ArticlePDF Available

Playing Like the Home Team: An Economic Investigation into Home Advantage in Football

Authors:

Abstract

Home advantage in football varies over time. Existing theories of home advantage struggle to explain this time-series variation. We argue that the decline in home advantage in English football since the mid-1980s was partly caused by the advent of televised football. We argue that the increase in live television coverage of football matches has worked to incentivize players to not to shirk when playing in away games, as supporters can now more effectively monitor their efforts. We test this hypothesis using both time-series and panel-data econometrics.
Playing Like the Home Team: An Economic Investigation into
Home Advantage in Football
Mark Koyama and J. James Reade
Department of Economics, University of Oxford
September 19, 2008
Abstract
Home advantage is well-established and long-lasting phenomeon in team sports. Home advantage in
football however varies over time. In particular, we find that home advantage in English football has
declined since the mid-1980s. Existing theories of home advantage struggle to explain this time-series
variation. We argue that this decline was partly caused by the advent of televised football. Previously
most spectators could only monitor the performance of individual players in home games. The advent
of televised games constituted an improvement in the monitoring technology available to spectators who
do not attend away games. We test this hypothesis using both time series and panal data econometrics.
1 Introduction
A phenomenon commonly observed in many sports is home advantage; a team competing in familiar sur-
roundings, controlling for ability, wins more than 50% of the time. Anecdotal evidence for this might come
from the strong performances of host nations in the football world cup, or the strong medal performance of
the country hosting a given Olympic Games. In football domestically, there remains a strong advantage
to being at home in all countries, subject to some variation, as Pollard (2006) shows. Explanations vary
considerably for football, from transportation costs, through implicit referee bias to levels of testosterone in
players.1
Despite the number of possible casual factors cited by researchers, there has been little research on the
mechanism generating an systematic advantage for home teams in football. Economics, and principle agent
analysis in particular, is well suited to analyzing this problem. What are the incentives affecting a football
player as he or she plays for his team in a given week? Do they contribute to making the team more or
less likely to succeed? We propose a mechanism through which supporters and television combine to affect
home advantage. This model suggests that the ability of supporters to monitor effort of football players in
both home and away matches in the TV and internet era means that no longer can players shirk on away
games and expect to still be a firm favorite with the supporters. We test this and other hypotheses by
exploiting recent advances in modelling football match outcomes to control for team ability in measuring
home advantage, before using this output in time-series and panel data models. We find that existing
explanations can only explain part of the variation in home advantage, and that our monitoring hypothesis
has support in the data.
The outline of the rest of the paper is as follows. In Section 2 the literature on home advantage is critically
reviewed, before Section 3 introduces our model of monitoring. Section 4 contains our econometric analysis
of home advantage motivated by the new and existing theories, before Section 6 concludes.
1In this paper, by football we refer throughout to association football or soccer, as opposed to American football.
1
2 Causes of Home Advantage
Home advantage is a complex phenomenon and it does not seem possible to identify a single factor responsible
for the pattern of home advantage we observe across different sports, and different leagues, through time.
Figures 1 and 2 introduce the kinds of issues to be discussed in this paper. Figure 1 plots, for all four English
soccer divisions, two common measures of home advantage.2On the top panel, the percentage of all available
points that are won by the home side proportional to the available points won by the away side is plotted,
and on the bottom, the number of soccer matches each season that are won by the home side is expressed as
a ratio of the number of matches won by the away side. Figure 2 then allows an international comparison,
plotting the English values against those of other major European footballing nations. Controlling for all
other factors, such as ability, one would expect that home teams win as many matches as they lose; as such,
even allowing for the existence of the drawn match outcome in soccer, both these ratios should be around
unity.
The four English leagues appear to show a fairly consistent picture at first glance; home advantage is declining
steadily; around the turn of the 20th century, and well into the inter-war years, the home side won more than
twice as many matches as the away side, and amassed nearly three times as many points. By the 1990s
and the turn of the 21st century, this ratio has dropped to only just over 1.5, and appears to be trending
downwards almost unremittingly throughout the history of soccer. This is true for the lower three divisions
of English football; however, it appears that for the Premiership, home advantage has been declining in
shifts, and has remained quite stable for lengthy periods of time. Since the late 1980s, the ratio of home
superiority appears to have stabilized, having fallen from a high in the mid-1970s.
Travel fatigue and rule advantages have been proposed as factors contributing towards explaining home
advantage. Players have to make long journeys, and particularly for lower divisions, the journey takes place
on the day of the match. In season 2007–08, Carlisle United, one of the most remotely located football clubs,
won 16 consecutive home matches, and Portsmouth, similarly remote on the South Coast, are perennially
strong at home. However, Figure 1 offers some initial evidence to the contrary; between 1921 and 1958,
the Third Division (labeled League One, its current name) was split into North and South sections, hence
reducing substantially the distance traveled by competing teams. However, from Figure 1, the League One
series is consistently higher than all other divisions, suggesting stronger home advantage in these regional
divisions over the period. After 1958, bar one season in the 1970s, the League One and Two series (the
North and South divisions became two national leagues in 1958) are indistinguishable from the top two
divisions, suggesting that home advantage fell back in line after the divisions stopped being regional, and
once distances traveled increased.
The most studied factor is crowd support and crowd size. Nevill et al. (1996), considering just one season of
English football, find attendance to be a significant factor, but Pollard and Pollard (2005), cast doubt on the
crowd hypothesis by considering more than one season, albeit without any statistical analysis. From Figure 1
home advantage in the Premiership, where crowds, on average, are largest, is not distinctly different from
other divisions, and additionally Figure 3 plots home advantage against average attendance in the English
Premiership between 1947 and 2007, and there is little relationship between the two.3It actually appears
like home advantage has slightly declined since the mid-1970s, while attendances have risen dramatically,
almost achieving 1950s levels.4
2For a more detailed historical exposition of the origins of the English soccer league, see Pollard and Pollard (2005). Addi-
tionally, in 1992 English football underwent a restructuring, after which the First Division became known as the Premiership.
Subsequent renamings mean that the division below the Premiership is called the Championship, with the third flight of English
football now known at League One, and the fourth division is League Two. In this paper the most recent names for divisions,
where possible, will be used throughout.
3An OLS regression of home advantage on attendance yields a tiny regression coefficient of 0.000003, which is insignificant
with a t-ratio of 0.552.
4This perhaps hints at a reverse effect documented in Dohmen (2008), that of football players choking given the pressure of
higher crowds.
2
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
1.5
2.0
2.5
Ratio of Total Home Points Won to Away Points Won
Premiership
League One Championship
League Two
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
2
3
Offside rule Substitutes 3 pointsPlay−offs
Ratio of Total Home Wins to Total Away Wins
Premiership
League One Championship
League Two
Figure 1: Plots of two common measures of home advantage for all four English Soccer Divisions, 1888-2007.
Top panel is the percentage of games won by the home team, and bottom panel is the percentage of total
points available won by the home side. Important dates in soccer history are noted on the plot also.
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
2
3
4England
Netherlands
Italy
Scotland
Germany
Spain
Figure 2: Plots of the percentage of matches won by the home side for various European countries.
Another possible cause is refereeing bias. As well as, or instead of, influencing the behaviour of players,
crowd size may also affect the decisions of the officials. Nevill et al. (1996) found that the number of sending-
offs and penalties favoured the home-side, Sutter and Kocher (2004) confirms a bias towards the home side
in decisions on penalty awards and end-of-game injury time in the German Bundesliga. Boyko et al. (2007)
find refereeing decisions and crowd size are the two significant variables that determine home advantage in
the English Premiership, although Johnston (2008) refutes these findings on a slightly more general dataset,
by additionally controlling for stadium capacity; many of the stadia considered in the Boyko et al. study
are full week in, week out. Additionally, there is a strong possibility of endogeneity in these conclusions.
Do teams have higher attendances because they are successful, and because they are successful they win
3
1.50
1.75
2.00
20000
30000
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Home Advantage and Attendance, English Premiership
Home to Away Points Ratio (left scale) Attendance (right scale)
Figure 3: Plot of home advantage (ratio of home points to away points won, left scale), and average atten-
dance (right scale) for the English Premiership, 1947–2007.
many home matches? Furthermore, the referee effect Boyko et al and Sutter and Kocher (2004) find may
be explained by the amount of attacking that good sides do; the more attacking a team does, the more likely
referees award decisions in their favour, particularly penalty kicks. So good sides win more home matches,
and also receive more decisions from the referee; the two may be correlated, but not necessarily causal.
Another possible hypothesis is that the degree of competition in a league will affect how much home advantage
exists; Forrest et al. (2005) suggest that that home advantage is positively related to competitiveness.
Competitive balance in a soccer league can be calculated using any of the competition measures developed
in industrial economics, and papers such as Brandes and Franck (2007) do just this. Figure 4 plots a
number of measures of competition in a division; the top panel is the standard deviation of points won by
each side, the middle panel the Herfindahl index, and the bottom panel the CR5 index, which measures how
dominant the top (biggest) five clubs (firms) in a division (industry) are.5All these measures show that since
the mid-1980s, the Premiership has become less competitive, while competitiveness in the Championship,
accounting for the three-points-for-a-win effect on the standard deviation measure, has remained stable,
but home advantage has fallen in both. This suggests that they may be something more than simply the
competitiveness argument dictating changes in home advantage.
Finally, it has been claimed that home advantage may have a deeper cause. Perhaps a biological one? Neave
and Wolfson (2003) find that players record higher levels of salivary testosterone before a home game then
they do prior to away games. Testosterone levels were also higher before important games between particular
rivals. The authors attribute this effect to an innate instinct males have to defend their own territory. But
their small-scale experiment has not been scaled up, nor do not establish a direct link between testosterone
levels and performance. Furthermore, static explanations of the existence of home advantage such as this
cannot explain why home advantage varies between sports, or why it has varied over time.
To our knowledge, only Pollard and Pollard (2005) and Jacklin (2005) consider the time dimension of home
advantage. The former investigates the home points ratio without any control for team ability, and simply
pursues a descriptive analysis. The latter uses post-war data from 1946 onwards, whereas in this paper
we consider data back to 1888. Jacklin specifically considers the effect of two events in English soccer
history: the introduction of substitutes in 1965 (before this teams could not replace injured players), and
the introduction of 3 points for a win in 1981. He finds that the introduction of substitutes increases the
home advantage, whereas the advent of three points for a win diminishes the effect. However, using the
technique Jacklin employs, he is unable to pin down the beginning of the decline in home advantage to a
5The patterns observed in the CR5 index appear to be strongly affected by the number of teams competing in a division,
and it is likely that the number of clubs in the top flight (which fluctuated between 22 and 20 in the late 1980s and early
1990s) is driving the rather odd patterns in this measure. Additionally, the standard deviation of points won is affected by the
introduction of three points for a win in 1981–82.
