The Impact of Consumer Perceptions of Tanking on National Basketball Association Attendance

University of South Carolina University of South Carolina

Scholar Commons Scholar Commons

Theses and Dissertations

Summer 2020

The Impact of Consumer Perceptions of Tanking on National The Impact of Consumer Perceptions of Tanking on National

Basketball Association Attendance Basketball Association Attendance

Hua Gong

Follow this and additional works at: https://scholarcommons.sc.edu/etd

Part of the Sports Management Commons

Recommended Citation Recommended Citation

Gong, H.(2020).

The Impact of Consumer Perceptions of Tanking on National Basketball Association

Attendance.

(Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/6061

This Open Access Dissertation is brought to you by Scholar Commons. It has been accepted for inclusion in

Theses and Dissertations by an authorized administrator of Scholar Commons. For more information, please

contact [email protected].

HE IMPACT OF CONSUMER PERCEPTIONS OF TANKING ON NATIONAL BASKETBALL

ASSOCIATION ATTENDANCE

Hua Gong

Bachelor of Science

University of Nottingham Ningbo China, 2013

Master of Science

Texas A&M University, 2015

Submitted in Partial Fulfillment of the Requirements

For the Degree of Doctor of Philosophy in

Sport and Entertainment Management

College of Hospitality, Retail, and Sport Management

University of South Carolina

2020

Accepted by:

Matt Brown, Major Professor

Nick Watanabe, Major Professor

Mark Nagel, Committee Member

Brain Soebbing, Committee Member

Cheryl L. Addy, Vice Provost and Dean of the Graduate School

All Rights Reserve

iii

DEDICATION

To my parents, thank you for being open-minded. Thank you for your unwavering

support along the way. Thank you for all the sacrifices.

ACKNOWLEDGEMENTS

I want to take this opportunity to acknowledge my mentors who helped me

tremendously over the past four years. First, I would like to thank Dr. Nick Watanabe for

your great mentorship. You set a high standard for me and shaped who I am as a scholar

today. Without your great mentorship, I would not be ready for future challenges in

academia. I would also like to thank Dr. Matt Brown for your tremendous guidance over

the past few years. Following your guidance, I learned how to be a great teacher and

researcher. Thank you for your patience along the way.

I would also like to acknowledge other members of my dissertation committee. Dr.

Mark Nagel, thank you for your great support over the past few years. Your knowledge

of the sport industry inspired me to be a better scholar in the field of sport management

and your work ethic set a great example for me to follow. Dr. Brian Soebbing, thank you

for always taking the time and effort to give me thoughtful suggestions on my research. I

learned a lot from your work.

Finally, I want to thank my fellow Ph.D. students who provided me mental and

spiritual support over the past few years. I wish everyone the best. I would also like to

acknowledge the Department of Sport and Entertainment at the University of South

Carolina and everyone in the program. Thank you for your service and help.

ABSTRACT

This dissertation studies the impact of consumer perceptions of tanking on

National Basketball Attendance (NBA) attendance. The prevalence of tanking in the

NBA raised concerns that some teams were purposely avoiding winning games in order

to improve their draft position. The majority of previous studies on tanking have focused

on developing empirical evidence of the existence of tanking in sport. Yet, no study

systematically explored the impact of perceived tanking behavior on consumer demand

for sport. As tanking teams rarely reveal their tanking strategy to the public, fans may not

correctly identify tanking behavior in sport, and thus are likely to rely on their

perceptions of tanking to make attendance decisions. The current dissertation employs

tanking discussions on the social media platform Twitter along with data mining tools to

quantify consumer perceptions of tanking. Econometric models are then utilized to

analyze the effect of the perceived tanking behavior on demand for NBA games. The

estimation results provide robust evidence that the increasing awareness of tanking for

home teams hurts NBA attendance in both the short and long term. This dissertation also

reveals that more negative attitudes toward visiting teams’ tanking behavior can

undermine consumer interest in attending NBA games. These findings offer important

managerial implications on the urgency of restraining tanking behavior as well as the

importance of maintaining integrity in sports competitions.

TABLE OF CONTENTS

DEDICATION ................................................................................................................... iii

ACKNOWLEDGEMENTS ............................................................................................... iv

ABSTRACT .........................................................................................................................v

LIST OF TABLES ........................................................................................................... viii

LIST OF FIGURES ........................................................................................................... ix

LIST OF ABBREVIATIONS ..............................................................................................x

CHAPTER 1 INTRODUCTION .........................................................................................1

1.1 DEFINITION OF TANKING............................................................................3

1.2 LINK BETWEEN TANKING AND THE DRAFT SYSTEM .........................4

1.3 AN OVERVIEW OF TANKING TACTICS AND STRATEGIES ..................5

1.4 TANKING IN THE NBA ..................................................................................9

1.5 NBA DRAFT REFORMS ...............................................................................11

1.6 PURPOSE OF STUDY....................................................................................13

CHAPTER 2 LITERATURE REVIEW ............................................................................15

2.1 TANKING AND OTHER SIMILAR CONCEPTS .......................................15

2.2 EVIDENCE OF TANKING IN SPORT..........................................................19

2.3 TOURNAMENT THEORY AND TANKING ...............................................24

2.4 DEMAND FOR ATTENDANCE ...................................................................28

2.5 SOCIAL MEDIA AND SPORT ......................................................................39

CHAPTER 3 METHODOLOGY ......................................................................................45

vii

3.1 OVERVIEW ....................................................................................................45

3.2 DATA DESCRIPTIONS .................................................................................46

3.3 DATA COLLECTION ....................................................................................53

3.4 MODEL SPECIFICATIONS ..........................................................................61

CHAPTER 4 RESULTS ...................................................................................................75

4.1 SUMMARY OF TWITTER DATA ................................................................75

4.2 REGRESSION ANALYSIS RESULTS ..........................................................79

CHAPTER 5 DISCUSSION ............................................................................................105

5.1 DISCUSSION ................................................................................................105

5.2 THEORETICAL CONTRIBUTIONS ...........................................................109

5.3 PRACTICAL IMPLICATIONS ....................................................................111

CHAPTER 6 CONCLUSION..........................................................................................114

6.1 SUMMARY ...................................................................................................114

6.2 LIMITATIONS ..............................................................................................116

6.3 FUTURE STUDIES.......................................................................................117

REFERENCES ................................................................................................................119

viii

LIST OF TABLES

Table 3.1 Variable Descriptions ........................................................................................65

Table 3.2 Summary Statistics (n= 5945) ...........................................................................67

Table 3.3 NBA Team Search Keywords ............................................................................69

Table 3.4 Tanking Tweet Samples.....................................................................................70

Table 4.1 Ten Highest Volume of Tanking Tweets by Team and Season ........................95

Table 4.2 Ten Lowest Sentiment of Tanking Tweets by Team and Season ......................96

Table 4.3 Regression Analysis Results for Main Models ..................................................97

Table 4.4 Regression Analysis Results for Robustness Check 1 and 2 .............................99

Table 4.5 Regression Analysis Results for Robustness Check 3 .....................................101

Table 4.6 Regression Analysis Results for Robustness Check 4 .....................................103

LIST OF FIGURES

Figure 1.1 NBA Weighted Lottery Percentages ................................................................14

Figure 3.1 Textual Data Pre-Processing Steps ...................................................................71

Figure 3.2 Feature Extraction Process ...............................................................................72

Figure 3.3 Data Analysis Process with Machine Learning Models and Twitter Data .......73

Figure 4.1 Short-term Volume of Tanking Tweets for the Philadelphia

Sixers from the 2013-2014 to 2017-2018 Seasons ................................................89

Figure 4.2 Volume of Tanking Tweets by Season .............................................................90

Figure 4.3 Volume of Tanking Tweets by Month .............................................................91

Figure 4.4 Short-term Sentiment of Tanking Tweets for the Chicago

Bulls from the 2013- 2014 to 2017-2018 Seasons .................................................92

Figure 4.5 Sentiment of Tanking Tweets by Season .........................................................93

Figure 4.6 Sentiment of Tanking Tweets by Month ..........................................................94

LIST OF ABBREVIATIONS

API ............................................................................... Application Programming Interface

BCSD ............................................................................ Bias-Corrected Standard Deviation

EPL ................................................................................................. English Premier League

FCI ................................................................................................................ Fan Cost Index

FIFA ....................................................... Fédération Internationale de Football Association

LOESS ................................................................ Locally Estimated Scatterplot Smoothing

LSVM .................................................................................Linear Support Vector Machine

MLB ................................................................................................. Major League Baseball

NASCAR ................................................. National Association for Stock Car Auto Racing

NBA ................................................................................... National Basketball Association

NCAA ...................................................................National Collegiate Athletic Association

NFL ............................................................................................... National Football League

NHL ...............................................................................................National Hockey League

NLP ........................................................................................ Natural Language Processing

OLS .................................................................................................. Ordinary Least Squares

RSD ..................... Ratio of Actual Standard Deviation to the Idealized Standard Deviation

UFC ................................................................................... Ultimate Fighting Championship

UOH ............................................................................. Uncertainty of Outcome Hypothesis

CHAPTER 1

INTRODUCTION

Game attendance has long been a vital part of the professional sports business

model (Hansen & Gauthier, 1989; Neale, 1964). Consumers attending sporting events

contribute substantial ticket revenue to sports teams and leagues (Szymanski & Késenne,

2004). In addition, ancillary revenues such as merchandise sales and parking fees can

also experience significant growth with high attendance (Mason, 1999). Key

determinants of demand for sport are well documented in the literature (Borland &

MacDonald, 2003; Villar & Guerrero, 2009). One such determinant is the quality of

competition that generally encompasses the strength of the teams on the field and the

uncertainty of game outcomes (Villar & Guerrero, 2009). For instance, prior studies of

consumer demand for sport provide consistent evidence that team strength, normally

measured by the team winning percentage, is positively linked to game attendance

(Borland & MacDonald, 2003; Noll, 1974; Villar & Guerrero, 2009). That is, winning

teams typically draw more fans to sporting events than losing teams. Furthermore, studies

on the uncertainty of game outcomes find some evidence to support the hypothesis that

demand for certain matches increases when game outcomes are unpredictable (Benz,

Brandes, & Franck, 2006; Rascher, 1999; Rascher & Solmes, 2007).

At the same time, the prevalence of tanking in sport threatens the quality of

competitions and may impact demand for games through both undermining team strength

and the uncertainty of game outcomes (Price, Soebbing, Berri, & Humphreys, 2010). In

North American sports leagues, tanking could occur when teams do not exert their best

winning effort. They often tank in order to improve their draft position in the upcoming

amateur entry draft (Soebbing & Mason, 2009; Soebbing, Humphreys, & Mason, 2013).

This form of tanking emerges as a result of the draft system where the ability to acquire

high draft picks is determined by teams’ regular-season win-loss record in reverse order

(Soebbing et al., 2013). Tanking teams often field less competitive rosters, which

therefore reduce team quality (McManus, 2019). With lower team quality, sports teams

are more likely to lose games and also draw fewer fans to games. For example, after

trading away high-quality players, the Miami Marlins drew mere 10,013 fans per game in

2018, the lowest attendance for any Major League Baseball (MLB) team since 2004

(Shafer, 2018). Teams like the Marlins also compromise the uncertainty of game

outcomes by not fielding talented players, thereby not exerting maximum effort to win

games and pursue a division, conference, or league championship (Soebbing & Mason,

2009; Taylor &Trogdon, 2002). Thus, if unpredictability attracts fans to games in sports

leagues, the tanking strategy employed by the Marlins may seriously hurt attendance

(Duvall, 2018; Kendall & Lenten, 2017; Soebbing & Mason, 2009). As the success of

sport business relies on fan support, the presence of tanking can damage the quality of

competitions, and thus impose significant threats to sports leagues and teams.

In addition to the compromised quality of competitions resulting from tanking,

perceptions of tanking may further affect consumer demand for sport (Soebbing & Mason,

2009; Price et al., 2010). If fans perceive the existence of tanking by teams and within

sports leagues, then they may decide not to attend games or stop following a sport in

general, regardless of whether teams are actually tanking or not (Soebbing & Mason,

2009; Price et al., 2010). This is certainly a possibility as most teams do not acknowledge

tanking to the public (Aldridge, 2018). With limited information on which teams are

deliberately tanking, consumers may not differentiate actual tanking from below-average

performance on the field or court. Therefore, it is likely that they rely on their perceptions

tanking to make attendance decisions (Soebbing & Mason, 2009). While some fans are

not interested in watching tanking teams, others may fully support tanking as it can

ultimately help teams gain competitive advantages in the long term through acquiring

higher draft picks (Lenten, Smith, & Boys, 2018). In light of various opinions on tanking,

there is a need to further understand the relationship between perceptions of tanking and

consumer interest in sport. The dissertation utilizes the National Basketball Association

(NBA), where tanking has been a concern for decades, as the research context to explore

this relationship.

Chapter 1 covers a variety of topics related to the background information on

tanking, ranging from the definitions of tanking to draft reforms. The introduction serves

as a basis for the later discussion on the relationship between consumer perceptions of

tanking and demand for NBA games. To begin with, several definitions of tanking are

compared and differentiated.

1.1 DEFINITION OF TANKING

Despite the prevalence of tanking in sport, its definition seems to be vague in

literature. In fact, a range of studies have proposed different definitions of tanking

(Balsdon, Fong, & Thayer, 2017; Lenten et al., 2018; Soebbing & Mason, 2009). To

better explore tanking behavior in this dissertation, it is critical to first clarify the

meaning of tanking. Soebbing and Mason (2009) defined tanking as the behavior of

teams deliberately losing games to improve draft position. This definition has been

widely employed in numerous studies (e.g., Borland, Chicu, & Macdonald, 2009; Lenten

et al., 2018; Motomura, Roberts, Leeds, & Leeds, 2016; Price et al., 2010; Soebbing et al.,

2013; Soebbing & Humphreys, 2013). Balsdon et al. (2007) had a slightly different

definition in their study examining tanking behavior in college basketball. They noted

tanking took place when participants did not exert the best effort in winning a conference

tournament championship. They suggested that this tanking behavior might arise from

two reasons: (a) preserving energy for the National Collegiate Athletic Association

(NCAA) tournament, and (b) maximizing conference revenue by conceding the title to

another conference team which then will receive an automatic bid to the NCAA

tournament. Kendall and Lenten (2017) offered a broader definition of tanking than the

previous two, considering tanking as an act of intentionally dropping points or losing

games in order to gain competitive advantages.

To better analyze the effect of consumer perceptions of tanking on demand for

NBA games, Soebbing and Mason’s (2009) definition of tanking is adopted in this

dissertation. As such, this dissertation focuses on examining tanking behavior where

NBA teams deliberately lose games in order to acquire higher draft picks. Other

definitions of tanking are not considered in this inquiry.

1.2 LINK BETWEEN TANKING AND THE DRAFT SYSTEM

The previous section noted some links between tanking and the draft system. In

fact, the emergence of the tanking behavior in the NBA since the early 1980s could be

traced back to the draft system that the league adopted in 1947 (Soebbing & Mason,

2009). Initially, the draft policy was designed to promote competitive balance, or create

equal playing strength between league members (Forrest & Simmons, 2002; Sanderson &

Siegfried, 2003). For instance, the reverse-order draft scheme helps the worst performing

teams improve team quality through assigning the best draft prospect to them. After

acquiring high quality young talent, the poor performing teams can shorten the team

quality gap with other franchises. As such, the draft system serves to improve

competitive balance in sports leagues, which is considered as a key determinant of

demand for sport (Sanderson & Siegfried, 2003). However, the implementation of the

draft system also creates an unintended consequence that teams may intentionally lose

games to acquire the top draft pick (Soebbing & Mason, 2009). For instance, a range of

studies have offered evidence that NBA teams are more likely to lose games after being

eliminated from playoff contention, signaling that some NBA franchises are deliberately

losing games for high draft picks (e.g., Price et al., 2010; Taylor & Trogdon, 2002).

1.3 AN OVERVIEW OF TANKING TACTICS AND STRATEGIES

The previous section established a clear link between the draft system and tanking.

This section further introduces three tanking tactics and two tanking strategies that teams

may use to strategically lose games.

1.3.1 TANKING TACTICS

McManus (2019) summarized three tanking tactics in sport. The first is to trade

away key players in exchange for promising young players or valuable draft picks that

have the potential to develop into superior players and contribute wins to teams

(McManus, 2019). In this process, teams intentionally assemble less competitive rosters

in the hopes of losing more games and improving their draft position. To ensure the

effectiveness of this tanking tactic, teams can also choose not to sign high-profile and

high-performing players in the free agent market. This practice can help teams save

valuable financial resources while also reinforcing their goal of losing more games.

Among the major professional sports leagues in North America, MLB teams often use

this tactic (Sheinin, 2018). Due to the absence of a salary floor in MLB, setting a

minimum amount of salary each team has to pay in a given season, it is easier for MLB

teams to avoid signing high priced players in the free agent market as compared to teams

in other leagues where a strict salary floor is a part of many league’s collective

bargaining agreements (Sheinin, 2018).

The second tanking tactic is resting healthy, key players. By resting top players

and assigning more playing time to young and less experienced players, tanking

franchises deliberately play a less competitive roster thereby making them more likely to

lose games (McManus, 2019). Resting healthy players does not only improve teams’

odds of losing games and winning the draft lottery, but also gives young players

opportunities to develop. These young players may have the potential to help teams gain

competitiveness in the long term (Soebbing et al., 2013).

The third tanking tactic is giving up games. For instance, Preston and Szymanski

(2003) provided an example of Barbados intentionally kicked the ball into their own goal

at the end of regulation in order to tie their match with Grenada at the Shell Caribbean

Cup. Barbados had to win the game by two goals to advance to the next round, but only

led by one goal when there were three minutes left on the game clock. Therefore,

Barbados had to force the match into overtime so that they would have more time to

score more goals.

1.3.2 TANKING STRATEGIES

The previous section described three tanking tactics that teams could employ to

purposely lose games. This section centers on discussing how teams may use these tactics

to successfully deploy tanking strategies. McManus (2019) noted two tanking strategies

that were widely employed by major professional sports teams. The first strategy is to

intentionally lose games late in the regular season when teams are already eliminated

from playing in the postseason (Price et al., 2010; Taylor & Trogdon, 2002). After being

eliminated from playoff contention, the benefit of winning additional games can be

minimal (Soebbing et al., 2013). However, if teams lose more games and lower their

standing in the league, they will have better opportunities to obtain a high draft pick

under the reverse-order draft system (Soebbing et al., 2013). Thus, it is not surprising that

teams may choose to tank once they lose the hope to compete in the postseason.

This tanking strategy is often accomplished by resting key players (McManus,

2019). Teams resting key players utilize less talented rosters on the court, which reduces

the odds of winning games. Clearly, teams can also deliberately lose games by asking

players not to exert the best effort in competition. However, given the competitive nature

of sporting contests, players generally exert full effort to win games (McManus, 2019).

For this reason, resting key players theoretically becomes the more effective tactic to

deliberately lose games. Without key players on the field or court, the disparity in team

quality between tanking teams and their opponents makes it more difficult for tanking

clubs to win, even though players expend their best effort in competitions.

Unlike the first tanking strategy that is often deployed toward the end of the

season, the second tanking strategy is more extreme whereby teams attempt to

deliberately lose games for multiple seasons (McManus, 2019). Using the second tanking

strategy often reflects management’s belief that rebuilding their franchise through the

draft system is the best path to win championships (McManus, 2019). Rather than

perpetually staying in the middle or bottom of a league’s standings, these teams

purposely endure losing seasons in the hopes of maximizing their odds of acquiring the

best young talent through the draft system (Motomura et al., 2016). After accumulating

sufficient talent over time, teams are well positioned to contend for championships.

This second strategy is often executed with tanking tactics such as trading away

key players or avoiding signing high-profile free agents (Vamplew, 2018). By removing

key players and not signing quality free agents, teams intentionally construct less

competitive rosters in order to access higher draft position. Not only does the reduction of

talent level improve draft position, but it also aids in acquiring additional draft picks and

young talent from other franchises through player trades (McManus, 2019). High draft

picks and young talent are invaluable assets because they may turn into high quality

players and contribute numerous wins to the franchise in the long term (McManus, 2019).

In practice, MLB and NBA teams had the history of using this tanking strategy to

rebuild franchises (Lenten, 2016). In MLB, the Houston Astros tanked from the 2011 to

2013 seasons. During this period of time, the Astros lost an average of 108 of 162 games

per season and, as a result, received three consecutive first overall draft picks, along with

other draft picks and young talent they acquired from team trades (Miller, 2018). In the

meantime, the Astros significantly cut player payroll by not signing any high-profile

player in the free agent market. After years of piling high draft picks and developing

young talent through tanking, the Astros won the World Series in 2017. In the NBA, the

Philadelphia Sixers, under the leadership of General Manager Sam Hinkie, employed this

tanking strategy from the 2013 to 2016 seasons (Steinberg, 2018). After trading away

quality players, the Sixers only won 47 of 246 games they played over these three

seasons. The poor win-loss record, however, allowed the Sixers to obtain two second and

one first draft pick in the 2014-2016 NBA drafts. The Sixers also purposely accumulated

high draft picks from trades they made with other NBA franchises. With these draft picks,

the Sixers selected promising young players such as Joel Embiid and Ben Simmons, who

helped improve the quality of the Sixers and led the team to the Eastern Conference

Semifinals in 2019.

1.4 TANKING IN THE NBA

While tanking is not rare in the sport industry, perhaps the NBA has dealt with

tanking more often than any other North American professional sports leagues (Soebbing

& Mason, 2009). For this reason, this dissertation uses the NBA as the research context to

examine the relationship between consumer perceptions of tanking and demand for sport.

Here, a brief explanation of why tanking occurs more frequently in the NBA than other

leagues and some notable NBA tanking incidents are discussed.

With basketball games featuring only five players per team on the court, an

individual basketball player can significantly influence game results (Sanderson &

Siegfried, 2003; Soebbing & Mason, 2009). As such, NBA teams may have greater

incentives to acquire high-quality talent via the draft than teams in other leagues as the

presence of a superior basketball player may drastically improve team quality (Soebbing

& Mason, 2009).

The NBA teams had a long history of employing tanking tactics and strategies

discussed before to deliberately lose games. For instance, the Houston Rockets were

alleged to have tanked late in the 1984 regular season in an attempt to acquire Hakeem

Olajuwon in the 1984 NBA draft (Webb, 2012). Desiring to acquire Olajuwon through

the draft, the Houston Rockets lost 14 of their last 17 regular-season games during the

1983-84 season and successfully improved their draft position. Based on their season

record, the Rockets finished the first half of the season winning nearly 50% of games. In

the second half of the season, the team gave more playing time to its bench players and

began losing games more frequently (Vaccaro, 2019). The Rockets eventually finished

the season with a 29-53 record and received the first draft pick to Hakeem Olajuwon after

winning the coin flip with the Portland Trail Blazers. The coin flip was the procedure

used to determine the first overall pick at the time. It was between the last place team in

each conference (Soebbing & Mason, 2009).

A decade later, the San Antonio Spurs allegedly tanked during the 1996-1997

season in order to draft Tim Duncan. They actively rested key players and completed the

season with a 20-62 record. While this was not the worst record in the 1996-1997 NBA

season, the Spurs fortunately won the draft lottery and used the first draft pick to select

Tim Duncan in the 1997 draft (Dorsey, 2012).

Generally, an NBA team’s decision to tank is not disclosed to the public.

However, in certain cases, team executives later reveal their tanking philosophy to the

media. For example, former Toronto Raptors general manager, Bryan Colangelo, publicly

admitted that the Toronto Raptors attempted to tank in part of the 2011-12 season (Amick,

2014). To improve their draft position, Colangelo noted that the Raptors sought to tank

by resting key players and assigning more playing time to young players. Mark Cuban,

the owner of the Dallas Mavericks, also openly admitted that the Mavericks tanked

during the 2016-2017 and 2017-2018 seasons (Bieler, 2018). To compete for better draft

picks, the team purposely fielded a less competitive roster to increase their odds of losing

games.

1.5 NBA DRAFT REFORMS

The NBA has well noted the tanking problem over the past decades. To address

the issue, the league has conducted a number of draft reforms since its first use of the

draft system in 1947 (NBA, 2017). However, none of these reforms seems to be effective

as tanking still frequently takes place in the NBA. This section highlights the draft

systems that the NBA has employed and explains why tanking continues to be a problem

in the league after multiple reform attempts.

1.5.1 REVERSE ORDER DRAFT

The first NBA draft system was established in 1947 when draft picks were

assigned to teams based on their win-loss records in reverse order. Teams also were

allowed to forfeit their first-round draft pick and select a territorial player who came from

the geographic area where the franchise was located. The use of territorial draft picks

aimed to attract more local fans as the presence of ‘local heroes’ could significantly

improve attendance (Brandes, Franck, & Nüesch, 2008; Yamamura, 2011). The territorial

draft system was replaced by a full reverse-order draft format in 1966. Under the reverse-

order draft system, the first draft pick was determined by a coin flip between two worst

performing teams in each of the West and East Conferences. The winner of the coin flip

would receive the first draft pick and the loser would receive the second. The rest of the

draft picks were assigned to teams based on their regular season win-loss records in

reverse order.

