The attraction of baseball games in a small-size league: are the effects of outcome uncertainties really important?
Jane, Wen-Jhan ; Kuo, Nai-Fong ; Wu, Jyun-Yi 等
Introduction
Previous research has indicated that one of the missions in
professional sports has been to achieve greater uncertainty. Sports
leagues have consistently justified restrictions on competition as being
needed to promote outcome uncertainty, thereby benefiting the consumer
by providing more attractive games. This kind of uncertainty includes a
competitive balance (hereafter, CB) and game uncertainty (hereafter,
GU), which are key factors in the demand function (Borland &
Macdonald, 2003). The former describes the degree of uncertainty in a
league which is comprised of members that are relatively competitive in
terms of strength, and the latter refers to the degree of uncertainty
which is made up of the relative capabilities of two teams that are
competing to win a game.
Studies in sport economics have attempted to discern the effect of
CB and other factors on league attendance. It is widely believed that
sporting events will be more attractive and more entertaining should
there be a greater degree of CB among the teams (Szymanski &
Kesenne, 2004). If this were not so, it is believed that attendees would
lose interest, and thus there would be significantly lower attendance at
the games involved. The theory of CB in team sports was first proposed
by Rottenberg (1956), who noted that "the nature of the industry is
such that competitors must be of approximately equal 'size' if
any are to be successful; this seems to be a unique attribute of
professional sports" (p. 242). In cases where consumer demand
depends to a large extent on inter-team competition, the necessary
interactions across teams define the special nature of sports. Games
with no balance of competition or contests among poorly matched
competitors will eventually cause fan interest to wane and industry
revenues to fall (Sanderson & Siegfried, 2003).
Forrest and Simmons (2002) define CB as a league structure which is
characterized by relatively equal playing strength between league
members. Michie and Oughton (2004) refer to balance between the sporting
capabilities of teams in a league. A comprehensive classification of the
definition of GU is proposed by Szymanski (2003). There are three types
of GU in professional sports studies, and they are match uncertainty,
seasonal uncertainty, and championship uncertainty. His research has
documented (1) the relationship between CB and league attendance between
seasons/years (e.g., championship uncertainty), (2) the relationship
within a season/year (e.g., seasonal uncertainty), and (3) the
relationship between GU and game-day attendance. However, the empirical
evidence in this area seems far from unambiguous (see Szymanski, 2003,
p. 1155).
Taiwan is one of the smallest countries to have had a professional
baseball league. Its national baseball team is considered to be one of
the strongest teams in the world. In Taiwan baseball retains a large
following and remains the most popular team sport. Several Taiwanese
players have enjoyed successful careers in Major League Baseball (hereafter, MLB) and Nippon Professional Baseball (hereafter, NPB).
Including some evidence of game uncertainty in professional team sports
and comparing the results with the MLB are important.
The three major purposes in this paper are as follows. First, the
relationships between league-level and game-level uncertainties and
game-day attendance are investigated. That is, the effect of CB among
league-level teams and the effect of match uncertainty between two teams
are included in the model. One technical difficulty in the analysis
concerns the measurement of CB and GU, and so we employ some indices to
measure the league-level and game-level uncertainties in a game while at
the same time investigating game-day attendance.
Second, ordinary least squares (OLS) and censored normal (Tobit)
regressions have been employed in previous research on this issue.
Researchers seem to have assumed that the regressors only affect the
location of the conditional mean. However, in the case of demand
variables at sporting events, the shape of the distribution will be
altered by different regressor values. The effect of outcome
uncertainties on a high-quantile/box-office game and a
low-quantile/unpopular game are different. It is therefore necessary to
ask whether such a relationship changes for both the box-office and
unpopular games.
Third, the ambiguity surrounding the hypothesis of outcome
uncertainties still needs more empirical evidence in the literature. We
focus on investigating the relationship between attendance and CB for
different conditions/quantiles related to game-day attendance. As such,
to fill this gap in the literature, we employ data from the Chinese
Professional Baseball League (hereafter, CPBL) to examine these
relationships.
