.
swartzm
12/03
Yeah, I agree that it's not a very relevant correlation given the sample size. I only thought to look at it because of what I learned about the ratio of triples/doubles for home teams, and how stadium quirks would seem to exacerbate those type of difficulties on the road.
Agent007
12/03
What about cheating? A number of teams have been detected (or accused with reasonable proof) of reading the catcher's signs through some in-park technology and relaying the information to batters. That's a lot harder to do on the road.
swartzm
12/03
Maybe, but this seemed to be an improvement across the board, so I doubt it. It seems like a lot to be explained by cheating anyway-- even if every team stole signs and swapped out balls every single game they played, would it really change the outcome of 1.5% of games? Now suppose that only half the teams did it and only half the time-- still a lot of cheating-- would that 6% of games from losses to wins?
hitmannls
12/03
Is it possible that teams are just getting better at building their rosters to maximize the attributes of their ballpark? Its anecdotal but as a west coast fan it sure seems that the Giants, A's, and Padres are all striving to field teams with more athletism to cover the large outfields in their home parks. Similarly teams with large ball parks have the luxury of buying low on fly ball pitchers.
swartzm
12/03
You know, I was looking for this effect a year or so ago, and it seemed like rosters were just built very poorly for their parks. But maybe that is starting to change more. That's probably just as likely as the quirky stadium effect, come to think of it. Worth looking at-- I just need to think about methodology because that's a tricky question to ask the data.
bsolow
12/06
Do you think you could use the availability of high quality spray chart type data as an instrument for roster optimization? I don't know how long this has been available to teams, but it seems that it's only been available to the public recently. In order to do an Angrist-Imbens-Rubin style IV analysis, though, you'd need for some teams to have access to the data sooner than others. Not sure whether or not that's the case, but at least that's a question that can be answered. If so, you could use a difference-in-difference strategy to identify the effect.

I would think that it's a good instrument because I can't think of any other channels through which knowing spray charts would help specifically home teams except through roster optimization. All of the other benefits (i.e. positioning of fielders) would accrue equally to both home and road teams. So, if it were the case that there was temporal variability across teams in their access to that kind of data, it might be a useful instrument to consider. The only assumption A-I-R's IV approach requires that would be left to verify is random assignment of the data to the teams.
BurrRutledge
12/03
This is what I was thinking, too. But you should see a more pronounced effect in the teams that are actively adjusting their roster. Or, do we assume that they are all doing it, but we're hyper-aware of only a few?
formersd
12/03
Could the increasing disparity between hitters and pitchers parks be a factor. Seems like we have less "neutral" parks than in the past.
swartzm
12/03
Hmm...I know that I tried to look at whether hitter's or pitcher's park would trend towards larger or smaller HFAs, but found virtually no difference though maybe a small tendency towards pitcher's parks. I'm not sure that I even thought about checking extreme vs. neutral parks...worth looking into also, thanks.
sensij
12/03
Is anyone else reading this familiar with Xbar and R control charts? A pretty good description is here:

http://www.itl.nist.gov/div898/handbook/pmc/section3/pmc321.htm

If the data from 1951 to 2007 are subgrouped in non-overlapping sets of three, 53.9% is the average with an upper control limit of 55.9%, a lower control limit of 51.9%, and a range upper control limit of 2.0%.

Adding 2008-2010 as the next group of 3 does not violate any of the control limits, and are unremarkable in that context.

Someone smarter than I will have to explain why a Student's T-test shows the last three years as significant, while Shewart's control chart scheme does not.
sensij
12/03
correction... range upper control limit of 5.1%, which says that over any consecutive three year period you should see less variation than that.
JayhawkBill
12/03
So if I understand what you're saying, Dan Fox joins the Pirates, and three years later we realize that during his tenure with the team they possess the highest home field advantage in MLB?

It's probably coincidence, but it's also exactly the sort of thing I'd expect Dan to be able to affect.
BillJohnson
12/05
The flip side of this (and no, I'm not blaming Dan...) is that the Pirates' epic ineptness on the road is responsible for a significant fraction of the observed effect all by itself, since that affects not just their own splits but their opponents'. How many other teams in the last fifty years have had a split that big? I'm curious.
mentalmeat
12/04
Great piece, Matt.