1.5.2 EQUAL-CHANCE DRAFT LOTTERY

The reverse-order draft system was used until 1985 when the NBA replaced it

with the equal-chance draft lottery system. The new draft scheme gave each non-playoff

team an equal chance of receiving the first draft pick. This draft reform arose from the

concerns that teams were purposely losing games in order to improve their draft position

under the reverse-order draft system (Gerchak, Mausser, & Magazine, 1995). These

concerns reached a new height during the 1983-1984 season when the Houston Rockets

deliberately lost their 14 of the last 17 regular-season games in order to draft Hakeem

Olajuwon, a future member of the Basketball Hall of Fame (Soebbing & Mason, 2009).

1.5.3 WEIGHTED DRAFT LOTTERY

The equal-change draft lottery format ultimately led to fears that it did not provide

sufficient help for poor-performing teams to regain competitiveness, which was the

original purpose of implementing draft systems in sports leagues (Bondy, 2007).

Therefore, the NBA reformed the draft system again in 1990 (Soebbing et al., 2013).

Specifically, the league launched a weighted lottery system that gave the worst

performing team the best chance of acquiring the first draft pick, which is close to

16.67%. The remaining non-playoff teams had decreasing odds to obtain the first pick

based on the inverse of their final regular season standings.

In 1993, the Orlando Magic won the draft lottery with only a 1.52% chance.

Witnessing the unfolding of such an improbable outcome, the NBA decided to revise the

draft lottery by increasing the probability for the worst-performing team to receive the

first draft pick from 16.67% to 25% in 1994 (Figure 1.1). The new weighted draft system

remained intact until 2017 when the public worried that it incentivized numerous teams

to tank for the entire season. Therefore, the NBA board of governors again voted to

reform the weighed lottery draft by decreasing the odds for the top seed to receive the

first draft pick in 2017 (Lowe, 2017). Under the new lottery draft system that was

officially instituted in 2019, the bottom three teams shared the same 14% chance to win

the draft lottery. The odds for the remaining non-playoff teams to win the lottery

gradually decreased based on win-loss records (Figure 1.1).

1.6 PURPOSE OF STUDY

The purpose of this dissertation is to analyze the link between consumer

perceptions of tanking and fan demand for NBA games. To explore this relationship, I

analyze social media posts from Twitter to quantify consumer perceptions of tanking.

Specifically, the two variables, namely, the volume and sentiment of tanking tweets, are

constructed to represent perceptions of tanking. After the quantification of consumer

perceptions of tanking, econometric models are employed to systematically explore how

perceived tanking behavior can affect NBA attendance.

Figure 1.1 NBA Weighted Lottery Percentages

CHAPTER 2

LITERATURE REVIEW

The following chapter discusses a wide range of literature that relates to the main

themes of this dissertation. The first section highlights the differences between tanking

and other similar concepts, including match fixing, rebuilding, and shirking. The second

section reviews the studies showing empirical evidence of tanking. The next section

discusses tournament theory, which has been employed to explain the rise of tanking in

sport. The fourth section notes the key determinants of demand for sport in order to set

the theoretical foundation for this dissertation. The last section considers social media

research in the sport industry.

2.1 TANKING AND OTHER SIMILAR CONCEPTS

In discussing tanking, concepts like match fixing, rebuilding, and shirking are

sometimes used interchangeably (Lenten et al., 2018; McManus, 2018; Vamplew, 2018).

While these concepts share some similarities with tanking, a clear distinction has to be

drawn here in order to better understand the nature of tanking in sport.

2.1.1 TANKING AND MATCH FIXING

Match fixing takes place when contestants attempt to fix game outcomes by

reducing their effort in competitions (Blair, 2018). Preston and Szymanski (2003)

summarized three types of match fixing, claiming tanking should be treated as a form of

match fixing. In their categorization of match fixing, the first type often involves bribery

that one team is willing to provide benefits to opponents or referees in order to obtain a

particular game outcome (Blair, 2018). For instance, in sports leagues using the

promotion and relegation system, teams on the verge of being relegated may pay

opponents or referees to ask for a win. In this case, match fixing occurs as marginal teams

bribe opponents or referees in order to save themselves from being relegated to lower

divisions.

The second type of match fixing relates to gambling activities where players or

officials gain financial benefits from a specific match result (Preston & Szymanski, 2003).

For instance, the 1919 World Series between the Chicago White Sox and Cincinnati Reds

was alleged to be controlled by gamblers. While the White Sox were considered as the

overwhelming favorite, they lost the series. The belief was some White Sox players

received money from gamblers to throw games. In this case, gamblers fixed game results

in order to gain financial returns from betting markets (Anderson, 2001).

The third type of match fixing occurs when manipulating game outcomes helps

contestants obtain competitive benefits in the long term (Preston & Szymanski, 2003).

Unlike the other two match-fixing practices, the third type may not be strictly prohibited

by sports leagues, although it is certainly against the spirit of sport (McManus, 2019).

Tanking fits this type of match fixing as it aims to produce a particular result that can

help teams gain a favorable position in the draft (Preston & Szymanski, 2003). For

instance, MLB teams have attempted to deliberately lose games in order to win high draft

picks (Sheinin, 2018). With high-quality draft prospects, tanking teams expect to regain

competitiveness or win championships in the near future. The Chicago Cubs tanked to

obtain talent and subsequently won the World Series in 2016 (Sheinin, 2018). Before

securing the title, the Cubs lost an average of 96 games per season from 2011 to 2013 as

they employed low-priced marginal talent.

2.1.2 TANKING AND TEAM REBUILDING

The word ‘tanking’ is often associated with the phrase ‘team rebuilding’ in the

press (McManus, 2019). In general, sports leagues, teams, and media believe that tanking

is an effective way to rebuild sports franchises (Motomura et al., 2016; Mullin, 2018;

Sheinin, 2018). Tanking teams intend to restore competitiveness through high draft picks

that they may acquire under the reverse-order draft system by deliberately losing games.

However, not all tanking practices are favored by leagues to rebuild sports franchises. For

instance, NBA commissioner Adam Silver noted that there was a clear distinction

between effort teams might put forth to rebuild franchises, such as trading players, and

actions teams might take on the court to lose games, such as deliberately missing shots or

giving up on defense (Aldridge, 2018). The former can be accepted while the latter is

clearly against league rules. According to Silver’s tanking comments, it appears the

tanking tactic of trading away quality players is regarded as an acceptable action under

league rules, whereas other tactics such as resting key players and throwing games seem

to be inappropriate (Amick, 2018).

Interestingly, Los Angeles Dodgers President Stan Kasten made similar

comments regarding the distinction between tanking and team rebuilding (Aldridge,

2018). He considered strategic tanking as a process of cutting payrolls and acquiring

bright young players who, through years of training, could be developed into impactful

athletes. In the rebuilding process, teams are willing to endure consecutive years of losses

in order to regain competitiveness by acquiring talent and draft picks. However, he

viewed purposely losing games on the field as improper. Clearly, Kasten treated the

tanking practice of trading superior players or not signing high-profile free agents as a

reasonable way to rebuild sports franchises, but deliberately giving up games on the field

as an unacceptable practice.

To further highlight the link between tanking and team rebuilding, the NBA

fining Dallas Mavericks owner Mark Cuban during the 2017-18 season for his tanking

comments provides a specific example to demonstrate the league’s stance on these two

notions (Bieler, 2018). Specifically, Mark Cuban publicly admitted the Dallas Mavericks

tanked during the 2016-2017 and 2017-2018 seasons. He also suggested losing was the

best and quickest option for the franchise to regain competitiveness in the NBA. Shortly

after the comments, the NBA fined Mark Cuban $600,000, stating that Cuban’s

comments were detrimental to the NBA. Yet, in a memo sent to all NBA teams following

Cuban’s tanking comments, the NBA noted it had no basis to conclude that Dallas

Mavericks exerted less than the best effort to compete on the court (Aldridge, 2018).

Mark Cuban also assured the league that he did not ask players or coaches to throw

games to tank. In fact, Mark Cuban was fined because his comments might jeopardize the

perceived integrity of competitions that the league attempted to protect by all means.

Given the fact that the integrity of sport is the cornerstone of any sport, actions that

undermine the actual and perceived integrity of the game should be penalized (McLaren,

2011). Nevertheless, this incident illustrates that while not exerting maximum effort on

the court is clearly against NBA rules, the league seems to consider strategic tanking as

an acceptable practice, since the Dallas Mavericks trading away experienced players to

tank did not receive penalties from the NBA. The rationale of fining Mark Cuban for his

tanking comments is consistent with what Adam Silver defined tanking that roster

changes to rebuild franchises are inherently distinct from throwing games.

2.1.3 TANKING AND SHIRKING

Although tanking sometimes serves as an alternative word to shirking, these two

notions are distinct from each other in the context of sport (Vamplew, 2018). In defining

shirking, Vamplew (2018) suggested that shirking took place when contestants

underperformed by not exerting their best effort in competition. Although tanking also

involves contestants not expending maximum effort in games, shirking contestants do not

necessarily aim to lose games (Vamplew, 2018). For instance, prior research sought to

explain why players tended to shirk after signing long-term guaranteed contracts

(Scroggins, 1993). This shirking behavior can be explained through the principle-agent

theory in which the principles (teams) may not recognize the nuance between shirking

and below-average performance, and thus allow the agents (players) to act

opportunistically (Krautmann, 1990). In other words, long-term contracts disincentivize

contestants to play with their best effort because stakeholders may not explicitly identify

shirking behavior in practice (Berri & Krautmann, 2006). Thus, shirking players

underperform because they do not have sufficient incentives to work hard, not that they

intend to deliberately lose games in order to gain long-term competitive advantages like

tanking teams do (Vamplew, 2018).

2.2 EVIDENCE OF TANKING IN SPORT

Tanking is a widespread phenomenon in sport (Kendall & Lenten, 2017). An

abundance of studies developed evidence of tanking (e.g., Price et al., 2010; Taylor &

Trogdon, 2002). In North American sports leagues, tanking is a particular concern in the

NBA (Soebbing & Mason, 2009). As NBA teams rarely reveal their tanking plan to the

general public, scholars have sought to provide empirical evidence that teams deliberately

lose games in the NBA. This section of the literature review will firstly review prior

research seeking to prove the existence of tanking in the NBA. The rest of the section

will cover empirical studies revealing the evidence of tanking in other sports leagues.

2.2.1 EVIDENCE OF TANKING IN THE NBA

The seminal study conducted by Taylor and Trogdon (2002) tested whether NBA

teams were more likely to lose games after being eliminated from postseason contention.

Looking at the 1983-1984, 1984-1985, and 1989-1990 NBA season data, Taylor and

Trogdon (2002) found evidence that teams eliminated from playoff contention tended to

lose more game late in the regular season under the reverse-order and weighted lottery

draft systems. However, during the 1984-1985 season when the league adopted the equal-

chance draft lottery, there was no strong evidence of tanking. That is, teams ceased to

compete for draft picks by intentionally losing games when each non-playoff team shared

the same odds of winning the draft lottery.

Price et al. (2010) explored the same research question as Taylor and Trogdon

(2002) while employing a longer period of time. Using NBA data spanning from 1978 to

2008, they drew similar conclusions as the Taylor and Trogdon’s (2002) study, except

under the reverse-order draft format. In contrast to the Taylor and Trogdon’s (2002)

finding that NBA teams tanked under the reverse-order draft system, Price et al. (2010)

did not find strong evidence of tanking under this draft mechanism. They explained that

this inconsistent result might emerge from different econometric models employed in

these two studies. Nevertheless, Price et al. (2010) consolidated the evidence of tanking

in the NBA under the weighted lottery draft system but questioned the previous finding

of tanking under the reverse-order draft format.

Soebbing and Humphreys (2013) followed the previous two studies and

developed evidence of tanking by using NBA betting odds data from the 2004-2005 to

2008-2009 seasons. The betting market is chosen as the research context because it

generally takes all available data, including any tanking information, into consideration

when making predictions. Thus, if gamblers believed that tanking takes place in the NBA,

they will adjust the betting odds accordingly. Following this premise, Soebbing and

Humphreys (2013) tested whether bookmarkers would set different point spreads for

games involving eliminated teams. They found strong evidence that there was a

systematic change in point spreads, suggesting that the betting markets believed that

tanking existed in the NBA.

To further understand tanking in the NBA, Soebbing et al. (2013) explored

whether tanking took place more often in NBA conference games, as compared to

nonconference games. Considering that teams move down their standings faster when

they lose to teams from the same conference than teams from the other conference, teams

may have greater incentives to tank when facing conference opponents. After analyzing

NBA data from the 1983-1984, 1984-1985, 1989-1990, and 1993-1994 seasons,

Soebbing et al. (2013) concluded that NBA teams were more likely to tank in conference

games than nonconference games under the weighted draft lottery adopted since 1994.

For other draft systems, including the reverse-order draft, equal-chance draft, and the

early weighed draft lottery imposed in 1990, there was no strong evidence that teams

tanked more often in conference games than nonconference games.

Prior studies provide evidence of tanking in the NBA, particularly toward the end

of the regular season (Soebbing et al. 2013). While the NBA attempted to eliminate

tanking by tweaking draft formats, incentives to tank remained as a big concern for the

league (Soebbing et al. 2013; Taylor & Trogdon, 2002). Recently, the NBA passed

another draft reform that would be implemented in 2019 (Ward-Henninger, 2017). The

new draft policy aimed to further diminish teams’ incentives to tank as the league has

observed certain extreme tanking cases (Greenstein, 2019). For example, unlike other

NBA teams that have been accused of tanking late in the regular season, the Philadelphia

Sixers were alleged to have tanked for several seasons from 2013 to 2016 (Steinberg,

2018). The prevalence of tanking today raises concerns that tanking may undermine fans’

interest in the NBA. In light of such concerns and limited research on the topic, there is a

need for further investigation on how tanking can affect consumer demand for NBA

games.

2.2.2 EVIDENCE OF TANKING IN OTHER SPORTS LEAGUES

Outside the NBA, there also exists the evidence of tanking. Balsdon et al. (2007)

noted that tanking might happen in college basketball where regular-season conference

champions were less likely to win conference tournament championships. They

suggested that such underperformance was plausibly motivated by two factors. First,

teams were saving themselves for the NCAA tournament. For regular-season conference

champions with guaranteed bids to the NCAA tournament, conference tournaments

become meaningless. In fact, the earlier teams finished in conference tournaments, the

more rest time they could receive. The second factor was financial gains from

participating in March Madness. As the winner of conference tournaments often receives

an automatic bid to March Madness, regular-season conference champions losing

conference tournaments to other conference members can allow more schools from the

same conference as conference champions to participate in the NCAA tournament. With

the NCAA distributing tournament revenue to conferences partially based on the number

of schools selected into the tournament, and considering that some conferences equally

distributed tournament revenue to their members, more participants in March Madness

implies increased revenue for each member of a conference. After investigating these two

possible motivations of tanking, Balsdon et al. (2007) concluded that the financial gain

might be the primary motivation for regular-season champions to underperform in

conference tournaments.

Kendall and Lenten (2017) pointed out that tanking occurred even in mega

sporting events like the Olympic Games. Specifically, they showed that women’s doubles

badminton teams deliberately lost games at the 2012 Olympic Games in order to gain

competitive benefits in the wider context of tournament play. Due to the timing of games,

when two tanking teams faced each other, they had known whom they would compete

against in the next rounds of the tournament. After knowing that the losing team would

face weaker opponents, both teams actively sought ways to lose games. Noticing two

teams not competing with their best winning efforts, the Badminton World Federation

quickly disqualified both teams, claiming that their behavior was against the spirit of

sports competition (Kendall & Lenten, 2017).

Research indicates that tanking is pervasive in sport. Not only does the NBA

suffer from tanking but also other sports leagues and tournaments around the world.

Although it is clear that tanking undermines the quality of competitions, the damage that

tanking brings to sports leagues may be worse than what the existing research suggests

(Soebbing & Mason, 2009). That is, fans who perceive the existence of tanking may

question sports leagues’ credibility, and thus choose not to attend games. Given the

pervasiveness of tanking in the sport industry, it is crucial to thoroughly investigate the

effects of tanking on demand for sport. With the findings from this dissertation, sports

leagues can better evaluate the consequences of tanking, and thus take actions to protect

sports leagues’ credibility as needed.

2.3 TOURNAMENT THEORY AND TANKING

Preston and Szymanski (2003) suggested that tanking was largely caused by ill-

designed tournament rules. For most tanking cases in the NBA, these rules are the draft

systems that the NBA has deployed since 1947. However, not all NBA draft systems give

teams incentives to tank. For instance, prior research offered consistent evidence that the

equal-chance draft system did not cause tanking in the NBA. In fact, the decision for

teams to tank depends on the marginal benefit of the draft picks (Price et al., 2010; Taylor

& Trogdon, 2002). When the value of the first draft pick is exceedingly higher than the

rest of the draft picks, teams will likely exert stronger losing efforts to tank. To

systematically explain why sports franchises have incentives to lose under certain NBA

draft formats, tournament theory is used (Price et al., 2010; Taylor & Trogdon, 2002).

2.3.1 TOURNAMENT THEORY AND DRAFT SYSTEMS

Tournament theory was originally developed to describe the rank-based

compensation structures in corporations (Lazear & Rosen, 1981). For instance, this

theory helps explain why CEOs often receive disproportionally higher salaries relative to

managers. In a rank-based reward system, wages are determined by the relative rank

instead of the absolute value of outputs employees produce. For instance, CEOs’ salary is

not calculated based on their absolute value of outputs. Rather, their rank in companies

determines salary. Tournament theory predicts that, when employees’ actual production

is difficult to monitor, the rank-based payment system is superior to the piece rate

payment scheme that compensates employees based on their outputs (Lazear & Rosen,

1981). For instance, it is difficult to quantify the absolute value of outputs that CEOs

create for companies in practice. Thus, paying CEOs based their ranks rather than their

absolute value of outputs is a rational compensation scheme. Not only does the rank-

based compensation system save the cost to monitor employees’ output, but it also

ensures that employees will exert sufficient effort in the workplace (Lazear & Rosen,

1981). Specifically, tournament theory suggests that employees’ effort level is

determined by the prize gap in a rank-based reward system. Hence, setting the proper

level of the prize gap between the winner and loser of a tournament can elicit ideal effort

levels from participants. However, when organizers of a tournament set an overly high

prize gap, such a reward structure may create a substantial financial burden for organizers.

On the contrary, if the prize gap is too small, contestants may not put forth enough effort

in competition.

Some draft systems in professional sports leagues resemble a rank-based

compensation system, with draft picks as rewards. For instance, the reverse-order draft

format allocates the rewards, draft picks, to teams based on their regular-season

performance in reverse order. Tournament theory predicts that teams will exert

significant losing effort to compete for draft picks when the marginal benefit of draft

picks exhibits a nonlinear structure (Price et al., 2010). Thus, the key to determining

whether draft systems motivate teams to tank is centered on the value of draft picks.

Soebbing and Humphreys (2013) noted the value of draft picks came from two main

areas. The first area is additional team wins and related revenue generated by draft

prospects. High draft picks, particularly the first draft pick, are expected to contribute

many wins to teams, and thus help the team generate more revenue through improved

team performance (Price et al., 2010). The second area lies in the advertising effect of top

draft prospects. Price et al. (2010) estimated the presence of the first draft pick itself

could bring $2.8 million in revenue to teams that selected the prospect. Given these

estimations of the value of NBA draft picks, Price et al. (2010) concluded the reverse-

order draft format did not create a nonlinear structure of the value of draft picks. Thus,

NBA teams will not have incentives to tank under the reverse-order draft scheme based

on tournament theory. Meanwhile, Price et al. (2010) illustrated the nonlinear structure of

the probabilities under weighted draft lottery which created incentives for teams to tank..

2.3.2 APPLICATION OF TOURNAMENT THEORY IN SPORT

Tournament theory is commonly used to study the effectiveness of policies in

sports leasgues (Connelly, Tihanyi, Crook, & Gangloff, 2014). In particular, some

scholars employ tournament theory to test whether the design of tournaments elicits the

optimal level of effort from participants. Outside major professional sports leagues,

McClure and Spector (1997) applied tournament theory to examine team behavior in the

NCAA basketball tournament. As tournament theory suggests that actors’ level of effort

solely depends on the prize spread, they speculated that contestants in March Madness

would expend different levels of effort in response to various prize structures that

conferences adopted to distribute tournament revenue. In particular, certain conferences

largely allocate tournament revenue to schools which take part in the tournament, while

other conferences tend to evenly distribute tournament revenue to all conference

members. Although schools had various tournament reward structures, McClure and

Spector (1997) did not find strong empirical evidence that schools subject to disparate

revenue-sharing rules performed differently in the tournament. Such an indifferent

response demonstrated that conference revenue sharing policies did not affect contestants’

effort in the tournament. McClure and Spector (1997) further explained that while

schools faced different reward structures, the real contestants, student athletes, would not

receive any extra compensation, regardless of their tournament performance. Hence,

conference revenue sharing policies did not create different incentives for players to

compete in March Madness.

Von Allmen (2001) employed tournament theory to explain the rationale behind

the National Association for Stock Car Auto Racing’s (NASCAR) reward scheme.

Compared to the PGA Tour, NASCAR adopted a relatively constant reward structure

where the reward for the winner of a race was only slightly higher than the prizes

received by other contestants. As tournament theory predicts that a contestant’s level of

effort depends on the prize gap, there were some concerns that NASCAR’s constant

reward structure might not elicit considerable effort from drivers. To address these

concerns, Von Allmen (2001) pointed out that the implementation of a linear reward

system in NASCAR was primarily aimed to restrain reckless driving behavior. When cars

crash during contests, car sponsors will miss advertising opportunities. Sponsorship as

the main revenue source for many NASCAR teams may be withdrawn if sponsored teams

cannot provide sufficient exposure to sponsors’ brands. To maintain the visibility of

advertisements, implementing a relatively linear prize structure can potentially reduce the

number of car accidents on the track, and thus help teams and leagues retain sponsorship

revenue. Thus, the relatively constant reward structure in NASCAR should be considered

as an efficient policy as it ensures the sustainable development of the league.

While the theoretical explanations of tanking are extensively studied, little

research sheds light on the connection between tanking and demand for attendance. To

explore the impact of tanking on attendance in this dissertation, a thorough literature

review on the economic theory of demand for sport is needed in order to establish a solid

theoretical foundation for this research (Borland & MacDonald, 2003; Villar & Guerrero,

2009).

2.4 DEMAND FOR ATTENDANCE

The economic theory of demand for sport has been widely applied to study

motives for fans attending sporting events (Borland & MacDonald, 2003; Villar &

Guerrero, 2009). The theory is an extension of the consumer demand model that

describes the choices of buying goods and services under constrained resources (Borland

& MacDonald, 2003). Most demand studies have employed attendance at sporting events

to measure consumer demand for sport (Borland & MacDonald, 2003). A growing line of

research uses televised viewership as a proxy for consumer demand for sport (e.g.,

Tainsky, 2010; Feddersen & Rott, 2011). The rise of studies using viewership to explore

demand for sport can be ascribed to the availability of broadcasting data (Feddersen &

Rott, 2011; Tainsky & McEvoy 2012; Tainsky, 2010; Salaga & Tainsky, 2015).

Compared with attendance data that only reflects ticket sales in the home market,

viewership data captures demand for sport in the home, away and neutral markets,

providing researchers with a more disaggregated dataset to examine (Tainsky, 2010).

While research using television ratings to study demand for sport has gained

momentum, this dissertation will employ attendance to examine the effect of consumer

perceptions of tanking on consumer interest in the NBA, considering the availability of

historical NBA attendance data. In reviewing consumer demand for attendance, Borland

and MacDonald (2003) summarized five major determinants of demand for attendance,

including consumer preferences, economic factors, quality of viewing, supply capacity,

and the quality of sporting contests. Given the importance of understanding the demand

model in this dissertation, the current section follows Borland and MacDonald’s

categorizations to review five major variable categories that will significantly affect fans’

decisions to attend sporting contests.

2.4.1 CONSUMER PREFERENCES

Consumer preferences can take several forms (Borland & MacDonald, 2003). For

instance, some fans develop a habit of attending games over the years. Such loyal fans

are likely to support home teams by attending games, regardless of the performance of

their teams. Lee (2006) pointed out that, in contrast to young sports leagues, consumers

from established sports leagues were less concerned with the quality of competitions. In

fact, historical sports franchises are often supported by loyal fans, some of whom

regularly attend games, even though teams they support struggle to win games (Coates &

Humphreys, 2005).

