The effect of monitoring and crowds on crime and law enforcement: A natural experiment from European football

Economics Faculty Working Papers Series Economics

10-2022

The effect of monitoring and crowds on crime and law The effect of monitoring and crowds on crime and law

enforcement: A natural experiment from European football enforcement: A natural experiment from European football

Brad R. Humphreys

West Virginia University

, [email protected]

Alexander Marsella

West Virginia University

, [email protected]

Levi Perez

University of Oviedo

Follow this and additional works at: https://researchrepository.wvu.edu/econ_working-papers

Part of the Economics Commons

Digital Commons Citation Digital Commons Citation

Humphreys, Brad R.; Marsella, Alexander; and Perez, Levi, "The effect of monitoring and crowds on crime

and law enforcement: A natural experiment from European football" (2022).

Economics Faculty Working

Papers Series

. 67.

https://researchrepository.wvu.edu/econ_working-papers/67

This Working Paper is brought to you for free and open access by the Economics at The Research Repository @

WVU. It has been accepted for inclusion in Economics Faculty Working Papers Series by an authorized

administrator of The Research Repository @ WVU. For more information, please contact

researchreposit[email protected]>.

The eﬀect of monitoring and crowds on crime and law

enforcement: A natural experiment from European

football

Brad R. Humphreys

West Virginia University

Alexander Marsella

West Virginia University

Levi Perez

University of Oviedo

October 17, 2022

Abstract

Technological advancements like the presence of smart phones and body cameras

have led to increased monitoring of police, but little evidence exists on their impact.

We address these problems using data on fouls from football matches in ﬁve European

football leagues over six seasons. This period contains exogenous changes in monitor-

ing rule enforcers through introduction of Video Assistant Referee review and limited

“bystanders” from Covid-19 restrictions. Results from diﬀerence-in-diﬀerences models

estimated separately for each league indicate that both events inﬂuenced the number

of fouls called with substantial heterogeneity across leagues and home/away teams.

(Word count: 4776, 18 tables.)

JEL Codes: H41 (Public Goods), K42 ( Illegal Behavior and the Enforcement of Law),

Z20 (General Sports Economics)

Keywords: crime, police monitoring, football fouls

Introduction

Crime represents an important societal problem around the world and a large literature

in economics analyzes the complex relationship between crime and policing. The seminal

paper by Becker (1968) emphasized the importance of decisions made by criminals and

police oﬃcers, the interaction between their actions, and the role that the presence of other

people nearby play in the commission of crime and apprehension of criminals. For example,

Becker’s model focuses on individual’s utility maximizing decisions about committing or not

committing crimes based on her constraints and preferences, what actions she expects the

police to take, and what consequences she faces if caught. The presence or absence of others

nearby also aﬀect the expected beneﬁt from committing a crime, the expected likelihood of

detection, the actions taken by police, and the likelihood a crime is committed.

Owens and Ba (2021) summarized the recent literature on monitoring of police oﬃcers,

the interaction between police and the general public, and systematic bias in policing. This

paper documented dramatic recent changes in monitoring police and how the public interacts

with police due to technological changes in smart phone use, CCTV cameras, and police

body cameras. However, little empirical evidence on the impact of the presence or absence

of nearby people on the commission and detection of crime exists, probably due to lack of

appropriate data.

The empirical economics of crime literature contains a long, rich tradition of using data

from sports to assess the eﬀectiveness of policing (Pope et al., 2018; Kitchens et al., 2019).

This paper uses data from games played in European professional football leagues and a nat-

ural experiment to assess the impact of direct monitoring of law enforcement and interaction

between law enforcement and the general public on a form of petty crime, fouls called in

football matches. We exploit two sources of exogenous variation in factors aﬀecting referees,

players, and fouls in football matches: the introduction of Video Assistant Referee (VAR)

oversight of oﬃciating decisions made in football matches and the full or partial elimination

of fans in stadiums due to Covid-19 pandemic restrictions. Both events should be plausibly

exogenous to unobservable factors aﬀecting policing and the commission of crime in this

setting. Cohen et al. (2021) take a similar approach, but focus only on football matches in

Germany.

We analyze outcomes in 10,818 matches played in the top professional football leagues

in England, France, Italy, Germany, and Spain over the 2015/16 to 2020/21 domestic league

competitions. We estimate separate regression models for each league to account for league

and country speciﬁc heterogeneity. Our diﬀerence-in-diﬀerences model results exploiting the

introduction of VAR and prohibition on fan attendance ﬁnd signiﬁcant, but heterogeneous

impacts of these changes on fouls called. The introduction of VAR monitoring reduced fouls

in Germany, Spain, and Italy but not in England or France. Elimination or reduction of

fans attending games reduced fouls committed by visiting teams in Germany and France,

both home and visiting teams in Italy, and by home teams in England. These results are

robust to the inclusion of game-level control variables that proxy for game importance and

intensity of play. Both monitoring of police and the presence of people nearby aﬀect the the

commission of crime and law enforcement.

Context

The presence or absence of people can aﬀect both potential criminals and law enforcement.

Some criminal acts are more likely to occur in crowded areas such as shopping districts, bus

stops, and on public transportation. Other criminal acts are more likely to occur in isolated

settings. For example, petty crimes like pickpocketing and shoplifting are more likely to go

unnoticed in crowded environments in which criminals tend to target victims or property

based on an appearance of inattention. Evidence also suggests that law enforcement person-

nel are not impartial enforcers of laws, but sometimes exhibit biased behavior depending on

the setting (Pope et al., 2018).

The fact that empirical researchers typically only observe outcomes where an individual

was arrested for a crime, and not the actions taken by the person arrested, the oﬃcer making

the arrest, their interactions prior to the arrest, and the presence or absence of observers

complicates empirical analysis of outcomes in these markets. Lack of information about

the setting in which arrests take place also complicates empirical analysis. Sporting events

represent a useful setting for empirical research on crime. Play in team sports reﬂects clear

rules governing conduct, law enforcers in the form of referees, and varying costs and beneﬁts

associated with committing an oﬀense. It also features interaction between law enforcers,

players and the general public (fans).

The analogy between full stadiums and crowded public spaces can be explained in terms of

the atmosphere, namely the noise fans produce in stadiums. The crowd’s cheering, chanting

and shouting for their teams can improve the performance of the home players and at the

same time adversely aﬀect the performance of players on the visiting team. In addition,

there is another mechanism at work in football stadiums: the home crowd clearly attempts

to put pressure on the referees to make decisions favorable to their team. In such a setting,

some players’ fouls and misconduct may go unnoticed (like pickpockets in crowded public

spaces). Additionally, in terms of referees’ behavior, fan pressure could distract referees or

make them more rash (prone to erroneous decisions, again, similar to what police experience

in crowded spaces).

We posit that referees and players can be inﬂuenced by the atmosphere in the stands. One

of the seminal papers on this topic, Garicano et al. (2005), showed that because referees have

discretion over the amount of extra time awarded in matches to compensate for stoppages

in play, referees systematically reduce time awarded when the home team is winning; in

matches with a greater payoﬀ for winning they modify their behaviour even more. In fact,

the authors argue that these results occur because of referees’ desire to satisfy the crowd

of home team fans in the stands. And the more fans in attendance, the larger the referee

response. Other research conﬁrms this result in other settings, and similar results for the

enforcement of other football rule infractions like fouls and red/yellow cards.

The Covid-19 pandemic represents an opportunity to study crime and policing using a

unique natural experiment. The 2019/20 football season in Europe was interrupted in mid-

March 2020 by coronavirus spread and the Covid-19 disease. The seasons resumed in June

2020 in the “Big-Five” professional football leagues in Europe, except for French Ligue 1

which terminated its season in April 2020. All remaining matches were played behind closed

doors under governments’ plans to combat coronavirus spread. With some minor exceptions,

European leagues started the next season (2020/21) in empty stadiums.

A number of recent studies discussed below also exploit the Covid-19 attendance bans

to understand the impact of fans on decisions made by players and referees. These papers

analyzed data from many diﬀerent competitions over diﬀerent sample periods using a umber

of diﬀerent match outcomes. Some focus on football results and performance variables like

wins and goals; others analyzed crime related outcomes like Red Cards, Yellow cards, and

fouls. Nearly all allow for eﬀect heterogeneity by estimating separate regression models for

home team and away team outcomes. The general approach used posits that the banning of

fans from matches represents a strong diﬀerence-in-diﬀerences natural experiment.

Reade et al. (2020) estimated the eﬀects of the absence of fans in men’s football compe-

titions in Europe beginning in the 2002/03 season, a relatively long pre-treatment period.