4
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
5
10
15
20 Standard Deviation of Points Won
Premiership Championship
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
1.05
1.10
1.15 Herfindahl Index
Premiership Championship
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
500
1000
CR5 Index
Premiership Championship
Figure 4: Measures of the Competitiveness of the First and Second Divisions of English Soccer
particular year, with estimates ranging between 1981 and 1987. Furthermore, from his plots of the ratio
of home wins to away wins (his Figure 2), it appears that the decline began during the early 1970s. We
thus question firstly his identification of a trend in home advantage, and secondly his interpretation of it;
if the decline was caused by three points for a win, then a simple shift change should have emerged, a new
equilibrium level once teams had adjusted to the new ball game in play.
The main difficulty in identifying a cause of home advantage is in specifying a mechanism that causes a
particular factor to affect home sides or away sides more. Taking the change in points system from two
points for a win to three, Jacklin (2005) argues that this significantly undermined home advantage. All
teams attempt to win their home games, and according to this argument, in the two-points era, the away
side often played for a draw. The increase in ‘cost’ of drawing a game as a result of the move to three points
for a win thus had an asymmetric effect because it encouraged attacking play for away teams.
Can this explain the decline of home advantage? The incentives facing the away side just as much apply
to the home side, so it is unclear that three points should have adversely affected home advantage. This
explanation depends on the away team changing tactics and playing more offensive football, and it would
predict less draws and more home and away wins but the number of home wins in the Premiership remained
unaffected until later in the 1980s, while the number of draws displays no obvious trend, from Figure 5.6
Figures 5 and 6 plot the percentage of home wins, away wins and draws for the top two divisions of English
soccer, alongside the goals per game for these divisions for all the years of their existence.7A solid vertical
6Figure 2, which plots the ratio of home wins to away wins for various European countries, lends even less support. Germany,
Italy and Scotland only adopted three points for a win in the mid-1990s, by which time the most drastic decline in home
advantage had taken place in these countries. In particular, Scotland’s home advantage has arguably been increasing since its
incorporation of three points for a win in 1994.
7The pattern in the third and fourth divisions closely resemble that of the second division in Figure 6 and so those plots are
5
1900 1950 2000
0.2
0.4
0.6 1981
Homewins
Draws Awaywins
1900 1950 2000
1
2
31981
Hgoals Agoals
Figure 5: Plot of the proportion of home wins, away wins and draws in the top division of English football
since 1888 (left panel), and the number of goals per game in the top division (right panel).
1900 1920 1940 1960 1980 2000
0.25
0.50
1981
Homewins
Draws Awaywins
1900 1920 1940 1960 1980 2000
1
2
31981
Hgoals Agoals
Figure 6: Plot of the proportion of home wins, away wins and draws in the second division of English football
since 1892 (left panel), and the number of goals per game in the second division (right panel).
bar represents the 1981-82 season, when three points for a win was introduced. The proportion of matches
finishing as home wins does not appear to be affected by the three point change; for the lower divisions the
steady decline in the proportion of home wins simply continues unabated, while in the top flight there is no
discernible affect of the change; there does appear to be a change in 1986-87 season, a level shift downwards,
but it would be hard to argue this was a result of the three-point introduction five years earlier. Furthermore,
the top flight upward trend in away wins that persists to this day had begun before 1975, while the number
of draws shows no effect of three points: it has been fluctuating around the highest mean level in its entire
history since the late 1960s. The same can be said for draws and away victories in the other three divisions,
and in those divisions home wins have been steadily declining throughout. An alternative hypothesis is that
there are more draws now compared to the 1950s because there is considerably fewer goals per game. All
plots show that the 1950s (and even more so the late 1920s and 1930s after the introduction of the offside
rule) were a golden age in terms of goalscoring. Another prediction of increased away attacking play would
be more away goals being scored (and also more home goals since away defences are less solid). But as
with wins, 1981–82 does not appear to affect either home or away goals. Rather, again 1986–87 appears
to be something of a watershed, although away goals are unaffected. As such, the three-points-for-a-win
hypothesis does not appear to have much support in the data.
not reported.
6
3 An Alternative Hypothesis Regarding Home Advantage
3.1 The Monitoring Hypothesis
No single factor can be expected to account for the decline in home advantage, and further most explanations
do not differentiate between home and away sides effectively. Various alternative hypotheses have been to
be examined, none of which seem able to explain home advantage satisfactorily. In this section, we draw
upon the economic literature on team performance to attempt to shed light upon a new channel through
which home advantage might affect performance.
We argue the mechanism through which the presence of home supports influencing performance is player
motivation. It is not so much that playing at home directly motivates players; this cannot be ruled out, but
also cannot be tested. Rather, our hypothesis is that in team games, each individual player has an incentive
to to put in slightly less effort than he or she would if being judged on individual performance alone.
This logic behind this hypothesis was first laid out by Alchian and Demsetz (1972). A football club is a
firm where the players are employees and the supporters are in effect consumers purchasing the performance
of the team (Sloane, 1971; Szymanski and Smith, 1997). This purchase is a bundled one. Supporters get
utility when they see their team put in a good performance but even if they missed the game they get utility
from being able to boast about the league position of their team, or if a rival team has been defeated. In team
sports it is difficult or impossible to parse out the marginal product of each individual player. This means
that team sports are characterized by production externalities. If one player improves his performance, this
not only benefits him but benefits his team members as well. Similarly if one player reduces the amount of
effort he is putting in, not only does his performance suffer but so does the performance of the other players
on the team. In football the externality problem is exacerbated by the fact that there is no common metric
for evaluating player performance. This means that it is impossible to supply the right individual incentives
to ensure optimal performance (Holmstrom, 1999). Production externalities of this kind create a ‘cover’ for
shirking as Miller (1992) puts it.
The phenomenon of shirking in group tasks is widely documented in the management and psychology litera-
ture where it is termed ‘social loafing’.8It is also prevalent in a number of team sports. Miles and Greenberg
(1993) for instance found that in competitive swimming performance within teams lagged behind individ-
ual performance confirming the shirking or ‘social loafing’ hypothesis. They found that this problem could
be ameliorated when poor performance was punished. The prevalence or perceived prevalence of shirking
amongst baseball players is examined by Krautmann (1990), Maxcy et al. (2002), Knowles et al. (2003),
Marburger (2003) and Berri and Krautmann (2006).9This phenomenon has not been extensively discussed
in the literature on football.10 The exception to this McMaster (1997) which discusses the incentive players
have to shirk in the context of a wider set of agency problems football clubs face. He argues that “despite
performance being assessed on the basis of team success, the contribution of each individual is relatively
easy to assess . . . there are constraints on the ability of players to shirk or free ride, since there is an absence
of informational asymmetries” (McMaster, 1997, 27). This argument however we suggest miss-states the
moral hazard problem created by externalities in production: The incentive to shirk results less from asym-
metric information than indescribable information (?).11 The compensation packages clubs offer to players
recognizes the problem of motivating player performance and deterring individual players from shirking.
Bonuses for wins, goals, assists and cleans sheets and other forms of performance related pay are attempts
elicit the highest level of performance possible over time. Objectively verifiable information can be built
into a contract. Overall performance however is difficult to measure since it cannot be described by a single
8See Latane et al. (1979); Karau and Williams (1981); Liden et al. (2004).
9Krautmann (1990, 1993) emphasizes how random shocks affect the performance of major league baseball players. He argues
that this accounts for the perception that successful baseball players shirk after receiving long-term contracts.
10Audas et al. (1997, 31) note that football is characterized by a team production function but they do not follow up this
idea.
11This means that performance related pay contracts are necessarily incomplete.
7
sufficient statistic such as goals scored.12 Performance is dependent on a host of different variables, some
of which are not amenable to statistical representation, and transaction costs preclude designing individual
contracts that condition on a larger set of performance measures.
The effort a player puts into a game is roughly observable. The problem of asymmetric information in this
sense is muted. The manager, the coaches, and the fans can often see when a player is shirking: a player
may not be making the runs off the ball that are required of him or an attacking player may fail to help out
in defence; defenders who are inattentive risk playing opposition players onside.13 Importantly however,
though this information is observable, it is not verifiable and therefore cannot form part of an incentive
contract. Players who fail to meet the stipulated terms of their contract can be fined or fired. Verifiable
forms of shirking such as a failing to attend a training session can be deterred by suitably specified contracts.
This means that an informal rather than formal system of incentives is required to deter shirking.
As Alchian and Demsetz (1972) argued, when production takes place in teams and inputs are costly to verify
it is necessary to monitor individual performance because each member of a team has an incentive to shirk,
i.e. to put in less effort than his fellow team members if he can get away with it. The problem this creates
is one of designing appropriate rewards for the monitor. In firms this is often done by allocating residual
property rights to the monitors. Monitoring performance is of course the purview of the manager. So is
motivation and it is the task of the manager to create a sense of camaraderie and team spirit.14 Successful
managers do precisely this. But in many cases the ability of managers to single out particular players for
criticism is limited.15 Another way of ensuring that players play well is supporter pressure. Football fans
watching a game live can easily detect when a player is shirking. They can see whether players are making
appropriate runs or whether or not they are shirking.
This argument does not rely upon discounting monitoring by managers or coaches; rather we claim that the
ability of the manager to monitor players is imperfect and his ability to sanction players he thinks are shirking
is limited. Therefore monitoring by fans augments monitoring by the manager. Fans are comparatively
cheap monitors since they have internalized the benefits of sporting success. Furthermore, they can impose
different kinds of informal sanctions of players they disapprove of. This, we suggest, is one of the roles fans
perform in professional football. As a result players put more effort in, and play better when they are being
watched by their own fans. Players who do visibly shirk are signalled out by the fans. Moreover players
who are unpopular with the fans almost inevitably leave the club.
How does this relate to the phenomenon of home advantage? Fans as a whole are better able to monitor
individual player performance in home games than in away games because a far higher proportion of fans
attend home games relative to away games. More specifically, they are better able to measure the effort
an individual player exerts in a home game. Therefore average player performance will be higher at home
games because shirking is more effectively deterred in home games. Our explanation does not discount the
role of managers. Rather we think that successful managers are those that can overcome these incentives.
But we think that these incentives do in fact exist.