Literature Review on Outcome Uncertainties and Attendance
Competitive balance is one of the key issues in sport economics. In
recent years, economic studies of CB have been performed on a variety of
sports, including baseball (Eckard, 2001; Schmidt & Berri, 2001;
Humphreys, 2002; Meehan et al., 2007), European football (Falter &
Perignon, 2000; Forrest & Simmons, 2002; Garcia & Rodriguez,
2002, 2007; Brandes & Franck, 2007), American college football
(Price & San, 2003; Depken & Wilson, 2006; Eckard, 1998) and
professional American football (Grier & Tollison, 1994). The nature
of CB has been outlined and research has found that league attendance is
positively influenced by an increase in CB.
In addition to the effect of CB in a league, GU between two teams
is also important. It is likely that as two teams rigorously compete
with each other, individual games become more attractive. There have
been a number of studies that examine the effect of this uncertainty on
sports attendance. In general, the literature on outcome uncertainty can
be separated into studies looking at aggregate attendance in a
year/season and those investigating attendance in specific games. The
hypothesis of GU has been verified in various professional sports, and
these include the Major League Baseball (Knowles, 1992; Rascher, 1999;
Meehan et al., 2007), the European football (Jennett, 1984; Peel &
Thomas, 1988; Buraimo & Simmons, 2008a, 2008b), the National
Basketball Association (Rascher & Solmes, 2007), the National Hockey
League (Jones & Ferguson, 1988), and the Rugby Union in New Zealand (Owen & Weatherston, 2004). As more appropriate data for measuring
GU, for example, the odds in the betting market, have become available
in recent decades, it has become possible to test the outcome
uncertainty hypothesis on a game-by-game basis. However, the results are
ambiguous (see Borland & MacDonald, 2003; Szymanski, 2003 for
detailed references), and further empirical analysis is needed to
clarify the figures.
Quantification of the CB of a league can be problematic (Utt &
Fort, 2002; Mizak et al., 2005). Since the league structure provides
competition, it is commonly suggested that the league should be treated
as a single entity, rather than making the club the unit of analysis.
Therefore, a standard technique is to calculate the percentage of
matches that each club wins in a season and to use the standard
deviation of this distribution, where each club has a perfectly balanced
chance of winning each match (= an 0.5 probability of winning each
match), to arrive at a single figure for the league (see Scully, 1989;
Quirk & Fort, 1992; Humphreys, 2002; Zimbalist, 2002). The higher
the index, the lower the degree of CB that is implied in a league. This
index, which was first applied by Scully (1989), has been used in
several studies to compare the closeness of win records within seasons
in professional leagues. Moreover, the range of win percentages and the
Hirschman-Herfindahl indexes in a season are also commonly used in the
literature.
Indices of CB have not been applied to matchday attendance.
Therefore, we have modified indices based to previous research to
measure the degree of CB on a matchday. However, the measurement of CB
has limitations (see Michie & Oughton, 2004 for a detailed
discussion), and it will still be an important issue in sport economics
in the future. (1) So we provide as many indices of CB as possible in
order to investigate the robustness of our empirical results and to
ensure the consistency of our main conclusions.
Empirical Methodology and Data Description
In this section we employ some indices to measure the CB for a
league which provides outcome uncertainty in the league, and the GU
which provides outcome uncertainty between rival teams. Then, the
hypothesis of uncertainties that games with a high degree of
league-level uncertainty and with a high degree of game-level
uncertainty lead to high attendance at a game is investigated.
Furthermore, the application of quantile regression can be used to
explore the difference between the effects of attraction for different
levels of outcome uncertainties on box-office games and low-attendance
games.
Empirical Methodology
The concepts of the outcome uncertainties include the relatively
equal playing strength between league members and the relative
capabilities of two competing teams for winning in a game. The basic
model of attendance demand for a match t between the host and guest
teams is as shown in equation (1).
ATT = f(CB, AD, DD, S, X), (1)
where ATT represents the attendance. The independent variables
include the measurement of competitive balance in a league (CB), the
uncertainty of sporting competition between two teams (AD), the dummy for the host-guest winning percentage (DD), the sum of the two
teams' winning percentage (S), and other control variables (X).
Teams compete with each other in order to win, and a professional
baseball league typically uses the winning percentage to calculate a
team's standing. In the case of a league-level competition among
team members, CB is measured by the standard deviation of win
percentages in a league at game t (CB_SD), the range of win percentages
(CB_R) in a league at game t, and the Herfindahl-Hirschman index of win
percentages (CB_HHI) in a league at game t. Therefore, the data on the
winning percentage for every team at every game needs to be collected
before these indices are calculated. The greater these indices are, the
greater the degree of competitive imbalance in a league at time/game t
will be. The calculation of HHI is, however, more complicated, and we
describe the data process as follows. The share of cumulative wins for a
team ([Share.sub.1] is calculated by the ith team's cumulative wins
divided by the total number of games played in the league before game t.