Competitions analyzed include a pooled sample of matches played in the UEFA Champions

League, the UEFA Europa League, French Ligue 1, Italian Serie A, B and C, and the Coppa

d’Italia. The authors found a signiﬁcant eﬀect absence of fans on the severity of referee

penalties given to visiting teams only. Visiting teams got fewer yellow cards.

Like Reade et al. (2020), Cueva (2020) analysed match outcomes and referee decisions in

matches played in 41 football leagues in 30 diﬀerent countries beginning in 1993. The pooled

analysis sample contained more than 230,000 matches with 2,749 matches played with no

fans, again a long pre-treatment period. The results indicated that home teams received

more red and yellow cards and were penalized with more fouls after the pandemic began.

Away team treatment in the pandemic was no diﬀerent than before.

Dilger and Vischer (2020) analyzed changes in home advantage in terms of the likelihood

the home team won matches, as well as Yellow cards and Red cards issued with and without

fans attending matches in Germany’s Bundesliga 1 in the 2019/20 season. Dilger and Vischer

(2020) reported signiﬁcantly fewer Yellow and Red cards for the away team, but no change

in home advantage.

Endrich and Gesche (2020) also focused on the eﬀects of missing fans on referees’ be-

havior and decision-making using an analysis sample of 1224 matches played in Germany’s

Bundesliga 1 and 2, the top two professional football leagues in Germany in the 2019/20

season. The results indicated that signiﬁcantly more fouls and yellow cards were called on

home teams when no fans were present relative to when fans were allowed to attend matches.

Ferraresi and Gucciardi (2020) analyzed match outcomes in the top league in France,

Germany, Italy, Spain, and the United Kingdom in 2019/20 season in a pooled sample,

focusing only on points earned by teams. They found home team performance deteriorated

during the pandemic, but away team performance did not.

McCarrick et al. (2020) analyzed team success, goals scored, Yellow cards, and Red

cards in a pooled sample of matches played in 15 diﬀerent leagues in 11 diﬀerent European

countries in the 2019/20 season. The results found a negative impact of the pandemic on

home team performance and a negative impact on away team Yellow cards.

Bryson et al. (2021) analyzed match outcomes in a pooled sample of 23 professional

football leagues in the 2019/20 season, including matches played by 369 football teams

and oﬃciated by 472 referees. Their data set included 6481 football matches played in

the 2019/20 season in 17 countries before and after the mid-season shutdown. The sample

contained data from 1498 matches, 23% of the sample, played without spectators. The

absence of fans reduced yellow cards called on visiting teams by 33%.

Cohen et al. (2021) analyzed the impact of missing fans and the introduction of exter-

nal Video Assistant Referee (VAR) review of in-game referee decisions matches played in

Germany’s Bundesliga 1 and 2. The Bundesliga 1 sample started with the 2009/10 season

and the Budesliga 2 sample with the 2013/14 season. Cohen et al. (2021) analyzed three

diﬀerent periods in the sample: the period before the introduction of VAR review of referee

decisions, the period between the introduction of VAR and the beginning of the pandemic,

and the period with both VAR review and no fan attendance. The paper focused on the

impact of mistakes made be referees, as identiﬁed by decisions subject to VAR review that

were overturned, on subsequent referee decisions in each match. The results showed that

referees called fewer Yellow cards on home teams following a call ruled incorrect by VAR

review. They also found that VAR oversight reduced the number of referee errors, but had

no eﬀect on the number of Yellow cards or the number of goals scored.

Scoppa (2021) analyzed diﬀerences in team wins and goals, all cards, and penalties called

in a pooled sample of matches played in ﬁrst and second division leagues in Germany, Spain,

England, Italy, and Portugal from the 2010/11 until the 2019/20 season. Again, this paper

used a long pre-treatment sample period. The paper reported evidence that more fouls were

called on both home and away teams in the pandemic period with fans absent.

Fischer and Haucap (2021) analyzed changes in home advantage, Yellow cards, and fouls

in matches played in the top three divisions in Germany (Bundesligas 1, 2, and 3) over the

2017/18 to 2019/20 seasons using separate regression models for each division. Results found

reduced home advantage in the pandemic period in the top division only. Fewer Yellow cards

were given to away teams and more to home teams. More fouls were called against home

teams.

Morita and Araki (2022) analyzed match outcomes, Yellow cards, and fouls in the top

two football leagues in Japan, JL1 and JL2, in the 2019/2020 season. The results: fewer

fouls called on home teams but no change in fouls called on away teams.

Several patterns emerge in this growing literature. First, a clear presence of heterogeneous

eﬀects of the pandemic on all outcomes throughout the literature, along with heterogeneous

outcomes for home and away teams. Second, most of the studies using matches played in

a large number of diﬀerent leagues use pooled samples, implicitly assuming homogeneous

eﬀects of the lack of fans in the leagues analyzed. Third, even the research focused on

outcomes in one country, Germany, tend to report heterogeneous eﬀects of missing fans

on diﬀerent outcomes. Fourth, many pooled studies contain relatively long pre-pandemic

periods. Finally, nearly all of the studies adopt a “kitchen sink” approach analyzing a large

number of outcomes.

These patterns appear related. Pooled samples across countries and leagues will not

account for underlying heterogeneity in individual leagues, and the results will reﬂect id-

iosyncratic features of the sample. Also, the use of long pre-treatment periods will tend to

attenuate estimates of the eﬀect of the pandemic on outcomes, and these studies also tend

to ignore the eﬀect of the introduction of VAR referee oversight in the seasons before the

pandemic. Our research design addresses these limitations by using a relatively short pre-

treatment sample period beginning in the 2015/16 season, accounting for the introduction

of VAR oversight, and estimating separate models for each league. It also focuses on a single

outcome, fouls called in matches, which we posit represents a good proxy for petty crimes,

unlike more serious oﬀenses like Yellow cards, Red cards, and penalties.

Research Questions

This paper tests whether playing behind closed doors with no fans in attendance aﬀected

the number of fouls called by referees, a proxy for the commission of petty crimes in settings

outside football matches. Without fans, pressure fouls and player misconduct will receive

more notice and a greater number of fouls might be called. On the other hand, in this

setting there may be less incentive for players to engage in misconduct or commit fouls since

they may not feel “protected” by a noisy environment (the likelihood of misconduct going

unnoticed is lower), so fewer fouls might be expected to be committed, and thus called.

Furthermore, in empty stadiums, the referee might be more attentive to play because of

reduced distractions.

Second, this paper examines the impact of Video Assistant Referee (VAR) monitoring

on the number of fouls called in matches, relative to matches with no VAR oversight. VAR

review of play was implemented in the “Big-Five” European football leagues beginning in the

2017/18 season. Note that VAR review applies only to “clear and obvious errors” or “serious

missed incidents” occurring in four match-changing situations: goals, penalty decisions,

direct red-card incidents, and cases of mistaken player identity. If the VAR monitor believes

the referee committed a clear and obvious error by only cautioning a player, rather than

sending him oﬀ, they can advise the referee to watch the incident again on video in the

Referee Review Area.

If the VAR review identiﬁes an error in any of these four situations they will intervene

in play, regardless of how marginal the decision was. In terms of subjective decisions, either

the referee informs the VAR oﬃcial that a decision should be reviewed or the VAR oﬃcial

identiﬁes a “clear and obvious error” in one of the four match-changing situations and com-

municates this to the referee. However, there is no consensus about what qualiﬁes as a “clear

and obvious error”.

In any event, knowledge that the game is under surveillance by an independent monitor

should aﬀect players’ behavior, including the likelihood of engaging in behavior that could

result in a foul and other forms of misconduct that could be reviewed. Referee behavior

should also change in light of VAR adoption. This could aﬀect many referee decisions

including the decision to call or not call a foul and the choice between issuing a caution or

sending a player oﬀ. Previous research by Lago-Pe˜nas et al. (2019) and Han et al. (2020)

found that VAR monitoring reduced oﬀ sides calls, Yellow cards, and overall foul calls by

referees after implementation.

This increased monitoring is expected to aﬀect players’ behavior regarding fouls and

misconduct (petty crimes are less likely to occur when police presence is felt) as well as

referee (police) behavior, much like body cameras aﬀect police actions. Jennings et al.

(2015) and Ariel et al. (2017) ﬁnd that the implementation of body-worn cameras (BWCs)

in police precincts reduced complaints against oﬃcers substantially, but Ariel et al. (2017)

goes further and found evidence of a “contagion” eﬀect where oﬃcers in the control group

in the experiment (those working on shifts in the same department at the same time but

not wearing body cameras) were almost identically aﬀected. Ready and Young (2015) found

that BWCs reduced arrests made by oﬃcers but increased the number of citations issued.