If a fan does not attend an away game, he can easily find the result of the game, read a match report and
perhaps watch highlights on television. All of this gives him an idea of how well the team as a whole
performed but less of an impression about how each individual player performed. Therefore it is likely that
some level of home advantage will always exist. But our framework also predicts that innovations such as
televised games and the internet will reduce the advantage a home side enjoys, at least to the extent that
such technologies enable fans to monitor the performance of individual players. In the next section we
develop this hypothesis more formally.
12The consequences of performance contracts conditioning on a single statistic is outlined in Nalebuff and Stiglitz (1983).
13A couple of issues of nomenclature; first, in this paper the terms spectator, supporter and fan will be used interchangeably
to refer to people that devote their allegiance to a particular team. Secondly, in the UK, the term manager is used to describe
what might elsewhere be called the coach, or the head coach, and so we use the term manager to describe the member of the
coaching team who has overall executive power.
14A sense of team cohesion has been found to deter shirking in other sports (Miles and Greenberg, 1993).
15Such behaviour is, appropriately or not, deemed by fans to be detrimental for team spirit if undertaken by a manager.
8
Pressure from supporters can also undercut performance. Our argument is perfectly consistent with Dohmen
(2008) finding that during penalty shootouts the probability of home players ‘choking’ or missing the penalty
altogether is higher than for visiting players. High pressure situations and high expectations can adversely
affect performance; the penalty shootout stage of a knock-out game is qualitatively very different from league
football. Therefore we find it unsurprisingly that teams can enjoy home advantage in league football and
home disadvantage in penalty competitions.
3.2 Predictions of the Monitoring Hypothesis
In the context of a sport like football, the performance of individual players is difficult to monitor. In
this section we develop a simple theoretic model that captures this insight. Drawing on insights from
multi-task principle agent theory, we show that when monitoring ability varies across environments, relative
performance is distorted (Holmstrom and Milgrom, 1991).
There are two environments: H ome and Away. The environments vary according to the monitoring tech-
nology. Since a higher proportion of fans attend home games than attend away games, they are better able
to evaluate the performance of individual players at Home than they can away from home. Note we do not
assume that fans care more about home performance than they do about performances away from home.
The only variation is in the monitoring technology.
Denote the amount of points a team expects to win in a league, by E(P) = E(PH)+E(PA) where superscripts
denote home and away respectively. Conditional on team ability, we argue that the number of points a number
obtains depends on the amount of the effort the players as a team put into the performance. Therefore we
can write, E(P) = E(P[e|x]) as a function of team effort ewhere eis a vector comprising {e1, e2, . . . , en},
conditional on a number of exogenous variables x, where there are nplayers in the team.
Players choose the level of effort to put into a game e(0,1). Effort however is costly. The cost of exerting
effort is given by c(ei) where c0(ei)<0. The utility of player iis therefore given by,
max ui=wc(ei) (1)
If effort is perfectly measurable and each individual’s wage packet depends solely on the effort they expend
and w=w(ei) then each player equates marginal benefits with marginal costs. In this case each player puts
in the optimal amount of effort into each game as given by: w0(ei) = c0(ei). Denote the amount of effort e
that solves this equation e
We assume that players are paid according to how much the fans value them and that this in turn depends
on how much effort they are perceived as putting in on the pitch.16 These assumptions are captured by the
following relationship between wages and effort.
w(e) = (1 µ)ei+µi.(2)
where iis an error term uniformly distributed along (0,1) and µis a variable between 0 and 1 reflecting
the extent to which a player’s performance can be monitored by the fans.
The effort a player iexpends in each environment can be denoted by: eH
iand eA
irespectively. The ability
the fans have to monitor the performance of individual players depends on whether or not they attend the
match, or whether or not it is possible to watch the game, or perhaps highlights of the game on television.
These factors will determine µ.
We are interested in how players distribute their effort between home and away games. For simplicity
consider the following polar cases. At Home we suppose that µ= 0 while at Away,µ= 1 Under these
16Players are paid by clubs but we assume that clubs only retain players that the fans want to see at the club. Players who
are perceived as shirking do not have their contracts renewed.
9
assumptions the first order condition for the optimal of effort a player puts in at home is the same as the
first-best efficient level: w0(eH
i) = c0(ei). Away from home: the relevant first order condition is: c0(ei) = 0.
Therefore players put in no effort away from home.
eH=e(3)
eA= 0 (4)
More generally, the marginal benefit of expending extra effort is falling in µ. Therefore, so long as µis lower
at home than it is away from from home: µH< µA, i.e. so long as it more difficult to accurately monitor
effort away from home than it is at home, the amount of effort eplayers exert in will be greater at home.
In order to complete this model we have to specify the relationship between the effort players expend in
a match and overall team performance. Effort is only one among many possible determinants of match
performance. But we expect that if the mechanism we have identified is plausible then, controlling for the
overall ability of a team, and other exogenous factors such as the competitiveness of the league, and the
distance travelled by each team, the relative advantage of the home team over the away time should depend
in part upon the extent to which supporters are able to monitor the performance of individual players. We
can formulate this prediction as follows.
Proposition 1 As the ability to monitor players performance improves, and µfalls due to increases in the
proportion of away fans attending matches, television coverage, or the internet, the ratio of points won at
home relative to points won away from home will fall.
This states that the proportion of points won at home: E(PH)/E(PA)=[E(P)E(PA)]/E(PA) is falling
in µ.
∂E(PH)/(PA)
∂µ <0.(5)
One advantage of this hypothesis is that it is compatible with the the finding that home advantage is
a phenomenon unique to team sports. Tennis players and other individual sportsmen do not experience
particular benefits from playing at home because their pay depends solely on their own performance. It is
also consistent with the finding that in American sports, home advantage appears strongest in indoor sports
such as Basketball where the supporters are close to the players.17
A second advantage of this explanation is that it is consistent with the views of many supporters as reported
by Wolfson et al. (2005). Wolfson et al. (2005) surveyed a large of number of supporters in an attempt
to assess their own view of home advantage and the role of crowd support in explaining home advantage.
Comments considered representative include: ‘Players try harder at home because they are in full view of
their own supporters’. And that ‘Players are probably keen to put on a better display’ when they play at
home (Wolfson et al., 2005, 369). Football fans clearly perceive their role to one of motivating the players in
this light. They do not only encourage good play. They also single out players viewed as “lazy” for special
abuse. In the next section, we detail the growth of televised games in the 1980s and 1990s to see whether
or not the timing of this expansion coincides with the relative decline of home advantage.
Our argument is also consistent with a slightly different hypothesis. First, note that effort is costly and
professional football in draining, physically and mentally. Since players can play up to 60 competitive games
a year, it makes sense for them to ration their strength. Second, supporters are the customers of a football
club; they ultimately pay the wages of the players. The phenomenon of home advantage arises because
football is essentially a bundled product. When a supporter buys a ticket she cares about an individually
weighted sum of the entertainment she is going to get simply from watching a good football match and the
result. Causal footballer supporters may attend a home game simply hoping for a good game. Other fans
are more interested in the result; they want to see their team perform well over the course of the season. To
17Pollard and Pollard (2005) finds that home advantage in basketball is currently about 60 percent.
10
1983 1985 1986 1988 1992 1997 2001 2004 2007
Length of contract (years) 2 0.5 2 4 5 4 3 3 3
Broadcaster BBC/ITV BBC BBC/ITV ITV BSkyB BSkyB BSkyB BSkyB BSyB/Setanta
Rights fee (£m) 5.2 1.3 6.3 44 191.5 670 1,100 1,024 1,700
Number of live matches per season 10 6 14 18 60 60 66 132 132
Table 1: Number of matches televised per season in the English Premiership. Source: FA Premier League
and BBC News
the extent that a proportion of football supporters fall with the former category, then players always have
an incentive to ration their effort in favour of home rather than away performance since more fans watch
home games. If a player invests an additional amount of effort in an away game then less fans of the former
type will appreciate it. This suggests that players are simply giving fans what they want by playing better
at home.
3.3 Television Coverage of Football in England
Are the predictions of this hypothesis consistent with the available data? As shown in Table 1 the number
of televised games increased throughout the 1980s and 1990s. The current television rights package offered
by the Premiership is for 132 live matches per season, of the 380 matches played in total. BSkyB paid
£1.7bn for the right to show 92 of those matches. In 1983-1985 the BBC and ITV paid just £5.2m to
the then Football League to cover 10 matches per season between 1983 and 1985. However no games were
televised in the 1985–86 season due to industrial action. Therefore it was only in 1986, that the number
of matches televised more than doubled from 6 to 14. From November 1988 onwards live football appeared
on the television most Sundays each season. As such we associate the period from around the mid-1980s
onwards as a new regime, one in which monitoring is less costly for supporters. Even if a team is not playing
live, the increased possibility of highlights being shown on television increases the probability shirking can be
detected, and hence we hypothesize that this period in English football is one characterised by a falling home
advantage, at least for the top division of English football, where the overwhelming majority of television
coverage and television money has been focussed.
There exists a complementary mechanism which would link increased television coverage to reduced home
advantage; notably the hugely increased volumes of money in football that has resulted from the phenomenal
inflation in the fee paid for television coverage of the Premiership since 1986. The financial cost of being
caught shirking has increased commensurately, and assuming that players were already investing maximal
effort in their home matches, the only arena in which they could increase their effort to reflect the increased
cost of shirking would be in away matches.
4 Econometric Analysis
Having considered a number of potential explanations for home advantage, we now turn to data analysis to
attempt to quantify the roles these various factors may have played. First the data used will be introduced,
before we analyse this data using formal time series and panel data methods to investigate the causes of
home advantage.
4.1 The Data
Football data is widely available on the internet; www.soccerbase.com provides data back to the beginning
of the English and Scottish Football Leagues, and hence provides a rich store of data on home advantage.
Data can also be found on the website for the major European leagues since around 2000, and further
11
back for Italy (1990) and Germany (1963).18 Time series of individual team performances for a given
season can be constructed once the data is extracted from the website. Attendance data is found at
www.european-football-statistics.co.uk/ back to 1947 for each team; this is only the season average.
Soccerbase stores attendances for individual Premiership and Football League matches back to 1992; an
extension of this work envisaged is to run the bivariate Poisson model above including attendance, and
distance visiting team has travelled, as regressors and test their significance. The distances that teams
travel are calculated using data from www.postcode.org.uk; the postcode area for each team (e.g. SW1 for
Fulham) can be used to get a measure of the distance between each football stadium in England.
As with Pollard and Pollard (2005) and Jacklin (2005), we consider the time-series dimension of the data.
Each season provides a treasure trove of information on home advantage; the Premiership in England is
played over 380 games, and by considering any of these three measures, some handle on home advantage can
be gained. Furthermore, by considering each division over a large number of years, one can detect whether
home advantage has been changing over the years. This can help in discussing possible factors in explaining
home advantage.