Empirical analyses of consumer demand for sport often employed the age of the

franchise or one-year lagged attendance as a proxy of fan loyalty or habit (Ahn & Lee,

2007; Winfree, McCluskey, Mittelhammer, & Fort, 2004). While a number of studies

provided consistent evidence that the age of the sports franchise and fan loyalty could

positively affect game attendance, the calculation of the age of the team requires special

attention (Ahn & Lee, 2007; Borland & Lye, 1992; Coates & Harrison, 2005).

Specifically, team relocation can complicate the calculation of the age of the sports

franchise. For instance, in the National Football League (NFL), the Oakland Raiders

moved to Los Angeles in 1982, but returned to Oakland in 1994. Although the Raiders

continually used the same nicknames over the years, scholars often considered them as

three distinct teams. Hence, in determining the age of the team, the 1994-1995 season

was often deemed as the inaugural season for the returning Oakland Raiders (Coates &

Humphreys, 2002).

2.4.2 ECONOMIC FACTORS

A wide range of economic factors are proposed as determinants of demand for

sport (Borland & MacDonald, 2003). First, the law of demand predicts that ticket prices

are negatively correlated with attendance. Yet, empirical research provided mixed results

in evaluating the impact of ticket prices on attendance. For instance, Rivers and

DeSchriver (2002) showed there was a positive relationship between season average

ticket prices and annual MLB attendance, while other research suggested fans bought

fewer tickets when ticket prices increased (Fort, 2006). To address why studies reached

contrary conclusions regarding the relationship between ticket prices and demand for

attendance, Krautmann and Berri (2007) noted that ticket prices might not be the actual

cost of attending a sporting event. Other related costs such as food and travel should be

added to the total cost of attending a game. Hence, a few studies begun to utilize the Fan

Cost Index (FCI) that contained a more complete description of the costs of attending

sporting events in analyzing the relationship between costs and demand (Coates &

Humphreys, 2007; Gitter & Rhoads, 2010; Krautmann & Berri, 2007). Yet, studies

employing the FCI as a measure of the cost of attending sporting events did not reach

consensus (Krautmann & Berri, 2007). For instance, Coates and Humphreys (2007)

concluded the FCI was negatively related to attendance in the NBA, but insignificant in

the NFL and MLB.

Second, market size can affect demand for sport. Big market teams with more

population face higher demand for sporting events than small market teams (Noll, 1974).

Less clear is the link between income levels and demand for sport. The consumer demand

model predicts that higher income consumers are willing to purchase more normal goods

and less inferior goods. In the context of sport, if sporting events are regarded as normal

goods, fans from affluent areas will have higher demand for sport than fans from poor

regions. However, studies examining the link between income levels and demand for

sport produced mixed results. For instance, Lee (2004) noticed that baseball was

considered as a normal good in Japan, Korea, and the USA but Fort and Rosenman (1999)

pointed out that baseball was only viewed as a normal good in the National League of

MLB, not in the American League. Noll (1974) also posited that baseball was an inferior

good as his empirical test showed a negative relationship between income levels and

MLB game attendance.

The presence of entertainment alternatives is another factor that can impact

demand for sport (Borland & MacDonald, 2003). With the availability of more

entertainment activities, sports teams will face fiercer competition in drawing fans to

games. Such competition may arise from multiple sources (Borland & MacDonald, 2003).

First, sports teams compete with clubs from other leagues that play the same sport. For

example, scholars uncovered that minor league and junior league hockey teams served as

primary cross-league substitutes for National Hockey League (NHL) fans during the

NHL lockout season in 2004 (Winfree & Fort, 2008). Second, sports teams from cross-

sports leagues also compete with NHL teams for consumers. Rascher, Brown, Nagel, and

McEvoy (2009) estimated that the NHL lockout in 2004 increased NBA attendance by

2%. Third, there is strong evidence that sports franchises compete with teams from the

same league and geographic area for fans. For instance, incumbent teams will experience

a significant increase or decrease in attendance when other sports clubs from the same

league enter or exit the territory, either through expansion or relocation (Winfree, 2009).

Lastly, other entertainment activities, such as theater or cinema, are often deemed as

substitutes for sporting events. While there is little academic research examining fan

substitution to other forms of entertainment activities, anecdotal evidence showed that the

MLB strike in 1994 considerably improved movie sales in that year (Winfree, 2009).

2.4.3 QUALITY OF VIEWING

The quality of viewing is a particular concern for many sports spectators. A range

of factors affect the quality of viewing in a sporting event. First, the age of the facility

can affect consumer experience at sporting events. A newly constructed stadium typically

attracts a fair amount of spectators in the beginning years of operation. However, such a

“novelty effect” can quickly vanish as attendance drops in the following years (Clapp &

Hakes, 2005; Coates & Humphreys, 2005; McEvoy, Nagel, DeSchriver, & Brown, 2005;

Leadley & Zygmont, 2005). It also appears that the attendance can spike in the last years

of operation of a stadium due to the “nostalgia effect” (Coates & Humphreys, 2007;

McEvoy et al., 2005).

Game day weather conditions also play a significant role in determining demand

for sport. Especially for outdoor sports such as baseball, soccer, and cricket, adverse

weather may discourage fans from attending contests because of low viewing quality.

Even for indoor sports, poor weather conditions can create an unpleasant customer

experience due to increased travel costs. Empirical studies of the impact of weather on

attendance found consistent evidence that rainy weather on game days would

significantly reduce attendance (Hynds & Smith, 1994).

Game time is critical to demand for sport (Borland & MacDonald, 2003). In

general, sporting contests scheduled on weekends or holidays will have larger crowds

than games scheduled on other days (Hansen & Gauthier, 1989; Hill, Madura & Zuber,

1982; McDonald & Rascher, 2000). Prior studies suggested that consumers had more

time to participate in leisure activities during weekends and holidays. Furthermore,

research showed that afternoon games would have lower attendance than night games

(Hansen & Gauthier, 1989). While afternoon games are uncommon in today’s sport

industry, before World War II, the majority of baseball games took place in the daytime

(Quinn, 2009). To meet the demand for baseball games from the working-class during

wartime, baseball games were then moved to the late afternoon and night when the

working class had enough time to attend games (Quinn, 2009).

2.4.4 SUPPLY CAPACITY

The fourth category is the supply capacity (Borland & MacDonald, 2003). In live

sporting events, the number of tickets teams can sell is constrained at the facility’s

capacity. At sellouts, the actual demand for sporting events cannot be observed as the

quantity of available tickets is limited. Among four major sports leagues in North

America, game tickets are often sold out except for MLB (Quirk & Fort, 1997). Sell-out

events are also fairly common in European top-level soccer leagues such as the English

Premier League (EPL; Koning, 2000). Given the pervasiveness of sellouts in sport, it is

crucial to consider censored attendance when analyzing determinants of demand for

attendance. Otherwise, statistical models may produce biased estimations (Humphreys &

Johnson, 2020). To correct censored attendance, some studies treated the capacity as an

explanatory variable in statistical models (Jones, 1984). However, such an adjustment

seems to be problematic as including the capacity as an independent variable in demand

models still produces biased estimators (Cairns, Jennett, & Sloane, 1986). DeSchriver,

Rascher, and Shapiro (2016) developed interval regression to account for the censor

attendance data. Other studies chose to correct for unobserved attendance by employing

the censored normal regression model (Amemiya, 1973; Welki & Zlatoper, 1994). It is

suggested that estimators from the censored normal regression model will be less biased

(Welki & Zlatoper, 1994). For instance, Humphreys and Johnson (2020) adopted the

censored normal regression model to estimate the effect of star power on attendance,

considering numerous sellouts in the NBA.

2.4.5 QUALITY OF SPORTING CONTESTS

One of the key components of the quality of sporting contests is team strength

(Borland & MacDonald, 2003). It is suggested that sports fans prefer to watch or attend

sporting events featuring high quality teams. Empirical analysis provided consistent

evidence that attendance soars when the quality of home teams is high (Villar & Guerrero,

2009). However, the impact of the quality of visiting teams on attendance is less clear.

Some studies showed a positive relationship between the quality of away teams and

attendance, whereas others did not find a significant connection between the quality of

visiting teams and demand for sport (Buraimo & Simmons, 2008; Villar & Guerrero,

2009). To quantify team strength, a variety of metrics were employed. The most common

one is the team winning percentage or standing (Forrest & Simmons, 2002). More

recently, a study conducted by Mills, Salaga, and Tainsky (2016) employed a team rating

metric named Elo that took more information such as home court advantage and margin

of victory into consideration. Therefore, the Elo rating system may better reflect true

team strength than the team winning percentage or standing (Mills et al., 2016).

Tanking teams are often associated with poor team quality as they purposely

assemble less talented rosters to compete (Taylor & Trogdon, 2002). This action of

assembling less talented rosters is particularly true for teams seeking to tank for entire

seasons. By trading away key players and avoid signing quality free agents, tanking

teams deliberately compromise their team strength, and thus may hurt fan interest in

games (McManus, 2019). In addition, some teams decide to tank toward the end of the

regular season. This tanking strategy is to deliberately field less competitive rosters that

may further deter fans from attending sporting events.

The quality of sporting contests is also determined by the presence of star players

(Villar & Guerrero, 2009). Considering that NBA teams only field five players on the

court and athletes do not wear gear such as helmets that may hide their faces from the

spotlight, NBA star players can draw considerable attention from fans (Berri, Schmidt, &

Brook, 2004; Hausman & Leonard, 1997). Empirical analyses of star power on NBA

attendance provided consistent evidence that the presence of star players was positively

related to game attendance (Berri et al., 2004; Hausman & Leonard, 1997; Jane, 2016).

Outside the NBA, Buraimo and Simmons (2015) found the significant impact of star

players on attendance in the EPL. In individual sports such as tennis and the Ultimate

Fighting Championship (UFC), star status was also an important predictor of demand for

sport (Chmait et al., 2020; Reams & Shapiro, 2017).

Uncertainty of game outcomes makes up the other dimension of the quality of

sporting contests. The Uncertainty of Outcome Hypothesis (UOH) predicts game

uncertainty positively relates to demand for sport (Rottenberg, 1956; Neale, 1964). In

other words, fans prefer more unpredictable events than certain outcomes. This

hypothesis was recently challenged by the study conducted by Coates, Humphreys, and

Zhou (2014) who suggeste home team fans might dislike uncertain home games due to

loss aversion. Notably, loss aversion assumes that the utility lost from a team loss is

higher than the utility gained from a team win. Based on this assumption, Coates et al.

(2014) concluded unpredictable games might lead to lower attendance than predictable

matches.

To measure the uncertainty of game outcomes, there exists two metrics that are

particularly relevant to this dissertation. The first metric attempts to quantify league win

distribution. For instance, the Ratio of Actual Standard Deviation to the Idealized

Standard Deviation (RSD) has been widely employed to describe the dispersion of team

wins in sports leagues for a given season (Noll, 1988). The high RSD implies an

unbalanced league where matches lack unpredictability (Fort & Quirk, 1995). Recently,

Lee, Kim, and Kim (2018) suggested that RSD could be a biased measure of competitive

balance when the number of games played within sports leagues was large. To correct the

bias, they proposed a new competitive balance metric, the Bias-Corrected Standard

Deviation (BCSD), which better captured league-wide competitive balance than RSD.

Unlike RSD, BCSD is not subject to the number of matches teams play in a given season,

and thus may be better employed to compare competitive balance across leagues and

seasons that may host various number of games (Lee et al., 2018).

The second measure of game uncertainty measures the closeness of individual

competitions. For example, the absolute difference between the home and away team’s

winning percentage prior to a game is considered as a valid proxy for game uncertainty

(Meehan, Nelson, & Richardson, 2007). In addition, betting odds or point spreads from

gambling markets are widely applied to measure game uncertainty (Buraimo & Simmons,

2008; Coates & Humphreys, 2010; Forrest & Simmons, 2002). While betting odds or

point spreads may not always fully reflect true game uncertainty, they are considered as

better estimators of the team quality gap than other metrics, as bookmakers of betting

odds or point spreads have strong financial incentives to make as accurate predictions as

possible (Forrest, Goddard, & Simmons, 2005).

In general, tanking undermines the uncertainty of game outcomes (Soebbing &

Mason, 2009). If outcome uncertainty is a driving force of game attendance, the presence

of tanking teams in sports leagues may hurt consumer demand for sport. An abundance of

studies found evidence that tanking teams are more likely to lose games than they would

have, implying that game uncertainty is compromised when participants do not exert their

best effort in contests (Price et al., 2010; Taylor & Trogdon, 2002).

While the literature clearly indicated that tanking undermined the quality of

sporting contests by threatening both team strength and the uncertainty of game outcomes,

none of the prior research examines the link between consumer perceptions of tanking

and their connections to demand for sport (Price et al., 2010). To address the gap, this

dissertation analyzes consumer behavior in response to the existence of tanking in sports

leagues. Such an inquiry provides useful insights in understanding the change of fan

interest in sport when sports leagues’ creditability is in jeopardy.

2.4.6 DEMAND FOR NBA ATTENDANCE

While five major determinants of demand of sport were well documented in

sections 2.4.1 to 2.4.5, recent work of analyzing demand for NBA attendance seem scant.

The most popular studies within the context of the NBA were to examine the effect of

star power on attendance. With the relatively small number of players required in the

game of basketball, the NBA served as a great context to understand the importance of

star players in drawing fans (Berri & Schmidt, 2007; Hausman & Leonard, 1997;

Humphreys & Johnson, 2020; Jane, 2016). In addition, a few studies investigating the

impact of game uncertainty on demand selected the NBA as the research context (Mills et

al., 2016). For instance, Rascher and Solmes (2007) examined NBA data and found

evidence to support the UOH that fans preferred more uncertain games than predicable

ones. The long-standing NBA was also considered as a great context to study the

relationship between facility age and demand for sport. For example, both Coates and

Humphreys (2005) and Leadley and Zygmont (2005) noted the novelty effects of new

areas in attracting NBA fans.

Among the aforementioned studies of NBA attendance, a few factors were widely

employed as control variables. For instance, most NBA attendance studies used team

quality, population, income level, market competition, and game time to control for their

effects on consumer interest in the NBA (Hansen & Gauthier, 1989). The results

generally showed that these factors were significant predictors of NBA attendance. For

instance, studies employed the team winning percentage as a proxy for team quality

revealed that both home and visiting team quality were positively related to NBA teams

demand (Mongeon & Winfree, 2012). Similarly, NBA teams located in markets with a

bigger population, higher income, and fewer other professional sports teams largely

experienced higher demand for NBA games (Coates & Humphreys, 2005; Coates &

Humphreys, 2007; Mongeon & Winfree, 2012). Lastly, weekend and holiday NBA

games could expect significantly higher demand than matches in other time slots

(Humphreys & Johnson, 2020). Given the above observations, it is therefore crucial to

control for these variables when examining the link between perceptions of tanking and

demand for NBA attendance in this dissertation.

2.5 SOCIAL MEDIA AND SPORT

The emergence of social media has drastically changed the way sports fans

consume sports media products (Billings, Broussard, Xu, & Xu, 2019). In the past, the

traditional media such as newspapers and TV mainly served as a means to distribute

sports content to broad audiences (Boyle & Haynes, 2002). That is, sports fans passively

received information from these traditional media firms. The advent of social media

platforms such as Facebook and Twitter allow fans to create their own sports content

(Pegoraro, 2014). Thus, in the era of social media, sports fans do not only consume sports

products from traditional media companies, but also actively generate sports content and

distribute it to other social media users. As such, social media data has been widely

studied and used in understanding consumer opinions on products and services. This

section sheds light on research using social media data to address research questions in

the sport industry.

Filo, Lock, and Karg (2015) noted a few types of sports social media research that

is fairly relevant to the present dissertation. First, a handful of studies examined

consumer behavior on social media. For instance, Watanabe, Yan, and Soebbing (2015)

examined key factors that affected the change of Twitter followers of MLB teams.

Through analyzing Twitter follower data and performing a regression analysis, they

concluded that factors such as the content of social media messages, certain calendar

events, and postseason appearances were significant in drawing Twitter followers.

Watanabe, Yan, and Soebbing (2016) further analyzed the statistical relationship between

the total number of followers of MLB teams on Twitter and team characteristics. They

offered evidence that the aggregate popularity (Twitter followers) of all starting players

had a significantly positive impact on the change of Twitter followers of a team in the

short term, implying that MLB players maintaining active Twitter accounts were crucial

to attracting fans to team Twitter accounts. However, in models using the total number of

Twitter followers of teams as the dependent variable, the authors did not find such a

relationship. This suggested that sports franchises needed strategic plans to connect with

fans in order to achieve lasting popularity on Twitter. The above two studies using

Twitter data were valuable in understanding how to effectively build sport brands on

digital platforms. Additionally, Watanabe, Yan, Soebbing, and Pegoraro (2017)

employed Twitter data to investigate discrimination issues in MLB. Through analyzing

the total number of player followers on Twitter, they showed that Hispanic MLB players

in general had much fewer followers than their counterparts on Twitter, indicating

potential consumer bias toward Hispanic baseball athletes.

The second line of social media research in sport centers on analyzing sentiments

expressed in social media posts. In particular, researchers took various methods to extract

sentiments from social media data in an attempt to understand consumer feelings on

products and services. First, a number of studies chose to manually analyze social media

content. For instance, Delia and Armstrong (2015) used Twitter data to analyze

sponsorship effectiveness in the 2013 French Open. After collecting 1,138 tweets

containing sponsor names, they manually labeled each with positive and negative

sentiments. They concluded that almost all tweets mentioning sponsors contain positive

sentiments. Burton (2019) sought to understand consumer sentiment toward ambush

marketing and used Twitter data from the 2018 Fédération Internationale de Football

Association (FIFA) World Cup Finals as the context to address the research questions. He

gathered 2,136 Twitter posts and asked two coders to manually label each tweet with one

of the sentiments, negative, neutral, and positive. After calculating the average sentiment

of tweets, he concluded that consumer attitudes toward non-sponsor campaigns during

the World Cup were exceedingly more positive than sponsor campaigns, highlighting that

ambush marketing might not be as immoral as previously thought.

Another popular method to estimate sentiments in social media posts is the

lexicon-based approach that uses a pre-defined dictionary of positive and negative words

(Hong & Skiena, 2010). The sentiment is derived by counting the number of positive

words minus the number of negative words in a message. A handful of studies in sport

have adopted this approach to estimate sentiments in social media posts (Chang, 2019;

Yu & Wang, 2015). For instance, Chang (2019) explored the relationship between sports

spectators’ emotional reaction and team performance in Super Bowl 50 through Twitter

data. He used a lexicon-based method to estimate sentiments in Twitter posts. In sum, he

noted fans showed strong positive emotions when their supporting teams scored and

negative emotions when opponents scored. While such emotions diminished in intensity

as teams continued to score, a surge of emotions was observed when supporting teams

scored after a touchdown from opposing teams.

Similarly, Yu and Wang (2015) studied American fans’ emotional responses

toward the 2014 FIFA World Cup matches by using real-time Twitter data. After

analyzing sentiments extracted from live tweets posted during the World Cup by using a

lexicon-based method, they found evidence American fans primarily exhibited negative

sentiments such as fear and anger throughout the games that Team USA played. Such

negative emotions faded away when Team USA scored but were enhanced when

opponents scored.

Schumaker, Jarmoszko, and Labedz (2016) employed Twitter data to predict

match outcomes and goal spreads in the English Premier League. Specifically, they used

a software named CentralSport, a lexicon-based program to collect Twitter data and

analyze sentiments in tweets. Evidence showed that sentiments hidden in tweets could

effectively predict game outcomes and goal spreads, implying how fans perceive games

was useful information in understanding the nature of sport competition.

The final approach to understand sentiments in social media content is the

machine learning method where statistical models are used to learn patterns in textual

data from a training set and then applied to predict sentiments in remaining datasets.

While scant sports social media studies adopted the machine learning approach, a wide

range of research from other fields have greatly relied on machine learning models to

analyze social media content. For instance, Kang, Yoo, and Han (2012) used the machine

learning model Naïve Bayes to estimate sentiments in restaurant reviews. The authors

collected 70,000 restaurant reviews from various websites and were able to correctly

identify sentiments in 81% of the reviews by using machine learning models. McGurk,

Nowak, and Hall (2019) analyzed inventor sentiment expressed in tweets and its

relationship with stock returns. After estimating sentiments in 3.9 million tweets by using

Naïve Bayes models, they found evidence that investor emotion embedded in social

media posts could successfully predict stock directions.

Despite diverse techniques researchers may use to identify sentiments in social

media content, the above sports social media studies demonstrate that social media data

contains valuable information regarding consumer attitudes toward products and services

(Filo et al., 2015). Following this observation, it is fair to argue that social media

platforms may also contain thoughts and opinions reflecting consumer perceptions of

tanking. As such, this dissertation will draw data from social media posts to gauge

consumer perceptions of tanking and study its relationship with demand for sport.

In sum, a thorough literature review shows that the prior studies of tanking have

focused on developing evidence of tanking. However, there has been limited research

examining the link between perceptions of tanking and demand for attendance. To

address this research gap, attention is placed on the NBA where concerns over tanking

have existed for decades (Soebbing & Mason, 2009). Following the economic theory of

demand for sport, this dissertation seeks to evaluate how perceptions of tanking will

impact NBA attendance. The social media research in sport management has offered

evidence that discussions on social media platforms may reflect how sports fans perceive

tanking. As a result, the current research collects data from the social media platform

Twitter to quantify consumer perceptions of tanking. Considering the possible long-term

effect of perceived tanking behavior on demand for sport, this dissertation also uses

Twitter data to construct variables measuring the past consumer perceptions of tanking

and examines their relationships with NBA attendance in the current season. Taken

together, this dissertation proposes the following four research questions:

RQ1: How do consumer perceptions of tanking for home teams affect NBA

attendance in the short term?

RQ2: How do consumer perceptions of tanking for home teams affect NBA

attendance in the long term?

RQ3: How do consumer perceptions of tanking for visiting teams affect NBA

attendance in the short term?

RQ4: How do consumer perceptions of tanking for visiting teams affect NBA

attendance in the long term?

CHAPTER 3

METHODOLOGY

3.1 OVERVIEW

Following the prior empirical examinations of demand for sport, this dissertation

employs a panel dataset to investigate the relationship between consumer perceptions of

tanking and NBA attendance. The panel data consists of 29 of the 30 NBA teams and all

regular season games played between the 2013-2014 and 2017-2018 NBA seasons. The

Toronto Raptors are not included in the dataset due to their location in Canada and the

differences between how federal governments calculate economic variables such as

population. In addition to gathering team and game data, social media platform Twitter is

employed to gauge consumer perceptions of tanking.

Twitter data is selected to quantify consumer perceptions of tanking for several

reasons. First, while other popular social media platforms such as Facebook and

Instagram are available, the underlying function of Twitter is different from others

(Waterloo, Baumgartner, Peter, & Valkenburg, 2018). In particular, Twitter emphasizes

the function of sharing ideas and topics, whereas Facebook highlights the ability to

connect people and Instagram underlies the capability of sharing photos with friends

(Davenport, Bergman, Bergman, & Fearrington, 2014; Van Dijck & Poell, 2013;

Watanabe et al., 2016). Given this observation, it is believed that Twitter is the platform

where users will express strong feelings and emotions toward tanking. Thus, Twitter may

be considered as a better channel to measure perceptions of tanking than other social

media websites. Second, Twitter has gained significant popularity since 2011. It was

reported that Twitter had 140 million daily users by 2012 (Kywe, Lim, & Zhu, 2012).

Hence, there exists a large quantity of data on Twitter that can be analyzed to understand

consumer options and feelings toward tanking (Vicente, Batista, & Carvalho, 2019).

The first section of this chapter describes key variables used in the data analysis,

including the two variables of interest measuring consumer perceptions of tanking and a

set of control variables. The next section presents detailed steps of acquiring Twitter data

and constructing the variables of interest. The last section discusses model specifications

and estimation methods used in data analysis.

3.2 DATA DESCRIPTIONS

3.2.1 UNIT OF ANALYSIS AND POPULATION

The current dissertation analyzes game-level NBA attendance data from the 2013-

2014 to 2017-2018 seasons. Using game-level data can explain more variances of game

attendance than aggregated season-level data (Bradbury & Drinen, 2006). In general,

NBA teams play 41 home and 41 away regular-season games in a given season. Thus,

there are 5,945 unique regular season NBA games from 29 NBA teams across five

seasons available to examine. However, subsets of games are used in different models

depending on the focus of research questions. For instance, for models exploring the

short-term effect of perceptions of tanking for home teams on attendance, 2,870 games

featuring non-playoff teams at home in a given are used. For analysis studying the impact

of perceptions of tanking for home teams on attendance in the long term, 2,829 matches

featuring home teams that did not qualify for the playoffs in the prior season are

identified. In addition, 2,769 and 2,774 NBA games are respectively investigated in

models analyzing the impact of perceptions of tanking for visiting teams on NBA

attendance in the short and long run. The reason to consider only non-playoff teams in

various models is due to the fact that playoff teams often undertake tanking for reasons

other than competing for higher draft picks. For example, teams after clinching the

playoff berth may begin to purposely lose games in order to gain better playoff seeds or

saving energy for the playoffs (Balsdon et al, 2007). However, neither of these behaviors

relates to the focus of the current dissertation, which is to examine the behavior of teams

purposely losing games to acquire high draft picks. Therefore, to accurately capture how

consumers perceive teams tanking for draft picks, only games involving non-playoff

teams are considered.