Empirical Analysis

We estimate reduced form empirical models explaining observed variation in fouls committed

by players and called by referees in football matches. The model takes the form

F ouls

mijt

= α

+ γ

+ β

V AR

+ β

LimF ans

+ β

NoF ans

+ β

mijt

+ ε

mijt

(1)

where F ouls

ijt

represents fouls committed by either home team i or away team j playing

in match m in season t. V AR

is a binary variable equal to 1 when match m used Video

Assistant Referee monitoring at time t. LimF ans

is a binary variable equal to 1 for

match m at time t where attendance in the stadium was limited but not zero due to Covid-

19 restrictions. NoF ans

is a binary variable equal to 1 for match m at time t where

no fans were allowed to attend due to Covid-19. The parameters on these three variables

represent the parameters of interest.

mijt

contains a vector of observable characteristics of matches. We include variables

that might aﬀect the nature of play in matches, including the ratio of the implied home win

and away win probability from betting odds on each match, the total number of shots by

the home and away teams in each match, total home and away team shots on target, and

total home and away team corner kicks. We estimate alternative models that exclude and

include A

mijt

to show robustness of the results to inclusion of match-level variables.

is a home team ﬁxed eﬀect capturing unobservable team-level heterogeneity. γ

is a

vector of day-of-week and season indicator variables. ε

mijt

is a mean zero, heteroskedastic

equation error term capturing other factors that aﬀect fouls in football matches.

Some previous research suggests that the number of fans attending games and their

proximity to the ﬁeld can aﬀect referee and player behavior. To investigate this idea, we also

estimate empirical models of the form

F ouls

mijt

= α

+ γ

+ β

V AR

+ β

attendance

+ β

lockdown

+β

mijt

+ β

attendance × lockdownε

mijt

(2)

where F ouls

ijt

again represents fouls committed by either home team i or away team j

playing in match m in season t and V AR

again is a binary variable equal to 1 when

match m used Video Assistant Referee monitoring at time t. attendance

is an indicator

variable identifying matches played after the pandemic lockdown aﬀected each league in the

2019/20 season. attendance × lockdown is the interaction between match attendance and

the lockdown indicator. This indicator variable allows for the impact of fans on fouls to vary

with match attendance.

Finally, to assess the relationship between fouls in football matches and the actual com-

mission of petty crimes, we estimate models that add variables reﬂecting the actual number

of petty crimes (assaults and thefts) committed in each country to Equation (1). This rep-

resents an assessment of the external validity of our regression results. The number of these

petty crimes committed varies only by country and year, not by match.

Data and descriptive analysis

Data on 10,818 matches played in the “Big-Five” European leagues were collected for seasons

2015/16 to 2020/21. Table 1 contains summary statistics on fouls called by league and other

important match characteristics for the full sample pooled over all ﬁve leagues.

Data for our match-level statistics, which include home/away fouls, home/away shots,

home/away shots on target, home/away corners, and odds ratios were taken from www.

football-data.co.uk. Odds ratios were calculated using betting odds from William Hill

Limited, as reported on football-data.co.uk. Attendance numbers were obtained from

www.soccerstats.com.

Table 1: Summary Statistics

Statistic Mean St. Dev. Min Max

Home Fouls 12.6 4.1 0 29

Away Fouls 12.9 4.1 0 32

VAR Matches 0.53 — 0 1

Matches With No Fans 0.222 — 0 1

Match Attendance (000) 25.24 19.49 0 81.2

Betting Odds Ratio 2.92 4.24 0.05 49.0

Home Shots 13.3 5.2 0 37

Away Shots 10.9 4.6 0 32

Home Shots on Target 4.8 2.6 0 18

Away Shots on Target 4.0 2.3 0 15

Home Corners 5.5 3.0 0 20

Away Corners 4.5 2.6 0 19

On average, slightly more fouls were called on visiting team players (12.9 per match)

than on home team players (12.6) in the sample. More than half the matches in the sample

featured VAR monitoring. About 20% were played with no fans in attendance.

The betting odds ratio reﬂects the diﬀerence between the implied probability that the

home team will win each match relative to the implied probability that the visiting team

will win the match based on the ﬁnal betting odds oﬀered on each match after accounting

for bookmaker over round. Betting odds based win probabilities have been used extensively

in the literature (Vandenbruaene et al., 2022). On average, betting odds implied that the

home team was 3 times more likely to win a match than the visiting team. Home teams

took more shots, more shots on target, and more penalty kicks than visiting teams.

Regression Results

Regression models explaining variation in the total number of fouls called per match were

estimated separately for each of the ﬁve top European football leagues. These diﬀerence-in-

diﬀerences models included indicator variables for the period when fans were barred from

matches (“Lockdown” matches) and also indicator variables for the period when VAR referee

oversight was in place in each league. The ratio of the home team probability to the away

team win betting probability based on betting odds for each match were included in some

models to control for expected match outcomes and intensity of play and referee oversight.

Other match level variables, including home and away team shots, shots on target, and

corner kicks, are included in some models to further control for intensity of play in matches

and to assess robustness of the results to inclusion of match-level variables.

The results from a baseline model containing only diﬀerence-in-diﬀerences indicator vari-

ables for the introduction of VAR monitoring and the banning of fan attendance or limiting

of fan attendance are shown on the left panel of Table 5 for England, Table 6 for France,

Table 2 for Germany, Table 4 for Italy, and Table 3 for Spain. These results come from

models omitting the vector of match characteristics (A

mijt

) from Equation (1). In keeping

with the practice in the literature, we estimate all models using three diﬀerent dependent

variables: home team fouls, visiting team fouls, and total fouls. All estimated standard

errors are cluster corrected at the season level.

We ﬁnd that the eﬀect of Lockdown measures and VAR oversight vary greatly depending

on the country and league. Fouls called in England’s Premier League and France’s Ligue 1

were completely unaﬀected by VAR Oversight. Italy’s Serie A and Germany’s Bundesliga

1 experienced a substantial impact from VAR oversight, reducing fouls by approximately 3

per match for Italy and 4 per match in the Bundesliga, roughly 15% reductions.

Based on the coeﬃcient estimates, it is not clear that either team was necessarily favored

after the imposition of VAR oversight. Spain’s La Liga experienced a much more modest

eﬀect from VAR, with the negative eﬀect on fouls called only being signiﬁcant for the home

team.

The eﬀect of lockdown measures and the lack of fans was substantially more heterogeneous

across the leagues than the eﬀect of VAR oversight, which either reduced fouls or did nothing

at all. Leagues in Italy and England saw substantial increases on fouls called when no fans

were present. In general, this eﬀect is more often signiﬁcant for the home team, especially

in England.

The fact that the eﬀect is more salient for the home team has an intuitive explanation:

there is an extra cost associated with calling fouls on a home team since the referee then faces

thousands of booing fans assailing him for several moments. The intuition for why limited

attendance would not have this same eﬀect on home fouls is simple as well: reducing boos

from a 50,000 person audience to a 2,000 person audience still leaves thousands of booing

voices in the crowd, whereas an empty stadium removes them completely.

The eﬀect of removing fans from matches in France diﬀered from other leagues in that it

resulted in almost two fewer fouls to be called on the away team, but had no eﬀect on the

home team or the total number of fouls called. Since away fouls are part of total fouls, this

is more diﬃcult to explain.

That being said, a lack of fans in French football games may cause referees to be more

lenient with the away team. This could be an inverse eﬀect from the aforementioned booing

fans. When a referee calls a foul on the opposing team, he is met with cheers from thousands

of home-team fans. This may increase the beneﬁt to the referee of calling a foul on the away

team, making them marginally more likely to call them when there are fans in the stadium.

Without fans, that eﬀect is gone, and referees have a marginally reduced incentive to call

fouls on the away team.

As for the interaction models, the results are generally consistent with those of the

baseline models. For teams where “No Fans” was signiﬁcant, the “Lockdown” variable

was usually signiﬁcant. However, in the case of Italy’s Serie A, the lockdown eﬀect was

insigniﬁcant despite the ”No Fans” eﬀect being signiﬁcant. Instead, the attendance and

lockdown interaction term is signiﬁcant. When limited fans were allowed, every additional

thousand of them increased total fouls called in Italian games by approximately 0.06.

The results including the actual annual number of assaults and thefts committed in each

country are shown on Table 12 for Germany, Table 13 for Spain, Table 14 for Italy, Table

15 for England, and on Table 16 for France. Adding these variables makes little diﬀerence

to the parameter estimates of interest in Equation 1, except in the case of Germany, where

Note that in Table 3, there is no Limited Attendance variable. This is because Spain only adopted full

attendance restrictions during Covid-19.