4.2 Measuring Home Advantage
A number of measures to capture home advantage have been used in the literature; Attrill et al. (2008) and
Nevill et al. (1996), use the percentage of a team’s matches that are won, while others, such as Pollard and
Pollard (2005), have used the ratio of points won to total available points because a draw (a tied match) is
worth a point in football. Boyko et al. (2007) use the goal differential in Premiership matches, controlled for
the average attacking strength of the home and away teams. Boyko et al. also model match outcome using
a Probit model to try to measure home advantage, while Clarke and Norman (1995) use a linear regression
model.
Measures of efficacy at home, such as ratio of total home points won, are not necessarily informative; if a
team is effective both at home and away, it is probably a strong side; taking the ratio of home success to
away success might enable us thus to get some idea of just how much more effective a team is at home, and
may control to some extent for team ability.19 Furthermore, this ratio of home to away performance seems
much closer to the relevant variable of interest: it is not interesting per se that a team wins many home
matches; the variation in home performance relative to away performance is what defines home bias.
Nonetheless, it is not clear how well the ability of the teams competing in any given match is controlled
for in measures that do not explicitly model match outcome. In recent years methods of modelling match
outcomes in football have progressed considerably; Dixon and Robinson (1998) considered a two-dimensional
birth process for goal arrival in football matches which they showed to work effectively for English football,
while Karlis and Ntzoufras (2003) developed bivariate Poisson regression models to capture the arrival of
goals. Goals in football matches appear to follow a Poisson process, as their variance is approximately equal
to their mean; Table 2 shows that indeed, for English football since 1990, the Poisson approximation appears
a very good one, as the mean and variance are very close to each other for each team’s goals (H and A), and
for total goals. The dependence between these two processes must be accounted for though: the probability
of one team scoring a goal depends on the opposition faced. The bivariate Poisson model captures this, and
provides a way of modelling football match outcomes that is relevant for the current context; by estimating
bivariate Poisson models for each football season since 1888, we can derive a measure of home advantage that
adequately controls for team ability. Because goals scored, and the winning margin in a football match,
are discrete variables, it makes sense to consider such distributions. A comparison between count data
modelling of match outcome and limited dependent variable modelling such as that of Boyko et al. would
make an interesting study.
18Another even more comprehensive source of international football results can be found at www.rsssf.com.
19If team ability is a multiplicative effect, then taking the ratio of home and away performance should mean the ability factors
cancel. Jacklin (2005) also uses the ratio of home-to-away points won.
12
Goals
Period Moment H A Total
1990-2008 Mean 1.5 1.1 2.6
1990-2008 Variance 1.5 1.1 2.7
Table 2: Mean and variance of goals scored in English football matches, 1990-2008.
We thus model a team’s rate of goal arrival in match ias a Poisson distributed variable, with an intensity
parameter λ. We specify three random variables, X1,i,X2,i and X3,i as independently Poisson distributed
with parameter λκ,i,κ= 1,2,3, and defining Xi=X1,i +X3,i and Yi=X1,i +X3,i , then Xiand Yijointly
follow a bivariate Poisson distribution BP (λ1,i, λ2,i, λ3,i ) with density function:
PXi,Yi(xi, yi) = P(Xi=xi, Yi=yi)
= exp {− (λ1,i +λ2,i +λ3,i)}λxi
1,i
xi!
λyi
2,i
yi!
min(xi,yi)
X
k=0 xi
kyi
kk!λ3,i
λ1,iλ2,i k
.(6)
Then E(Xi) = λ1,i +λ3,i, and E(Yi,i ) = λ2,i +λ3,i , while Cov(Xi, Yi) = λ3,i . Hence λ3,i captures the
dependence between the goal arrival rates of the two teams competing in a match. With λ3,i = 0, then
(6) reduces to two independent Poisson distributions, a double Poisson distribution. To capture additional
factors that might affect the goal arrival rates of the teams, such as weather conditions, vectors of explanatory
variables wκ,i can be introduced, and a bivariate Poisson regression can be run:
(Xi, Yi)BP (λ1,i , λ2,i, λ3,i ),(7)
log (λκ,i) = βκwκ,i , κ = 1,2,3.
The wκ,i are specific to the match and parameter being estimated, while βκis the vector of regression
coefficients. The regression models run for the three parameters are, where hidenotes the home team, and
aithe away team:
log(λ1,i) = µ1+β1,1hi+β2,1ai,(8)
log(λ2,i) = µ2+β1,2ai+β2,2hi,(9)
log(λ3,i) = µ3+γ1hi+γ2ai.(10)
So λ1,i is estimated based on home goals in a game, and hence gives parameter estimates for β1,1and β2,1,
home attacking ability and away defensive ability respectively, while λ2,i is estimated based on the away goals
scored in a game, giving equivalent parameter estimates for home defence and away attack. The covariance
parameter λ3is allowed to vary based on which team is playing at home or away in the given match. The
specification of the regression model can be altered to differing effects; by setting µ3=γ1=γ2= 0 in (10)
no dependence between the two goal arrival rates due to characteristics of the competing teams is assumed.
In football, this seems unlikely, and as such, in our models, we impose no restrictions on (10).
The measure of home advantage for a given season derived from this model is HBP
t=bµ1bµ2, notably the
difference in the intensity rates for home and away goals, when the attacking and defensive proficiency of
each team has been controlled for. This measure can be criticised; the constant term in any regression can
be badly estimated for many reasons, not least omitted variables. Furthermore, Table 3 in the Appendix
shows that in the early years of English football, the Poisson assumption was less valid: pre-WW2 goals were
considerably over-dispersed, although after the war the mean and variance of goals converge.20 Nonetheless,
this measure does come from what is an approximately appropriate statistical model for goal arrival, and
explicitly controls for home and away team ability in both attack and defence. Further, as Table 3 only
20Mixture-distribution models have been constructed to account for different types of dependence, notably by McHale and
Scarf (2007)
13
1900 1950 2000
1
2
3Premiership
(H/A)t HBPt
1900 1950 2000
1
2
Championship
(H/A)t HBPt
Figure 7: Comparison of measures of the home advantage in the top two divisions of English football. (H/A)t
is the home-to-away-win ratio, and HBP
tis the bivariate Poisson measure. Premiership (top division) is in
the left panel, Championship (second division) right panel.
contains the mean and variance of goals, it checks the validity of a simple univariate Poisson model, not
accounting for team strengths or the dependence between the individual teams’ scoring intensities that is
possible with the bivariate Poisson regression model.
Using every season since each division was formed in the English football league this model was estimated,
and the resulting time series for the top two divisions are plotted in Figure 7, alongside the home-to-away
win ratio. The two measures show considerable similarity throughout, although HBP
thas a smaller variance.
For both the divisions, the upward trend apparent between the wars in the home-to-away points ratio is
absent in HBP
t, and throughout, HBP
tseems less erratic than the home-to-away win ratio, perhaps because
it better controls for ability. Nonetheless, the plateau effect noted earlier in the Premiership remains when
ability is controlled for, while the continuous decline in lower division home advantage also remains. Overall,
the Bivariate Poisson measure appears effective at measuring home advantage, and as such we use this new
measure in our time series econometric model in the next section.
4.3 Time Series Analysis
Aggregating over divisional clubs and seasons, time series analysis can be used to attempt to explain the
variation in home advantage witnessed over the history of English football. Our main hypothesis is that
increased TV coverage enabled supporters to much better monitor shirking; to measure this, a dummy
variable taking the value zero before 1987, and unity afterwards, is created. While a variable containing
the number of matches televised per season from Table 1, or a variable with the price paid by television
companies could be used, we take a simple binary variable to reflect our believe that the increased televising
of football ushered in a new era of monitorability on the part of supporters; we believe that regular live
television, as became a feature after around 1987, constitutes a structural break in home advantage, and this
hypothesis can be investigated by testing the significance of the TV variable. Naturally, other interpretations
are possible for a simple dummy variable like our TV variable; other plausible interpretations are discussed
in Section 5.
The models in this section are selected using Autometrics (Doornik, 2006), which is a model selection
algorithm that builds on the ‘general-to-specific’ statistical modelling theory of (Hendry, 1995), and on
the experience automating the process in, inter alia, Hendry and Krolzig (2001) and Hendry and Krolzig
(2005). The algorithm requires a general well-specified econometric model that covers all possible theories
of determination for the variable of interest (here, home advantage). The model must be well specified
in that the residuals satisfy the assumption of Normality placed upon them; then confidence can be had
14
that nothing systematic has been missed in attempting to explain the phenomena.21 Then variables can
be omitted if they are insignificant (using tand Ftests), and provided that the resulting model passes all
the diagnostic checks. When no further reduction is possible, a candidate model has been found. At this
point one can either choose to average remaining models, or select between them; we choose to select one
particular model (Hendry and Reade, 2007).
At an aggregate level, the hypotheses set out in earlier sections can now be tested; the first model considered
regresses our Bivariate Poisson measure of home advantage in the Premiership on its own lag, seasonal
average attendance, average total distance between clubs in the Premiership each season, a measure of the
competitiveness of the Premiership each season (the CR5 measure), and our TV variable capturing the
entering of the new TV age in English football. We focus on the Premiership because the vast majority
of TV attention has been focussed on this division.22 Because attendance data is only available from 1947
onwards, the sample is restricted, with the lagged dependent variable, to 60 observations 1948–2007:
HBP
t= 0.46
(0.21) 0.03
(0.13)HBP
t10.12
(0.04)TVt+ 0.000002
(0.000002)Attendancet
+ 0.00007
(0.0002) CR5t+ 0.00004
(0.00005)Distancet+bεt,(11)
bσ= 0.08,R2= 0.32, F (5,54) = 5.24[0.001]∗∗,DW = 1.92,
FAR(2)(2,52) = 0.62[0.54],FARCH(1)(1,52) = 0.60[0.44],
χ2
Normality(2) = 0.18[0.92],FHetero(9,44) = 1.28[0.26].