Therefore, the degree of CB for each team can be calculated by
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where n is the
number of teams within a league and [Share.sub.i] is the ith team's
share of wins. The extent of CB is seriously biased because of the
uncompleted games in the first couple of rounds, and so the games in the
former rounds in the data must be dropped when the model is regressed.
For the game-level competition between two teams (m and n) in game
t, ADWinr, DDWinr, and SWinr are calculated via winning percentages as
two rivals meet for the last time before game t. ADWinr is the absolute
value of the differential in the winning percentage for team m and team
n in the latest game, and is used to measure the two rivals' game
certainty before game t starts. DDWinr is the dummy of the host-guest
winning percentage (host>guest=1, otherwise=0) for team m and team n
in the latest game and we employ it to test the host team's winning
effect on fans. SWinr is the sum of the two teams' winning
percentages before game t and it is used to measure the quality of a
game.
Other control variables include the characteristics of the host
cities, stadiums, and teams. Besides, we also include the environmental
factors and the game (Game) and year (Year) dummies. The former three
characteristics include variables such as the household's average
disposable income in the host city (Income), the population of the host
city (Pop), the location of the stadium (Distance), the dummy for turf
(Turf), the dummy for a public-viewing screen (Screen), the size of the
field in square feet (Field), and the dummies for rival-team
combinations (Pair). The latter variables include the dummy for the
holiday (Holiday) and the rainfall (Rainfall).
To empirically assess the effects of CB and GU, we build up the
following econometric model based on Equation (1):
ATT = C + F[beta] + X[gamma] + u, (2)
where C is the constant term, F=[CB, AD, DD, S] is the matrix of
the sporting competition, X is a set of control variables, and u is the
unobserved disturbance. A standard approach estimates the unknown
parameters of equation (2) using the method of ordinary least squares
(OLS). The OLS estimation is known to be the best linear unbiased
estimator under well-defined conditions. However, the OLS approach only
measures the "average" behavior of a conditional distribution.
It can not provide enough information to describe the entire conditional
distribution.
Quantile regression (QR) as introduced by Koenker and Bassett
(1978) is a good approach to adopt at this point. This estimation
permits us to estimate various quantile regressions on their conditional
distribution, and it can provide richer information and enable us to
obtain a more comprehensive and robust analysis. Therefore, we employ
the QR to obtain a complete picture of the conditional distribution on
attendance. Equation (2) can be re-written as:
ATT = C + F[[beta].sub.q.] + X[[gamma].sub.q] + [u.sub.q], (3)
where [[beta].sub.q] and [[gamma].sub.q] are matrices of
coefficients on sporting competition and control variables which are
associated with the q-th quantile, and [u.sub.q] is an unknown error
term. It is assumed that [u.sub.q] satisfies the constraint,
[Quant.sub.q]([u.sub.q]|F,X) = 0. To obtain the estimators of
([[beta].sub.q], [[gamma].sub.q]), given a particular value for q, we
solve
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (4)
which can be shown in the form of a linear programming problem and
is computationally straightforward. The special case q=0.5 is referred
to as the median regression estimator. A more detailed discussion can be
found in Koenker (2005).
Data Description
Our data contain information for six teams on over 1,860 individual
matches in 15 stadiums within the period 2001-2007 in the CPBL. (2)
There were 180 games for the seasons for the years 2001 and 2002 as each
team played 90 games, and 300 games were scheduled as each team played
100 games for the seasons after 2003. The average number of attendees
was 2,798.
The degree of sporting competition that included league-level CB
and game uncertainties are calculated using previous definitions.
Control variables including the characteristics of the host cities,
stadiums, and teams are employed. All variables are collected from the
Baseball Record Books, the official website of the CPBL, the
Directorate-General of Budget, Accounting and Statistics (DGBAS), the
URMAP, and the Central Weather Bureau. (3)
Sporting competitions will become attractive when the extent of the
CB among the teams is high. Otherwise, fans lose interest and there is
significantly lower attendance (Depken & Wilson, 2006). To attract
potential competition in a game, the relationship between the lack of
competitive balance in a league (CB_SD, CD_R, or HHI) and attendance is
thus expected to be negative. Similarly, the coefficient of game
certainty for two teams (ADWinr) is also expected to be negative. As to
the coefficient of DDWin, it is expected to be positive because the host
team's wins bring more host attendees. Moreover, fans are
interested in high quality play, and so SWinr is expected to positively
affect attendance.