The "quirky stadium" theory may be enhanced slightly due to the irregular inter-league play schedule. OF are playing in certain road parks for the first time.

Another thought - have there been more OF who have switched leagues the past three seasons? Could it be that defensive horrors the likes of Man-Ram, Gomes, Ibanez, etc. have not only stopped DHing but also had to learn a new set of parks?

Are (especially) NL teams using an increasing number of lousy defensive OF in an effort to get a bat in the lineup? In addition to the above trio, consider Bradley, Burrell, Lee, Dunn was an OF until 2010.

Finally, more players are playing at a later age which means more old/DH dudes like Matsui and Vlad playing OF in NL parks. Lots of 1B do that, too, but it's been happening since 1997.

All peanuts, but it might add up to 1%.
MJMcC0
12/04
I'm not sure I buy the conclusion that there's significant evidence of a change in park effect in these data. I did a very simple Z score analysis, using both the full data set and running 20-year samples. By both measures, there's not been a Z score over 2.0 since 1978. And over the full 111-year history of these data, only two years in which the 20-year Z falls outside the +/-2 band (1948 as well as 1978) and just 8 using the mean & std deviation for the whole (most before 1932). And this despite a relatively steady decline in the 20-year standard deviation throughout.
swartzm
12/04
Z-score analysis on one-year samples requires a 56% Home-Field Advantage (up from 54%) to show up as statistically significant. There simply are not enough games in one season to confirm a switch to 55-55.5% Home-Field Advantage either way, so you will always discover no effect. It seems that you also may have used the standard deviation of the data, which is not the way to analyze this at all. We don't need sample standard deviation-- this is a binary variable. The standard deviation is sqrt(p*(1-p)/n) in any set of n games with a HFA of p. I think that using 3 years is going to be the best way to look at this, since we're asking about the last 3 years, and it's worth looking at 4 or 5 (like I did in the article) just to check if I'm cherry-picking data. You also would have found a significance in the 8-year period of 2003-2010 had you used the correct standard deviation formula rather than the sample standard deviation.
adamst
12/05
At the start of the article, you say the likelihood of such a high home field advantage being random variation is 1-2% depending what you measure. At the end, you say you're skeptical that a change has taken place. Are you saying you don't trust the numbers?

I don't have THE theory why, but I suspect a lot of things could cause a .1 to .2% change and put together could move HFA 1%. Teams and managers seem to be specializing more with 7-man bullpens and playing matchups. There seems to be a trend toward pitching your closer in a tie game in the top of the 9th or extra innings; that philosophical change would help home teams.

Is home field advantage greater in lower scoring periods? What's the correlation between league R/G and HFA?
swartzm
12/06
What I'm basically saying is that a 1% chance of such an extreme three-year period happening by random chance is still a chance, and that I'd bet this is just a fluke. I'm not saying I don't trust the numbers-- I calculated them myself. I'm saying that there's still a chance that it was a random fluctuation, and that I'm guessing that's what it will end up being. I'm sure that there is some hitter who is hitting over .500 in Tuesday day games, despite how improbable that is, and I don't think he's actually an over .500 hitter. But 1% of hitters will be 2.5 standard deviations above their true talent level in Tuesday day games, which would yield false positives. That's my best guess about what's happening here, but I'm very unsure.

I don't think HFA and R/G are correlated at all, given that the numbers have barely moved for 60 years. Also, the R/G has fallen since 10 years ago but risen since 20 years ago, so we're not in a unique period in terms of R/G.
TangoTiger1
12/06
Even if the observed .550 is significantly different from .540, that does not mean that the true level is now .550. It just means that it's above .540. It could now be a true .541. After all, if you make the baseline .545, an observed .550 is no longer statistically significant, and therefore, we have to accept .545 to still be the true.

In other words, the observed performance only indicates that the true has moved, but not that it has moved completely to the new observed level.