First, the regression model comfortably satisfies all the assumptions placed upon it; the null hypothesis that
the residuals are independently Normally distributed cannot be refuted. This tells us that the remaining
output of the model we can be confident about, not least the standard errors for the coefficients. Second,
the only significant variable is the TV effect; consideration of Figure 7 should give some indication of
why: there does appear to be a break after about 1986; after this point, the mean home-to-away points
ratio drops to around 1.5 from about 1.75. As can be seen, attendance has no discernable affect at this
level of aggregation, and neither do distance travelled, or the competitiveness of the division. Support
of our monitoring hypothesis is offered by the significant TV parameter; home advantage does appear to
have undertaken a structural shift at around the same time that TV became more and more predominant in
English football and hence the threat of monitoring. The model can satisfactorily be reduced down; omitting
the insignificant variables does not alter the diagnostic test output, giving a post-war model depending solely
on the TV dummy. However, if attendance can safely be omitted, then our sample can be extended back to
1888, and our regression re-run.23
The full-sample model, which is selected using Autometrics, is:
HBP
t= 0.70
(0.03) 0.11
(0.02)PostWart0.10
(0.02)TVt0.00004
(0.00001)Distancet
+ 0.29
(0.07)1901t0.22
(0.07)1908t+ 0.19
(0.07)1965t+bεt,(12)
F(6,93) = 36.44[0.000]∗∗,R2= 0.68,DW = 2.06,
FAR(2)(2,99) = 0.04[0.96],FARCH(1)(1,99) = 0.002[0.97],
χ2
Normality(2) = 0.21[0.90],FHetero(7,93) = 0.32[0.94].
The model is well specified, and explains twice as much of the variation in home advantage than the post-war
model could. It is notable how easily the diagnostic tests are passed by our model; this seems to suggest
21This does not necessarily mean no further significant variables will found if one searches hard enough. It means though
that the inference on the included variables will satisfy the confidence intervals placed around it.
22The TV variable is also significant in the Championship model, but smaller due to the smaller TV exposure in that division,
and also due to the different dynamics of home advantage in that division apparent in Figure 7.
23This of course assumes that were attendance available as a variable back to 1888, it would not be relevant even over
the longer sample. The experience with the distance variable suggests that this kind of stationarity might not be a good
assumption.
15
that any variation above these constant levels is white noise, perhaps explaining why many investigators
have found difficulty attempting to explain home advantage. There is a post-war effect, which can be seen
in the plots in Figure 7, reducing home advantage by about 16%, while the TV effect appears to be there
again, and reduces home advantage by around a further 16%. The measure of competitiveness was found
to be insignificant, but the measure of distance is significant and negative, which is slightly surprising. It
seems likely however that any causality found here is spurious; when the English football league began,
the twelve competing teams were all from the Midlands and the North West of England, and the average
distance travelled was 916 miles; by the First World War this number reached 2500 miles, and by 1921 this
rose to above 3000 miles, and post-Second World War the number appears to oscillate around 3500 miles
per team. During this same period home advantage has declined; it was very strong pre-First World War,
and appears to have declined somewhat by the inter-war period (when ability is controlled for, see Figure 7),
before falling again post-Second World War. It seems difficult to imagine any causal link between these two
variables, given the sign of the coefficient, and hence it appears safer to assume the correlation is reflecting
some other, omitted, variable.24 Finally, three outliers were found (by searching for standardised residuals
larger than 2.5) for the seasons 1901–02, 1908–09, and 1965–66, the latter coinciding with the advent of
substitutions in English football.25
Thus our time series analysis has suggested that the TV monitoring hypothesis may have some tentative
support in the data, as a structural break does exist around the time we hypothesise television began to
exert a serious effect on footballers in England. We have also confirmed the difficulty in explaining home
advantage; there appears to be little variation above simple white noise, when structural breaks for the Second
World War, and the television era are accounted for. Although we are unable to explain the mechanism
behind the Second World War shift, we have provided a plausible mechanism for explaining the TV shift.
4.4 Panel Data Analysis
Various levels of aggregation are possible, and previous studies have considered the overall home advantage
on display in a given country or division within a country; Figures 1 and 2 do just this, plotting a single
number for a given season, and the models of the previous Section do the same. However, it may well be
that much of explanatory power of the data is dulled by aggregation. The data sources mentioned above
afford a panel dataset with 155 cross-sectional units (football clubs), with up to 61 time-series observations
(1947–2007), although for a number of the clubs, not all 61 seasons have data available on attendance or
playing record, but overall the unbalanced panel dataset has 4769 observations.26
Ideally, individual matches would be modelled to try and capture the effect of television, by including a
dummy variable taking unity for every televised match. Dohmen (2008) is able to do this for German
football since 1963, but sadly to the authors’ best knowledge no resource exists that logs all televised
matches in English soccer back to 1986 (and earlier).27 Additionally, attendance data is difficult to find for
individual matches much further back than 1992, particularly for lower divisions. As a result, our measure
is somewhat blunter, and plausibly suffers from a lack of power. Countering this somewhat, we suggest
that just the presence of television enacted a regime change in English soccer. Post-1986 players knew
that the probability of observation had increased since one of their team’s away matches at any point might
be televised, hence we feel such a season-long dummy is appropriate. Our regression analysis ought to be
24If the home-to-away points ratio is used atsthe dependent variable in place of the Bivariate Poisson measure, the only
difference is that this variable is found to be insignificant, since again from Figure 7, the inter-war period shows a sizeable
increase in home advantage.
25Dummy variables are introduced to cover outliers as their magnitude may mask variation in the remaining observations,
distorting inference.
2636% of clubs have all 61 observations, while 68% have between 30 or more observations, while only 6% of clubs have 10 or
fewer observations.
27The actual results exist, and have been used above to get the Poisson measure of home advantage for a given season back
to 1888.
16
viewed bearing this in mind; it is quite likely that any significant effect reported here understates the true
impact of TV.
While using the bivariate Poisson measure of home advantage constructed in this paper for panel analysis
would be preferable, the Poisson measure is at the season-long, division-wide level which is inappropriate
as we use club-level analysis in our panel dataset.28 The distribution of home advantage measured by
the ratio of home points to away points, due to the non-negativity constraint, is quite non-Normal. For
regression purposes, as this variable will be the dependent variable in the panel data analysis, it seems wise
to transform this variable to induce a more Normal distribution; the logarithm of the home-to-away points
ratio, achieves this. Additionally, the log-linear model gives potentially more interesting interpretations of
regression coefficients in terms of percentage changes and elasticities.
For each football club and season, a number of variables can be constructed from the data already amassed
and analysed thus far. Average attendance for the season, a variable which varies greatly from season to
season within and between clubs, distance travelled by a club in a particular season, the divisional home
advantage for a particular season (does the club simply follow the divisional trend in its home performances?),
how competitive the division is in a given season, and whether or not the club is under television exposure
(defined as the Premiership after 1986, as in the time-series model above). Ability can be controlled for in
this model by including the team’s attacking and defensive strength parameters from the bivariate Poisson
model estimated earlier for every season and every division.29
The resulting panel data model is estimated using the Within Groups transformation using DPD in OxMetrics
(Doornik and Hendry, 2007, Section IV), and was arrived at after estimating a model with lagged dependent
variables and lags of all the independent variables alongside their contemporanous values. All the variables
dated t1 were very insignificant, and hence were omitted, and the resulting model is:
log Ht= 0.096
(0.008)
log GenHomeAdvt+ 0.019
(0.007)
log Distancet+ 0.602
(0.007)
log HwinRt
0.596
(0.006)
log AwinRt0.019
(0.007)
PremTVt
0.017
(0.007)
Div2t0.014
(0.008)
Div3t0.021
(0.008)
Div4t
σ= 0.13,R2= 0.86,RSS = 74.97,TSS = 541.11,
χ2
Joint sig.(12) = 1.539e+ 04[0.00]∗∗ , χ2
J.dummy sig.(162) = 447.8[0.000]∗∗ ,
AR(1)N(0,1) =0.380[0.70], AR(2)N(0,1) =0.115[0.91].
The model output shows that the model is reasonably well specified; there is no evidence of autocorrelation
in errors. Furthermore, evidence of the usefulness of allowing variation between clubs is shown by the Wald
test of joint significance of dummy variables for each club, which at 447.8 overwhelmingly rejects the null
that these effects are insignificant.30 The model is also able to explain 86% of the variation in seasonal
home advantage per club, shown by the R2statistic.
Having argued that the panel data model is of reasonable quality, we can now discuss the implications of the
model, as unlike in the time-series model, a number of interesting variables are significant. It appears that
teams follow the general trend for a given season; if other teams around them are displaying strong home
advantage, controlling for ability, they also will; for a 10% increase in the Poisson home advantage measure
for that team’s division, the team’s home-to-away-win ratio will increase by 0.8%. In addition distance
matters, with a one percent increase in the distance travelled in a given season by the average visiting club
28While constructing a home advantage measure for each club is surely feasible, we relegate this undertaking to future
research.
29This is done using the βparameters in (9) and (9).
30Time dummies are also jointly significant, but much of this effect is captured by the TV effect after the mid-1980s; the small
number of time dummies that were significant came in the 1950s and 1960s predominantly. For brevity, the (155) individual
and (61) time specific dummies are not reported.
17
to the home team stadium increasing the ratio of home-to-away points by 0.046%; teams in remote areas
are generally renowned for being strong at home but not travelling well. Both these effects are small but
significant, particularly given the variation in the distance travelled measure.
Attendance is very insignificant, arguing against the hypothesis of Nevill et al. (1996) in favour of an atten-
dance effect, but supports Pollard and Pollard (2005). It also suggests that the choking effects found by
Dohmen (2008) might be confined to more high pressure knock-out Cup competition matches. Moreover, it
doesn’t argue for strong attendance effects via the referee-decision mechanism promoted by Dohmen (2008)
or Boyko et al. (2007); although we are using season-long data, not individual matches.
The effects of winning home and away matches offset each other; the dependent variable is the home-to-away
points ratio; a 1% increase in the home win increases the home-to-away points ratio by 0.6%, while a 1%
increase in the away win ratio decreases the home-to-away points ratio.31
The controlling factors for team ability, attacking (AttStrength) and defensive strength (DefStrength) actually
come out very insignificant, which is somewhat surprising.32 However, ability is factored into the general-
home-advantage regressor, which is significant. Also of interest is the lack of a significant competitiveness
effect; all three measures plotted in Figure 4 are insignificant and omitted from
The model also documents the historic degree of home advantage in a particular division by the Div pa-
rameters, which take unity if the club is in a particular division; the reference division is the Premiership,
so the reported coefficients show the difference between the top flight and that particular division. The
Premiership seems to have the strongest home-to-away point ratios; 3.5% higher than the Championship,
4% higher than League One, and 5% higher than League Two.33 This stands in contrast to the absence of
any divisional effect that Clarke and Norman (1995) find, but seems plausible from a careful inspection of
Figure 7.