The population in the current dissertation consists of 29 NBA teams. The Toronto

Raptors are excluded due to their geographic location in Canada. Additionally, the

rebranding of the Charlotte Bobcats after the 2013-2014 season needs special attention.

The Charlotte Bobcats joined the NBA in 2004 as an expansion team and used the

nickname Bobcats until 2014 when team owners decided to rebrand the franchise from

the Charlotte Bobcats to Charlotte Hornets which was the original name for a franchise in

the city from 1988 to 2002 prior to the franchise moving to New Orleans. While the team

adopted a new team name and logo upon the conclusion of the 2013-2014 season, the

ownership and physical location of the franchise remained the same. For this reason, the

Charlotte Bobcats and Charlotte Hornets are treated as the identical team in this

dissertation.

3.2.2 VARIABLES

To examine the effect of perceptions of tanking on NBA attendance, a set of

variables is employed in econometric models (Table 3.1). The summary statistics can be

found in Table 3.2. The dependent variable, game attendance (Attendance), for NBA

matches played from the 2013-2014 through 2017-2018 regular seasons is collected from

basketball-reference.com. While attendance is an effective measure of consumer demand

for sporting events, it is important to note that the reported game attendance reflects the

number of tickets sold, not the actual number of fans attended NBA games. In some cases,

fans may buy tickets but decide not to attend matches for some reason (Forrest &

Simmons, 2002). As such, the actual attendance may be slightly lower than the reported

one. The average attendance for games included in this dissertation is 17,734 per game

and 45.8% of games are sold out.

Moving to the independent variables, the two variables of interest, the volume and

sentiment of tanking tweets, are created to approximate perceptions of tanking. The

volume of tanking tweets reflects consumer awareness of tanking and is measured by the

quantity of tanking tweets posted over a certain period of time. The increasing discussion

of tanking on Twitter may signify that people become more aware of the concept of

tanking (Hutter, HautzDennhardt, & Füller, 2013). The sentiment of tanking tweets

represents consumer sentiments (i.e., positive, negative, and neutral) toward tanking. If

fans consider tanking as a viable strategy, then positive sentiments are likely to be

detected on Twitter. However, Twitter users may express disappointment or

dissatisfaction if they dislike the idea of tanking. Furthermore, if people feel indifferent to

tanking, neutral sentiments shall be observed on Twitter.

To address four research questions proposed in Section 2.5, the two variables of

interest are constructed based on two timeframes, the short term and long term. First, the

short-term variables calculate the volume and sentiment of tanking tweets posted within

31 days before the tipoff of an NBA game.

It is likely that consumer perceptions of

tanking formed prior to an NBA match will affect fans’ decisions to buy tickets.

Therefore, the short-term variables are used to quantify how fans perceive tanking

behavior approaching an NBA game. Second, the long-term variables estimate the

volume and sentiment of tanking tweets posted from the previous season

(July 1 as the

season cutoff

). These lagged variables are created in an attempt to examine whether the

past consumer perceptions of tanking affect attendance decisions in the current season.

When constructing the two variables of interest, the volume and sentiment of

tanking tweets for home teams and visiting teams are also differentiated. It is likely that

local fans who intend to attend NBA games do not only consider tanking behavior for

home teams but also for visiting teams. In other words, consumer perceptions of tanking

related to both home and away teams can potentially affect attendance decisions.

Therefore, the two variables of interest for opponents are also added in this dissertation.

In sum, a total of eight variables measuring consumer perceptions of tanking are

31 days is chosen as the timespan for short-term variables. Other timespans such as 15

days, 62 days, and 93 days will be used in the robustness checks.

The previous season is chosen as the timespan for long-term variables. Other timespans

such as the entire regular season will be used in the robustness check.

July 1st is chosen as the cutoff between the current NBA season and the previous NBA

season because the NBA considers July 1

as the beginning of a new fiscal year.

examined in this dissertation (Short_Volume_Home, Short_Sentiment_Home,

Long_Volume_Home, Long_Sentimnet_Home, Short_Volume_Away,

Short_Sentiment_Away, Long_Volume_Away, Long_Sentiment_Away).

The demand for sport literature also documents a range of control variables that

may affect attendance decisions. These control variables include economic factors, the

quality of viewing, and the quality of sporting contests as summarized by Borland and

MacDonald (2003). For economic factors, the first variable concerns the Metropolitan

Statistical Area (MSA) population (Population), which is a proxy for the market size of

the city where an NBA team locates. A range of empirical studies shown that more

populated regions would experience higher demand for sporting contests (Noll, 1974).

The MSA population data is drawn from the U.S. Bureau of Economic Analysis website.

The second economic factor, namely, Coincident Indexes (Coincident_Index) is a

continuous variable, measuring monthly economic conditions in the state where an NBA

team resides (Oga, 1998). Previous research revealed that people living in wealthier areas

tended to have higher demand for games if sporting events were considered as normal

goods (Carson, Cenesizoglu, & Parker, 2011; Hansen & Gauthier, 1989). Coincident

Index data is retrieved from the Philadelphiafed.org website. Demand for attendance also

depends on pressure from competitors (Rascher et al., 2009). Fierce competition for the

same market may lead to low demand for certain individual sports teams and leagues

(Winfree & Fort, 2008). To account for this competition, a continuous variable counting

the number of teams from other professional sports leagues (Competition) sharing the

same market is considered. This data is acquired from sports-reference.com.

One aspect of the quality of viewing concerns the condition of the sports facilities.

For instance, facility age (Facility_Age) can affect consumer experience at sporting

events, and thus determines demand for sport. Previous research showed a convex

relationship between facility age and game attendance (Coates & Humphreys, 2005).

That is, attendance seems higher in the early years of the operation of sport facilities and

begins to decrease over time. However, attendance may surge again in later years of the

facility’s lifespan. As such, a quadratic term for facility age is added into models

(Facility_Age

). Facility age data is retrieved from NBA media guides.

Another aspect of the quality of viewing is the timing of sporting events, which

plays a crucial role in determining consumer demand for sport (Schofield, 1983). Time-

related dummy variables, such as holidays (Holiday), the day of a week (Day), the month

of a year (October, November, December, January, February, March, and April), and

NBA seasons are employed in statistical models (Hansen & Gauthier, 1989). Following

previous studies on demand for sport, the following holidays are used: Christmas Day,

Memorial Day, Labor Day, Independence Day, New Years’ Day, Thanksgiving Day, and

Martin Luther King Day (Hansen & Gauthier, 1989). It is commonly assumed that teams

face higher demand for sporting events on holidays and weekends where fans have more

time to attend or watch games.

Turning to variables measuring the quality of sporting contests, it is well

documented that team quality and game uncertainty are vital determinants of demand for

attendance. In this dissertation, Elo ratings created by FiveThirtyEight.com are employed

to measure team strength. Compared with other measures of team quality such as the

winning percentage, Elo ratings account for more performance information, and thus may

be regarded as a better measure of team quality. For instance, the calculation of the Elo

ratings does not only consider current win-loss records, but also the expectation of game

outcomes (Mills et al., 2016). Such consideration provides a more objective assessment

of team quality than other scales (Silver & Fischer-Baum, 2015). The average Elo rating

for NBA teams is 1,500. The score increases when teams win games and declines when

they lose games over the course of the season. The Elo rating system is a particularly

useful metric at the beginning of the season (Mills et al., 2016). In contrast to the winning

percentage that may not truly reflect team quality when teams only play a few games, Elo

ratings incorporate team performance from the previous season in calculating the Elo

rating for the current season, and thus may better reflect team success than other metrics

in the early of the season (Silver & Fischer-Baum, 2015). In this dissertation, both home

and visiting team Elo rating variables (Elo_Home, Elo_Away) are included. A handful of

studies offered evidence that home and visiting teams’ quality were positively linked to

fan interest in attending sporting events (Cairns et al., 1986; Humphreys & Johnson, 2020;

Leadley & Zygmont, 2005).

Point spreads collected from betting markets are used to quantify game

uncertainty (Soebbing & Humphreys, 2013). The closing point spread (Point_Spread) is

a continuous variable, estimating the score difference between two competing teams in a

given match. For instance, bookmakers set the following point spreads in the game

between the Washington Wizards and Atlanta Hawks on October 27, 2016,

(Goldsheet.com):

Washington Wizards

Atlanta Hawks -4

In this example, the negative sign of point spread indicates that the Atlanta Hawks were a

four-point favorite to win the game. If the Hawks eventually win by more than four

points, then gamblers who bet that win the wager. The point spread is considered as a

better estimator of the team quality gap than other measures such as the absolute

difference between the home and away team winning percentage, as bookmakers who

rely on setting point spreads to make profits have strong motivations to make accurate

predictions (Forrest et al., 2005). Considering a probable curvilinear relationship between

game uncertainty and attendance, a quadratic term of the point spread variable is also

added in models (Point_Spread

; Coates et al., 2014). Point spread data is retrieved from

Goldsheet.com, a website offering betting odds for varying types of sports competitions.

3.3 DATA COLLECTION

The previous section noted that Twitter data was used to construct the two

variables of interest, the volume and sentiment of tanking tweets. This section offers

more details regarding the procedure of creating these two variables. The first part of this

section delineates specific steps taken to acquire and clean Twitter data. The second part

concerns the process of using Twitter data to build the two variables of interest.

3.3.2 TWITTER DATA COLLECTION

Both tanking tweets and NBA team tweets are needed in order to calculate the

volume and sentiment of tanking tweets. First, to search for tanking tweets for individual

NBA teams, the keywords ‘tanking’ and the NBA team nickname(s) such as ‘Lakers’ are

used in the Twitter Application Programming Interface (API), an access point where

users retrieve Twitter posts. It is critical to note that the reason that NBA team nicknames

are chosen here as search terms is that these words can produce a more complete list of

tanking tweets, than full NBA team names such as the ‘Los Angeles Lakers’. All NBA

team nicknames used are listed in Table 3.3. Among these names, the Philadelphia Sixers’

nickname can be refereed as either ‘Sixers’ or ‘76ers’. Thus, both ‘Sixers’ and ‘76ers’ are

searched in gathering tanking tweets. Also, the keyword ‘tanking’ has to be used with

caution. While ‘tanking’ may not be the sole word to describe the behavior of teams

purposely losing games, other search terms such as ‘rebuilding’ generated a considerable

amount of irrelevant results. For this reason, this dissertation only employs ‘tanking’ as

the search term in collecting tanking tweets.

In addition to drawing tanking tweets from the Twitter API, NBA team tweets

measuring the popularity of NBA franchises are acquired. The purpose of retrieving team

tweets lies in the possibility that the number of tanking tweets may not fully reflect the

intensity of tanking discussions on Twitter. For instance, big market teams such as the

Los Angeles Lakers and New York Knicks may have bigger fan base than small market

teams. Consequently, more tanking discussions related to big market teams tend to appear

on Twitter. However, the absolute quantity of tanking tweets pertaining to big market

teams does not effectively quantify the intensity of tanking discussions, which intends to

measure how well fans are aware of the idea of tanking. The number of team tweets

posted on Twitter signals the general popularity of NBA teams, and thus can be used to

normalize tanking tweets in order to better measure the intensity of tanking discussions.

Similar to the process of retrieving tanking tweets, I collect NBA team tweets

through the Twitter API. However, a different set of search keywords are used.

Specifically, I use the keywords ‘#NBA’ and the hashtag with an NBA team nickname

such as ‘#Lakers’ to collect team tweets. Considering computer processing time and the

possible millions of tweets for each team, this combination of keywords can return

sufficient number of tweets to represent team popularity.

In sum, a total of 166,875 tanking tweets and 6,389,698 NBA team tweets posted

between July 1, 2012, to June 30, 2018, are collected. The downloaded Twitter data

contains a few pieces of information. As shown in Table 3.4, each tweet contains a

unique Twitter ID, publication date, tweet text, the number of retweets, and related NBA

team names. However, after a careful examination of Twitter data, it seems that several

search results are not correctly identified. For instance, with the search terms ‘warriors’

and ‘tanking’, a list of tanking tweets related to the video game World of Warcraft

emerged in search results (See Table 3.4). To ensure the quality of tanking tweets, I took

a few steps to tease out irrelevant tanking tweets from the downloaded dataset.

The procedure of removing unrelated tanking tweets is drawn in Figure 3.3.

Overall, Natural Language Processing (NLP) combined with machine learning models is

implemented to eliminate irrelevant tweets based on the analysis of tweet content. The

raw tweets containing noisy information cannot be directly used in machine learning

models. Thus, a series of steps is performed here to process raw tweets (Figure 3.1). The

first step concerns data cleaning. Specifically, special symbols and characters including

hashtags, punctuations, digits, URLs, ‘@’, NBA team city names, NBA team nicknames,

are removed from tweets. Also, all capital letters in tweets are transformed into lowercase

letters. The second step involves tokenization, which functions to split sentences into

individual words. The third step aims to remove all stop words such as ‘the’ and ‘a’ that

are commonly used in sentences but not useful in understanding tweet content. The fourth

step is stemming and lemmatization, which intends to convert words into their base form.

For instance, lemmatization transforms words such as ‘saw’ to the base form ‘see’ and

stemming removes the derivational affixes in a word.

Next, feature extraction converting terms in a document into features that can be

used in machine learning models is employed. In this step, I adopted the n-gram model

with a unigram. In essence, the n-gram model is used to count the frequency of n

consecutive words in a given text. The word frequency is then treated as features in

machine learning models. For instance, a unigram employed in this model only considers

individual words in a text, while a bigram examines the sequence of two consecutive

words (See examples in Figure 3.2).

After raw tweets are cleaned and feature extraction is completed, a random

sample of 5,000 tweets are selected from the dataset and manually labeled by a coder

with one of the classes 1 and 0, with 1 meaning the tweet relates to the NBA and 0

indicating it does not. Next, the labeled tweets are fitted into several machine learning

models. After comparing the prediction performance across all models, the best

performing model is chosen to estimate the class in all remaining tweets. In this case, I

select the XGBoost algorithm that yields 98.8% accuracy in predicting the class of

tanking tweets.

In sum, of the original 166,875 tanking tweets collected through the Twitter API,

I removed 5,173 irrelevant tweets based on the predication of the XGBoost models. As

such, a total of 161,702 tanking tweets are used in this dissertation. With cleaned Twitter

data, I proceed to construct the two variables of interest, the volume and sentiment of

tanking tweets that are used to approximate consumer perceptions of tanking.

3.3.3 VARIABLE OPERATIONALIZATION

While the Twitter posts gathered in the previous steps contained tweet publication

date and text, neither could be directly used in econometric models. Thus, an additional

step is needed to transform the Twitter data into a useful variable that can be applied in

statistical models. In this section, I describe the detailed steps taken to construct the two

variables of interest, the volume and sentiment of tanking tweets.

To begin with, the volume of tanking tweets concerns the number of tanking

tweets posted within a certain period of time, adjusted for team tweets. Specifically, the

short-term volume variables count the number of tanking tweets posted within 31 days

before an NBA contest, divided by the quantity of team tweets published in the same

period. Here, the calculation of the volume variables takes team tweets into consideration

for normalization. As noted before, big market teams tend to draw more attention on

social media than small market teams. Hence, the absolute quantity of tanking tweets

may not truly reflect the intensity of tanking discussions on Twitter. Similarly, the long-

term volume variable calculates the quantity of all tanking tweets published from the

previous NBA season, adjusted for team tweets. It is critical to note that the number of

retweets is also considered in computing volume variables. For instance, if a tanking

tweet is retweeted 10 times, then 11, including 10 retweets and the tweet itself, will be

added into the calculation of the volume variables.

The second variable of interest, the sentiment of tanking tweets, concerns the

average sentiment expressed in tanking tweets. To extract sentiments from tweets,

sentiment analysis is conducted. Sentiment analysis is the process of understanding

people’s opinions and feelings by analyzing textual data (Kumar & Jaiswal, 2020).

Recently, a significant portion of sentiment analysis focuses on detecting valance on the

positive-negative scale (Farhadloo & Rolland, 2013). That is, one of the three sentiment

classes, positive, neutral, and negative is often assigned to entities based on the analysis

of textual data.

The field of sentiment analysis has developed a range of methods to identify

opinions and feelings in text messages (Feldma, 2013). As noted in the literature review

section, perhaps the most popular sentiment analysis approach adopted in sport

management research is the lexicon-based method, which estimates the sentiment of a

document by using a pre-defined dictionary of positive and negative words and counting

the quantify of net positive words.

The present dissertation utilizes machine learning models to perform sentiment

analysis. A range of advantages of using machine learning methods to perform sentiment

analysis have to be noted. First, machine learning models can automatically extract

sentiments from textual data with good accuracy (Neethu & Rajasree, 2013). This merit

is particularly useful in processing a large dataset such as social media posts where

thousands or even millions of data points may exist. With the increasing popularity of

social media, the sheer number of messages posted on social media platforms everyday is

growing faster than ever (Witkemper, Lim, & Waldburger, 2012). In the meantime, social

media posts contain valuable information regarding consumer sentiment, which is useful

data in understanding behavior (Watanabe et al., 2016). Compared with other measures

of consumer sentiment such as surveys, using machine learning methods can quickly

detect sentiments in a large set of social media data, and thus may better help capture

emotions from customers (Aydoğan & Akcayol, 2016).

Second, machine learning methods take the context into consideration when

attempting to estimate sentiments (Pang & Lee, 2004). Unlike the lexicon-based

approach, the machine learning method does not rely on a pre-defined dictionary. Instead,

a training set that consists of a randomly selected sample from the entire dataset with

manually labeled sentiments is used. This merit is especially important for sports research

as certain words may have special meanings in the context of sport. For instance,

previous sentiment analysis of Twitter posts using the lexicon-based method labeled the

following sentience with the sentiment disgust, ‘NFL is sick but I have no idea what's

happening’ (Chang, 2019). The literal interpretation of the word ‘sick’ might regard the

sentence as the expression of disgust. However, considering the context of sport, the

author of this post seemed to show affection for the NFL, rather than disgust. As such,

machine learning methods relying on statistical models rather than pre-defined

dictionaries may better measure sentiments in social media posts under various contexts.

The detailed steps of conducting sentiment analysis using machine learning

models are listed in Figure 3.3. The first step, data pre-processing, and the second step,

feature extraction, are the same as the previous work of removing irrelevant tanking

tweets, except that I employ the n-gram model with both unigram and bigram, not just

unigram as in the previous work.

After raw tweets are cleaned and feature extraction is completed, the random

sample of 5,000 tweets selected in the prior work of filtering out unrelated tanking tweets

are manually labeled by coders with one of the sentiment classes, positive (1), negative (-

1), and neutral (0) (See Table 3.4).

Next, the labeled tweets are fitted into machine learning models.

Methodologically, sentiment analysis resembles a multiclass classification problem

where models were trained to predict sentiments in new tweets. After comparing the

prediction performance across all models, the best performance model is selected to

estimate sentiment in all remaining tweets. For this sentiment analysis, I choose the

Linear Support-Vector Machine algorithm (LSVM) that yields 70.1% accuracy in

predicting sentiments in tanking tweets. In general, 70% prediction accuracy is

considered as a benchmark for successful sentiment analysis (Kirilenko, Stepchenkova,

Kim, & Li, 2018).

After identifying sentiments in all tweets, the next step is to construct the

sentiment variables. This variable is derived by calculating the average sentiment from

tweets posted within a certain period of time. The formula for calculating the sentiment

of tanking tweets is shown below:

 



















(1)

where S denotes the sentiment value, 1, -1, and 0; i indicates three sentiment categories,

positive, negative, and neutral; P(S) measures the percentage of each type of sentiments

in the sample tweets. It is important to note that P(S) accounts for the number of retweets.

For instance, if there are two tweets with one being labeled as negative and the other one

being labeled as positive and have three retweets, then the average sentiment will be 0.6,

not 0.

In sum, the value of the sentiment of tanking tweets spans from -1 to 1, with -1

indicating all discussions on Twitter are negative toward tanking and 1 implying that all

discussions on Twitter exhibit positive sentiments toward tanking. Furthermore, similar

to the volume of tanking tweets, both short-term and long-term sentiment of tanking

tweets are derived. The short-term sentiment variable estimates the average sentiment of

Twitter posts published within 31 days before an NBA match, while the long-term

sentiment variable calculates the average sentiment of all tanking tweets posted from the

prior season.

3.4 MODEL SPECIFICATIONS

Based on the literature review and collected Twitter data, the following

econometric models are employed to evaluate the relationship between perceptions of

tanking and NBA attendance. The equation (2) below shows that the dependent variable,

the natural log of game-level NBA attendance (Attendance), is a function of a vector of

explanatory variables (Table 3.1)











 !



"

!



"



$ !



"

$!



"

 '(



"

)

'(



"

$ '(



"

$'(



"

-(



"

.

-(



"



/



"



/



"

#

0!



"



0'(



"

%

/ 



" 

&

123



"

)

1/



"

*

!(



"

+

4(



" 

+

5





"8





(2)

where i represents teams in season t at game g. βs represent estimators for variables. 6



unobserved time effects and 7



is team-specific effects. 8



is the error term.

To estimate parameters in econometric models, this dissertation adopts a censored

normal regression model (Humphreys & Johnson, 2020). Due to the overwhelming

number of sell-out games in the NBA, most NBA attendance data is censored at the

capacity level (Coates & Humphreys, 2005). The data analysis shows that 45.8% of NBA

games in the sample were sold out. If the model would be estimated using the Ordinary

Least Squares (OLS) regression, the estimators may be biased and inconsistent

(Wooldridge, 2010). To overcome such an issue, the censored normal regression model

that takes censored attendance data into consideration was employed. The model takes

the form:

 9

  : ;

 < ;

(3)

where c is the facility capacity. In some cases, the reported game attendance exceeds the

arena capacity. Over-selling can emerge when NBA teams sell additional standing room

tickets or temporarily expand the capacity of luxury boxes (Coates & Humphreys, 2005).

Nevertheless, attendance is treated as censored data if the reported attendance is larger

than capacity (Buraimo & Simmons, 2008). If attendance is smaller than capacity, the

actual attendance was observed at the level of Attendance*.

The model also adds team and season fixed effects to account for unobserved

factors that can affect demand for sport. Despite concerns that non-linear regression

models with fixed effects may produce inconsistent estimators due to the incidental

parameters problem (Lancaster, 2000; Neyman & Scott, 1948), Greene (2004) conducted

the Monte Carlo simulation and noted that estimators from the censored normal

regression model with fixed effects were consistent.

Additionally, the unobserved error term in econometric models may be serially

correlated with one another within clusters (teams). To address this serial correlation

concern and derive efficient estimators, cluster-robust standard errors are calculated

(Woodridge, 2010).

In summary, a total of seven models are estimated and compared. The first model

attempts to reveal the general relationship between consumer perceptions of tanking and

game attendance by employing the entire dataset of 5,945 NBA matches. As noted before,

to ensure Twitter data measuring consumer perceptions of tanking is related to the

behavior of teams deliberately losing games for draft picks, not for other purposes, NBA

games featuring non-playoff teams which have incentives to tank for draft picks are

considered in other main models.

The second main model addresses the first research question that explores the

connection between perceptions of tanking for home teams and demand for sport in the

short term. Specifically, 2,870 games featuring home teams that are not eligible for the

playoffs in the observed season are examined in the second model. The third model

focuses on studying the long-term effect of consumer perceptions of tanking for home

teams on attendance. To do this, 2,829 matches involving home teams that did not

compete in the playoffs in the prior season are considered. The fourth model is tested in

an attempt to answer the third reason question whether consumer perceptions of tanking

for visiting teams will affect attendance in the short term. As such, 2,769 games featuring

away teams that do not enter the playoffs in the observed season are analyzed. The fifth

model investigates the long-term impact of consumer perceptions of tanking for away

teams on attendance. Here, 2,774 matches featuring visiting teams that do not compete in

the playoffs in the previous season are selected to test the relationship. The last two

models perform falsification tests by adding the variables gauging the future volume and

sentiment of tanking tweets for home teams and away teams respectively. Falsification

tests aim to examine whether an unlikely event is statistically significant in models. In

this case, if models are correctly specified, the future volume and sentiment of tanking

tweets should not impact NBA attendance.