Market size is usually included as a determinant of attendance, and
we use the city population in which the game is played (Pop) as a proxy.
Population data are available at various levels in the DGBAS, including
the county, local authority, and ward. We construct our market size
measure by using the smallest level. Winfree et al. (2004) have
indicated that baseball is a normal good and that the income elasticity
is inelastic (=0.175) for the MLB case. Thus, we would expect the
coefficients of Income and Pop to be positively significant.
The variable Distance equals the distance in meters from the
stadium to the nearest transportation hub. (4) Long transportation
increases the cost for a game, and so it is expected to be negatively
significant. Data for the other control variable, weather conditions,
which is measured by the rainfall (Rainfall), have been collected from
the Central Weather Bureau. (5) It is expected to negatively affect
attendance in open-air stadiums.
Previous studies (Forrest & Simmons, 2002; Buraimo &
Simmons, 2008a) use some classifications for revealing the Derby effect,
and they have shown that these games tend to attract more attendees,
ceteris paribus. In this study, we use the variables Pair_i, the dummies
for a combination of two rivals, to capture matches of historical
rivalry. (6) Compared to the abundant literature on European and
American sports, the advantage of our setting is that it is more
objective under the situation of no reference in the CPBL regarding
historical rivalry. The statistics and the corresponding expectation of
coefficients are listed in Table 1.
Empirical Results and Discussion
The results of the OLS and quantile regressions are presented in
Table 2.7 We find that the OLS estimators and the QR estimators for
different quantiles are quite different. Cameron and Trivedi (2009)
indicate that the differences in the coefficients across quantiles are
caused by the heteroskedastic errors. For this reason, we apply the
Breusch-Pagan (1979) test to investigate the heteroskedasticity. These
p-values of the test are 0.000, and they are listed in the bottom row of
Tables 2, 3, and 4. The results support the conclusion that the QR
method appropriately investigates the relationship between attendance
and outcome uncertainties.
In terms of the sporting competition variables, the league-level
(CB_SD) and game-level (ADWinr) uncertainties are significantly
negatively related to attendance. As for the proxy of quality play
(SWin), it is highly significant and positively related to attendance,
and the effects are consistent in the regressions. For the effects of
the host-city factors, the coefficients of income (Income) and
population (Pop) are significantly positive. Other significant and
consistent factors in the regressions are the provision of
public-viewing screen facilities (Screen), the dummy for holidays
(Holiday), Rainfall, and dummies for the rival-team combinations
(Pair_i). (8)
Outcome Uncertainties for a League and a Game
Previous research suggests that balanced competition is an
important factor. In the literature, however, there is no consistent
conclusion in the empirical studies on game-day attendance. Our results
support the hypotheses of outcome uncertainty. That is, both the
unbalanced degrees of the league-level competition and the game-level
certainty negatively affect the game-day attendance. On average, a 0.1%
decrease in the standard deviation of winning percentages in a league
brings 334 fans to a game, and a 1% decrease in the absolute value of
the differential between the two rivals' winning percentages
increases the number of fans at a game by 562, ceteris paribus.
Moreover, fans also care about match quality. On average, two quality
rivals who have an additional 0.1% of the total winning percentage bring
about an increase of 233 in attendance, other things being equal. In
these sporting competitions, the results correspond to our expectations.
Quality plays an important role in the demand for sports
competitions. Fans are not only interested in uncertain games with close
competition between two teams, but are also attracted by a league where
the teams are evenly matched. Moreover, the effects of higher quantiles
are greater than the lower ones for these factors. That is, both
uncertainties and quality play have a larger effect when a box-office
game is played. Regressions using alternative measures of competitive
balance are included in Tables 3 and 4.9 To provide a better
understanding of the effects for different quantiles, Figures 1, 2, and
3 display all the coefficients of the sporting competition variables for
all quantiles.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
Host-Win Effect, Location, Derby Effect
As to the effect of a host-team win on attendance, it is
insignificant and the result reflects the reality of Taiwanese baseball.