The sign is the same and the magnitude of the eﬀect is similar for total fouls, it is just a matter of the

estimated standard error preventing signiﬁcance at a bare minimum 5% level.

the impact of VAR introduction loses signiﬁcance when thefts are included. The parameter

estimate on the actual thefts variable are also positive in Germany, suggesting that crime

in football matches are associated with petty crime commission in the country. In Spain,

however, annual assaults are signiﬁcant and positively associated with total fouls called.

Table 17 shows simple correlations between total football fouls and total petty crimes by

country. The positive correlation exists in the unconditional correlations in Germany as well,

and to a lesser extent in Italy between thefts and football fouls only.

Conclusions

The results clearly show that VAR and crowd presence have a signiﬁcant eﬀect on the

way referees and players act in football matches. Perhaps even more intriguing, it appears

that the eﬀects diﬀer across countries, which could reﬂect some sort of cultural mechanism

through which rule changes aﬀect both individuals and law enforcers. The literature on

policing has long recognized that cultural norms can aﬀect the behavior of law enforcers

(Brehm and Gates, 1993). Our results contribute to this literature by developing evidence

from a novel setting containing two natural experiments that should be plausibly exogenous

to other unobservable factors aﬀecting policing outcomes.

If we assume that the heterogeneity in VAR and Crowd-Size eﬀects across European

Football fans in several diﬀerent countries can be extrapolated to overarching cultural factors,

then we can draw some interesting conclusions. For example, we have established that VAR,

which allows third-party analysis of sanctions on players, substantially reduces the number

of sanctions given in German and Italian football games. This might imply that body-worn

cameras for police in Germany and Italy will signiﬁcantly reduce the number of sanctions

they levy on citizens. In countries where the VAR eﬀect is absent, such as England and

France, body-worn cameras may have no eﬀect on police sanctions due to some cultural

mechanism speciﬁc to those countries, reﬂected in referee behavior.

In summary, two major conclusions can be drawn from the results. First, public pressure,

proxied by the presence of noisy crowds and reﬂecting interaction on social media, newspa-

pers, and gatherings of protesters, impact law enforcement, proxied by referees representing

police oﬃcers, “enforcing the rules.” Second, third-party visual monitoring, proxied by VAR

and representing BWCs also aﬀect the behavior of law enforcement oﬃcials.

Table 2: Regression Results - Bundesliga 1 (Germany)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −2.109

∗∗

−2.323

∗∗

−4.433

∗∗

−2.026

∗∗

−2.299

∗∗

−4.326

∗∗

(0.591) (0.799) (1.370) (0.578) (0.784) (1.340)

Limited Att. 0.569 0.213 0.782 0.445 0.210 0.655

(0.477) (0.718) (1.185) (0.475) (0.722) (1.191)

No Fans 0.778 0.618 1.395 0.673 0.648 1.321

(0.679) (0.706) (1.342) (0.707) (0.693) (1.348)

Odds Ratio 0.020 −0.036 −0.016

(0.042) (0.057) (0.069)

Home Shots −0.067 0.065 −0.002

(0.054) (0.041) (0.080)

Away Shots 0.004 −0.076 −0.073

(0.039) (0.040) (0.052)

Home Shots on Target 0.046 −0.096 −0.050

(0.045) (0.055) (0.087)

Away Shots on Target −0.066 −0.073 −0.139

∗

(0.053) (0.060) (0.059)

Home Corners −0.078 −0.066 −0.144

∗

(0.045) (0.039) (0.059)

Away Corners −0.091

∗∗

−0.001 −0.092

(0.032) (0.043) (0.049)

Observations 1,768 1,768 1,768 1,767 1,767 1,767

0.202 0.140 0.222 0.213 0.155 0.234

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 3: Regression Results - La Liga (Spain)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.489

∗∗∗

−0.266 −0.755

∗∗∗

−0.622

∗∗∗

−0.287 −0.909

∗∗∗

(0.166) (0.192) (0.237) (0.177) (0.176) (0.179)

No Fans 0.050 −0.246 −0.196 −0.307 −0.500

∗∗

−0.806

∗∗

(0.178) (0.193) (0.259) (0.263) (0.210) (0.340)

Odds Ratio −0.115

∗∗∗

0.045 −0.070

(0.027) (0.038) (0.051)

Home Shots −0.090

∗∗∗

−0.025 −0.115

∗∗

(0.029) (0.037) (0.056)

Away Shots −0.025 −0.148

∗∗∗

−0.173

∗∗∗

(0.025) (0.041) (0.038)

Home Shots on Target −0.013 −0.111

∗∗∗

−0.124

∗∗

(0.051) (0.032) (0.059)

Away Shots on Target −0.071 −0.070 −0.141

∗∗

(0.062) (0.076) (0.067)

Home Corners −0.116

∗∗∗

−0.071

∗∗∗

−0.187

∗∗∗

(0.041) (0.026) (0.058)

Away Corners −0.085

∗∗∗

−0.110

∗∗

−0.195

∗∗∗

(0.031) (0.043) (0.051)

Observations 2,280 2,280 2,280 2,280 2,280 2,280

0.130 0.080 0.100 0.159 0.124 0.151

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 4: Regression Results - Serie A (Italy)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −1.459

∗∗∗

−1.382

∗∗

−2.841

∗∗∗

−1.717

∗∗∗

−1.611

∗∗∗

−3.328

∗∗∗

(0.491) (0.553) (1.038) (0.381) (0.418) (0.779)

Limited Att. 0.209 0.375 0.584 0.165 0.325 0.490

(0.259) (0.420) (0.615) (0.335) (0.424) (0.661)

No Fans 0.669

∗∗

0.578 1.248

∗∗

0.542

∗∗

0.512 1.054

∗∗

(0.261) (0.382) (0.576) (0.248) (0.356) (0.537)

Odds Ratio −0.024 −0.050

∗∗

−0.074

∗∗∗

(0.020) (0.024) (0.007)

Home Shots −0.120

∗∗∗

−0.062

∗

−0.182

∗∗∗

(0.018) (0.037) (0.044)

Away Shots −0.028 −0.104

∗∗∗

−0.133

∗∗∗

(0.033) (0.008) (0.039)

Home Shots on Target 0.022 0.001 0.023

(0.042) (0.060) (0.089)

Away Shots on Target −0.045 −0.082 −0.127

(0.068) (0.069) (0.107)

Home Corners 0.008 −0.060

∗∗

−0.052

(0.014) (0.026) (0.037)

Away Corners −0.160

∗∗∗

−0.108

∗∗∗

−0.268

∗∗∗

(0.023) (0.029) (0.041)

Observations 2,261 2,261 2,261 2,260 2,260 2,260

0.128 0.111 0.153 0.153 0.141 0.195

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 5: Regression Results - Premier League (England)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.218 0.059 −0.158 −0.171 0.130 −0.041

(0.207) (0.271) (0.429) (0.210) (0.247) (0.396)

Limited Att. 0.563

∗∗

−0.562

∗

0.001 0.353

∗

−0.758

∗∗

−0.405

(0.247) (0.314) (0.281) (0.187) (0.345) (0.295)

No Fans 1.275

∗∗∗

0.107 1.382

∗∗∗

1.145

∗∗∗

0.023 1.168

∗∗∗

(0.136) (0.178) (0.278) (0.162) (0.187) (0.311)

Odds Ratio −0.137

∗∗∗

−0.091

∗∗∗

−0.228

∗∗∗

(0.039) (0.032) (0.050)

Home Shots −0.021 0.018 −0.003

(0.024) (0.023) (0.022)

Away Shots 0.016 −0.049

∗∗

−0.033

(0.011) (0.023) (0.031)

Home Shots on Target 0.083

∗∗

−0.051

∗

0.032

(0.034) (0.031) (0.038)

Away Shots on Target −0.001 0.001 0.0004

(0.015) (0.032) (0.032)

Home Corners −0.036 −0.026 −0.062

(0.048) (0.038) (0.065)

Away Corners −0.073

∗∗

−0.071 −0.144

∗∗

(0.033) (0.046) (0.057)

Observations 2,330 2,330 2,330 2,330 2,330 2,330

0.090 0.107 0.106 0.110 0.121 0.129

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 6: Regression Results - Ligue 1 (France)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight 0.048 −0.096 −0.047 0.101 −0.051 0.050

(0.160) (0.376) (0.525) (0.155) (0.366) (0.507)

Limited Att. 0.290

∗∗∗

−0.341

∗∗

−0.051 0.138

∗∗

−0.442

∗∗

−0.304

∗

(0.049) (0.159) (0.135) (0.054) (0.172) (0.166)

No Fans 0.111 −1.677

∗∗∗

−1.566

∗∗∗

0.094 −1.738

∗∗∗

−1.645

∗∗∗

(0.378) (0.280) (0.332) (0.395) (0.272) (0.327)