Table 3.1 Variable Descriptions

Variable Description

Attendance Number of tickets sold in a game

Short_Volume_Home Short-term (31 days before games) volume of tanking tweets for home teams

Short_Sentiment_Home Short-term (31 days before games) sentiment of tanking tweets for home teams

Long_Volume_Home Long-term (Prior season) volume of tanking tweets for home teams

Long_Sentiment_Home Long-term (Prior season) sentiment of tanking tweets for home teams

Short_Volume_Away Short-term (31 days before games) volume of tanking tweets for away teams

Short_Sentiment_Away Short-term (31 days before games) sentiment of tanking tweets for away teams

Long_Volume_Away Long-term (Prior season) volume of tanking tweets for away teams

Long_Sentiment_Away Long-term (Prior season) sentiment of tanking tweets for away teams

Future_Volume_Home Future (31 days after games) volume of tanking tweets for home teams

Future_Sentiment_Home Future (31 days after games) sentiment of tanking tweets for home teams

Future_Volume_Away Future (31 days after games) volume of tanking tweets for away teams

Future_Sentiment_Away Future (31 days after games) sentiment of tanking tweets for away teams

Facility_Age Age of sports facilities (in years)

Facility_Age

Age of sports facilities Squared (in years)

Point_Spread Expected score differences between two competing teams

Point_Spread

Expected score differences between two competing teams Squared

Elo_Home Elo ratings for home teams

Elo_Away Elo ratings for away teams

Population Population of city (in ten thousands)

Coincident_Index Monthly coincident index for each of the 50 states

Competition Number of MLB, NFL, and NHL teams in the same market as a NBA team

Holiday Games held on National holidays (1=yes)

Monday Games held on Monday (1=yes)

Tuesday Games held on Tuesday (1=yes)

Wednesday Games held on Wednesday (1=yes)

Thursday Games held on Thursday (1=yes)

Friday Games held on Friday (1=yes)

Saturday Games held on Saturday (1=yes)

Sunday Games held on Sunday (1=yes)

October Games held in October (1=yes)

November Games held in November (1=yes)

December Games held in December (1=yes)

January Games held in January (1=yes)

February Games held in February (1=yes)

March Games held in March (1=yes)

April Games held in April (1=yes)

2014 Games held in the 2013-2014 NBA season (1=yes)

2015 Games held in the 2014-2015 NBA season (1=yes)

2016 Games held in the 2015-2016 NBA season (1=yes)

2017 Games held in the 2016-2017 NBA season (1=yes)

2018 Games held in the 2017-2018 NBA season (1=yes)

Table 3.2 Summary Statistics (n= 5945)

Variable Mean St. Dev. Min Max

Attendance 17,733.5 2,294.8 7,244 23,152

Short_Volume_Home 0.074 0.183 0.000 2.755

Short_Sentiment_Home -0.628 0.264 -1.000 1.000

Long_Volume_Home 0.042 0.079 0.0002 0.597

Long_Sentiment_Home -0.674 0.115 -0.909 -0.197

Short_Volume_Away 0.072 0.176 0.000 2.685

Short_Sentiment_Away -0.629 0.265 -1.000 1.000

Long_Volume_Away 0.041 0.078 0.0002 0.597

Long_Sentiment_Away -0.673 0.116 -0.909 -0.197

Future_Volume_Home 0.081 0.180 0.000 2.582

Future_Sentiment_Home -0.632 0.247 -1.000 1.000

Future_Volume_Away 0.082 0.184 0.000 2.429

Future_Sentiment_Away -0.627 0.256 -1.000 1.000

Facility_Age 19.8 10.1 1 52

Point_Spread 1.7 7.0 -18.5 21.0

Elo_Home 1,502.8 112.4 1,174.7 1,835.7

Elo_Away 1,504.5 111.6 1,175.5 1,838.6

Population 55.6 50.8 5.3 203.2

Coincident_Index 120.5 9.6 98.3 144.8

Competition 2.2 1.9 0 7

Holiday 0.016 0.125 0 1

Monday 0.14 0.347 0 1

Tuesday 0.117 0.321 0 1

Wednesday 0.205 0.404 0 1

Thursday 0.075 0.264 0 1

Friday 0.191 0.393 0 1

Saturday 0.15 0.357 0 1

Sunday 0.121 0.326 0 1

October 0.038 0.191 0 1

November 0.181 0.385 0 1

December 0.185 0.388 0 1

January 0.184 0.387 0 1

February 0.136 0.342 0 1

March 0.191 0.393 0 1

April 0.085 0.28 0 1

2014 0.2 0.4 0 1

2015 0.2 0.4 0 1

2016 0.2 0.4 0 1

2017 0.2 0.4 0 1

2018 0.2 0.4 0 1

Table 3.3 NBA Team Search Keywords

Number

NBA Franchises Search Keywords (Nicknames)

1 Atlanta Hawks Hawks

2 Boston Celtics Celtics

3 Brooklyn Nets Nets

4 Charlotte Hornets (2014-2015 – Present) Hornets

Charlotte Bobcats (Before 2013-2014) Bobcats

5 Chicago Bulls Bulls

6 Cleveland Cavaliers Cavaliers

7 Dallas Mavericks Mavericks

8 Denver Nuggets Nuggets

9 Detroit Pistons Pistons

10 Golden State Warriors Warriors

11 Houston Rockets Rockets

12 Indiana Pacers Pacers

13 Los Angeles Clippers Clippers

14 Los Angeles Lakers Lakers

15 Memphis Grizzlies Grizzlies

16 Miami Heat Heat

17 Milwaukee Bucks Bucks

18 Minnesota Timberwolves Timberwolves

19 New Orleans Pelicans Pelicans

20 New York Knicks Knicks

21 Oklahoma City Thunder Thunder

22 Orlando Magic Magic

23 Philadelphia 76ers 76ers and Sixers

24 Phoenix Suns Suns

25 Portland Trail Blazers Blazers

26 Sacramento Kings Kings

27 San Antonio Spurs Spurs

28 Utah Jazz Jazz

29 Washington Wizards Wizards

Table 3.4 Tanking Tweet Samples

Twitter ID Date Text Retweet Team Class Sentiment

xxxx 7872 2013-03-06 Whether by design or through incompetence (more

likely) I love that @Sixers are tanking! #SixersTalk

#SixersNation #Sixers @SixersCEOAdam"

0 Sixers 1 Positive (1)

xxxx 5472 2016-01-26 I hate to admit it as a diehard laker fan but @Lakers I

know your tanking. I hate it cause its not within our

blood. I'll never support it

2 Lakers 1 Negative (-1)

xxxx 3424 2013-07-08 @JakeJ29 @zero_chill I'm fine with tanking. But we

wont be top 5 worst teams either. We wont land

Wiggins or Glen Robinson III. #Bucks

0 Bucks 1 Neutral (0)

xxxx 0672 2014-06-17 @Celestalon if protection warriors get a talent to

make then DPS can there be a talent to make Frost

DKs tank again I loved tanking frost

0 Warriors 0 -

Figure 3.1 Textual Data Pre-Processing Steps

Figure 3.2 Feature Extraction Process

Figure 3.3 Data Analysis Process with Machine Learning Models and Twitter Data

CHAPTER 4

RESULTS

Chapter 4 describes the data analysis results. The first section of this chapter

provides a detailed summary of the two variables of interest, the volume and sentiment of

tanking tweets, which are derived from Twitter data. It is critical to note that the

summary only includes non-playoff teams as they have incentives to tank for draft picks,

while other teams do not. With the current dissertation focusing on examining the

behavior of teams deliberately losing games for draft picks, attention is therefore placed

on non-playoff teams. The second section of this chapter reports regression analysis

results, along with robustness checks that test the strength of estimation results.

4.1 SUMMARY OF TWITTER DATA

4.1.1 VOLUME OF TANKING TWEETS

The volume of tanking tweets reflects the fans’ awareness of tanking and is

calculated by using the number of tanking tweets divided by the quantity of team tweets.

Here, I summarize the volume of tanking tweets at the game, team, month, and season

levels. First, in the Twitter dataset spanning from the 2013-2014 to 2017-18 NBA

seasons, the highest short-term volume of tanking tweets for an individual game was the

Chicago Bulls on March 23, 2018 when they played against the Milwaukee Bucks at

home. Such extensive discussion of tanking could be attributed to the player rest strategy

that the Bulls had adopted since the beginning of February that year. In particular, the

Bulls rested Robin Lopez and Justin Holiday, both of whom were starters for the team

and had a significant amount of playing time before being requested to sit out (NBA,

2018). The NBA subsequently issued a warning to the Bulls on March 6, stating that their

behavior violates the league player rest policy (NBA, 2018). The warning caused strong

speculation among fans that the Bulls rested healthy players in order to tank for draft

picks. As such, it was not surprising that the high volume of tanking tweets appeared

prior to the Bulls’ game on March 23.

At the team level, the 2014-2015 Philadelphia Sixers had the highest volume of

tanking tweets among all team-season observations. With the hiring of Sam Hinkie, an

advocate of using tanking to rebuild NBA franchises, on May 10, 2013 as the General

Manager (ESPN, 2013), the Sixers implemented a tanking strategy (Hinkie, 2016). After

only 18 wins out of 82 games played in the 2014-2015 NBA season, it was not surprising

to see fans raise strong concerns that the Sixers were tanking. The second highest volume

of tanking tweets at the team-season level was the previous Sixers’ season (2013-2014).

This season was the Sixers’ first season under the leadership of Sam Hinkie. It appeared

that a large quantity of tanking discussions related to the Sixers posted on Twitter

throughout the entire 2013-2014 season. The ten highest season volume of tanking tweets

and corresponding teams and seasons are listed in Table 4.1. Figure 4.1 plots the short-

term volume of tanking tweets for the Sixers with Locally Estimated Scatterplot

Smoothing (LOESS) that shows a general trend of volume over the sample period. Plots

for other teams are available upon request.

At the season level, the 2013-2014 NBA season experienced a significant increase

of tanking discussions on Twitter. The volume of tanking posts on Twitter seemed to

progressively decline in the following seasons. However, tanking discussions spiked

again in the 2017-2018 NBA season. As Figure 4.2 showes, the volume of tanking tweets

in the 2013-2014 and 2017-2018 NBA seasons were considerably higher than other

seasons. Notably, the 2013-2014 NBA season marked the first season when Sam Hinkie

became the general manager of the Sixers and planned to undertake the extreme tanking

strategy by deliberately losing numerous games in consecutive seasons (Paxton, 2018).

The 2017-2018 season was surrounded by several tanking-related incidents that caused

the extensive discussion of tanking on Twitter. For example, in addition to the warning

issued to the Bulls regarding resting healthy players, the NBA league office fined the

Dallas Mavericks owner Mark Cuban $600,000 for publicly admitting tanking during the

2017-2018 NBA season (NBA, 2018).

Additionally, the volume of tanking tweets varies from month to month during the

NBA regular season. Figure 4.3 summarizes and plots the volume of tanking tweets by

month. It seems the volume of tanking tweets is exceedingly higher in February and

March than other months during the NBA regular season. These findings are not

surprising, given the fact that NBA teams are more likely to deliberately lose games

toward the end of the NBA season, especially when they are eliminated from playoff

contention (Price et al, 2010; Taylor & Trogdon, 2002).

4.1.2 SENTIMENT OF TANKING TWEETS

The second variable of interest, the sentiment of tanking tweets, measures

consumer attitudes toward tanking and is estimated by the average sentiment expressed in

tanking tweets. Recall that sentiment analysis is performed to extract sentiments from

tanking tweets. Specifically, each tweet is assigned with one of the classes, negative (-1),

neutral (0), and positive (1). When Twitter posts contain information supporting the

concept of tanking, these tweets are labeled as positive. Neutral tweets may include any

posts that are in the middle of supporting and disliking tanking. Negative tweets often

represent the fans’ disappointment toward tanking.

After conducting sentiment analysis on tanking tweets, the overall sentiment

toward tanking is clear. Specifically, of 161,702 tanking tweets collected, 75% of Twitter

posts exhibit negative views toward the idea of tanking. Of the rest, 17.7% of posts are

labeled as neutral and 7.3% of tweets show some positive attitudes toward the concept of

tanking. Here, I report the summaries of the sentiment of tanking tweets at the team,

month, and season levels. First, sentiment analysis revealed that 2015-2016 Chicago

Bulls season had the strongest negative sentiment (-0.881) on tanking among all team-

season observations in the dataset. Table 4.2 lists the ten worst sentiments on tanking and

the related seasons and teams. Additionally, Figure 4.4 plots the short-term sentiment of

tanking tweets for the Bulls, as an example, with LOESS that displays a general trend of

sentiments over the seasons.

At the season level, it is interesting to note that while tanking tweets

predominately exhibit negative views, the sentiment toward tanking becomes more

positive over the five-season period. For instance, the sentiment of tanking tweets is

scored at -0.703 in the 2013-2014 season but raises to -0.600 in the 2017-2018 season

(Figure 4.5). A simple linear regression of the sentiment on NBA seasons also shows a

positive relationship between the NBA seasons and sentiment toward tanking, indicating

that fans seem to become more acceptable to the idea of tanking over the years. Another

interesting result is that the 2013-2014 NBA seasons have the lowest sentiment on

tanking among five seasons examined in this dissertation. The extremely low sentiment

might come from the fact that the Sixers were alleged to have tanked for the entire season.

Specifically, as noted before, Sam Hinkie who was renowned for employing the tanking

strategy to rebuild NBA franchises embarked on his career as the general manager of the

Sixers in the 2013-2014 season. While the Sixers did not publicly admit to tanking, the

appointment of Sam Hinkie and his subsequent public statements might make fans

speculate that the team would be committed to tanking (Rappaport, 2017).

The sentiment of tanking tweets is also examined at the month level (Figure 4.6).

Figure 4.6 suggests sentiment regarding tanking seems to improve first and then

deteriorates over the course of the season. Early in the season (October and November),

the sentiment on tanking is scored at -0.670 and -0.638 respectively. Then, the sentiment

becomes more positive as the season progresses and reaches the peak in December.

However, the sentiment appears to aggravate toward the end of the season (March and

April). This observation is expected as abundant studies have shown that teams are more

likely to tank toward the end of the NBA regular season (Price et al, 2010; Taylor &

Trogdon, 2002). With an increasing number of tanking teams approaching the conclusion

of an NBA season, more fans may express disappointment and displeasure in March and

April (Price et al, 2010; Taylor & Trogdon 2002).

4.2 REGRESSION ANALYSIS RESULTS

The previous section summarized the two variables of interest and revealed some

interesting observations. This section focuses on reporting regression analysis results for

four research questions proposed in Chapter 2.

Table 4.3 reports the main findings. Column (1) presents the results for models

using the entire dataset of 5,945 games. While all 5,945 games played between the 2013-

2014 to 2017-2018 NBA seasons are available to examine, some of which are not useful

in addressing the proposed research questions. The reason rests on the fact that not all

tanking teams aim to deliberately lose games to improve draft position. For instance,

certain playoff teams may tank for other purposes such as reserving energy for the

playoffs or competing for better playoff seeds (McManus, 2019). The current dissertation

focuses on analyzing the behavior of tanking for draft picks and how fans perceive such

behavior. Thus, subsets of games featuring non-playoff teams that have incentives to tank

for draft picks are used in models.

Table 4.3, Column (2) contains the findings for the first research question whether

consumer perceptions of tanking for home teams will affect attendance in the short run.

To ensure the Twitter data can effectively reflect consumer perceptions of tanking for

draft picks, the analysis employs 2,870 games featuring home teams that do not enter the

playoffs in the observed season. The estimated coefficient on the short-term volume of

tanking tweets for home teams is negative and significant at the 0.05 level, meaning that

the higher volume of tanking discussions for home teams prior to an NBA game will

result in lower demand for sport. The parameter estimate on the short-term sentiment of

tanking tweets for home teams is insignificant, indicating that fan attitudes toward

tanking related to home teams do not affect demand for NBA games in the short term.

The second research question investigates the possibility that consumer

perceptions of tanking may have a prolonged impact on consumer interest in NBA games.

To address this question, the same model specifications as Column (2) is used but a

different set of 2,829 NBA games featuring home teams that did not qualify for playoffs

in the previous season are examined. Recall that the long-term variables measure the

volume and sentiment of tanking from the previous season, while the short-term variables

are calculated by using tweets posted within 31 days prior to the start of an NBA game.

The estimation results in Column (3) show that the extensive discussion of tanking for

home teams from the prior season can negatively affect NBA attendance as the estimated

parameter on the long-term volume of tanking tweets for home teams is negative and

significant at the 0.05 level. Yet, the estimated coefficient on the long-term sentiment of

tanking tweets for home teams is not significant, implying that fan views of tanking for

home teams formed in the past do not influence their attendance decisions.

The third and fourth questions aim to explore whether consumer perceptions of

tanking for visiting teams will affect attendance decisions in both the short-term and

long-term. To answer these questions, 2,769 games featuring away teams that are not

eligible for playoffs in the observed season and 2,774 games featuring away teams that

did not compete in playoffs in the prior season are employed in models in Column (4)

and Column (5) respectively. The findings in Column (4) reveal that the parameter

estimate on the short-term sentiment of tanking tweets for away teams is positive and

significant at the 0.05 level, indicating that how fans view visiting teams’ tanking

behavior prior to an NBA match positively affect their attendance decisions. However,

the estimate on the short-term volume of tanking tweets for visiting teams is not

significant, meaning the decision to attend an NBA game does not hinge on how well

local fans are aware of tanking related to away teams. Additionally, the results in Column

(5) suggest estimated coefficients on the long-term volume and sentiment of tanking

tweets for away teams are not significant at the 0.05 level, signaling that consumer

perceptions of tanking for visiting teams do not influence local fans’ attendance decisions

in the long run.

Lastly, two falsification tests are performed to test whether model specification

are correctly identified (Goldhaber & Chaplin, 2015). To perform falsification tests, I add

a set of variables that represents future consumer perceptions of tanking. Specifically, the

future volume and sentiment of tanking tweets for both home and visiting teams are

calculated by using tweets posted within 31 days after an NBA game. If the model

specifications are correct, future perceptions of tanking should not have any effects on

attendance decisions. The estimation results are displayed in Table 4.3, Column (6) and

Column (7). The parameter estimates on future perceptions of tanking for home and

visiting teams are all insignificant at the 0.05 level. Thus, my models pass falsification

tests.

Moving to control variables, the estimated parameter on home team quality is

significant and positive in all models in Table 4.3, suggesting that high quality home

teams tend to attract more fans to games. The different estimation results are observed on

away team quality variables. Specifically, the findings in Column (1), (2), and (3) show a

positive and significant relationship between away team quality and NBA attendance,

while the estimations in Column (4), (5) do not.

Economic factors such as population and income level may be important

predictors of attendance in the NBA (Borland & MacDonald, 2003). The results in

Column (1), (3), and (4) show that MSA population is positively associated with demand

for NBA games. However, the finding in Column (2) concludes that there is a statistically

negative link between population and NBA attendance. Additionally, the estimation in

Column (5) suggests that population may not be a determining factor of consumer

interest in sport. Unlike the population variables revealing different relationships under

various models, the parameter estimates on coincident indexes measuring monthly

economic conditions in states produce consistent outcomes across all models.

Specifically, the findings suggest coincident indexes have a statistically positive

connection with demand for sport at the 0.05 level, indicating a stronger economy leads

to higher attendance at NBA games. The number of professional sports teams in the same

market as NBA teams does not seem to affect attendance. Only the results in Column (1)

display a negative relationship between Competition and consumer interest in NBA

games, meaning that stronger market competition will lead to lower demand for

attendance. All other models suggest that market competition is not a key determinant of

consumer interest in the NBA.

NBA game attendance may also be determined by game uncertainty (Mills & Fort,

2014). The current dissertation uses point spreads from betting markets to approximate

game uncertainty. The model specifications also include a quadratic term on point

spreads in order to catch the nonlinear relationship between game uncertainty and

demand for sport. The estimation results suggest that the parameter estimate on the

squared point spreads is positive and significant in Column (1), (2), (3), and (5), while the

estimated coefficient on point spreads is not significant, except in Column (1). These

findings suggest that the lowest attendance will appear when point spreads are 0, where

game outcomes are most unpredictable. Thus, this dissertation does not offer evidence

that more uncertain NBA games will increase ticket sales.

The literature on demand for sport also suggested that the facility age might play

an important role in changing NBA attendance (Coates & Humphreys, 2005). The present

dissertation produces inconsistent conclusions regarding the connection between facility

age and consumer interest in NBA games. The findings in Column (3) and (5) suggest

that facility age does not affect NBA attendance, while the result in Column (2) reveals a

concave relationship between these two variables. Additionally, the parameter estimate

on facility age is negative in Column (1), implying that demand will decline after the

opening of a new sports facility. Furthermore, the result in Column (4) indicates a convex

relationship between facility age and NBA attendance, meaning that attendance drops

after a sports facility opens but increases toward the end of the lifespan of the facility.

The timing of NBA games seems to be a significant predictor of NBA attendance

(Watanabe, Yan, Soebbing, & Fu, 2019). In this dissertation, the estimated coefficients

on holiday games are positive and significant at the 0.05 level across all models. For

instance, the estimation in Column (2) indicates that holiday games will attract 8.3%

more fans than non-holiday games. The estimated coefficients on other timing variables

such as the day of a week and the month of a year produce same conclusions as the prior

studies of demand for sport. Due to limited space, their estimation results are not reported

here.

4.3 ROBUSTNESS CHECKS

To test the strength of the findings, I perform a number of robustness checks. First,

the short-term volume and sentiment variables for both home teams and away teams are

recalculated by using tweets posted within 15, 62, and 93 days before the tipoff of an

NBA game rather than 31 days used in the main models. Originally, the construction of

the two variables of interest assumes that fans on average book tickets to NBA games

within a month prior to the start of an NBA game (Mills, Salaga, & Tainsky, 2016),

However, some fans may choose to buy tickets earlier or later than the one-month

window. Thus, it is important to check the robustness of the findings under different

timeframes. As such, I re-estimated models by using variables calculated from 15-day,

62-day and 93-day data.

The estimation results are shown in Table 4.4. In general, all models produce

similar results as the main models in terms of the sign and significance. For models

investigating the effect of perceptions of tanking for home teams on attendance in Table

4.4, Column (1), (2), and (3), the estimated parameters on the short-term volume of 15-

day, 62-day and 93-day tanking tweets for home teams are negative and significant at the

0.05 level, meaning the increasing discussion of tanking related to home teams prior to an

NBA game undermines consumer interest. Comparing the magnitude of the parameter

estimates across three models, it appears that tanking discussions occurred within 93 days

before the start of an NBA match has the highest impact on attendance and such effects

diminish as the time window shortens. In other words, the awareness of tanking formed

within at least 93 days before an NBA game can negatively affect attendance decisions.

Moving to the short-term sentiment of tanking tweets for home teams, the findings are

consistent across three robustness check models in Table 4.4, Column (1), (2), and (3)

that the estimated coefficients are not significant at the 0.5 level. Despite different ways

of calculating sentiments, all models draw the same conclusion that fan attitudes toward

tanking for home teams measured prior to an NBA match do not affect consumer

behavior in attending NBA games.

For models studying the impact of perceptions of tanking for visiting teams on

attendance in Table 4.4, Column (4), (5), and (6), the estimated coefficients on the short-

term sentiment of 15-day, 62-day and 93-day tanking tweets for visiting team are all

positive and significant at the 0.5 level. These estimation results are consistent as the

main models in terms of the sign and significance. The magnitude of these coefficients is

slightly different. Table 4.4, Column (4) shows that the sentiment of 15-day tanking

tweets for visiting teams has the smallest positive impact on attendance, while the

sentiment of 93-day tanking tweets for away teams has the largest positive impact on

attendance among three models. Additionally, the estimated coefficients on the short-

term volume of 15-day, 62-day, and 93-day tanking tweets for visiting teams are not

significant at the 5% for all models. These findings are similar to the main models that

the broad discussion of tanking for away teams prior to an NBA game will not affect

local fans’ attendance decisions.

The second robustness check concerns the long-term impact of consumer

perceptions of tanking on demand for NBA games. Recall that the long-term volume and

sentiment of tanking tweets are calculated from tanking tweets posted in the previous

NBA season and are employed to quantify the past consumer perceptions of tanking. To

calculate the long-term variables, an entire NBA season spanning from July 1 to June 30

is considered. However, Twitter posts over the entire NBA may not adequately quantify

the past consumer perceptions of tanking. For instance, it is likely that fans may develop

their perceptions of tanking over the course of the season when they watch games (Aday

& Phelan, 2016). Therefore, the second robustness check uses tanking tweets published

during the prior regular season, rather than the entire season, to construct the long-term

variables. The models in Table 4.3, Column (3) and (5) are re-estimated with the revised

long-term variables and the results are reported in Table 4.4, Column (7) and (8).