Taiwan covers a land area of 35,280 square kilometers, which is
relatively small in the world. The effect of location is thus less
important in such a small geographic area with just 15 stadiums. In the
absence of clear demarcations of market territories for the teams, plus
the fact that fans do not entertain a strong sense of geographical
division, the scheduling and assignment of game locations is done in
such a way that the area factor does not distinguish the host team from
the guest team. Rather, the host-or-guest designation is determined by
equal odds in terms of taking turns playing the host or guest roles at a
given location. This results in a totally different type of operation
when compared to the MLB. Unlike teams that have their own stadium in
their city, the localization of CPBL teams is not as strong as that of
teams in the MLB. In effect, CPBL teams play in stadiums and cities
which do not belong to any team. The evidence shows that attendance is
not influenced by the appellation of the "titular" host team.
This is mainly due to the small land area, the ease of travel, and the
absence of a strong sense of geographical division among fans. The
finding that the effect of distance is insignificant also echoes this
point.
[FIGURE 3 OMITTED]
Historic rivalries are also an important factor for fans. For the
Derby effect, 9 of 14 coefficients (Pair_i) are significantly positively
related to attendance compared to the base combination. The top five
coefficients of a paired game are Pair_2, Pair_3, Pair_4, Pair_5, and
Pair_7, and only one team (Brother Elephants) is involved in all of
them. Just like the effects of the New York Yankees in the MLB,
Manchester United in European soccer, or the Giants in the NPB, there is
a large effect, e.g., a 1,521 to 3,623 increase in audiences compared to
the base game on average, when the Brother Elephants play in a CPBL
game.
With respect to other significant coefficients, the positive
coefficients of income and local population show that baseball is a
normal good in relation to the consumers' expenditure and the
market size has a positive effect on attendance. The income elasticity
ranges from 0.398 to 0.426 in the semi-log model, and these estimations
are higher than in the case of the MLB. (10) In addition, the provision
of public-viewing screen facilities (Screen) and the dummy for holidays
(Holiday) have a positive effect on attendance, while rainfall has a
negative effect. Two positive effects of Screen and Holiday are
magnified in a box-office game and these effects result in a popular
game becoming more popular. Conversely, the rainfall reduces the number
of fans at low-attendance games and makes an unpopular game even more
unpopular.
From the cost and benefit view, the cost of an indoor stadium is
much larger than the cost of public-viewing screens. However, the gap
between the marginal effects of screen (509) and rainfall (-8) is huge
as shown in Table 2. The policy implications of constructing a stadium
facility in the short run is that the public-viewing screens are needed,
but the indoor stadiums are not justified financially.
Conclusions
Unlike most research on the determinants of attendance, we include
measures of CB in a league and an index for GU for two rivals at the
same time in our quantile analysis. The results support the hypothesis
of outcome uncertainties. More uncertainty in sporting competitions
between two rivals in a game and more uncertainty regarding who will
emerge as the champion among teams in a league can increase the
fans' interest, and thus bring more fans to a game. The effects are
especially enhanced in high box-office (high-quantile) games. An
imbalance in competition reduces the number of fans at low-attendance
games and leads to bigger losses in popular games. The robust
regressions also support the hypothesis. In addition, our results
support the view that the quality of a game is also an influential
factor in game-day attendance, and the evidence of the Derby effect
shows that there is a Brother-team effect on CPBL attendance.
The bandwagon effect can explain the results of the quantile
regressions. That is, fans often attend and believe in high box-office
games merely because many other fans attend and believe in the same
things. Therefore, uncertainty strongly affects attendance, especially
when more fans go to the field.
As Rottenberg (1956) suggested, an approximately equal size for
each team is needed if any are to be successful. Therefore, the
implementation of regulations to prevent big-market teams from using
their financial advantage to buy success is important.
Institutional barriers that govern the limitations on the right to
trade players and the team's expenditure on human resources have
been implemented to different degrees in the league. In the long run,
the policy implications of the uncertainty effect for the league are
condensed as follows. On the one hand, in order to increase attendance,
the league should consider more enforcement of existing rules, e.g.,
salary caps and draft rules, to increase the CB in the league.
Restricted revenue sharing which is only used for investing in human
resources is also a good way to accomplish this. On the other hand,
increasing the number of potential competitors is important. Like the
Premier League in England, which uses relegation and promotion of
team(s) between first-tier and second-tier teams, an effective policy
may include more second-tier teams in the CPBL to promote balanced
competition.