Odds Ratio −0.080

∗∗∗

−0.003 −0.083

(0.031) (0.065) (0.056)

Home Shots −0.087

∗∗∗

−0.038

∗∗∗

−0.125

∗∗∗

(0.031) (0.014) (0.039)

Away Shots −0.002 −0.115

∗∗∗

−0.117

∗∗∗

(0.025) (0.032) (0.034)

Home Shots on Target 0.025 −0.011 0.014

(0.048) (0.049) (0.054)

Away Shots on Target −0.032 −0.065 −0.097

(0.041) (0.066) (0.096)

Home Corners −0.056

∗∗∗

−0.095

∗∗∗

−0.151

∗∗∗

(0.021) (0.033) (0.050)

Away Corners −0.109

∗∗∗

−0.042 −0.151

∗∗∗

(0.032) (0.027) (0.053)

Observations 2,178 2,178 2,178 2,176 2,176 2,176

0.064 0.080 0.083 0.084 0.105 0.114

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 7: Regression Results, Interaction Model - Bundesliga 1 (Germany)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −2.106

∗∗∗

−2.327

∗∗∗

−4.433

∗∗∗

−2.025

∗∗∗

−2.304

∗∗∗

−4.330

∗∗∗

(0.602) (0.792) (1.375) (0.586) (0.777) (1.341)

Limited Att. 0.007 −0.009 −0.002 0.005 −0.011 −0.006

(0.012) (0.016) (0.012) (0.009) (0.018) (0.016)

No Fans 0.720 0.085 0.805 0.556

∗∗

−0.001 0.555

(0.447) (1.242) (1.206) (0.269) (1.324) (1.423)

Odds Ratio 0.013 −0.034 −0.021

(0.042) (0.060) (0.072)

Home Shots −0.067 0.064 −0.002

(0.054) (0.040) (0.080)

Away Shots 0.005 −0.078

∗

−0.073

(0.037) (0.040) (0.052)

Home Shots on Target 0.048 −0.098

∗

−0.050

(0.045) (0.056) (0.088)

Away Shots on Target −0.066 −0.072 −0.138

∗∗

(0.053) (0.060) (0.058)

Home Corners −0.078

∗

−0.065

∗

−0.143

∗∗

(0.046) (0.038) (0.059)

Away Corners −0.091

∗∗∗

−0.0002 −0.091

∗

(0.031) (0.043) (0.049)

Attendance x Lockdown 0.017 0.005 0.022

∗∗

0.018

∗∗∗

0.007 0.024

(0.011) (0.020) (0.011) (0.006) (0.021) (0.016)

Observations 1,768 1,768 1,768 1,767 1,767 1,767

0.204 0.140 0.222 0.215 0.155 0.234

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 8: Regression Results, Interaction Model - La Liga (Spain)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.503

∗∗∗

−0.288

∗

−0.791

∗∗∗

−0.655

∗∗∗

−0.314

∗∗

−0.968

∗∗∗

(0.167) (0.168) (0.211) (0.188) (0.153) (0.162)

Limited Att. 0.012 0.018

∗∗

0.030

∗∗

0.024

∗∗∗

0.012 0.036

∗∗

(0.008) (0.009) (0.014) (0.008) (0.008) (0.014)

No Fans 0.338 0.158 0.495 0.299 −0.249 0.050

(0.255) (0.306) (0.324) (0.307) (0.224) (0.351)

Odds Ratio −0.130

∗∗∗

0.036 −0.094

∗

(0.024) (0.040) (0.054)

Home Shots −0.088

∗∗∗

−0.024 −0.112

∗∗

(0.030) (0.036) (0.056)

Away Shots −0.026 −0.149

∗∗∗

−0.175

∗∗∗

(0.024) (0.041) (0.039)

Home Shots on Target −0.015 −0.110

∗∗∗

−0.125

∗∗

(0.051) (0.032) (0.060)

Away Shots on Target −0.072 −0.070 −0.143

∗∗

(0.062) (0.076) (0.067)

Home Corners −0.114

∗∗∗

−0.070

∗∗∗

−0.184

∗∗∗

(0.040) (0.026) (0.057)

Away Corners −0.083

∗∗∗

−0.108

∗∗∗

−0.191

∗∗∗

(0.032) (0.041) (0.049)

Attendance x Lockdown −0.001 0.002 0.001 −0.011 0.006 −0.005

(0.006) (0.008) (0.014) (0.008) (0.008) (0.013)

Observations 2,280 2,280 2,280 2,280 2,280 2,280

0.131 0.082 0.102 0.161 0.125 0.153

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 9: Regression Results, Interaction Model - Serie A (Italy)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −1.396

∗∗∗

−1.362

∗∗

−2.758

∗∗∗

−1.641

∗∗∗

−1.594

∗∗∗

−3.235

∗∗∗

(0.500) (0.564) (1.060) (0.391) (0.427) (0.802)

Limited Att. −0.023

∗∗

−0.005 −0.028 −0.024

∗

−0.005 −0.029

(0.010) (0.015) (0.023) (0.013) (0.016) (0.027)

No Fans −0.226 0.389 0.163 −0.364 0.343 −0.021

(0.470) (0.705) (1.150) (0.523) (0.675) (1.168)

Odds Ratio −0.021 −0.050

∗∗

−0.072

∗∗∗

(0.020) (0.025) (0.013)

Home Shots −0.119

∗∗∗

−0.063

∗

−0.182

∗∗∗

(0.017) (0.037) (0.043)

Away Shots −0.025 −0.105

∗∗∗

−0.130

∗∗∗

(0.032) (0.010) (0.042)

Home Shots on Target 0.018 0.002 0.020

(0.037) (0.060) (0.085)

Away Shots on Target −0.046 −0.082 −0.127

(0.069) (0.070) (0.110)

Home Corners 0.010 −0.059

∗∗

−0.049

(0.014) (0.026) (0.037)

Away Corners −0.164

∗∗∗

−0.108

∗∗∗

−0.272

∗∗∗

(0.024) (0.030) (0.044)

Attendance x Lockdown 0.053

∗∗∗

0.005 0.058

∗

0.053

∗∗∗

0.003 0.056

∗

(0.012) (0.021) (0.032) (0.014) (0.020) (0.033)

Observations 2,261 2,261 2,261 2,260 2,260 2,260

0.131 0.111 0.155 0.156 0.141 0.196

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 10: Regression Results, Interaction Model - Premier League (England)

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.239 0.025 −0.214 −0.200 0.091 −0.110

(0.210) (0.275) (0.435) (0.213) (0.251) (0.401)

Limited Att. −0.0001 0.014 0.014 0.006 0.016 0.022

(0.007) (0.013) (0.017) (0.005) (0.012) (0.015)

No Fans 1.054

∗∗∗

0.443 1.497

∗∗

1.129

∗∗∗

0.432 1.561

∗∗∗

(0.291) (0.503) (0.665) (0.202) (0.478) (0.597)

Odds Ratio −0.138

∗∗∗

−0.094

∗∗∗

−0.232

∗∗∗

(0.039) (0.032) (0.051)

Home Shots −0.020 0.020 −0.0001

(0.024) (0.024) (0.023)

Away Shots 0.016 −0.048

∗∗

−0.032

(0.011) (0.023) (0.031)

Home Shots on Target 0.088

∗∗∗

−0.046 0.042

(0.032) (0.031) (0.034)

Away Shots on Target 0.001 0.001 0.002

(0.016) (0.031) (0.030)

Home Corners −0.035 −0.025 −0.060

(0.049) (0.037) (0.064)

Away Corners −0.072

∗∗

−0.070 −0.142

∗∗

(0.034) (0.046) (0.058)

Attendance x Lockdown 0.010 −0.001 0.009 0.007 −0.002 0.005

(0.009) (0.011) (0.016) (0.007) (0.011) (0.013)

Observations 2,330 2,330 2,330 2,330 2,330 2,330

0.089 0.108 0.105 0.109 0.121 0.128

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 11: Regression Results, Interaction Model - Ligue 1 (France)

Dependent variable: Fouls

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight 0.057 − 0.084 −0.028 0.108 −0.037 0.071

(0.162) (0.373) (0.526) (0.158) (0.365) (0.510)

Limited Att. −0.016 −0.013 −0.029 −0.015 −0.018 −0.033

∗

(0.011) (0.014) (0.019) (0.010) (0.015) (0.018)

No Fans −0.507

∗∗

−1.615

∗∗∗

−2.122

∗∗∗

−0.625

∗∗∗

−1.863

∗∗∗

−2.487

∗∗∗

(0.204) (0.476) (0.606) (0.139) (0.480) (0.563)