Overall, the findings with the new long-term variables are largely the same as the

main models, indicating the robustness of findings pertaining to the long-term effect of

perceptions of tanking on demand for sport. Specifically, the results in Table 4.4, Column

(7) show that the long-term volume of tanking tweets for home teams is negatively

related to attendance. This conclusion is similar to the previous analysis. However, the

findings related to the long-term sentiment of tanking tweets for home teams are different

from the main models. As shown in Column (7), the robustness check model suggests

that the long-term sentiment of tanking tweets for home teams negatively relates to

attendance as the estimated coefficient is negative and significant at the 0.05 level, while

the main models revealed an insignificant relationship. One possible explanation for this

difference is that fans who developed negative attitudes toward tanking for home teams

during the prior regular season are curious about what teams have changed in the current

season. As such, local fans may express higher demand for home games this year,

although they may hold negative attitudes on tanking for home teams in the previous

season. Focusing on the estimations pertaining to the long-term variables for visiting

teams, the results in Column (8) are the same as prior examinations that neither the

awareness nor attitudes toward tanking for visiting teams can influence local fans’

attendance decisions in the long run.

The third robustness check re-estimated the effect of the volume of tanking tweets

on attendance without adjusting for team tweets. Recall that NBA team tweets are

collected to normalize the quantity of tanking tweets since big market teams such as Los

Angeles Lakers and New York Knicks may intrinsically have more social media posts,

and thus have a higher quantity of tanking tweets than other NBA teams (Robinson &

DeSchriver, 2003; Késenne, 2000). Here, the first five models listed in Table 4.3 are re-

estimated without counting team tweets in the volume variables. The results are presented

in Table 4.5. Overall, the significance and sign of the parameters of interest are consistent

as the main models. The only exception is that the estimate parameters on the short-term

volume of tanking tweets for visiting teams become positive and significant at the 0.05

level across five modes in Table 4.5 but are insignificant in the main models. This

difference may arise from the fact that the volume of tanking tweets is not normalized,

and thus is not a proper proxy for the fan awareness of tanking. Nevertheless, this

robustness check generally reconfirms the finding that the increasing awareness of

tanking for home teams can significantly harm consumer interest in both the short term

and long term. In sum, the above three robustness checks largely show consistent results

as the main models. Thus, I am confident of the robustness of the findings in these

models.

The fourth robustness check relates to the concern that Elo ratings may be highly

correlated with the volume and sentiment of tanking tweets variables, as the calculation

of Elo ratings may take teams’ tanking behavior into consideration (Silver & Fischer-

Baum, 2015; Soebbing & Humphreys, 2013). To perform the fourth robustness check,

seven main models are re-estimated by using team winning percentage prior to the start

of NBA game for home and away teams (Win_home, Win_away) instead of Elo ratings.

The estimation results can be found in Table 4.6. In sum, the parameter estimates on the

variables of interest in Table 4.6 have similar significance and sign as estimated

coefficients in main models, confirming the robustness of main findings.

Figure 4.1 Short-term Volume of Tanking Tweets for the Philadelphia Sixers from the 2013-2014 to 2017-2018 Season

2014 2015 2016 2017 2018

0.0

0.5

1.0

1.5

2.0

Game Date

Short-term Volume of Tanking Tweets

Figure 4.2 Volume of Tanking Tweets by Season

Season

Volume of Tanking Tweets

2014 2015 2016 2017 2018

0.00

0.04

0.08

0.12

0.16

0.20

Figure 4.3 Volume of Tanking Tweets by Month

Month

Volume of Tanking Tweets

10 11 12 1 2 3 4

0.00

0.04

0.08

0.12

0.16

0.20

Figure 4.4 Short-term Sentiment of Tanking Tweets for the Chicago Bulls from the 2013-2014 to 2017-2018 Seasons

2014 2015 2016 2017 2018

-1.0

-0.5

0.0

0.5

1.0

Game Date

Short-term Sentiment of Tanking Tweets

Figure 4.5 Sentiment of Tanking Tweets by Season

Season

Sentiment of Tanking Tweets

-0.75

-0.70

-0.65

-0.60

-0.55

-0.50

2014 2015 2016 2017 2018

Figure 4.6 Sentiment of Tanking Tweets by Month

Month

Sentiment of Tanking Tweets

10 11 12 1 2 3 4

-0.75

-0.70

-0.65

-0.60

-0.55

-0.50

Table 4.1 Ten Highest Volume of Tanking Tweets by Team and Season

Rank Team Season Volume

1 Philadelphia 76ers 2014-2015 0.597

2 Chicago Bulls 2017-2018 0.588

3 Philadelphia 76ers 2013-2014 0.485

4 Dallas Mavericks 2017-2018 0.431

5 Utah Jazz 2013-2014 0.378

6 Atlanta Hawks 2017-2018 0.369

7 Memphis Grizzlies 2017-2018 0.367

8 Milwaukee Bucks 2013-2014 0.285

9 Orlando Magic 2017-2018 0.250

10 Phoenix Suns 2017-2018 0.206

League Average 0.101

Table 4.2 Ten Lowest Sentiment of Tanking Tweets by Team and Season

Rank Team Season Sentiment

1 Chicago Bulls 2015-2016 -0.881

2 New York Knicks 2013-2014 -0.847

3 New York Knicks 2015-2016 -0.805

4 Orlando Magic 2015-2016 -0.802

5 Minnesota Timberwolves 2016-2017 -0.795

6 Utah Jazz 2015-2016 -0.787

7 Sacramento Kings 2014-2015 -0.783

8 Brooklyn Nets 2016-2017 -0.762

9 Philadelphia 76ers 2013-2014 -0.759

10 Charlotte Hornets 2014-2015 -0.757

League Average -0.653

Table 4.3 Regression Analysis Results for Main Models

Full

Non-playoff

(home)

Non-playoff

(home last year)

Non-

playoff

(away)

Non-playoff

(away last year)

Non-playoff

(home)

Non-playoff

(away)

(1) (2) (3) (4) (5) (6) (7)

Short_Volume_Home -0.075

***

-0.074

***

-0.102

***

-0.057

-0.054

-0.078

***

-0.058

(0.015) (0.016) (0.021) (0.021) (0.022) (0.018) (0.021)

Short_Sentiment_Home -0.010 -0.010 0.005 -0.001 -0.015 -0.009 -0.001

(0.009) (0.013) (0.012) (0.012) (0.012) (0.013) (0.012)

Long_Volume_Home -0.213

***

-0.088 -0.299

***

-0.195

***

-0.163

-0.086 -0.196

***

(0.039) (0.046) (0.044) (0.056) (0.057) (0.046) (0.056)

Long_Sentiment_Home -0.017 -0.055 0.011 0.028 0.024 -0.060 0.027

(0.025) (0.043) (0.044) (0.035) (0.036) (0.044) (0.035)

Short_Volume_Away 0.009 0.004 -0.012 -0.0002 -0.029 0.002 -0.012

(0.015) (0.018) (0.019) (0.015) (0.022) (0.018) (0.018)

Short_Sentiment_Away 0.011 0.005 0.009 0.042

0.036

0.005 0.042

(0.009) (0.012) (0.012) (0.014) (0.013) (0.012) (0.014)

Long_Volume_Away 0.006 0.014 0.032 -0.046 -0.052 0.014 -0.047

(0.034) (0.044) (0.045) (0.037) (0.038) (0.044) (0.037)

Long_Sentiment_Away -0.036 -0.030 -0.041 -0.017 0.021 -0.031 -0.015

(0.020) (0.026) (0.027) (0.030) (0.029) (0.026) (0.031)

Future_Volume_Home

0.011

(0.019)

Future_Sentiment_Home

-0.017

(0.015)

Future_Volume_Away

0.024

(0.016)

Future_Sentiment_Away

0.002

(0.015)

Facility_Age -0.011

***

0.015

***

0.004 -0.016

***

-0.006 0.015

***

-0.016

***

(0.003) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004)

Facility_Age

0.0002 -0.001

***

-0.0003 0.0003

0.00001 -0.001

***

0.0003

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Point_Spread 0.001

0.001 0.001 -0.0001 0.0003 0.001 -0.00005

(0.0005) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)

Point_Spread

0.0003

***

0.0003

***

0.0004

***

0.0001 0.0002

0.0003

***

0.0001

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Elo_Home 0.001

***

0.0003

***

0.001

***

0.001

***

0.001

***

0.0003

***

0.001

***

(0.00004) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Elo_Away 0.0003

***

0.0003

***

0.0003

***

0.0001 0.0001 0.0003

***

0.0001

(0.00003) (0.00004) (0.00004) (0.0001) (0.00005) (0.00004) (0.0001)

Population 0.0003

-0.001

***

0.002

***

0.0004

0.0004 -0.001

***

0.0004

(0.0001) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)

Coincident_Index 0.004

0.008

***

0.004

0.006

***

0.004

0.008

***

0.006

***

(0.001) (0.002) (0.001) (0.002) (0.002) (0.002) (0.002)

Competition -0.052

0.015 0.063 -0.050 -0.035 0.017 -0.050

(0.023) (0.031) (0.037) (0.030) (0.030) (0.031) (0.030)

Holiday 0.111

***

0.083

0.096

0.118

***

0.102

***

0.083

0.119

***

(0.021) (0.030) (0.030) (0.029) (0.027) (0.030) (0.029)

Observations 5945 2870 2829 2769 2774 2870 2769

Note: The dependent variable is log(attendance).

p<0.05;

p<0.01;

***

p<0.001.

Table 4.4 Regression Analysis Results for Robustness Check 1 and 2

Non-playoff (home) Non-playoff (away)

Non-

playoff

(home last

year)

Non-playoff

(away last

year)

15 days 62 days 93 days 15 days 62 days 93 days Regular season

(1) (2) (3) (4) (5) (6) (7) (8)

Short_Volume_Home -0.040

-0.139

***

-0.185

***

-0.022 -0.115

***

-0.177

***

-0.177

***

-0.127

***

(0.013) (0.021) (0.025) (0.016) (0.027) (0.031) (0.025) (0.028)

Short_Sentiment_Home

-0.013 -0.003 0.0002 -0.007 -0.009 -0.001 0.018 -0.025

(0.010) (0.015) (0.017) (0.010) (0.014) (0.015) (0.014) (0.014)

Long_Volume_Home -0.081 -0.091

-0.093

-0.189

***

-0.193

***

-0.195

***

-0.212

***

-0.090

(0.046) (0.046) (0.046) (0.056) (0.056) (0.056) (0.033) (0.044)

Long_Sentiment_Home

-0.049 -0.058 -0.062 0.033 0.026 0.026 -0.205

***

0.014

(0.043) (0.044) (0.044) (0.035) (0.035) (0.035) (0.042) (0.036)

Short_Volume_Away 0.010 0.003 -0.013 0.009 0.00003 -0.010 -0.015 -0.034

(0.013) (0.023) (0.027) (0.012) (0.020) (0.023) (0.023) (0.025)

Short_Sentiment_Away

0.002 -0.008 -0.003 0.027

0.037

0.050

0.003 0.046

(0.009) (0.013) (0.014) (0.011) (0.015) (0.016) (0.013) (0.015)

Long_Volume_Away 0.013 0.013 0.020 -0.050 -0.046 -0.038 0.044 -0.023

(0.044) (0.044) (0.044) (0.037) (0.038) (0.038) (0.037) (0.031)

Long_Sentiment_Away

-0.029 -0.032 -0.034 -0.018 -0.015 -0.019 -0.043 0.015

(0.026) (0.026) (0.026) (0.030) (0.030) (0.030) (0.026) (0.029)

Facility_Age 0.015

***

0.015

***

0.015

***

-0.017

***

-0.016

***

-0.015

***

0.008

-0.006

(0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004)

Facility_Age

-0.001

***

-0.001

***

-0.001

***

0.0003

-0.0004

-0.00001

100

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Point_Spread 0.001 0.001 0.001 -0.0002 -0.0001 -0.0001 0.001 0.0003

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)

Point_Spread

0.0003

***

0.0003

***

0.0003

***

0.0001 0.0001 0.0001 0.0004

***

0.0002

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Elo_Home 0.0003

***

0.0003

***

0.0003

***

0.001

***

0.001

***

0.001

***

0.001

***

0.001

***

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Elo_Away 0.0003

***

0.0003

***

0.0003

***

0.0001 0.0001 0.0001 0.0003

***

0.0001

(0.00003) (0.00004) (0.00004) (0.0001) (0.0001) (0.0001) (0.00004) (0.00005)

Population -0.001

***

-0.001

***

-0.001

***

0.0004

0.002

***

0.0004

(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)

Coincident_Index 0.008

***

0.008

***

0.008

***

0.006

***

0.006

***

0.006

***

0.004

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001) (0.002)

Competition 0.014 0.017 0.017 -0.052 -0.050 -0.052 0.077

-0.034

(0.031) (0.031) (0.031) (0.030) (0.030) (0.030) (0.037) (0.030)

Holiday 0.083

0.081

0.078

0.117

***

0.119

***

0.120

***

0.093

0.104

***

(0.030) (0.030) (0.030) (0.029) (0.029) (0.029) (0.030) (0.027)

Observations 2870 2870 2870 2769 2769 2769 2829 2774

Note: The dependent variable is log(attendance).

p<0.05;

p<0.01;

***

p<0.001.

101

Table 4.5 Regression Analysis Results for Robustness Check 3

Full

Non-playoff

(home)

Non-

playoff (home

last year)

Non-playoff (away

year)

Non-

playoff (away

last year)

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

Short_Volume_Home -0.035

***

-0.026

***

-0.034

***

-0.027

***

-0.028

***

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

Short_Sentiment_Home -0.012 -0.011 -0.002 -0.005 -0.017

(0.009) (0.012) (0.012) (0.012) (0.012)

Long_Volume_Home -0.006

***

-0.002 -0.006

***

-0.005

-0.004

(0.001) (0.001) (0.001) (0.002) (0.002)

Long_Sentiment_Home -0.025 -0.052 -0.008 0.024 0.018

(0.025) (0.043) (0.044) (0.035) (0.036)

Short_Volume_Away 0.009

0.010 0.010 0.006 -0.002

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

Short_Sentiment_Away 0.010 0.003 0.006 0.041

0.036

(0.009) (0.011) (0.012) (0.014) (0.013)

Long_Volume_Away 0.005

***

0.007

***

0.006

***

0.005

***

0.004

***

(0.001) (0.001) (0.001) (0.001) (0.001)

Long_Sentiment_Away -0.032 -0.025 -0.036 -0.008 0.022

(0.020) (0.026) (0.027) (0.030) (0.029)

Facility_Age -0.011

***

0.015

***

0.007 -0.016

***

-0.006

(0.003) (0.004) (0.004) (0.004) (0.004)

102

Facility_Age

0.0002 -0.001

***

-0.0004

0.0003

0.00000

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Point_Spread 0.0005 0.0005 -0.00004 -0.001 -0.001

(0.0005) (0.001) (0.001) (0.001) (0.001)

Point_Spread

0.0002

***

0.0003

***

0.0003

***

0.0001 0.0001

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Elo_Home 0.001

***

0.0003

***

0.001

***

0.001

***

0.001

***

(0.00004) (0.0001) (0.0001) (0.0001) (0.0001)

Elo_Away 0.0004

***

0.0003

***

0.0004

***

0.0002

0.0001

(0.00003) (0.00003) (0.00003) (0.0001) (0.00005)

Population 0.0004

-0.001

***

0.001

***

0.0004

(0.0001) (0.0002) (0.0002) (0.0002) (0.0002)

Coincident_Index 0.004

***

0.009

***

0.006

***

0.006

***

0.005

(0.001) (0.002) (0.001) (0.002) (0.002)

Competition -0.063

0.001 0.008 -0.059

-0.046

(0.023) (0.031) (0.037) (0.030) (0.030)

Holiday 0.108

***

0.082

0.094

0.113

***

0.097

***

(0.021) (0.030) (0.030) (0.029) (0.027)

Observations 5945 2870 2829 2769 2774

Note: The dependent variable is log(attendance).

p<0.05;

p<0.01;

***

p<0.001.

103

Table 4.6 Regression Analysis Results for Robustness Check 4

Full Non-playoff (home)

Non-

playoff (home

last year)

Non-playoff (away)

Non-playoff (away

last year)

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

Short_Volume_Home -0.131

***

-0.093

***

-0.142

***

-0.116

***

-0.129

***

(0.015) (0.016) (0.021) (0.021) (0.022)

Short_Sentiment_Home -0.011 -0.012 0.005 -0.004 -0.018

(0.009) (0.013) (0.012) (0.013) (0.013)

Long_Volume_Home -0.307

***

-0.120

-0.361

***

-0.291

***

-0.267

***

(0.040) (0.046) (0.045) (0.057) (0.058)

Long_Sentiment_Home -0.002 -0.075 0.011 0.034 0.015

(0.026) (0.044) (0.045) (0.036) (0.037)

Short_Volume_Away -0.013 -0.010 -0.032 -0.013 -0.038

(0.015) (0.017) (0.018) (0.015) (0.022)

Short_Sentiment_Away 0.012 0.004 0.011 0.041

0.032

(0.009) (0.012) (0.012) (0.014) (0.013)

Long_Volume_Away -0.074

-0.047 -0.017 -0.079

-0.070

(0.034) (0.043) (0.044) (0.037) (0.039)

Long_Sentiment_Away -0.023 -0.017 -0.028 -0.001 0.037

(0.021) (0.026) (0.028) (0.031) (0.030)

Facility_Age -0.016

***

0.012

0.006 -0.020

***

-0.011

(0.003) (0.004) (0.005) (0.004) (0.004)

104

Facility_Age

0.0004

***

-0.001

***

-0.0003 0.0005

0.0002

(0.0001) (0.0001) (0.0002) (0.0001) (0.0001)

Point_Spread 0.002

***

0.001 0.001 0.001 0.002

(0.0005) (0.001) (0.001) (0.001) (0.001)

Point_Spread

0.0003

***

0.0004

***

0.0004

***

0.0003

***

0.0004

***

(0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Win_Home 0.180

***

0.049

0.161

***

0.179

***

0.181

***

(0.018) (0.025) (0.025) (0.026) (0.026)

Win_Away 0.155

***

0.125

***

0.135

***

0.043 0.075

(0.015) (0.019) (0.020) (0.026) (0.025)

Population 0.0003 0.0004 0.002

0.0003 0.0003

(0.0001) (0.001) (0.001) (0.0002) (0.0002)

Coincident_Index 0.002 0.004

0.003 0.004

0.002

(0.001) (0.002) (0.002) (0.002) (0.002)

Competition -0.057

0.053 0.079

-0.054 -0.037

(0.024) (0.033) (0.039) (0.032) (0.031)

Holiday 0.114

***

0.086

0.098

0.130

***

0.103

***

(0.022) (0.030) (0.031) (0.030) (0.028)

observations 5866 2832 2791 2733 2740

Note: The dependent variable is log(attendance).

p<0.05;

p<0.01;

***

p<0.001.

105

CHAPTER 5

DISCUSSION

The previous chapter presented some interesting results from the regression

analysis regarding the impact of consumer perceptions of tanking on demand for NBA

games. Chapter 5 focuses on interpreting these findings. First, I provide a thorough

discussion of the analysis of outcomes reported in Chapter 4. Next, theoretical

contributions as well as practical implications of the findings are discussed.

5.1 DISCUSSION

The present dissertation examining the impact of consumer perceptions of tanking

on demand for NBA games has a few important findings. First, the estimation results in

Table 4.3, Column (2) suggest that the short-term volume of tanking tweets for home

teams is negatively related to attendance in the NBA. Recall that the short-term variable

quantifies the intensity of tanking discussions within 31 days before the tipoff of an NBA

game. Hence, this result provides evidence that when fans become more aware of tanking

for home teams prior to a match, they tend to have lower demand for NBA games. This

finding is not surprising, considering the fact that tanking teams purposely losing games

jeopardizes the spirit of sports competition, which requires participants to exert the best

effort in sporting contests (Tayade, Bhamare, Kamble, & Jadhav, 2013). Therefore, when

fans perceive the existence of tanking, they may lose interest in attending these games.

The above conclusion is also slightly different from the observation provided by

Preston and Szymanski (2003) that cheating behavior in sport may not significantly

106

undermine consumer interest in sport. Preston and Szymanski (2003) listed some

anecdotal evidence that betting-related match fixing in MLB and the English Premier

League (EPL) did not seem to hurt demand for sport as the popularity of these sports

leagues continued to grow after gambling scandals. While tanking is distinct from other

match-fixing behaviors in that it is not strictly prohibited by league rules (Lenten et al.,

2018), empirical evidence from this dissertation shows that perceived tanking behavior

does negatively affect consumer demand for sport.

Another important finding from this dissertation is the long-term volume of

tanking tweets for home teams is negatively related to NBA attendance. Recall that the

long-term variable gauges the intensity of tanking discussions from the previous season.

This conclusion implies that tanking discussions from the past has a persistent influence

on present attendance decisions. In other words, teams with a tradition of tanking may not

only suffer from lower attendance in the current season but also in the future seasons due

to perceived tanking behavior.

The above findings that the increasing awareness of tanking undermines

consumer interest in both the short term and long term are worth further discussion. The

prior studies of tanking noted that tanking behavior in sport damaged the quality of sports

competitions, which generally encompasses the quality of sports teams and game

uncertainty (McManus, 2019; Price et al., 2010; Soebbing & Mason, 2009; Vamplew,

2018). The current dissertation offers additional evidence of the adverse impact of

perceived tanking on consumer demand for sport. Considering teams do not publicly

admit tanking, fans may not always correctly recognize tanking behavior from poor team

performance (Paxton, 2018). Thus, how fans perceive the idea of tanking may become a

107

critical factor in their attendance decisions. The findings from the regression analysis

draw upon the previous research of demand for sport and offer evidence that the more

fans are aware of the existence of tanking, the lower demand for sport will appear.

Second, my models conclude that neither the short-term nor long term sentiment

of tanking tweets for home teams has an impact on attendance in the NBA. Recall that

sentiment variables measure the fan attitudes toward the idea of tanking. These findings

suggest that even if fans hold more positive attitudes toward tanking for home teams,

game attendance will remain the same. Similarly, fans with more negative attitudes on

tanking for home teams do not seem to lower game attendance. This finding seems

surprising as numerous studies concluded that consumer sentiment is a key predictor of

purchase behavior (Huth, Eppright, & Taube, 1994; Nguyen & Claus, 2013). One

possibility is that some fans develop habit to attend sporting events even they hold

negative sentiment on tanking behavior (Ge, Humphreys, & Zhou, 2020). Another

possible explanation of this result is that fans may not be interested in attending games

featuring tanking teams, despite their support of the idea of tanking for home teams. For

instance, fans may root for home teams to deliberately lose games for draft picks which

can help regain competitiveness in the near future (Wade, 2013). Yet, such positive views

on tanking do not create more incentives for fans to attend games in the current and next

season.

Finally, this dissertation concludes that consumer perceptions of tanking for

visiting teams will affect attendance decisions. Specifically, the data analysis reveals that

there is a statistically positive relationship between the short-term sentiment of tanking

tweets for visiting teams and demand for NBA games. There were concerns that tanking

108

teams may create negative externalities when they play road games (Whitney, 2005).

That is, tanking teams undermining consumer interest in NBA games may incur

additional costs to the rest of the league without being penalized (Humphreys & Nowak,

2017). Both positive and negative externalities are common in the sport industry

(Downward & Rasciute, 2011). For instance, Humphreys and Johnson (2020) noted that

NBA super stars such as Michael Jordan, Magic Johnson, and Larry Bird could produce

significant positive externalities to the rest of the league. When NBA star players play

road games, the number of tickets sold in opponents’ arenas are expected to spike. Yet,

opponents benefiting from more ticket sales do not pay salaries for these star players

(Humphreys & Johnson, 2020). Similarly, the current dissertation finds evidence of

negative externalities of perceived tanking behavior related to visiting teams. If local fans

have strong negative views toward the opponent’s tanking strategy prior to an NBA

match, then attendance will be lower even though tanking is not directly linked to home

teams.

In addition to the two variables of interest, this dissertation reveals important

findings in control variables that are worth discussion. First, despite some variations in

results showing the impact of facility age on NBA attendance, the general trend suggests

that consumer interest in NBA games declines as sports facilities age. In particular, the

full model utilizing the entire dataset of NBA games ranging from the 2013-14 to 2017-

2018 seasons shows the positive estimator on the linear term of facility age and the

insignificant coefficient on the quadratic term (Leadley & Zygmont, 2005). This finding

shows NBA attendance will fall as NBA arenas becomes older, which is similar to what

Coates and Humpreys (2005) estimated. Second, most models in this dissertation show

109

the estimated coefficients on the quadratic term of point spreads are positive and

significant and are insignificant on the linear term. Such results suggest a convex

relationship between game uncertainty and demand for NBA games. In other words, this

conclusion indicates that local fans may like predictable games more than unpredictable

matches, which is consistent with what Coates et al. (2014) and Demmert (1973)

suggested but does not support the UOH. Third, the market competition, quantified by the

number of MLB, NFL, and NHL teams in the same market as an NBA team, does not

show a significant relationship with NBA attendance in most models. On the contrary,

the full model reveals a negative link, which is consistent with the conclusion drawn from

many prior studies (e.g., Demmert, 1973; Whitney, 2005). The different estimation

outcomes may arise from the fact that some models in this dissertation use subsets of

NBA games. Considering that the market competition variable does not change

significantly across the seasons, these models using subsets of data may not contain

sufficient variations in the market competition variable, and thus produce insignificant

estimators. Nevertheless, the results in the full model offer similar findings as preceding

research that market competition may limit consumer interest in NBA games (Morse,

Shapiro, McEvoy, & Rascher, 2007). Fourth, this inquiry finds control variables such as

team quality, population, economic conditions, and holiday games largely have positive

relationships with consumer demand in the NBA. These findings have similar

conclusions as previous studies (Hansen & Gauthier, 1989; Jane, 2016; Leadley &

Zygmont, 2005).