The TV effect is marginally significant, with a t-ratio of around 2; the home-to-away points ratio is 1.7%
lower for each team competing in the Premiership after the big increase in TV coverage in the late 1980s.34
So regardless of the team in the Premiership, it appears that the increased TV coverage incentivizes players
to put increased effort in away from home. A less significant estimate for the TV variable is perhaps not
surprising given the earlier discussion about the potential lack of power in the measure. Given this, the TV
effect in our panel model lends more support to the hypothesis proposed in this paper of a monitoring effect
for home advantage.
5 Discussion
We have now presented two econometric models, both of which lend support to the hypothesis of a structural
break taking place in home advantage at about the time that live television coverage of English football
substantially increased. However, we first need to establish that the structural break cannot be accounted
for by any other possible explanation. The clear break in the data series on home advantage, also visible
in the studies of Jacklin (2005) and Pollard and Pollard (2005), could be interpreted in a number of other
ways. At least three competing explanations could be offered:
1. The end of terracing in English stadia. The Hillsbrough disaster of 1989, when 96 Liverpool supporters
were killed, led to the banning of standing areas at stadia in the top two divisions in English football. It
may be the case that the huge investment in stadia after this ruling led to much more pleasant surrounds
31In the Premiership, where a team plays 19 matches currently at home in a season, then an extra win means a 5.3% increase
in the home win ratio, and so a 3.2% increase in the home-to-away points ratio.
32This result holds regardless of the panel-data model specification.
33The difference between the Conference and the Premiership is 6%, but there are only 10 seasons of Conference football
recorded in the dataset, and hence this is a less representative number.
34For clarity we should re-emphasise that the TV variable takes 1 if a team is in the Premiership after 1986, and zero
otherwise.
18
for visiting teams to perform in; certainly Pollard (2005) finds that moving to a new stadium, an extreme
version of this investment, reduces home advantage. However, the ban on standing areas only came
into effect in 1994 in the top two divisions, with the phasing out beginning in 1989, which suggests any
effect of this incident would have come later than around 1987, ruling it out as an explanation.
2. The three-points-for-a-win rule taking effect. This is the Jacklin (2005) hypothesis, and it is extensively
discussed in Section 2. However, as mentioned above, the break does appear much later than the
introduction of three points for a win, and there appears to be little evidence in support of an effect
of three points for a win, either theoretically, or empirically from other European football leagues.
3. The introduction of play-offs to decide promotion (primarily) and relegation (less often). In 1986–87,
instead of the top three sides in the Championship being promoted to the Premiership, the top two
were promoted, with the subsequent three teams competing in a play-off competition with the third-
from-last placed team in the Premiership, at the completion of the standard league season.35 The
most documented effect of this change was that often teams of poorer quality were promoted, because
promotion became dependent on a team winning a single match, as opposed to gaining the most points
over 46 league matches. However, this doesn’t seem a likely explanation for the sudden change in the
attacking ability of home sides, and the defensive ability of away sides, as the play-offs only affected one
side competing in the top division each season. It is perhaps plausible that having one less guaranteed
promotion spot meant re-promotion after demotion from the top flight became less probable, hence
causing sides to be more cautious; however again, it seems odd that this might affect home sides and
not away sides.
As such, given the arguments against the other explanations, we favour our television hypothesis as providing
some of the driving force towards the lower home advantages now witnessed in Premiership football in
England. Thus we can attach a plausible mechanism to explain the significance of what is simply a post-
1986 dummy variable in our regressions, and rule out competing explanations. In terms of implications
of this, we might expect that the decline in home advantage that took place in the late 1980s and early
1990s will not be reversed; there is little sign of any reversal in TV coverage nor the appetite for it in the
future. Furthermore, the results do give some indication of the power that supporters do actually wield;
while supporters do not have any actual executive power, this evidence, and numerous anecdotal stories
suggest that in football, the consumer/supporter is sovereign. Whether it is productive for football clubs
to act upon this and attempt to improve their away records by, for example, providing free live TV feed to
their supporters for every game, is not clear, since if all clubs attempt similar measures, then if successful,
this means that all teams will win less home matches as a result. There is only a certain number of points
available to football clubs competing in a league; if only drawn matches are turned into home or away
victories as a result of action by clubs, then there may be some overall gain, since a drawn match produces
two points, but a win produces three. The rewards would be more unequal though, since the two points in
a drawn match are shared, whereas the winner takes all three points.
However, we should emphasise that we don’t believe our explanation to be the single defining factor relating
to home advantage. The different evolutions of home advantage in lower divisions, and abroad (from
Figure 2 German teams stand out as displaying remarkably strong home advantage during the 1970s before
a drastic drop around 1990) suggest that other factors are at work. It is not necessarily the case that if clubs
acted upon this TV finding that any effect would be seen due to other potential causes of home advantage
interacting with the monitoring hypothesis.
There are plausibly a whole plethora of additional, crowd-related (and non-crowd-related) factors that explain
the different evolutions of home advantage in different divisions. The make-up of crowds has surely changed
over the post-war years, particularly as traditional regional industries decline, and entry prices increase.
The latter factor has undoubtedly helped turn spectators away from being supporters, willing to encourage
their team on to greater things, into consumers, expectant of teams to do great things for them, given how
35From the 1988–89 season onwards, the third-last-placed Premiership club was replaced by the sixth placed Championship
club in the competition.
19
much they are paying in real terms. It may also be lower divisions, and smaller crowds, where this effect is
felt strongest, as vocal views of individual fans are more audible, making for a more intimidating atmosphere
for home players. Nonetheless, such hypothesised effects are simply that, and it seems unlikely that data
to investigate such hypotheses will materialise.
It is also important to note that our hypothesis merely offers a plausible explanation of a fall in home
advantage, not its actual level. Nonetheless, the hypothesis can explain some of the difference in home and
away performances, notably that monitoring is still imperfect even with television: movement of players off
the ball is harder to observe when vision is restricted to where television cameras are focussed, compared to
actually being at a game. Additional unmeasurable factors that we have not ruled out, such as familiarity,
as well as factors such as distance travelled, surely impact upon the level of home advantage.
6 Conclusions
In this paper the causes of home advantage in football have been investigated. Much analysis has already
been done in this area, although little of it in economics. We propose an economic mechanism for linking
attendance at football matches and match outcome that is starkly different from that proposed by others
in the literature; we suggest that the increase in live television coverage of football matches has worked to
incentivise players no longer to shirk when playing in away games, as supporters can now more effectively
monitor their efforts. We also introduce a novel and promising method of modelling match outcome,
which provides an alternative measure of home advantage, along with measures of defensive and attacking
capabilities of teams in each season. The TV effect does appear to have some support in the data when
considering both aggregated time series, and panel data methods, suggesting support for our alternative
explanation for home advantage. Naturally, as more and more data becomes available, and as methods for
modelling football match outcomes continue to improve, we expect that clearer measures of this hypothesis
will be put forward; for example we suspect that the non-shirking hypothesis will gain more support if data
can be found on numbers of away supporters attending football matches historically. We excitedly await
such future studies.
A Appendix A: Tables
References
Alchian, A.A and H. Demsetz (1972), ‘Production , Information Costs, and Economic Organization’, Amer-
ican Economic Review 62(5), 777–795.
Attrill, M.J., K.A. Gresty, R.A. Hill and R.A. Barton (2008), ‘Red shirt colour is associated with long-term
team success in English football’, Journal of Sports Sciences 26(6), 577–582.
Audas, R., S. Dobson and J. Goddard (1997), ‘Team Performance and Managerial Change in the English
Football League’, Economic Affairs 17(3), 30–37.
Berri, D.J. and A.C. Krautmann (2006), ‘Shirking on the Court: Testing for the Incentive Effects of Guar-
anteed Pay’, Economic Inquiry 44(3), 536–546.
Boyko, R.H., A.R. Boyko and M.G. Boyko (2007), ‘Referee Bias Contributes to Home Advantage in English
Premiership Football’, Journal of Sports Sciences 25(11), 1185–1194.
Brandes, L. and E. Franck (2007), ‘Who Made Who Who? An Empirical Analysis of Competitive Balance
in European Soccer Leagues’, Eastern Economic Journal 33(3), 379–403.
20
Goals
Period Moment H A Total
1888-1899 Mean 2.4 1.3 3.7
1888-1899 Variance 3.4 1.6 4.5
1900-1909 Mean 1.9 1.0 3.0
1900-1909 Variance 2.4 1.2 3.4
1910-1919 Mean 1.8 1.0 2.9
1910-1919 Variance 2.2 1.2 3.2
1920-1929 Mean 2.0 1.1 3.1
1920-1929 Variance 2.5 1.2 3.9
1930-1939 Mean 2.2 1.2 3.4
1930-1939 Variance 2.7 1.4 4.1
1940-1949 Mean 1.8 1.1 2.9
1940-1949 Variance 2.2 1.3 3.4
1950-1959 Mean 2.0 1.2 3.2
1950-1959 Variance 2.3 1.3 3.6
1960-1969 Mean 1.9 1.2 3.1
1960-1969 Variance 2.1 1.3 3.5
1970-1979 Mean 1.6 1.0 2.5
1970-1979 Variance 1.6 1.0 2.7
1980-1989 Mean 1.6 1.1 2.7
1980-1989 Variance 1.7 1.1 2.8
1990-2008 Mean 1.5 1.1 2.6
1990-2008 Variance 1.5 1.1 2.7
Table 3: Mean and variance of goals scored in English football matches, 1888-2008.
Clarke, S.R. and J.M. Norman (1995), ‘Home Ground Advantage of Individual Clubs in English Soccer’,
The Statistician 44(4), 509–521.
Dixon, M.J. and M.E. Robinson (1998), ‘A Birth Process Model for Association Football Matches’, The
Statistician 47(3), 523–538.
Dohmen, T.J. (2008), ‘The Influence of Social Forces: Evidence from the Behavior of Football Referees’,
Economic Inquiry .
URL: http://www.blackwell-synergy.com/doi/abs/10.1111/j.1465-7295.2007.00112.x
Doornik, J.A. (2006), Autometrics: Further Applications of Automatic Model Selection. Mimeo, Nuffield
College.
Doornik, J.A. and D.F. Hendry (2007), Modelling Dynamic Systems Using PcGive 12: Volume III, Timber-
lake Consultants Press, London.
Forrest, D., J Beaumont, J. Goddard and R. Simmons (2005), ‘Home Advantage and the Debate About
Competitive Balance in Professional Sports Leagues’, Journal of Sports Sciences 23(4), 439–445.
Hendry, D.F. (1995), Dynamic Econometrics, Advanced Texts in Econometrics, Oxford University Press,
Oxford.
Hendry, D.F. and H.-M. Krolzig (2001), Automatic Econometric Model Selection Using PcGets, Timberlake
Consultants Ltd, London.