Odds Ratio −0.079

∗∗

−0.002 −0.080

(0.032) (0.065) (0.053)

Home Shots −0.087

∗∗∗

−0.040

∗∗∗

−0.127

∗∗∗

(0.030) (0.014) (0.038)

Away Shots −0.004 −0.116

∗∗∗

−0.119

∗∗∗

(0.025) (0.032) (0.033)

Home Shots on Target 0.025 −0.010 0.015

(0.048) (0.048) (0.053)

Away Shots on Target −0.030 −0.065 −0.095

(0.040) (0.065) (0.095)

Home Corners −0.056

∗∗∗

−0.096

∗∗∗

−0.152

∗∗∗

(0.021) (0.033) (0.050)

Away Corners −0.108

∗∗∗

−0.042 −0.150

∗∗∗

(0.032) (0.027) (0.053)

Attendance x Lockdown 0.131

∗∗∗

0.248

∗∗∗

0.380

∗∗∗

0.132

∗∗∗

0.266

∗∗∗

0.398

∗∗∗

(0.032) (0.043) (0.054) (0.032) (0.043) (0.045)

Observations 2,178 2,178 2,178 2,176 2,176 2,176

0.065 0.079 0.083 0.084 0.105 0.114

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 12: Regression Results Including Actual Crimes - Bundesliga (Germany)

Dependent variable:

Total Fouls

(1) (2) (3) (4)

VAR −4.170

∗∗∗

−4.316

∗∗∗

−3.766

∗∗

−0.983

(1.257) (1.379) (1.490) (2.328)

No Fans 1.798

∗

(0.992)

Assaults 0.084

(0.080)

Thefts 0.012

∗

(0.007)

Home Team FE? X X X X

Week of Season FE? X X X X

Day of Week FE? X X X X

Observations 1,461 1,461 1,317 1,317

0.244 0.246 0.238 0.248

Adjusted R

0.207 0.209 0.196 0.206

Note:

∗

p<0.1;

∗∗

p<0.05;

∗∗∗

p<0.01

Season-level SE. Controls included.

Table 13: Regression Results Including Actual Crimes - La Liga (Spain)

Dependent variable:

(1) (2) (3) (4)

VAR −0.851

∗∗∗

−0.885

∗∗∗

−1.085

∗∗∗

−1.127

∗∗∗

(0.178) (0.176) (0.138) (0.121)

No Fans 0.239

(0.530)

Assaults 0.029

∗∗

(0.012)

Thefts 0.006

(0.004)

Home Team FE? X X X X

Week of Season FE? X X X X

Day of Week FE? X X X X

Observations 1,900 1,900 1,700 1,700

0.148 0.148 0.153 0.152

Adjusted R

0.112 0.111 0.112 0.112

Note:

∗

p<0.1;

∗∗

p<0.05;

∗∗∗

p<0.01

Season-level SE. Controls included.

Table 14: Regression Results Including Actual Crimes - Serie A (Italy)

Dependent variable:

(1) (2) (3) (4)

VAR −3.220

∗∗∗

−3.295

∗∗∗

−2.404

∗∗∗

−3.483

∗∗

(0.794) (0.762) (0.806) (1.466)

No Fans 1.238

∗∗∗

(0.313)

Assaults −0.326

∗

(0.184)

Thefts −0.001

(0.005)

Home Team FE? X X X X

Week of Season FE? X X X X

Day of Week FE? X X X X

Observations 1,881 1,881 1,681 1,681

0.218 0.220 0.231 0.230

Adjusted R

0.185 0.186 0.194 0.192

Note:

∗

p<0.1;

∗∗

p<0.05;

∗∗∗

p<0.01

Season-level SE. Controls included.

Table 15: Regression Results Including Actual Crimes - Premier League (England)

Dependent variable:

(1) (2) (3) (4)

VAR 0.489 −0.033

(0.432) (0.391)

No Fans 2.268

∗∗∗

(0.345)

Assaults −0.004

(0.004)

Thefts −0.001

(0.002)

Home Team FE? X X X X

Week of Season FE? X X X X

Day of Week FE? X X X X

Observations 1,881 1,881 1,321 1,321

0.134 0.139 0.149 0.148

Adjusted R

0.096 0.101 0.097 0.095

Note:

∗

p<0.1;

∗∗

p<0.05;

∗∗∗

p<0.01

Season-level SE. Controls included.

2019 Crime data missing.

Table 16: Regression Results Including Actual Crimes - Ligue 1 (France)

Dependent variable:

(1) (2) (3)

VAR −0.016 0.402 0.211

(0.520) (0.691) (0.566)

Assaults −0.006

(0.012)

Thefts 0.001

(0.001)

Home Team FE? X X X

Week of Season FE? X X X

Day of Week FE? X X X

Observations 1,797 1,705 1,705

0.129 0.125 0.126

Adjusted R

0.090 0.083 0.084

Note:

∗

p<0.1;

∗∗

p<0.05;

∗∗∗

p<0.01

Season-level SE. Controls included.

2019 Crime data missing.

Table 17: Within Country Correlation Coeﬃcients - Crimes and Football Fouls

League Cor(Assaults,Total Fouls) Cor(Thefts, Total Fouls)

Germany 0.14 0.35

Spain 0.04 0.03

Italy -0.21 0.18

England -0.08 -0.06

France -0.02 0.06

References

Ariel, B., Sutherland, A., Henstock, D., Young, J., Drover, P., Sykes, J., Megicks, S., and

Henderson, R. (2017). “contagious accountability”: A global multisite randomized con-

trolled trial on the eﬀect of police body-worn cameras on citizens’ complaints against the

police. Criminal Justice and Behavior, 44(2):293–316.

Becker, G. S. (1968). Crime and punishment: An economic approach. Journal of Political

Economy, 76(2):169–217.

Brehm, J. and Gates, S. (1993). Donut shops and speed traps: Evaluating models of super-

vision on police behavior. American Journal of Political Science, pages 555–581.

Bryson, A., Dolton, P., Reade, J. J., Schreyer, D., and Singleton, C. (2021). Causal eﬀects

of an absent crowd on performances and refereeing decisions during covid-19. Economics

Letters, 198:1–5.

Cohen, A., Neeman, Z., and Auferoth, F. (2021). Judging under public pressure. Working

paper, National Bureau of Economic Research.

Cueva, C. (2020). Animal spirits in the beautiful game. testing social pressure in professional

football during the covid-19 lockdown. Working paper, OSF Preprints.

Dilger, A. and Vischer, L. (2020). No home bias in ghost games. Discussion paper, Institute

for Organisational Economics.

Endrich, M. and Gesche, T. (2020). Home-bias in referee decisions: Evidence from “ghost

matches” during the covid19-pandemic. Economics Letters, 197:1–4.

Ferraresi, M. and Gucciardi, G. (2020). Team performance and audience: experimental

evidence from the football sector. Working paper, Societ`a italiana di economia pubblica.

Fischer, K. and Haucap, J. (2021). Does crowd support drive the home advantage in profes-

sional football? evidence from german ghost games during the covid-19 pandemic. Journal

of Sports Economics, 22(8):982–1008.

Garicano, L., Palacios-Huerta, I., and Prendergast, C. (2005). Favoritism under social pres-

sure. Review of Economics and Statistics, 87(2):208–216.

Han, B., Chen, Q., Lago-Pe˜nas, C., Wang, C., and Liu, T. (2020). The inﬂuence of the video

assistant referee on the chinese super league. International Journal of Sports Science &

Coaching, 15(5-6):662–668.

Jennings, W. G., Lynch, M. D., and Fridell, L. A. (2015). Evaluating the impact of police of-

ﬁcer body-worn cameras (bwcs) on response-to-resistance and serious external complaints:

Evidence from the orlando police department (opd) experience utilizing a randomized

controlled experiment. Journal of Criminal Justice, 43(6):480–486.

Kitchens, C., Makofske, M. P., and Wang, L. (2019). “Crime” on the ﬁeld. Southern

Economic Journal, 85(3):821–864.

Lago-Pe˜nas, C., Rey, E., and Kal´en, A. (2019). How does video assistant referee (VAR)

modify the game in elite soccer? International Journal of Performance Analysis in Sport,

19(4):646–653.

McCarrick, D., Bilalic, M., Neave, N., and Wolfson, S. (2020). Home advantage during the

covid-19 pandemic in european football. Working paper, PsyArXiv.

Morita, H. and Araki, S. (2022). Social pressure in football matches: an event study of

‘remote matches’ in japan. Applied Economics Letters, pages 1–4.

Owens, E. and Ba, B. (2021). The economics of policing and public safety. Journal of

Economic Perspectives, 35(4):3–28.