110

5.2 THEORETICAL CONTRIBUTIONS

This dissertation makes several theoretical contributions. First, it contributes to

the literature of demand for sport by exploring the impact of tanking in sport, behavior

that may undermine the integrity of competition (Borland & MacDonald, 2003; Villar &

Guerrero, 2009). While prior studies noted the importance of understanding the effect of

unethical team behavior such as tanking and match-fixing on consumer demand

(McManus, 2019; Preston & Szymanski, 2003), none of them attempted to quantitatively

measure such impacts. The controversial but prevailing topic tanking serves as a great

means to understand how sports fans will respond to unethical behavior in sport

(Howman, 2012; Wolfers, 2006). Hence, the current dissertation develops a new set of

factors estimating perceptions of tanking and uses the NBA as a context to systematically

analyze how these variables may affect NBA attendance. As such, the findings from the

current disseratation strengthen the theoretical understanding of demand for sport in

relation to unethical team behavior.

Second, the current dissertation extends previous examinations of demand for

NBA attendance (Hausman & Leonard, 1997; Leadley & Zygmont, 2005; Mills & Fort,

2014). Despite research interests in understanding factors that affect consumer demand

for sport, the recent studies of demand for NBA attendance seem scant. This dissertation

employs the attendance data from five recent NBA seasons between 2013-2014 to 2017-

18 and re-examines a wide range of variables that are deemed as key determinants of

NBA attendance. The findings suggest that variables such as facility age quality, game

uncertainty, team quality, population, economic conditions, and holiday games are

significant predictors of attendance decisions for NBA games, largely echoing previous

111

test results (Hansen & Gauthier, 1989; Morse et al., 2007). As such, the present

dissertation advances the literature of NBA attendance by examining key determinants of

demand for NBA games with more recent data.

Third, this dissertation provides theoretical contributions in understanding tanking

behavior in sport. Prior tanking research primarily focused on developing evidence of

tanking in sport (Fornwagner, 2019; Price et al., 2010; Soebbing & Humphreys, 2013;

Taylor & Trogdon, 2002). For instance, the seminal tanking study conducted by Taylor

and Trogdon (2002) concluded NBA teams would attempt to purposely lose games after

being removed from playoff contention. Despite the extensive examinations of tanking in

sport, none of the earlier research studies how consumers may respond to the idea of

tanking (Borland et al., 2009; Price et al., 2010; Soebbing & Humphreys, 2013; Taylor &

Trogdon, 2002). The present dissertation explores the relationship between perceptions of

tanking and consumer interest in sport and is the first one to quantitatively explore

consumer reaction to tanking behavior in sport. As such, this dissertation adds significant

theoretical contributions to the tanking literature.

Fourth, this dissertation advances sport management research of sentiment

analysis on social media. With the increased use of social media, a range of sports

research have begun to examine content from social media such as Twitter, Facebook,

and Instagram, in order to form a better understanding of consumer sentiment (Abeza,

O’Reilly, & Reid, 2013; Filo et al., 2015; Yan, Pegoraro, & Watanabe, 2018; Yan,

Watanabe, Shapiro, Naraine, & Hull, 2019). Unlike previous sport social media studies

that predominately use lexicon-based methods to estimate sentiments from social media

posts, this dissertation adopts machine learning models to extract sentiments from social

112

media content (Chang, 2019). Using social media Twitter combined with sentiment

analysis techniques, my work reveals the change of consumer perceptions of tanking over

time, and offers a rigorous measure of fans’ reactions to underperformance in sport.

5.3 PRACTICAL IMPLICATIONS

Through constructing short-term and long-term variables quantifying consumer

perceptions of tanking, this dissertation develops robust evidence regarding how

consumer interest in sport will be affected by perceived tanking behavior. Such evidence

may provide several important practical implications for policymakers. First, my work

indicates that both the increasing awareness of tanking for home teams and negative

attitudes toward tanking related to visiting teams can significantly undermine consumer

interest in attending sporting events. Furthermore, it is evident that not only does the

extensive discussion of tanking prior to NBA matches damage fan interest, but also the

tanking discussions from the last season damage interest too. This finding suggests the

widespread discussion of tanking around the league can produce both instant and

persistent negative impacts on demand for sport. To maintain consumer interest in

sporting contests, it is, therefore, critical for policy makers to restrain the speculation of

tanking and retain a less negative image of tanking around the league. For instance, the

NBA fined the Dallas Mavericks owner Mark Cuban $600,000 for publicly admitting

tanking in the 2017-2018 NBA season (NBA, 2018). Such penalizations may effectively

reduce the broad discussion of tanking among fans and create a more positive league

image, and thus can help avoid a drastic falloff of demand for NBA games due to

perceived tanking behavior (Paxton, 2018).

113

Second, this dissertation offers a tool for practitioners to monitor consumer

perceptions of tanking. Using social media platform Twitter, my work concludes that fans

in general hold negative views toward tanking and the sentiment seems to improve over

the seasons. Additionally, compared to other NBA seasons included in the dataset, the

2013-2014 and 2017-2018 NBA seasons witnessed the extensive discussion of tanking.

The above conclusions show Twitter could be considered as an effective medium to

understand how fans perceive the concept of tanking (Yu & Wang, 2015). Furthermore,

my work also provides detailed steps of calculating the volume of tanking tweets as well

as the process of using machine learning models to extract sentiments from Twitter posts

(Schuckert, Liu, & Law, 2015; Xiang, Du, Ma, & Fan, 2017). Taking the volume and

sentiment of tanking tweets together, practitioners can form a holistic view of consumer

perceptions of tanking that may help improve decision makings related to tanking

behavior in sport.

114

CHAPTER 6

CONCLUSION

6.1 SUMMARY

This dissertation centers on examining the behavior of sports teams tanking for

draft picks and thoroughly analyzes the impact of consumer perceptions of tanking on

demand for sport. To successfully measure perceptions of tanking, the current

dissertation used social media posts from Twitter, an online platform where people

express thoughts and opinions on various topics (Morstatter, Pfeffer, Liu, & Carley,

2013). By analyzing tanking-related Twitter posts, the two variables are proposed to

represent consumer perceptions of tanking, the volume and sentiment of tanking tweets.

Notably, the volume of tanking tweets reflects the fans’ awareness of tanking and is

calculated by counting the number of tanking tweets posted over a period of time. The

data analysis of this variable indicates the 2013-2014 and 2017-2018 NBA seasons

contained considerately more tanking discussions on Twitter than other seasons examined

within this dissertation. This observation signals that fans are well aware of the existence

of tanking in these two NBA seasons.

The second variable, the sentiment of tanking tweets, represents consumer

sentiment on tanking and is quantified by calculating the average sentiment expressed in

tanking tweets. Sentiment analysis with machine learning models is performed to classify

each tanking tweet into one of the sentiment classes, positive, neutral, and negative

(Kouloumpis, Wilson, & Moore, 2011). The sentiment variable then counts the average

115

sentiment from tanking tweets posted within a particular period of time. Overall, the

Twitter data suggests that a considerable portion of tanking tweets contain negative

sentiments, meaning fans in general hold negative views toward tanking. Additionally,

the season average sentiment on tanking shows continues improvement, implying that

fans become more acceptable to the idea of tanking over the years.

After building two variables measuring consumer perceptions of tanking, I

estimate their relationship with NBA attendance. The regression analysis offers clear

evidence that the short-term volume of tanking tweets for home teams prior to an NBA

game is negatively related to attendance decisions. This evidence suggests the extensive

discussion of tanking related to home teams significantly affect ticket sales in the short

run. This dissertation also finds a lasting effect of the volume of tanking discussions for

home teams on demand for NBA games. Specifically, the results conclude that the broad

discussion of tanking related to home teams from the prior NBA season can significantly

reduce consumer demand for games in the current season, indicating that the adverse

effect of the volume of tanking discussions on attendance can persist for a while.

While the volume of tanking tweets for home teams has significant impacts on

NBA attendance in both the short term and long term, the sentiment of tanking

discussions for home teams does not appear to have an effect on demand for NBA games.

The results from a number of regression models suggest that neither the short-term nor

long-term sentiment of tanking tweets for home teams is statistically related to attendance

decisions. Although Twitter data collected in this dissertation shows that fans develop

more positive attitudes toward tanking over the seasons, the findings from statistical

116

models predict that improved sentiment on tanking does not seem to help attract fans to

NBA arenas.

This dissertation further analyzes how consumer perceptions of tanking for

visiting teams can affect NBA attendance. Specifically, econometric models reveal that

only the short-term sentiment of tanking discussions for visiting teams seems to be

significant in positively changing demand for NBA matches among all variables

examined. That is, local fans who hold more negative views toward visiting teams’

tanking behavior prior to an NBA match are less likely to attend games. This suggests

that perceived tanking behavior for away teams can produce the adverse impact on

demand for home games.

6.2 LIMITATIONS

The current dissertation is not without limitations. First, selection bias may exist

in Twitter data (Culotta, 2014). While the number of Twitter users has grown

tremendously over the past few years, the cohort may not represent the general

population (Bae & Lee, 2012). For instance, it is estimated the Twitter users are overall

younger and wealthier than the U.S. public (Wojcik & Hughes, 2019). As such, Twitter

posts may not fully reflect consumer perceptions of tanking. Despite this concern, a

thorough literature review shows that Twitter posts can effectively gauge consumer

sentiment (Daniel, Neves, & Horta, 2017). For instance, a wide range of research

studying the stock market revealed that investor sentiment calculated from Twitter posts

could successfully predict stock price (McGurk et al., 2019). These inquiries provide

evidence that opinions and feelings on social media platforms can as least partially

117

exhibit market sentiment. Therefore, I strongly believe that tanking posts gathered from

Twitter well represent the fans’ feelings of tanking.

Second, in performing sentiment analysis, the initial step of manually labeling the

sentiment of tanking tweets may involve some subjectivity (Liu, 2010). To ensure the

quality of manual classification, a few cautious steps were taken in the labeling process

(McGurk et al., 2019). First, I invite an expert with strong domain knowledge to label

tanking tweets. Second, the coder is asked to spend at least 30 seconds on labeling each

tweet. Third, repeated tweets are randomly assigned to label. Any repeated tweets with

inconsistent labels will be removed from the sample. The results show that only 0.4% of

tweets are wrongly labeled and thus discarded. Fourth, I asked another coder who does

not have a strong background in tanking to label the same set of tweets as the main coder.

Comparing labeling results shows that 42% of labels are matched, meaning there is some

degree of consensus between two coders on the sentiment of tanking tweets. However, as

the expert has more domain knowledge in tanking than the other coder, its labels are

eventually used in the training set. Despite these precautions, it is critical to note that

some subjectivity may still exist in the labeling process.

6.3 FUTURE STUDIES

There are a number of areas for future research that emerges from the topics in

this dissertation. Future study could explore new draft policies that will mitigate tanking

issues while reserving their original functions of maintaining competitive balance in

sports leagues. The present dissertation develops evidence that consumer perceptions of

tanking can hurt attendance. As such, it is crucial for sports leagues to amend existing

draft policies to restrain tanking behavior in sport. However, it is equally important to

118

note that draft systems were initially designed to improve competitive balance in sport,

which is a key determinant of consumer demand for sport (Grier & Tollison, 1994;

Sanderson & Siegfried, 2003). Therefore, future studies may carefully design new draft

systems that can not only serve as a scheme to maintain competitive balance but also

discourage tanking behavior in sport.

Future studies may also explore the effectiveness of the new draft policy instituted

in 2019. Recall in Figure 1.1, the new NBA draft system reduces the possibility for the

worst record teams to receive the No.1 draft pick from 25% to 14%. While the new

system intends to diminish tanking incentives in the NBA, it is unclear whether it will

work as planned. Therefore, future research may use team performance data to further

investigate whether lowering the probability of acquiring the No.1 draft pick will

discourage tanking behavior in the NBA.

In addition, future research can follow this dissertation to examine the relationship

between league policies designed to discourage tanking behavior and consumer sentiment

regarding tanking. For instance, the NBA league office instituted the play rest policy in

2017 and fined Mark Cuban $600,000 for publicly admitting tanking in 2018. Such

actions aim to preserve a positive image of the league regarding tanking. Yet, it is unclear

whether these efforts are effective in changing fans’ minds regarding the league’s tanking

problem. The present dissertation uses the sentiment of tanking tweets to quantify

consumer sentiment toward tanking and can also be used to examine the impact of the

league’s tanking related policies on fans’ sentiment toward tanking.

119

REFERENCES

Abeza, G., O’Reilly, N., & Reid, I. (2013). Relationship marketing and social media in

sport. International Journal of Sport Communication, 6, 120-142.

doi:10.1123/ijsc.6.2.120

Aday, J. B., & Phelan, K. V. (2016). Attitudes, perceptions, and motivations of eco-

sustainable NBA fans. e-Journal of Social & Behavioural Research in Business, 7,

1-16.

Ahn, S.C., & Lee, Y.H. (2007). Life-cycle demand for Major League Baseball.

International Journal of Sport Finance, 2, 79-93.

Aldridge, D. (2018, March 05). Mark Cuban's tanking talk (and NBA-issued fine) speaks

of larger issues in league. NBA. Retrieved from

http://www.nba.com/article/2018/03/05/state-nba-adam-silver-memo-tanking-

2017-18-mark-cuban-fined-600000

Amemiya, T. (1973). Regression analysis when the dependent variable is truncated

normal. Econometrica: Journal of the Econometric Society, 41, 997-1016.

Amick, S. (2014, February 08). Ex-Raptors GM Bryan Colangelo: 'I tried to tank'. USA

Today. Retrieved from:

https://www.usatoday.com/story/sports/nba/raptors/2014/02/28/bryan-colangelo-

admits-tanking-toronto-raptors/5908983/

Amick, S. (2018, February 28). Adam Silver cautions NBA teams to protect integrity of

game. USA Today. Retrieved from

120

https://www.usatoday.com/story/sports/nba/2018/02/28/adam-silver-cautions-nba-

teams-against-tanking-protect-integrity-game/380197002/

Anderson, W. B. (2001). Saving the national pastime's image: Crisis management during

the 1919 Black Sox scandal. Journalism History, 27, 105-111.

Aydoğan, E., & Akcayol, M. A. (2016). A comprehensive survey for sentiment analysis

tasks using machine learning techniques. 2016 International Symposium on

INnovations in Intelligent SysTems and Applications (INISTA), 1-7.

Bae, Y., & Lee, H. (2012). Sentiment analysis of Twitter audiences: Measuring the

positive or negative influence of popular twitterers. Journal of the American

Society for Information Science and Technology, 63, 2521-2535.

doi:10.1002/asi.22768

Balsdon, E., Fong, L., & Thayer, M. A. (2007). Corruption in college basketball?

Evidence of tanking in postseason conference tournaments. Journal of Sports

Economics, 8, 19-38. doi:10.1177/1527002505275095

Benz, M. A., Brandes, L., & Franck, E. (2006). Heterogeneity in fan demand-new results

on uncertainty of outcome from quantile regression (Working Paper No.

48). Institute for Strategy and Business Economics, University of Zurich.

Retrieved from https://www.researchgate.net/profile/Men-

Andri_Benz/publication/229004136_Heterogeneity_in_fan_demand-

new_results_on_uncertainty_of_outcome_from_quantile_regression/links/54db6e

360cf261ce15d0225f.pdf

Berri, D. J., & Krautmann, A. C. (2006). Shirking on the court: Testing for the incentive

effects of guaranteed pay. Economic Inquiry, 44, 536-546. doi:10.1093/ei/cbj033

121

Berri, D. J., & Schmidt, M. B. (2006). On the road with the National Basketball

Association's superstar externality. Journal of Sports Economics, 7, 347-358.

doi:10.1177/1527002505275094

Berri, D. J., Schmidt, M. B., & Brook, S. L. (2004). Stars at the gate: The impact of star

power on NBA gate revenues. Journal of Sports Economics, 5, 33-50.

doi:10.1177/1527002503254051

Bieler, D. (2018 February 20). Mark Cuban fined $600,000 for saying his Mavericks plan

on ‘tanking’. The Washington Post. Retrieved from

https://www.washingtonpost.com/news/early-lead/wp/2018/02/20/mark-cuban-

admits-that-tanking-is-the-best-option-for-his-18-40-

mavericks/?utm_term=.28efc25f4992

Billings, A. C., Broussard, R. M., Xu, Q., & Xu, M. (2019). Untangling international

sport social media use: Contrasting US and Chinese uses and gratifications across

four platforms. Communication & Sport, 7, 630-652.

doi:10.1177/2167479518790014

Blair, K. (2018). The 2012 Olympic badminton scandal: Match-fixing, code of conduct

documents, and women’s sport. The International Journal of the History of Sport,

35 (2-3), 264-276. doi:10.1080/09523367.2018.1451330

Bondy, F. (2007). Tip-off: How the 1984 NBA Draft Changed Basketball Forever.

Cambridge, MA: Da Capo Press.

Borland, J., & MacDonald, R. (2003). Demand for sport. Oxford Review of Economic

Policy, 19, 478-502. doi:10.1093/oxrep/19.4.478

122

Borland, J., Chicu, M., & Macdonald, R. D. (2009). Do teams always lose to win?

Performance incentives and the player draft in the Australian Football League.

Journal of Sports Economics, 10, 451-484. doi:10.1177/1527002509331615

Boyle, R., & Haynes, R. (2002). New media sport. Sport in Society, 5, 96-114.

doi:10.1080/911094209

Bradbury, J. C., & Drinen, D. (2006). The designated hitter, moral hazard, and hit batters:

New evidence from game-level data. Journal of Sports Economics, 7, 319-329.

doi:10.1177/1527002504273391

Brandes, L., Franck, E., & Nüesch, S. (2008). Local heroes and superstars: An empirical

analysis of star attraction in German soccer. Journal of Sports Economics, 9, 266-

286. doi:10.1177/1527002507302026

Buraimo, B., & Simmons, R. (2008). Do sports fans really value uncertainty of outcome?

Evidence from the English Premier League. International Journal of Sport

Finance, 3, 146–155.

Cairns, J., Jennett, N., & Sloane, P. J. (1986). The economics of professional team sports:

A survey of theory and evidence. Journal of Economic Studies, 13, 3-80.

Carson, R. T., Cenesizoglu, T., & Parker, R. (2011). Forecasting (aggregate) demand for

US commercial air travel. International Journal of Forecasting, 27, 923-941.

doi:10.1016/j.ijforecast.2010.02.010

Chang, Y. (2019). Spectators’ emotional responses in tweets during the Super Bowl 50

game. Sport Management Review, 22, 348-362. doi:10.1016/j.smr.2018.04.008.

123

Chmait, N., Robertson, S., Westerbeek, H., Eime, R., Sellitto, C., & Reid, M. (2020).

Tennis superstars: The relationship between star status and demand for tickets.

Sport Management Review, 23, 330-347. doi:10.1016/j.smr.2019.03.006

Clapp, C.M., & Hakes, J. K. (2005). How long a honeymoon? The effect of new stadiums

on attendance in Major League Baseball. Journal of Sports Economics, 6, 237-

263. doi:10.1177/1527002504265957

Coates, D., & Humphreys, B. R. (2002). The economic impact of postseason play in

professional sports. Journal of Sports Economics, 3, 291-299.

doi:10.1177/1527002502003003005

Coates, D., & Humphreys, B. R. (2005). Novelty effects of new facilities on attendance at

professional sporting events. Contemporary Economic Policy, 23, 436-455.

doi:10.1093/cep/byi033

Coates, D., & Humphreys, B. R. (2007). Ticket prices, concessions and attendance at

professional sporting events. International Journal of Sport Finance, 2, 161-170.

Coates, D., Humphreys, B. R., & Zhou, L. (2014). Reference‐dependent preferences, loss

aversion, and live game attendance. Economic Inquiry, 52, 959-973.

doi:10.1111/ecin.12061

Connelly, B. L., Tihanyi, L., Crook, T. R., & Gangloff, K. A. (2014). Tournament theory:

Thirty years of contests and competitions. Journal of Management, 40, 16-47.

doi:10.1177/0149206313498902

Culotta, A. (2014). Reducing sampling bias in social media data for county health

inference. Joint Statistical Meetings Proceedings, 1-12.

124

Daniel, M., Neves, R.F., & Horta, N. (2017). Company event popularity for financial

markets using Twitter and sentiment analysis. Expert Systems with Applications,

71, 111-124. doi:10.1016/j.eswa.2016.11.022

Davenport, S. W., Bergman, S. M., Bergman, J. Z., & Fearrington, M. E. (2014). Twitter

versus Facebook: Exploring the role of narcissism in the motives and usage of

different social media platforms. Computers in Human Behavior, 32, 212-220.

doi:10.1016/j.chb.2013.12.011

Demmert, H. G. (1973). The economics of professional team sports. Lexington, Mass:

Lexington Books.

DeSchriver, T. D., Rascher, D. A., & Shapiro, S. L. (2016). If we build it, will they come?

Examining the effect of expansion teams and soccer-specific stadiums on Major

League Soccer attendance. Sport, Business and Management, 6, 205-227.

Dorsey, J. (2012, March 27). The biggest tank jobs in NBA history. Bleacher Report.

Retrieved from https://bleacherreport.com/articles/1120733-the-biggest-tank-jobs-

in-nba-history

Downward, P., & Rasciute, S. (2011). Does sport make you happy? An analysis of the

well‐being derived from sports participation. International Review of Applied

Economics, 25, 331-348. doi:10.1080/02692171.2010.511168

Duvall, A. (2018, April 17). Major League Baseball has a tanking issue that needs to be

addressed. SB Nation. Retrieved from

https://www.royalsreview.com/2018/4/17/17245390/major-league-baseball-has-a-

tanking-issue-that-needs-addressing

125

ESPN. (2013 May 14). 76ers formally hire Sam Hinkie. ESPN. Retrieved from

https://www.espn.com/nba/story/_/id/9275350/sam-hinkie-officially-introduced-

philadelphia-76ers-team-president-general-manager

Farhadloo, M., & Rolland, E. (2013). Multi-class sentiment analysis with clustering and

score representation. 2013 IEEE 13th international conference on data mining

workshops, 904-912

Filo, K., Lock, D., & Karg, A. (2015). Sport and social media research: A review. Sport

Management Review, 18, 166-181. doi: 10.1016/j.smr.2014.11.001

Fornwagner, H. (2019). Incentives to lose revisited: The NHL and its tournament

incentives. Journal of Economic Psychology, 75(Part B), 102088.

doi:10.1016/j.joep.2018.07.004

Forrest, D., & Simmons, R. (2002). Outcome uncertainty and attendance demand in sport:

The case of English soccer. Journal of the Royal Statistical Society: Series D (The

Statistician), 51, 229-241. doi:10.1111/1467-9884.00314

Forrest, D., Goddard, J., & Simmons, R. (2005). Odds-setters as forecasters: The case of

English football. International Journal of Forecasting, 21, 551-564.

doi:10.1016/j.ijforecast.2005.03.003

Fort, R. D. (2006). Sports economics. Upper Saddle, NJ: Prentice Hall.

Fort, R. D., & Quirk, J. (1995). Cross-subsidization, incentives, and outcomes in

professional team sports leagues. Journal of Economic Literature, 33, 1265-1299.

Fort, R., & Rosenman, R. (1999). Streak management. In J. Fizel J, E. Gustafson, L.

Hadley (Eds.), Sports economics: Current research (pp. 119-134). Westport, CT:

Greenwood.

126

Fort, R., Maxcy, J., & Diehl, M. (2016). Uncertainty by regulation: Rottenberg’s

invariance principle. Research in Economics, 70, 454-467.

doi:10.1016/j.rie.2016.06.004

Ge, Q., Humphreys, B.R., & Zhou, K. (2020). Are fair weather fans affected by weather?

Rainfall, habit formation, and live game attendance. Journal of Sports Economics,

21, 304-322.

Gerchak, Y., Mausser, H. E., & Magazine, M. J. (1995). The evolution of draft lotteries

in professional sports: Back to moral hazard?. Interfaces, 25, 30-38.

doi:10.1287/inte.25.6.30

Gitter, S. R., & Rhoads, T. A. (2010) Determinants of minor league baseball attendance.