In the short run, the policy implications of our findings provide
an arrangement for game schedules for the league. Such a method involves
creating some special games for the best two teams during the regular
season. Generally speaking, the schedule cannot be changed once the
season starts. Since the degree of league-level CB, the GU, and the game
quality all affect game-day attendance, the special games that are
inserted in the regular season should be determined based on these
measures. Moreover, the results of the quantile regressions also
indicate that the CB effect of the high quantile on attendance is larger
than the lower one. Therefore, a highly balanced competition should be
arranged in a bigger stadium. The revenue from box-office games should
be able to totally cover the losses in low-quantile games. More
specifically, the special games should be held on weekends every two or
three weeks during the season. The top two teams at each point should
compete with each other in these special games. As a result, attendance
in the league will increase.
Compared to host-guest systems such as MLB, in the CPBL the
insignificant coefficients of the host-team win and the distance reflect
a different type of operation. The results of this study show that
attendance is not affected by the titular host team, and the sense of
belonging for fans of the host team is negligible. The findings also
show that transportation costs do not seem to affect the fans'
demand for game attendance. This is mainly due to the small geographical
area and the absence of a strong sense of geographical division for fans
in Taiwan.
References
Borland, J., & Macdonald, R. (2003). Demand for sport. Oxford
Review of Economic Policy, 19(4), 478-502.
Brandes, L., & Franck, E. (2007). An empirical analysis of
competitive balance in European soccer leagues. Eastern Economic
Journal, 33(3), 379-403.
Bresusch, T., & Pangn, A. (1979). A Simple Test for
heteroscedasticity and random coefficient variation. Econometrica, 47,
1287-1294.
Buraimo, B., & Simmons, R. (2008a). Do sports fans really value
uncertainty of outcome? Evidence from the English Premier League.
International Journal of Sport Finance, 3, 146155.
Buraimo, B., & Simmons, R. (2009). A tale of two audiences:
Spectators, television viewers and outcome uncertainty in Spanish
football. Journal of Economics and Business, 61(4), 326-338.
Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics
usingstata. College Station, TX: Stata Press.
Depken, C. A., & Wilson, D. P. (2006). The uncertainty of
outcome hypothesis in Division IA college football. Working Paper.
Eckard, E. W. (1998). The NCAA cartel and competitive balance in
college football. Review of Industrial Organization, 13(3), 347-369.
Eckard, E. W. (2001). Free agency, competitive balance, and
diminishing returns to pennant contention. Economic Inquiry, 39(3),
430-443.
Falter, J. M., & Perignon, C. (2000). Demand for football and
intramatch winning probability: An essay on the glorious uncertainty of
sports. Applied Economics, 32, 1757-1765.
Forrest, D., & Simmons, R. (2002). Outcome uncertainty and
attendance demand in sport: The case of English soccer. The
Statistician, 51(2), 229-241.
Garcia, J., & Rodriguez, P. (2002). The determinants of
football match attendance revisited: Empirical evidence from the Spanish
football league. Journal of Sports Economics, 3(1), 18-38.
Grier, K. B., & Tollison, R. D. (1994). The rookie draft and
competitive balance: The case of professional football. Journal of
Economic Behavior and Organization, 25, 293-298.
Humphreys, B. (2002). Alternative measures of competitive balance
in sporting leagues. Journal of Sports Economics, 3(2), 133-148.
Jennett, N. (1984). Attendance, uncertainty of outcome and policy
in Scottish league football. Scottish Journal of Political Economy, 31,
176-198.
Jones, J. C. H., & Ferguson, D. G. (1988). Location and
survival in the National Hockey League. Journal of Industrial Economics,
36(4), 443-457.
Knowles, G., Sherony, K., & Haupert, M. (1992). The demand for
Major League Baseball: A test of the outcome uncertainty hypothesis. The
American Economist, 36, 72-80.
Koenker, R., & Bassett, G. S. (1978). Regression quantile.
Econometrica, 48, 33-50.
Koenker, R. (2005). Quantile regression. Cambridge, MA: Cambridge
University Press.
Meehan, Jr., J. W., Nelson, R. A., & Richardson, T. V. (2007).
Competitive balance and game attendance in Major League Baseball.
Journal of Sports Economics, 8(6), 563-580.
Michie, J., & Oughton, C. (2004). Competitive balance in
football: Trends and effects. Research Paper 2004 No.2 in Football
Governance Research Centre.