Hendry, D.F. and H.-M. Krolzig (2005), ‘The Properties of Automatic GETS Modelling’, Economic Journal
115(502), C32–C61.
21
Hendry, D.F. and J.J. Reade (2007), Modelling and Forecasting Using Model Averaging, Working Papers,
Department of Economics, University of Oxford.
Holmstrom, B. (1999), ‘The Firm as a Subeconomy’, Journal of Law, Economics and Organization 15(1), 74–
102.
Holmstrom, B. and P. Milgrom (1991), ‘Multitask Principal-Agent Analyses: Incentive Contracts, Asset
Ownership, and Job Design’, Journal of Law, Economics and Organization 7(0), 24–52.
URL: http://ideas.repec.org/a/oup/jleorg/v7y1991i0p24-52.html
Jacklin, P.B. (2005), ‘Temporal Changes in Home Advantage in English Football Since the Second World
War: What Explains Improved Away Performance?’, Journal of Sports Sciences 23(7), 669–679.
Johnston, R. (2008), ‘On Referee Bias, Crowd Size, and Home Advantage in the English Soccer Premiership’,
Journal of Sports Sciences 26(6), 563–568.
Karau, S.J. and K.D. Williams (1981), ‘Social Loafing: a Meta-Analytic Review and Theoretical Integration’,
Journal of Personality and Social Psychology 40, 681–706.
Karlis, D. and I. Ntzoufras (2003), ‘Analysis of Sports Data Using Bivariate Poisson Models’, Journal of the
Royal Statistical Society (Statistician) 52, 381–393.
Knowles, G., J. Murray, K. Sherony and M. Haupert (2003), ‘Shirking in Major League Baseball in the Era
of the Reserve Clause’, NINE: A Journal of Baseball History and Culture 12(1), 59–71.
Krautmann, A.C. (1990), ‘Shirking or Stochastic Productivity in Major League Baseball?’, Southern Eco-
nomic Journal 56(4), 961–968.
Krautmann, A.C. (1993), ‘Shirking or Stochastic Productivity in Major League Baseball: Reply’, Southern
Economic Journal 60(1), 241–243.
Latane, B., K. Williams and S. Harkins (1979), ‘Many Hands Make Light the Work: the Causes and
Consequences of Social Loafing’, Journal of personality and social psychology 37(6), .
Liden, R.C., S.J. Wayne, R.A. Jaworski and N. Bennett (2004), ‘Social Loafing: A Field Investigation’,
Journal of Management 30(2), 285–304.
Marburger, D.R. (2003), ‘Does the Assignment of Property Rights Encourage or Discourage Shirking?:
Evidence From Major League Baseball’, Journal of Sports Economics 4(1), 19–34.
Maxcy, J.G., R.D. Fort and A.C. Krautmann (2002), ‘The Effectiveness of Incentive Mechanisms in Major
League Baseball’, Journal of Sports Economics 3(3), 246–255.
McHale, I. and P. Scarf (2007), ‘Modelling Soccer Matches Using Bivariate Discrete Distributions with
General Dependence Structure’, Statistica Neerlandica 61(4), 432–445.
McMaster, R. (1997), ‘The Market for Corporate Control in Professional Football: Is There an Agency
Problem?’, Economic Affairs 17(3), p25 –.
Miles, J.A. and J. Greenberg (1993), ‘Using Punishment Threats to Attenuate Social Loafing Effects among
Swimmers’, Organizational Behavior and Human Decision Processes 56(2), 246–265.
Miller, G.J. (1992), Managerial Dilemmas, the Political Economy of Hierarchy, Cambridge University Press,
Cambridge.
Nalebuff, B.J. and J.E. Stiglitz (1983), ‘Prizes and Incentives: Towards a General Theory of Compensation
and Competition’, The Bell Journal of Economics 14(1), 21–43.
Neave, N. and S. Wolfson (2003), ‘Testosterone, Territoriality, and the ‘Home Advantage’’, Physiology &
Behavior 78(2), 269–275.
22
Nevill, A.M., S.M. Newell and S. Gale (1996), ‘Factors Associated With Home Advantage in English and
Scottish Soccer Matches’, Journal of Sports Sciences 14(2), 181–186.
Pollard, R. (2005), ‘Evidence of a Reduced Home Advantage when a Team Moves to a New Stadium’, Journal
of Sports Sciences 20(12), 969 – 973.
Pollard, R. (2006), ‘Worldwide Regional Variations in Home Advantage in Association Football’, Journal of
Sports Sciences 24(3), 231 – 240.
Pollard, R. and G Pollard (2005), ‘Long-Term Trends in Home Advantage in Professional Team Sports in
North America and England (1876 - 2003)’, Journal of Sports Sciences 23(4), 337 – 350.
Sloane, P.J. (1971), ‘The Economics of Professional Football: The Football Club as a Utility Maximiser’,
Scottish Journal of Political Economy 18(2), 121–46.
Sutter, M. and M.G. Kocher (2004), ‘Favoritism of Agents - the Case of Referees’ Home Bias’, Journal of
Economic Psychology 25(4), 461–469.
URL: http://ideas.repec.org/a/eee/joepsy/v25y2004i4p461-469.html
Szymanski, S. and R. Smith (1997), ‘The English Football Industry: Profit, Performance and Industrial
Structure’, International Review of Applied Economics 11(1), 135–153.
Wolfson, S., D. Wakelin and M. Lewis (2005), ‘Football supporters’ perceptions of their role in the home
advantage’, Journal of Sports Sciences 23(4), 365 – 374.
23
... While, in contrast to other countries, there were no artificial fan chants in Germany, some teams used 'carton spectators' to imitate a full stadium. Still, we are confident that this does not really affect our results, as the literature considers fans to affect home team success mainly via two channels: The reduction of moral hazard that players do not fully invest into the match if the feel unobserved (Koyama & Reade, 2008) and the actual noise of fans and their impact on the referee (Dohmen, 2008;Dohmen & Sauermann, 2016). Both factors are either not affected by artificial spectators or -if at all -less than for the inclusion of actual spectators. ...
Article
Fan support has often been considered to be a main driver of the home advantage in sports. Using the “natural exper‐ iment” of ghost games during the COVID‐19 pandemic in German professional soccer and a dataset of 6,120 matches ranging over the seasons 2011/2012 to 2020/2021, we test this claim. Indeed, applying a difference‐in‐differences setup, we identify a reduction in the home performance ‐ though with much heterogeneity across leagues and teams. We, moreover, observe the home advantage to recover over time. In analysing whether betting markets anticipated this drop in the home advantage and its recovery, we test the efficient market hypothesis. In the short‐run, betting odds do not properly reflect the effect of ghost games regarding changes in the home advantage but they suitably account for it in the long‐run. Also, league heterogeneity is efficiently captured in the odds. Based on these findings, we show how simple, short‐run betting strategies could have exploited the discussed phenomenon. Our findings also have implications for agents' behavior on financial markets in the presence of unforeseen and unfamiliar shocks.
... Throughout the history of professional team sports, there has been a tendency for a team to win more often when playing in their home stadium, city or country (e.g., Schwartz and Barsky, 1977;Courneya and Carron, 1992;Nevill and Holder, 1999;Koyama and Reade, 2009;Peeters and van Ours, 2020). As summarised in the review by Dohmen and Sauermann (2016), referee bias may be one significant cause of this home advantage, with much of the evidence on this from the field coming from professional football. ...
Article
Full-text available
Using the assignment of referees to European international association football matches played between 2002 and 2016, we ask whether judgements were biased according to the legacy of the Cold War. Referees from post-communist states favoured teams from non-communist states, but there was no evidence of favouritism in the other direction. This out-group bias of referees born behind the Iron Curtain was statistically significant for relatively less important and more subjective decisions, namely the awarding of yellow cards for foul play. The bias was particularly large among referees from the former Soviet Union. It has also diminished over time, perhaps due to increased professionalism in European refereeing, or because memories of the Cold War era have diminished among active referees.
... Moreover, other issues could also play a role for the home advantage, such as travel fatigue (Clarke & Norman, 1995;Goumas, 2014;Pollard & Gomez, 2014b), altitude differences (van Damme & Baert, 2019), the three-point rule (Pollard, 1986), or TV coverage (Koyama & Reade, 2008). In addition, note that the home advantage is mediated by the competitive balance of the opposing teams. ...
Article
This paper examines the relation between crowd support and home advantage in men’s professional football in making use of a unique “natural experiment” induced by restrictions due to the COVID-19 pandemic: the so-called ghost games in the top three German football divisions at the end of the 2019/2020 season. We find that there is a reduced home advantage in the first division, whereas no change is observed in the second and third divisions. Our regression analysis indicates that the decrease in the home advantage and the heterogeneity across divisions are not sensitive to a variety of performance, location, and team covariates and are best explained through the lower occupancy rate in the stadia. Hence, the decrease in occupancy to zero at the ghost games has been less dramatic for teams that have been used to low occupancy rates. We cannot find strong evidence for a change in referee behavior or teams’ tactics as main impact channels of occupancy rates on the home advantage. Hence, we argue that psychological reasons are of higher importance.
... In an early study, Schwartz and Barsky (1977) pointed toward greater home advantage within indoor sports, such as basketball and ice hockey, than outdoor, finding that the support of the home crowd appeared to be a major factor. In the English Football League, since its founding in 1888, there were only small variations in home advantage between seasons for a long time (Pollard, 1986), but in recent decades there has been a persistent and substantial decline in home advantage, that otherwise exists independently of the large cross-league differences in stadium attendances (Koyama and Reade, 2009;Peeters and van Ours, 2020). This has led to the suggestion that crowd size may not play a substantial role in football, and instead familiarity with home conditions could be a more important factor. ...
Preprint
Full-text available
We use a series of historical natural experiments in association football to test whether social pressure from a home stadium crowd affected behaviour and outcomes. The standout effect of an empty stadium was that referees cautioned visiting players less often, by over a third of a yellow card per match or once for every twenty-two fouls committed. Stadium crowds caused referees to favour the home team in their decision making. Empty stadiums appear to have reduced the overall home advantage in the final outcomes of football matches, but we cannot statistically reject no effect.
... In an early study, Schwartz and Barsky (1977) pointed toward greater home advantage within indoor sports, such as basketball and ice hockey, than outdoor, finding that the support of the home crowd was a major factor. In the English Football League, since its founding in 1888, for a long time there were only small variations in home advantage between seasons (Pollard, 1986), but in recent decades there has been a persistent and substantial decline in home advantage, that otherwise exists independently of the large cross-league differences in stadium attendances (Koyama and Reade, 2009;Peeters and van Ours, 2020). 1 This has led to the suggestion that crowd size may not play a substantial role in football, and instead familiarity with home conditions could be a more important factor. Although Boyko et al. (2007) found some evidence that home advantage in English football may relate to the crowd and its influence on referees, this was contradicted by Johnston (2008) in a replication study using more recent data. ...