Pope, D. G., Price, J., and Wolfers, J. (2018). Awareness reduces racial bias. Management

Science, 64(11):4988–4995.

Reade, J., Schreyer, D., and Singleton, C. (2020). Eliminating supportive crowds reduces

referee bias. Working paper 3743972, SSRN.

Ready, J. and Young, J. T. N. (2015). The impact of on-oﬃcer video cameras on police–citizen

contacts: ﬁndings from a controlled experiment in Mesa, AZ. Journal of Experimental

Criminology, 11:445–458.

Scoppa, V. (2021). Social pressure in the stadiums: Do agents change behavior without

crowd support? Journal of Economic Psychology, 82:102344.

Vandenbruaene, J., Ceuster, M. D., and Annaert, J. (2022). Eﬃcient spread betting markets:

A literature review. Journal of Sports Economics, pages 1–43.

Appendix: Robustness Checks (Removal of 2020/21 sea-

son data.)

Table 18: Regression Results - Bundesliga 1 (Germany) - 2020/21 Season Excluded

Dependent variable: Fouls by Team

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −2.119

∗∗∗

−2.295

∗∗∗

−4.414

∗∗∗

−2.056

∗∗∗

−2.260

∗∗∗

−4.316

∗∗∗

(0.611) (0.818) (1.413) (0.597) (0.797) (1.379)

No Fans 1.340

∗∗∗

0.523 1.863

∗

1.299

∗∗∗

0.499 1.798

∗

(0.390) (0.720) (1.042) (0.392) (0.679) (0.992)

Odds Ratio −0.015 −0.036 −0.051

(0.043) (0.071) (0.079)

Home Shots −0.070 0.054 −0.016

(0.066) (0.048) (0.094)

Away Shots 0.033 −0.089

∗

−0.056

(0.029) (0.047) (0.060)

Home Shots on Target 0.041 −0.110

∗

−0.069

(0.056) (0.061) (0.097)

Away Shots on Target −0.087

∗

−0.094 −0.181

∗∗∗

(0.050) (0.071) (0.042)

Home Corners −0.040 −0.052 −0.092

∗∗

(0.036) (0.045) (0.041)

Away Corners −0.096

∗∗

−0.003 −0.099

∗

(0.044) (0.055) (0.059)

Observations 1,462 1,462 1,462 1,461 1,461 1,461

0.220 0.146 0.235 0.230 0.164 0.246

Adjusted R

0.186 0.108 0.201 0.191 0.123 0.209

Residual Std. Error 3.775 4.092 6.051 3.759 4.058 6.016

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 19: Regression Results - La Liga (Spain) - 2020/21 Season Excluded

Dependent variable:

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.515

∗∗∗

−0.221 −0.735

∗∗∗

−0.653

∗∗∗

−0.232 −0.885

∗∗∗

(0.170) (0.176) (0.236) (0.184) (0.155) (0.176)

No Fans 0.547

∗∗

0.185 0.732

∗

0.273 −0.034 0.239

(0.277) (0.293) (0.417) (0.333) (0.370) (0.530)

Odds Ratio −0.127

∗∗∗

0.060

∗

−0.066

(0.029) (0.036) (0.054)

Home Shots −0.094

∗∗∗

−0.024 −0.118

∗

(0.033) (0.043) (0.068)

Away Shots −0.039 −0.134

∗∗∗

−0.174

∗∗∗

(0.024) (0.043) (0.045)

Home Shots on Target 0.006 −0.111

∗∗∗

−0.105

(0.052) (0.040) (0.066)

Away Shots on Target −0.033 −0.126

∗∗

−0.159

∗∗

(0.055) (0.056) (0.073)

Home Corners −0.105

∗∗

−0.065

∗∗

−0.170

∗∗∗

(0.048) (0.027) (0.060)

Away Corners −0.097

∗∗∗

−0.087

∗∗

−0.184

∗∗∗

(0.037) (0.040) (0.053)

Observations 1,900 1,900 1,900 1,900 1,900 1,900

0.142 0.087 0.097 0.171 0.133 0.148

Adjusted R

0.108 0.051 0.062 0.136 0.096 0.111

Residual Std. Error 4.015 4.016 5.745 3.953 3.921 5.591

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 20: Regression Results - Serie A (Italy) - 2020/21 Season Excluded

Dependent variable:

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −1.447

∗∗∗

−1.313

∗∗

−2.760

∗∗∗

−1.751

∗∗∗

−1.544

∗∗∗

−3.295

∗∗∗

(0.522) (0.541) (1.058) (0.398) (0.386) (0.762)

No Fans 1.132

∗∗∗

0.178 1.311

∗∗∗

1.025

∗∗∗

0.213 1.238

∗∗∗

(0.192) (0.371) (0.466) (0.211) (0.275) (0.313)

Odds Ratio −0.028 −0.044 −0.071

∗∗∗

(0.023) (0.027) (0.007)

Home Shots −0.119

∗∗∗

−0.063 −0.182

∗∗∗

(0.020) (0.046) (0.053)

Away Shots −0.052

∗∗

−0.111

∗∗∗

−0.163

∗∗∗

(0.022) (0.004) (0.022)

Home Shots on Target −0.005 −0.002 −0.008

(0.043) (0.071) (0.095)

Away Shots on Target −0.014 −0.120 −0.134

(0.069) (0.077) (0.132)

Home Corners 0.016 −0.063

∗∗

−0.046

(0.011) (0.027) (0.036)

Away Corners −0.163

∗∗∗

−0.123

∗∗∗

−0.286

∗∗∗

(0.030) (0.034) (0.053)

Observations 1,881 1,881 1,881 1,881 1,881 1,881

0.140 0.133 0.174 0.167 0.168 0.220

Adjusted R

0.106 0.099 0.141 0.131 0.132 0.186

Residual Std. Error 3.974 4.164 6.187 3.918 4.087 6.023

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 21: Regression Results - Premier League (England) - 2020/21 Season Excluded

Dependent variable:

Home Away Total Home Away Total

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.170 0.017 −0.153 −0.126 0.094 −0.033

(0.209) (0.287) (0.419) (0.219) (0.264) (0.391)

No Fans 1.571

∗∗∗

0.837

∗∗∗

2.409

∗∗∗

1.488

∗∗∗

0.779

∗∗∗

2.268

∗∗∗

(0.314) (0.144) (0.323) (0.327) (0.187) (0.345)

Odds Ratio −0.111

∗∗∗

−0.104

∗∗∗

−0.215

∗∗∗

(0.033) (0.034) (0.054)

Home Shots −0.041

∗∗

0.020 −0.021

(0.017) (0.025) (0.018)

Away Shots 0.018 −0.055

∗∗

−0.037

(0.015) (0.022) (0.034)

Home Shots on Target 0.111

∗∗∗

−0.054 0.058

(0.020) (0.040) (0.043)

Away Shots on Target −0.001 −0.010 −0.011

(0.017) (0.043) (0.039)

Home Corners −0.046 −0.029 −0.074

(0.060) (0.046) (0.080)

Away Corners −0.088

∗∗

−0.055 −0.143

∗∗

(0.038) (0.051) (0.069)

Observations 1,881 1,881 1,881 1,881 1,881 1,881

0.096 0.107 0.115 0.117 0.123 0.139

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 22: Regression Results - Ligue 1 (France) - 2020/21 Season Excluded

Dependent variable:

HF AF TF HF AF TF

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.015 −0.126 −0.141 0.049 −0.065 −0.016

(0.151) (0.390) (0.530) (0.153) (0.386) (0.520)

Odds Ratio −0.099

∗∗∗

0.021 −0.078

(0.024) (0.083) (0.078)

Home Shots −0.108

∗∗∗

−0.040

∗∗

−0.149

∗∗∗

(0.029) (0.017) (0.040)

Away Shots −0.008 −0.134

∗∗∗

−0.142

∗∗∗

(0.029) (0.032) (0.029)

Home Shots on Target 0.053 −0.009 0.044

(0.050) (0.060) (0.054)

Away Shots on Target −0.0001 −0.060 −0.060

(0.031) (0.079) (0.103)

Home Corners −0.064

∗∗∗

−0.083

∗∗

−0.147

∗∗

(0.022) (0.041) (0.063)

Away Corners −0.123

∗∗∗

−0.037 −0.161

∗∗

(0.034) (0.038) (0.066)

Home Team FE? X X X X X X

Week of Season FE? X X X X X X

Day of Week FE? X X X X X X

Observations 1,798 1,798 1,798 1,797 1,797 1,797

0.073 0.085 0.097 0.098 0.113 0.129

Adjusted R

0.036 0.048 0.060 0.058 0.073 0.090

Residual Std. Error 3.769 3.926 5.773 3.725 3.873 5.680

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 23: Regression Results, Interaction Model - Bundesliga 1 (Germany) - 2020/21 Season