Mizak, D., Stair, A., & Rossi, A. (2005). Assessing alternative
competitive balance measures for sports leagues: A theoretical
examination of standard deviations, gini coefficients, the index of
dissimilarity. Economics Bulletin, 12(5), 1-11.
Owen, P. D., & Weatherston, C. R. (2004). Uncertainty of
outcome and super 12 rugby union attendance: Application of a
general-to-specific modeling strategy. Journal of Sports Economics,
5(4), 347-370.
Peel, D. A., & Thomas, D. A. (1988). Outcome uncertainty and
the demand for football: An analysis of match attendances in the English
football league. Scottish Journal of Political Economy, 35, 242-249.
Peel, D. A., & Thomas, D. A. (1996). Attendance demand: An
investigation of repeat fixtures. Applied Economics Letters, 3, 391-394.
Powell, J. L. (1986). Censored regression quantiles. Journal of
Econometrics, 32, 143-155.
Price, D., & San, K. (2003). The demand for game day attendance
in college football: An analysis of the 1997 Division 1-A season.
Managerial and Decision Economics, 24, 35-46.
Quirk, J., & Fort R. (1992). Pay dirt: The business of
professional team sports. Princeton, NJ: Princeton University Press.
Rascher, D. (1999). A test of the optimal positive production
network externality in Major League Baseball. In E. Gustafson & L.
Hadley (Eds.), Sports economics: Current research. Westport, CT: Praeger
Press.
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.
Rottenberg, S. (1956). The baseball players' labor market.
Journal of Political Economy, 64, 242-258.
Sanderson, A. R., & Siegfried, J. J. (2003). Thinking about
competitive balance. Journal of Sports Economics, 4(4), 255-279.
Schmidt, M., & Berri, D. (2001). Competitive balance and
attendance: The case of Major League Baseball. Journal of Sports
Economics, 2(2), 145-167.
Scully, G. (1989). The business of Major League Baseball. Chicago:
University of Chicago Press.
Szymanski, S. (2003). The economic design of sporting contests.
Journal of Economic Literature, 41, 1137-1187.
Szymanski, S., & Kesenne, S. (2004). Competitive balance and
gate revenue sharing in team sports. The Journal of Industrial
Economics, 52(1), 165-177.
Utt, J., & Fort, R. (2002). Pitfalls to measuring competitive
balance with gini coefficients. Journal of Sports Economics, 3(4),
367-373.
Winfree, J. A., McCluskey, J. J., Mittelhammer, R. C., & Fort,
R. (2004). Location and attendance in Major League Baseball. Applied
Economics, 36, 2117-2124.
Zimbalist, A. S. (2002). Competitive balance in sports leagues: An
introduction. Journal of Sports Economics, 3(2), 111-121.
Zimbalist, A. S. (2003). Competitive balance conundrums: Response
to Fort and Maxcy's comments. Journal of Sports Economics, 4,
161-163.
Endnotes
(1) The complexity of measuring CB is well summarized by Zimbalist
(2003, p. 163). He indicated that, "In the end, it may be that the
best measure of competitive balance is a multivariate index, that it is
nonlinear or constrained, and/or that it differs league by league"
(2) The 15 stadiums included the Taipei City (Tienmou), Taipei
County (Hsinchuang), Longtan, Hsinchu, Taichung, Douliou, Chiayi City,
Chiayi County, Tainan, Kaohsiung City, Kaohsiung County (Cheng Ching Lake), Pingtung, Yilan, Hualien, and Taitung stadiums. We use the
addresses of these stadiums to calculate the location of each stadium
via a professional location system based on a Geographic Information
System (GIS), i.e., the URMAP.
(3) The website of the CPBL is
http://www.cpbl.com.tw/html/stadium.asp, the Directorate-General of
Budget, Accounting and Statistics is
http://61.60.106.82/pxweb/Dialog/statfile9.asp, the URMAP is
http://www.urmap.com/, and the Central Weather Bureau is
http://www.cwb.gov.tw/.
(4) The hubs include railway stations, subway stations, and highway
interchanges. We use the system of the General Packet Radio Service
(abbreviated as GPRS) location in the URMAP to calculate the shortest
distance.
(5) For all stadiums, we choose the nearest weather station to
obtain the information.
(6) There are six teams in the data, and the combination is C26=15.
Therefore, we use 14 dummies to investigate the Derby effect.