Journal of Sports Economics, 11, 614-628. doi:10.1177/1527002509359758

Goldhaber, D., & Chaplin, D. D. (2015). Assessing the “Rothstein Falsification Test”:

Does it really show teacher value-added models are biased?. Journal of Research

on Educational Effectiveness, 8, 8-34. doi:10.1080/19345747.2014.978059

Greene, W. (2004). Fixed effects and bias due to the incidental parameters problem in the

Tobit model. Econometric Reviews, 23, 125-147. doi:10.1081/ETC-120039606

Greenstein, T. (2019 May 25). Who is Duke superstar Zion Williamson? 'That dude is a

franchise-changer'. Chicago Tribune. Retrieved from

https://www.chicagotribune.com/sports/ct-spt-cb-who-is-zion-williamson-duke-

nba-draft-20190207-story.html

Grier, K. B., & Tollison, R. D. (1994). The rookie draft and competitive balance: The

case of professional football. Journal of Economic Behavior & Organization, 25,

293-298. doi: 10.1016/0167-2681(94)90016-7

127

Hansen, H., & Gauthier, R. (1989). Factors affecting attendance at professional sport

events. Journal of Sport Management, 3, 15-32. doi:10.1123/jsm.3.1.15

Hausman, J. A., & Leonard, G. K. (1997). Superstars in the National Basketball

Association: Economic value and policy. Journal of Labor Economics, 15, 586-

624. doi:10.1086/209839.

Hill, J. R., Madura, J., & Zuber, R. A. (1982). The short run demand for Major League

Baseball. Atlantic Economic Journal, 10, 31-35.

Hinkie, S. (2016, April 6). To the equity partners of Philadelphia 76ers, L.P. Retrieved

from: https://www.espn.com/pdf/2016/0406/nba_hinkie_redact.pdf.

Hong, Y., & Skiena, S. (2010). The wisdom of bookies? Sentiment analysis versus. the

NFL point spread. Fourth International AAAI Conference on Weblogs and Social

Media, 251-254.

Howman, D. (2012) Supporting the integrity of sport and combating corruption.

Marquette Sports Law Review, 23, 245-248.

Humphreys, B. R. (2002). Alternative measures of competitive balance in sports leagues.

Journal of Sports Economics, 3, 133-148. doi:10.1177/152700250200300203

Humphreys, B. R., & Johnson, C. (2020). The effect of superstars on game attendance:

Evidence from the NBA. Journal of Sports Economics, 21, 152-175.

doi:10.1177/1527002519885441

Humphreys, B. R., & Nowak, A. (2017). Professional sports facilities, teams and property

values: Evidence from NBA team departures. Regional Science and Urban

Economics, 66, 39-51. doi:10.1016/j.regsciurbeco.2017.06.001

128

Huth, W. L., Eppright, D. R., & Taube, P. M. (1994). The indexes of consumer sentiment

and confidence: Leading or misleading guides to future buyer behavior. Journal

of Business Research, 29, 199-206. doi:10.1016/0148-2963(94)90004-3

Hutter, K., Hautz, J., Dennhardt, S., & Füller, J. (2013). The impact of user interactions

in social media on brand awareness and purchase intention: The case of MINI on

Facebook. Journal of Product & Brand Management, 22, 342-351.

doi:10.1108/JPBM-05-2013-0299

Hynds, M., & Smith, I. (1994). The demand for test match cricket. Applied Economics

Letters, 1, 103-106. doi:10.1080/135048594358078

Jane, W. J. (2016). The effect of star quality on attendance demand: The case of the

National Basketball Association. Journal of Sports Economics, 17, 396-417.

doi:10.1177/1527002514530405

Jones, J. C. H. (1984). Winners, losers and hosers: Demand and survival in the National

Hockey League. Atlantic Economic Journal, 12, 54-63.

Kang, H., Yoo, S. J., & Han, D. (2012). Senti-lexicon and improved Naïve Bayes

algorithms for sentiment analysis of restaurant reviews. Expert Systems with

Applications, 39, 6000-6010.

Kendall, G., & Lenten, L. J. (2017). When sports rules go awry. European Journal of

Operational Research, 257, 377-394. doi:10.1016/j.ejor.2016.06.050

Késenne, S. (2000). The impact of salary caps in professional team sports. Scottish

Journal of Political Economy, 47, 422-430. doi:10.1111/1467-9485.00171

129

Kirilenko, A. P., Stepchenkova, S. O., Kim, H., & Li, X. (2018). Automated sentiment

analysis in tourism: Comparison of approaches. Journal of Travel Research, 57,

1012-1025. doi:10.1177/0047287517729757

Koning, R. H. (2000). Balance in competition in Dutch soccer. Journal of the Royal

Statistical Society: Series D (The Statistician), 49, 419-431. doi:10.1111/1467-

9884.00244

Kouloumpis, E., Wilson, T., & Moore, J. (2011, July). Twitter sentiment analysis: The

good the bad and the omg!. Paper presented at the 5th International AAAI

conference on weblogs and social media.

Krautmann, A. C. (1990). Shirking or stochastic productivity in Major League

Baseball?. Southern Economic Journal, 56, 961-968.

Krautmann, A. C., & Berri, D. J. (2007). Can we find it at the concessions?

Understanding price elasticity in professional sports. Journal of Sports

Economics, 8, 183-191. doi:10.1177/1527002505275093

Kumar, A., & Jaiswal, A. (2020). Systematic literature review of sentiment analysis on

Twitter using soft computing techniques. Concurrency and Computation:

Practice and Experience, 32, e5107. doi:10.1002/cpe.5107

Kywe, S. M., Lim, E. P., & Zhu, F. (2012). A survey of recommender systems in Twitter.

International Conference on Social Informatics, 7710, 420-433. doi:10.1007/978-

3-642-35386-4_3

Lancaster, T. (2000). The incidental parameter problem since 1948. Journal of

Econometrics, 95, 391-413. doi:10.1016/S0304-4076(99)00044-5

130

Lazear, E. P., & Rosen, S. (1981). Rank-order tournaments as optimum labor

contracts. Journal of Political Economy, 89, 841-864. doi:10.1086/261010

Leadley, J. C., & Zygmont, Z. X. (2005). When is the honeymoon over? National

Basketball Association attendance 1971-2000. Journal of Sports Economics, 6,

203-221. doi:10.1177/1527002504263399

Lee, Y. H. (2004). Competitive balance and attendance in Japanese, Korean and U.S.

professional base-ball leagues. In R. Fort & J. Fizel (Eds.), International sports

economics comparisons (pp. 281–292). Westport, CT: Praeger Publishers.

Lee, Y. H. (2006). The decline of attendance in the Korean Professional Baseball League:

The major league effects. Journal of Sports Economics, 7, 187-200.

doi:10.1177/1527002504271355

Lee, Y. H. (2009). The impact of postseason restructuring on the competitive balance and

fan demand in Major League Baseball. Journal of Sports Economics, 10, 219-235.

doi:10.1177/1527002508325785

Lee, Y. H., & Fort, R. (2008). Attendance and the uncertainty-of-outcome hypothesis in

baseball. Review of Industrial Organization, 33, 281-295. doi:10.1007/s11151-

008-9193-9

Lee, Y. H., Kim, Y., & Kim, S. (2019). A bias-corrected estimator of competitive balance

in sports leagues. Journal of Sports Economics, 20, 479–508.

doi:10.1177/1527002518777974

Lenten, L. J. (2016). Mitigation of perverse incentives in professional sports leagues with

reverse-order drafts. Review of Industrial Organization, 49, 25-41.

doi:10.1007/s11151-015-9494-8

131

Lenten, L. J., Smith, A. C., & Boys, N. (2018). Evaluating an alternative draft pick

allocation policy to reduce ‘tanking’ in the Australian Football League. European

Journal of Operational Research, 267, 315-320. doi:10.1016/j.ejor.2017.11.029

Liu, B. (2010). Sentiment analysis and subjectivity. Handbook of Natural Language

processing, 2, 627-666.

Lowe, Z. (2017, September 28). Blatant tanking to get slightly less valuable for some

NBA teams. ESPN. Retrieved from

http://www.espn.com/nba/story/_/id/20849861/zach-lowe-nba-draft-lottery-

reform-potential-ramifications

Mason, D. (1999). What is the sports product and who buys it? The marketing of

professional sports leagues. European Journal of Marketing, 33, 402-419.

doi:10.1108/03090569910253251

McClure, J. E., & Spector, L. C. (1997). Tournament performance and “Agency”

problems: An empirical investigation of “March Madness”. Journal of Economics

and Finance, 21, 61-68. doi:10.1007/BF02929023

McDonald, M., & Rascher, D. (2000). Does bat day make cents? The effect of

promotions on the demand for Major League Baseball. Journal of Sport

Management, 14, 8-27. doi:10.1123/jsm.14.1.8

McEvoy, C. D., Nagel, M. S., DeSchriver, T. D., & Brown, M. T. (2005). Facility age

and attendance in Major League Baseball. Sport Management Review, 8, 19-41.

doi:10.1016/S1441-3523(05)70031-0

132

McGurk, Z., Nowak, A. & Hall, J.C. (2019). Stock returns and investor sentiment:

Textual analysis and social media. Journal of Economics and Finance, 44, 458-

485. doi:10.1007/s12197-019-09494-4

McLaren, R. H. (2011). Is sport losing its integrity?. Marquette Sports Law Review, 21,

551-573.

McManus, J. (2019). Ethical considerations & the practice of tanking in Sport

Management. Sport, Ethics and Philosophy, 13, 145-160.

doi:10.1080/17511321.2018.1483418

Meehan, J. W. Jr, Nelson, R. A., & Richardson, T. V. (2007). Competitive balance and

game attendance in Major League Baseball. Journal of Sports Economics, 8, 563-

580. doi:10.1177/1527002506296545

Miller, S. (2018 March 23). In 2018, what is a win? A shift in philosophy has changed

MLB forever. ESPN. Retrieved from

http://www.espn.com/mlb/story/_/id/22815820/houston-astros-chicago-cubs-set-

path-winning-losing

Mills, B. M., Salaga, S., & Tainsky, S. (2016). NBA primary market ticket consumers:

Ex ante expectations and consumer market origination. Journal of Sport

Management, 30, 538-552. doi:10.1123/jsm.2015-0230

Mills, B., & Fort, R. (2014). League‐level attendance and outcome uncertainty in US pro

sports leagues. Economic Inquiry, 52, 205-218. doi:10.1111/ecin.12037

Mongeon, K., & Winfree, J. (2012). Comparison of television and gate demand in the

National Basketball Association. Sport Management Review, 15, 72-79.

doi:10.1016/j.smr.2011.09.001

133

Morse, A. L., Shapiro, S. L., McEvoy, C. D., & Rascher, D. A. (2007). The effects of

roster turnover on demand in the National Basketball Association. International

Journal of Sport Finance, 3, 8-18. doi:10.2139/ssrn.1690885

Morstatter, F., Pfeffer, J., Liu, H., & Carley, K. M. (2013, June). Is the sample good

enough? Comparing data from Twitter's streaming api with Twitter's firehose.

Paper presented at the 7th international AAAI conference on weblogs and social

media.

Motomura, A., Roberts, K. V., Leeds, D. M., & Leeds, M. A. (2016). Does it pay to build

through the draft in the National Basketball Association?. Journal of Sports

Economics, 17, 501-516. doi:10.1177/1527002516641169

Mullin. J. (2018 July 05). 'Rebuilding?' Tanking by any other name…. Bears set a better

example. NBC. Retrieved from

https://www.nbcsports.com/chicago/bears/rebuilding-tanking-any-other-name-

bears-set-better-example

NBA. (2017 October 1) NBA draft lottery history. NBA. Retrieved from

https://www.nba.com/history/draft/lottery

NBA. (2018 Feb 21). Dallas Mavericks owner Mark Cuban fined $600K for public

comments on tanking. NBA. Retrieved from

https://www.nba.com/article/2018/02/21/dallas-mavericks-owner-mark-cuban-

fined-comments-tanking

NBA. (2018 March 06). Reports: NBA warns Chicago Bulls about resting healthy players.

NBA. Retrieved from https://www.nba.com/article/2018/03/06/reports-nba-warns-

chicago-bulls-about-resting-healthy-players

134

Neale, W. C. (1964). The peculiar economics of professional sports. The Quarterly

Journal of Economics, 78, 1-14. doi:10.2307/1880543

Neethu, M. S., & Rajasree, R. (2013). Sentiment analysis in Twitter using machine

learning techniques. 2013 Fourth International Conference on Computing,

Communications and Networking Technologies (ICCCNT), 1-5.

Neyman, J., & Scott, E. L. (1948). Consistent estimates based on partially consistent

observations. Econometrica: Journal of the Econometric Society, 16, 1-32.

Nguyen, V. H., & Claus, E. (2013). Good news, bad news, consumer sentiment and

consumption behavior. Journal of Economic Psychology, 39, 426-438.

doi:10.1016/j.joep.2013.10.001

Noll, R. (1974). Government and the sports business. Washington, DC: Brookings

Institution.

Oga, J. (1998). Business fluctuation and the sport industry in Japan: An analysis of the

sport industry from 1986 to 1993. Journal of Sport Management, 12, 63-75.

doi:10.1123/jsm.12.1.63

Pang, B., & Lee, L. (2004). A sentimental education: Sentiment analysis using

subjectivity summarization based on minimum cuts. Proceedings of the 42nd

annual meeting on Association for Computational Linguistics, 271-278.

doi:10.3115/1218955.1218990

Paxton, A. T. (2018). Trust the process: How the NBA can combat its tanking problem in

court. Iowa Law Review, 104, 1551-1580.

Pegoraro, A. (2014). Twitter as disruptive innovation in sport communications.

Communication & Sport, 2, 132-137. doi:10.1177/2167479514527432

135

Preston, I., & Szymanski, S. (2003). Cheating in contests. Oxford Review of Economic

Policy, 19, 612-624. doi:10.1093/oxrep/19.4.612

Price, J., Soebbing, B. P., Berri, D., & Humphreys, B. R. (2010). Tournament incentives,

league policy, and NBA team performance revisited. Journal of Sports

Economics, 11, 117-135. doi:10.1177/1527002510363103

Quinn, K. G. (2009). Sports and their fans: The history, economics and culture of the

relationship between spectator and sport. Jefferson, N.C: McFarland & Co.

Quirk, J., Fort, R. D (1997). Pay dirt: The business of professional team sports. Princeton,

NJ: Princeton University Press.

Rappaport, M. (2017, August 23). The definitive history of `trust the process’. Bleacher

Report. Retrieved from: https://bleacherreport.com/articles/2729018-the-

definitive-history-of-trust-the-process

Rascher, D. (1999). A test of the optimal positive production network externality in major

league baseball. In J. Fizel, E. Gustafson, & L. Hadley (Eds.), Sports economics:

Current research (pp. 27-45). Westport, CT: Greenwood.

Rascher, D. A., Brown, M. T., Nagel, M. S., & McEvoy, C. D. (2009). Where did

National Hockey League fans go during the 2004-2005 lockout? An analysis of

economic competition between leagues. International Journal of Sport

Management and Marketing, 5, 183-195. doi:10.1504/IJSMM.2009.021758

Rascher, D., & Solmes, J. (2007). Do fans want close contests? A test of the uncertainty

of outcome hypothesis in the National Basketball Association. International

Journal of Sport Finance, 2, 130-141.

136

Reams, L., & Shapiro, S. (2017). Who's the main attraction? Star power as a determinant

of Ultimate Fighting Championship pay-per-view demand. European Sport

Management Quarterly, 17, 132-151. doi:10.1080/16184742.2016.1257039

Rivers, D. H., & Deschriver, T. D. (2002). Star players, payroll distribution, and Major

League Baseball attendance. Sport Marketing Quarterly, 11, 164-173.

Robinson, M. J., & DeSchriver, T. D. (2003). Consumer differences across large and

small market teams in the National Professional Soccer League. Sport Marketing

Quarterly, 12, 80-87.

Rottenberg, S. (1956). The baseball players' labor market. Journal of Political

Economy, 64, 242-258. doi:10.1086/257790

Sanderson, A. R., & Siegfried, J. J. (2003). Thinking about competitive balance. Journal

of Sports Economics, 4, 255-279.doi:10.1177/1527002503257321

Schofield, J. A. (1983). Performance and attendance at professional team sports. Journal

of Sport Behavior, 6, 196-206.

Schuckert, M., Liu, X., & Law, R. (2015). Hospitality and tourism online reviews: Recent

trends and future directions. Journal of Travel & Tourism Marketing, 32, 608-621.

doi:10.1080/10548408.2014.933154

Schumaker, R. P., Jarmoszko, A. T., & Labedz, C. S., Jr. (2016). Predicting wins and

spread in the Premier League using a sentiment analysis of Twitter. Decision

Support Systems, 88, 76-84. doi:10.1016/j.dss.2016.05.010

Shafer, J. (2018 February 23). Marlins' and Rays' embarrassing selloffs show MLB can't

survive in Florida. Bleacher Report. Retrieved from

137

https://bleacherreport.com/articles/2761058-marlins-and-rays-embarrassing-

selloffs-show-mlb-cant-survive-in-florida

Sheinin, D. (2018 June 22). Tanking in MLB has led to too many great teams, too many

awful ones, and a decline in fans. The Washington Post. Retrieved from

https://www.washingtonpost.com/news/sports/wp/2018/06/22/tanking-in-mlb-

has-led-to-too-many-great-teams-too-many-awful-ones-and-a-decline-in-

fans/?utm_term=.770e383703f4

Silver, N., & Fischer-Baum, R. (2015). How we calculate NBA Elo Ratings.

FiveThirtyEight. Retrieved from https://fivethirtyeight.com/features/how-we-

calculate-nba-elo-ratings/

Soebbing, B. P. and Humphreys, B. R. (2013). Do gamblers think that teams tank?

Evidence from the NBA. Contemporary Economic Policy, 31, 301–313.

doi:10.1111/j.1465-7287.2011.00298.x

Soebbing, B. P., & Mason, D. S. (2009). Managing legitimacy and uncertainty in

professional team sport: The NBA's draft lottery. Team Performance

Management: An International Journal, 15, 141-157.

doi:10.1108/13527590910964928

Soebbing, B. P., Humphreys, B. R., & Mason, D. S. (2013). Exploring incentives to lose

in professional team sports: Do conference games matter?. International Journal

of Sport Finance, 8, 192-207.

Steinberg, L. (2018, April 13). Was 'Tanking' the key to Philadelphia 76ers' success?.

Forbes. Retrieved from

138

https://www.forbes.com/sites/leighsteinberg/2018/04/13/was-tanking-the-key-to-

philadelphia-76ers-success/

Szymanski, S., & Késenne, S. (2004). Competitive balance and gate revenue sharing in

team sports. The Journal of Industrial Economics, 52, 165-177.

doi:10.1111/j.0022-1821.2004.00220.x

Tayade, M. C., Bhamare, S. M., Kamble, P., & Jadhav, K. (2013). Doping in sports:

Current review. International Journal of Current Research and Review, 5, 83-86.

Taylor, B. A., & Trogdon, J. G. (2002). Losing to win: Tournament incentives in the

National Basketball Association. Journal of Labor Economics, 20, 23-41.

doi:10.1086/323930

Vaccaro, M. (2019 March 01). Knicks’ disgraceful effort can’t be washed away by

tanking talk. New York Post. Retrieved from

https://nypost.com/2019/03/01/knicks-disgraceful-effort-cant-be-washed-away-

by-tanking-talk/

Vamplew, W. (2018). Tanking, shirking, and running dead: The role of economics and

large data sets in identifying competition corruption and its causes. The

International Journal of the History of Sport, 35, 141-156.

doi:10.1080/09523367.2018.1485652

Van Dijck, J., & Poell, T. (2013). Understanding social media logic. Media and

Communication, 1, 2–14.

Vicente, M., Batista, F., & Carvalho, J. P. (2019). Gender detection of Twitter users

based on multiple information sources. In L.T. Kóczy, J. Medina-Morena, & E.

139

Ramírez-Poussa (Eds.), Interactions Between Computational Intelligence and

Mathematics Part 2 (pp. 39-54). Cham, Switzerland: Springer.

Villar, J. G., & Guerrero, P. R. (2009). Sports attendance: A survey of the literature 1973-

2007. Rivista di Diritto e di Economia dello Sport, 5, 112-151.

Von Allmen, P. (2001). Is the reward system in NASCAR efficient?. Journal of Sports

Economics, 2, 62-79. doi:10.1177/152700250100200106

Wade, J. (2013 March 07). The argument for and against NBA Tanking. Bleacher Report.

Retrieved from https://bleacherreport.com/articles/1557217-the-argument-for-

and-against-nba-tanking

Ward-Henninger, C. (2017 September 28). Report: NBA approves draft lottery reform to

prevent tanking and player resting rules. CBS. Retrieved from

https://www.cbssports.com/nba/news/report-nba-approves-draft-lottery-reform-to-

prevent-tanking-and-player-resting-rules/

Watanabe, N. M., Yan, G., & Soebbing, B. P. (2016). Consumer interest in Major League

Baseball: An analytical modeling of Twitter. Journal of Sport Management, 30,

207-220. doi:10.1123/jsm.2015-0121

Watanabe, N. M., Yan, G., Soebbing, B. P., & Fu, W. (2019). Air pollution and

attendance in the Chinese Super League: Environmental economics and the

demand for sport. Journal of Sport Management, 33, 289-302.

doi:10.1123/jsm.2018-0214

Watanabe, N. M., Yan, G., Soebbing, B. P., & Pegoraro, A. (2017). Is there economic

discrimination on sport social media? An analysis of Major League

Baseball. Journal of Sport Management, 31, 374-386. doi:10.1123/jsm.2016-0244

140

Watanabe, N., Yan, G., & Soebbing, B. P. (2015). Major League Baseball and Twitter

usage: The economics of social media use. Journal of Sport Management, 29,

619-632. doi:10.1123/JSM.2014-0229

Waterloo, S. F., Baumgartner, S. E., Peter, J., & Valkenburg, P. M. (2018). Norms of

online expressions of emotion: Comparing Facebook, Twitter, Instagram, and

WhatsApp. New Media & Society, 20, 1813-1831.

doi:10.1177/1461444817707349

Webb, R. (2012, April 17). When tankers tell the truth. ESPN. Retrieved from

http://www.espn.com/blog/truehoop/post/_/id/40780/when-tankers-tell-the-truth

Welki, A. M., & Zlatoper, T. J. (1994). US professional football: The demand for game‐

day attendance in 1991. Managerial and Decision Economics, 15, 489-495.

doi:10.1002/mde.4090150510

Whitney, J. D. (2005). The peculiar externalities of professional team sports. Economic

Inquiry, 43, 330-343. doi:10.1093/ei/cbi022

Winfree, J. (2009). Fan substitution and market definition in professional sports

leagues. The Antitrust Bulletin, 54, 801-822.

Winfree, J. A., & Fort, R. (2008). Fan substitution and the 2004-05 NHL lockout. Journal

of Sports Economics, 9, 425-434. doi:10.1177/1527002508316266

Winfree, J. A., McCluskey, J. J., Mittelhammer, R. C., & Fort, R. (2004). Location and

attendance in Major League Baseball. Applied Economics, 36, 2117-2124.

doi:10.1080/0003684042000287664

141

Witkemper, C., Lim, C. H., & Waldburger, A. (2012). Social media and sports marketing:

Examining the motivations and constraints of Twitter users. Sport Marketing

Quarterly, 21, 170-183.

Wojcik, S., & Hughes, A. (2019 April 24). Sizing up Twitter users. Pew Research Center.

Retrieved from https://www.pewresearch.org/internet/2019/04/24/sizing-up-

twitter-users/

Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. Boston,

MA: Massachusetts Institute of Technology.

Xiang, Z., Du, Q., Ma, Y., & Fan, W. (2017). A comparative analysis of major online

review platforms: Implications for social media analytics in hospitality and

tourism. Tourism Management, 58, 51-65. doi:10.1016/j.tourman.2016.10.001

Yamamura, E. (2011). Game information, local heroes, and their effect on attendance:

The case of the Japanese Baseball League. Journal of Sports Economics, 12, 20-

35. doi:10.1177/1527002510369128

Yan, G., Pegoraro, A., & Watanabe, N. M. (2018). Student-athletes’ organization of

activism at the University of Missouri: Resource mobilization on Twitter. Journal

of Sport Management, 32, 24-37. doi:10.1123/jsm.2017-0031

Yan, G., Watanabe, N. M., Shapiro, S. L., Naraine, M. L., & Hull, K. (2019). Unfolding

the Twitter scene of the 2017 UEFA Champions League Final: Social media

networks and power dynamics. European Sport Management Quarterly, 19, 419-

436. doi:10.1080/16184742.2018.1517272

142

Yu, Y., & Wang, X. (2015). World Cup 2014 in the Twitter world: A big data analysis of

sentiments in US sports fans’ tweets. Computers in Human Behavior, 48, 392-400.

doi:10.1016/j.chb.2015.01.075