Excluded

Dependent variable:

HF AF TF HF AF TF

(1) (2) (3) (4) (5) (6)

VAR Oversight −2.105

∗∗∗

−2.374

∗∗∗

−4.479

∗∗∗

−2.054

∗∗∗

−2.342

∗∗∗

−4.397

∗∗∗

(0.639) (0.852) (1.476) (0.625) (0.816) (1.425)

Attendance 0.032 −0.131

∗

−0.099 0.014 −0.142

∗∗

−0.128

(0.053) (0.077) (0.128) (0.045) (0.063) (0.107)

Lockdown −0.111 2.263

∗∗

2.152

∗∗

0.090 2.360

∗∗∗

2.450

∗∗∗

(0.159) (0.914) (0.968) (0.244) (0.901) (0.918)

Odds Ratio −0.015 −0.034 −0.049

(0.044) (0.071) (0.081)

Home Shots −0.070 0.054 −0.017

(0.066) (0.048) (0.093)

Away Shots 0.032 −0.087

∗

−0.054

(0.029) (0.046) (0.061)

Home Shots on Target 0.041 −0.108

∗

−0.067

(0.055) (0.060) (0.095)

Away Shots on Target −0.086

∗

−0.101 −0.186

∗∗∗

(0.050) (0.069) (0.044)

Home Corners −0.037 −0.058 −0.095

∗∗

(0.034) (0.044) (0.041)

Away Corners −0.093

∗∗

−0.009 −0.103

∗

(0.043) (0.056) (0.061)

Attendance x Lockdown 0.034

∗∗∗

−0.040

∗∗∗

−0.006 0.028

∗∗∗

−0.042

∗∗∗

−0.014

(0.008) (0.008) (0.014) (0.009) (0.009) (0.015)

Observations 1,462 1,462 1,462 1,461 1,461 1,461

0.222 0.149 0.235 0.230 0.167 0.247

Adjusted R

0.186 0.110 0.200 0.191 0.124 0.208

Residual Std. Error 3.775 4.089 6.054 3.760 4.054 6.019

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 24: Regression Results, Linear Interaction - Serie A (Italy) - 2020/21 Season Excluded

Dependent variable:

HF AF TF HF AF TF

(1) (2) (3) (4) (5) (6)

VAR Oversight −1.326

∗∗

−1.252

∗∗

−2.579

∗∗

−1.603

∗∗∗

−1.451

∗∗∗

−3.054

∗∗∗

(0.554) (0.578) (1.130) (0.435) (0.443) (0.864)

Attendance −0.040

∗∗∗

−0.021 −0.061

∗

−0.052

∗∗∗

−0.032 −0.084

∗∗∗

(0.014) (0.023) (0.033) (0.014) (0.023) (0.033)

Lockdown −0.926

∗∗∗

−1.166

∗∗

−2.093

∗∗∗

−1.061

∗∗∗

−0.964

∗∗

−2.025

∗∗∗

(0.228) (0.484) (0.533) (0.241) (0.426) (0.358)

Odds Ratio −0.021 −0.040 −0.061

∗∗∗

(0.025) (0.025) (0.006)

Home Shots −0.123

∗∗∗

−0.066 −0.188

∗∗∗

(0.021) (0.048) (0.056)

Away Shots −0.051

∗∗

−0.111

∗∗∗

−0.162

∗∗∗

(0.021) (0.004) (0.021)

Home Shots on Target −0.002 −0.00001 −0.002

(0.041) (0.071) (0.095)

Away Shots on Target −0.010 −0.117 −0.127

(0.074) (0.079) (0.138)

Home Corners 0.017 −0.063

∗∗

−0.046

(0.011) (0.028) (0.036)

Away Corners −0.168

∗∗∗

−0.126

∗∗∗

−0.294

∗∗∗

(0.032) (0.036) (0.058)

Attendance:lockdown 0.076

∗∗∗

0.049

∗∗∗

0.125

∗∗∗

0.077

∗∗∗

0.044

∗∗∗

0.120

∗∗∗

(0.006) (0.008) (0.008) (0.005) (0.006) (0.005)

Observations 1,881 1,881 1,881 1,881 1,881 1,881

0.144 0.135 0.178 0.172 0.169 0.224

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 25: Regression Results, Linear Interaction - Premier League (England) - 2020/21

Season Excluded

Dependent variable:

HF AF TF HF AF TF

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.145 0.109 −0.036 −0.108 0.175 0.067

(0.211) (0.257) (0.393) (0.218) (0.243) (0.371)

Attendance −0.022 −0.063

∗∗∗

−0.085

∗∗∗

−0.016 −0.058

∗∗

−0.074

∗∗

(0.015) (0.020) (0.024) (0.012) (0.024) (0.029)

Lockdown 1.343

∗∗∗

2.663

∗∗∗

4.006

∗∗∗

1.477

∗∗∗

2.491

∗∗∗

3.967

∗∗∗

(0.249) (0.279) (0.200) (0.246) (0.358) (0.328)

Odds Ratio −0.109

∗∗∗

−0.096

∗∗∗

−0.205

∗∗∗

(0.032) (0.032) (0.052)

Home Shots −0.042

∗∗

0.015 −0.027

∗

(0.017) (0.024) (0.016)

Away Shots 0.018 −0.054

∗∗∗

−0.036

(0.014) (0.020) (0.031)

Home Shots on Target 0.110

∗∗∗

−0.055 0.055

(0.019) (0.039) (0.041)

Away Shots on Target 0.0004 −0.008 −0.007

(0.018) (0.040) (0.037)

Home Corners −0.046 −0.029 −0.075

(0.060) (0.046) (0.080)

Away Corners −0.089

∗∗

−0.059 −0.148

∗∗

(0.039) (0.051) (0.067)

Attendance:lockdown 0.006 −0.046

∗∗∗

−0.040

∗∗∗

0.0003 −0.043

∗∗∗

−0.043

∗∗∗

(0.006) (0.004) (0.006) (0.005) (0.005) (0.007)

Observations 1,881 1,881 1,881 1,881 1,881 1,881

0.097 0.114 0.120 0.117 0.129 0.143

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level

Table 26: Regression Results, Interaction Model - La Liga (Spain) - Season 6 Excluded

Dependent variable:

HF AF TF HF AF TF

(1) (2) (3) (4) (5) (6)

VAR Oversight −0.390

∗∗∗

−0.161 −0.551

∗∗

−0.553

∗∗∗

−0.145 −0.699

∗∗∗

(0.138) (0.190) (0.261) (0.146) (0.167) (0.199)

Attendance −0.082

∗∗

−0.041 −0.123

∗∗∗

−0.060 −0.056

∗∗

−0.116

∗∗∗

(0.039) (0.026) (0.021) (0.040) (0.027) (0.025)

Lockdown 1.066

∗∗∗

0.165 1.231

∗∗

1.083

∗∗∗

−0.098 0.985

∗

(0.250) (0.404) (0.550) (0.339) (0.356) (0.591)

Odds Ratio −0.122

∗∗∗

0.067

∗

−0.055

(0.027) (0.036) (0.055)

Home Shots −0.095

∗∗∗

−0.025 −0.120

∗

(0.033) (0.042) (0.067)

Away Shots −0.040

∗

−0.136

∗∗∗

−0.176

∗∗∗

(0.021) (0.045) (0.045)

Home Shots on Target 0.006 −0.108

∗∗∗

−0.102

(0.049) (0.040) (0.062)

Away Shots on Target −0.033 −0.124

∗∗

−0.157

∗∗

(0.053) (0.056) (0.071)

Home Corners −0.106

∗∗

−0.065

∗∗

−0.172

∗∗∗

(0.048) (0.027) (0.060)

Away Corners −0.096

∗∗∗

−0.084

∗∗

−0.180

∗∗∗

(0.036) (0.041) (0.053)

Attendance:lockdown −0.016

∗∗∗

0.002 −0.014 −0.027

∗∗∗

0.004 −0.023

∗

(0.006) (0.008) (0.013) (0.009) (0.009) (0.013)

Observations 1,900 1,900 1,900 1,900 1,900 1,900

0.144 0.088 0.099 0.173 0.134 0.150

Adjusted R

0.110 0.051 0.063 0.137 0.096 0.113

Residual Std. Error 4.011 4.017 5.740 3.950 3.921 5.586

Notes:

∗∗

p<0.05;

∗∗∗

p<0.01. All models contain home team, week of season, and day of

week ﬁxed eﬀects. Std. Errors Clustered at Season Level