... In an early study, Schwartz and Barsky (1977) pointed toward greater home advantage within indoor sports, such as basketball and ice hockey, than outdoor, finding that the support of the home crowd was a major factor. In the English Football League, since its founding in 1888, for a long time there were only small variations in home advantage between seasons (Pollard, 1986), but in recent decades there has been a persistent and substantial decline in home advantage, that otherwise exists independently of the large cross-league differences in stadium attendances (Koyama and Reade, 2009;Peeters and van Ours, 2020). 1 This has led to the suggestion that crowd size may not play a substantial role in football, and instead familiarity with home conditions could be a more important factor. Although Boyko et al. (2007) found some evidence that home advantage in English football may relate to the crowd and its influence on referees, this was contradicted by Johnston (2008) in a replication study using more recent data. ...
Preprint
Full-text available
We use a series of natural experiments in association football (soccer) to test whether the lack of social pressure from spectators affected behaviour and outcomes. We observe that the normal advantage to the home team from playing in their own stadium was on average eroded when they played behind closed doors, with no supporters. Among the various effects from no fans being present, visiting players were cautioned significantly less often by referees. This suggests that closed doors matches are different because referees favour the home team less in their decision making. We discuss these findings in the light of the Covid-19 pandemic that has led to the remainder of the 2019/20 European football season playing out in empty stadiums.
... The remaining results are almost evenly divided between draws and away wins, at 26% and 27% respectively. Research suggests that home advantage can be attributed to a number of factors such as local crowd support, pitch/stadium familiarity, referee bias favouring home team decisions, away team travel fatigue and increased home team player motivation [8] and in Figure 1 using time series analysis we can visually assess the effect of home advantage based on total points earned in the Premier League, which has remained consistent over the past ten years. Although there is a clearly uneven distribution of outcomes, for data mining purposes there should be enough examples of each outcome (label) in the football data set, however we will experiment with bootstrap sampling techniques to see if any predictive accuracy improvements can be achieved. ...
Conference Paper
Full-text available
Predicting the outcome of football matches is a popular research area primarily due to the popularity of the sport worldwide and it is therefore unsurprising that there have been numerous previous attempts both commercially and academically to forecast Premier League match results with varying degrees of success. In this research paper we apply the CRISP-DM (CRoss-Industry Standard Process for Data Mining) methodology to create and explore a variety of predictive classification models in an attempt to accurately and objectively predict the outcome of English Premier League football matches using freely available online data that is based only on previous Premier League match results from 2002 to 2012. Using an iterative approach, numerous data visualisation, pre-processing and transformation techniques have been applied and evaluated to determine the most effective predictive model, which has been assessed by its ability to correctly forecast the match outcome of the final 50 matches in the 2011/2012 Premier League season as per the initial business and data mining objectives. The result of this study showed that a number of models outperform human expert match predictions.
... Nevill, Balmer, and Williams (2002) found that referees are influenced by crowd noise, while Garicano, Palacios-Huerta, and Prendergast (2005) reported that close matches were shortened by referees when the home team was ahead, and lengthened when the home team was behind. Koyama and Reade (2009) focused on how the crowd reduce shirking attitudes of home teams and how televised football has contributed to lessening the home advantage because it allows viewers to monitor the effort of away teams. ...
Article
Full-text available
This paper tests for the existence of an order effect in competitive situations using the natural experiment of the introduction of the “away-goals rule” in CONMEBOL club competitions. This rule states that in a case of a two-legged fixture finishing level on goals, the team that scores more goals as a visitor will qualify for the next round. Fixed-effects logit analyses for the period 1988-2014 provide evidence that, after the application of the rule in 2005, teams that played the second leg as visitors had an increased probability of winning in regulation time. This phenomenon was especially significant when the “visitor in the second leg” was the stronger team of an evenly matched game, or conceded no goals in the first match. Several explanations related to behavioral biases, such as psychological pressure, time inconsistency, and loss aversion, are proposed.
... As the literature on soccer has concluded and our summary statistics in Table 2 point out, home-bias is a very influential parameter on the result of a match (see e.g. Clarke and Norman (1995), Carmichael and Thomas (2005), Koyama and Reade (2009) Table 2 shows that the matches that end in a draw are more likely to be the ones where the home team does not hold the market value advantage. This is a clear indication that the home-bias enters in the probability of winning as ...
Article
Full-text available
A contest success function (success function) maps the level of efforts into winning and losing probabilities in contest theory. We aim to assess the empirical performance of success functions for draws and analyze the differences between European soccer leagues in terms of home bias, return on talent (ROT), and talent inequality. We use a data set with 10,569 matches acquired manually from transfermarkt.co.uk containing club-based average market values of the lineup of teams for each match played through 12 seasons from 7 major European soccer leagues. The results are obtained estimating the parameters of the success functions with a general maximum-likelihood method, and the hypotheses suggested by success functions are controlled with a probit regression. Two of the success functions outperform one conclusively. The difference in the performance between these two groups results from the contrast in the main determinant of the success function in allocating the probability of a draw. The high-performing success functions take difference in aggregate talent levels as the main determinant in drawing, while the other takes the aggregate talent as the main determinant. The results also show that there are major differences across leagues in terms of ROT, home bias, and talent inequality, despite the similarities in economic environment and the homogeneity in the rules of the game imposed across leagues. Our analysis sheds light on the contributions and implications of microeconomic theory to model sports and presents the differing characteristics of the European soccer leagues that impact match results significantly.
Book
Full-text available
Proceedings of the 12th Information Technology and Telecommunication Conference
Article
Full-text available
After reviewing the simulation performance of general-to-specific automatic regression-model selection, as embodied in PcGets, we show how model selection can be non-distortionary: approximately unbiased ‘selection estimates’ are derived, with reported standard errors close to the sampling standard deviations of the estimated DGP parameters, and a near-unbiased goodness-of-fit measure. The handling of theory-based restrictions, non-stationarity and problems posed by collinear data are considered. Finally, we consider how PcGets can handle three ‘intractable’ problems: more variables than observations in regression analysis; perfectly collinear regressors; and modelling simultaneous equations without a priori restrictions.
Article
Least squares is used to fit a model to the individual match results in English football and to produce a home ground advantage effect for each team in addition to a team rating. We show that for a balanced competition this is equivalent to a simple calculator method using only data from the final ladder. The existence of a spurious home advantage is discussed. Home advantages for all teams in the English Football League from 1981-82 to 1990-91 are calculated, and some reasons for their differences investigated. A paired home advantage is defined and shown to be linearly related to the distance between club grounds.
Article
This article analyzes the role of competitive compensation schemes (in which pay depends on relative performance) in economies with imperfect information. These compensation schemes have desirable risk, incentive, and flexibility properties; they provide for an automatic adjustment of rewards and incentives in response to common changes in the environment. When environmental uncertainty is large, such schemes are shown to be preferable to individualistic reward structures; in the limit, as the number of contestants becomes large, expected utility may approach the first-best (perfect information) level. We study the design of contests, including the optimal use of prizes versus punishments and absolute versus relative performance standards. The analysis can also be viewed as a contribution to the multiagent, single-principal problem.
Article
Cambridge Core - Political Economy - Managerial Dilemmas - by Gary J. Miller
Article
NINE: A Journal of Baseball History and Culture 12.1 (2003) 59-71 Shirking has become an issue in professional sports with the advent of long-term guaranteed contracts, particularly because pay is largely uncoupled from performance. Baseball player-owner relations are a classic example of the principal-agent problem, which investigates incentive problems between contracting parties. The agent (player) is said to shirk if he chooses not to undertake efficient actions measured in terms of on-field performance. From the viewpoint of the principal (owner), actions are considered efficient when they afford the greatest opportunity for profit. Shirking, or inefficient action, is most likely to occur when contracts do not satisfy the incentive compatibility constraint. The constraint is satisfied when the contract spells out incentives that cause the agent to take efficient actions. In the case of both one-year contracts with a reserve clause and long-term guaranteed contracts in Major League baseball (MLB), there may be inadequate financial incentives for players to perform to the best of their ability, that is, efficiently. Krautmann (1990) has argued against the popular view held by some owners and fans that after signing a long-term contract, a player's performance tends to fall off due to the disincentive effect of the long-term contract. He concludes that systematic, league-wide shirking does not occur: "It is the stochastic nature of productivity that is the source of this alleged disincentive problem," and that "what owners are interpreting as shirking by players is more likely the expected outcome of a stochastic process." In his comment on Krautmann's article, Scoggins (1993) uses a different performance measure (total bases instead of slugging average) in rejecting the hypothesis that shirking does not occur. To establish whether individual shirking has occurred in professional sports would require some information about the specific contract and an observation of a drop-off in performance (an inefficient outcome) that might be attributed to actions by the player not in the best interests of the owner. The analysis should be based on individual-specific conditions (contracts, incentive clauses, and uncontrollable factors) that would be examined on a case-by-case basis. Differentiating between shirking and stochastic productivity should also be done on an individual, case-by-case basis. To conclude that systematic or league-wide shirking and not stochastic productivity exists may be an insurmountable task in the modern era examined by Krautmann and Scoggins due to the complex nature of long-term contracts. Each individual contract should be examined for incentives, and for each drop-off in performance actions by the principal and agent should be investigated. Given the nature of the null hypothesis in the above studies, one would have to agree with Krautmann that there is no evidence for systematic shirking. However, it does not follow that individual shirking does not occur or that variations in player performance are solely due to stochastic productivity. Nor is it necessary to observe shirking only in the presence of multiyear contracts. A player has a propensity to shirk when his contract with the owner is not performance based. In the era of the reserve clause, players would have a propensity to shirk, especially for those whose salaries were held far below their marginal revenue product. This was often the case because of the reserve clause in all player contracts, which assigned property rights of players to teams in perpetuity and was motivated by monopsonistic collusion. With players facing a monopsonistic employer and the inability to negotiate with other teams or other leagues, players engaged in shirking as a strategy to extract salaries closer to their marginal revenue product. Throughout most of the reserve era, the primary form of shirking by players was to hold out. By withholding their services until they negotiated a contract for a higher salary, players denied the team owner the ability to profit from their performances. In addition many players used the holdout strategy as an excuse to avoid spring training, for which players in the early history of MLB were not compensated. In either case this would be considered inefficient, since players would miss games or not be at peak performance at the beginning...