(7) Censoring occurs when the number of tickets is sold out. Thus,
in some research, the censored model is employed to deal with the
empirical problem. We also consider this situation and apply a censored
quantile regression proposed by Powell (1986). However, only two games
reach the capacity of the stadium in the regression. Therefore, there is
no need to consider the problem and we list the results for the QR in
Table 2.
(8) i=2, 3, 4, 5, 6, 7, 12, 15, 21.
(9) Both of the p-values of the Breusch-Pagan test are 0.000 for
Tables 3 and 4, respectively. The results support the conclusion that
the QR method appropriately investigates the relationship between
attendance and outcome uncertainties.
(10) The income elasticity is 0.175 for the MLB case (Winfree et
al., 2004). Their explanation of the income inelasticity is that the
consumers' leisure horizons widen when real incomes increase. The
results of the estimation are provided by the author if needed.
Authors' Note
The authors would like to thank the editor and the anonymous
referees for their helpful comments on the manuscript. All remaining
errors are our own. Jane is grateful to the National Science Council for
its financial support (NSC 98-2410-H-128 -016).
Wen-Jhan Jane [1], Nai-Fong Kuo [1], Jyun-Yi Wu [2], and Sheng-Tung
Chen [3]
[1] Shih Hsin University
[2] National Central University
[3] Feng Chia University
Wen-Jhan Jane is an assistant professor in the Department of
Economics. His current research focuses on the applied econometrics,
especially the topics of peer effects, competitive balance, and wage
theory in professional sports.
Nai-Fong Kuo is an associate professor in the Department of
Finance. His research interests include the empirical finance, spatial
econometrics, and applied general equilibrium.
Jyun-Yi Wu earned his doctorate in economics. His current research
interests include empirical econometrics, time series analysis, and
industrial economics.
Sheng-Tung Chen is an assistant professor in the Department of
Public Finance. His research focuses on applied econometrics and public
finance.
Table 1: Descriptive Statistics of the Data (n = 1860)
Variable Description Mean
ATT Number of attendees 2798.991
SD Standard deviation 0.165
of win percentages
in a league
R Range of win 0.414
percentages in a
league
HHI Hirschman-Herfindahl 3351.197
indexes for
competitive balance
in a league
ADWinr Absolute value of 0.194
differential between
two teams' winning
percentages before
game t started
DDWinr Dummy of host-guest 0.46
winning percentages
before game t started
(host>guest =1,
otherwise=0)
SWinr Sum of two teams' 0.964
winning percentage
before game t started
Income Household's average 903530.6
disposable income in
city j in a year
(NT Dollars)
Pop Population of host 1612421
team (thousands)
Distance The distance to the 2864.154
nearest transportation
hub (meters)
Turf Dummy of turf (yes=1) 0.8278
Screen Dummy of public- 0.63
viewing screen (yes=1)
Field The area of the field 132778.2
([feet.sup.2])
Rain Rainfall (millimeters) 3.295
Holiday Dummy of holiday 0.56
(weekend =1)
Variable Std. Dev. Expectation
ATT 2170.456
SD 0.093 Negative
R 0.217 Negative
HHI 4554.196 Negative
ADWinr 0.196 Negative
DDWinr 0.498 Negative
SWinr 0.263 Positive
Income 162267.6 Positive
Pop 1245908 Positive
Distance 1140.883 Negative
Turf 0.378 ?
Screen 0.483 ?
Field 3565.973 ?
Rain 12.25 Negative
Holiday 0.496 Positive
Table 2: Estimation Results of OLS and Quantile Regressions (CB_SD)
ATT OLS
25-th
Coef. P>t Coef. P>t
CB_SD -3343.3 *** (0.002) -1219.6 * (0.100)
ADWinr -562.8 * (0.088) -361.8 * (0.096)
DDWinr -107.6 (0.168) -64.73 (0.216)
SWinr 2331.3 *** (0.000) 1675.3 *** (0.000)
Income 0.001 * (0.067) 0.001 *** (0.004)
Pop 0.000 * (0.015) 0.000 ** (0.035)
Distance -0.07 (0.335) -0.03 (0.461)
Field 0.003 (0.802) 0.008 (0.376)
Turf -5.55 (0.966) 28.59 (0.765)
Screen 509.3 *** (0.001) 186.5 ** (0.042)
Pair2 2517.2 *** (0.000) 1667.8 *** (0.000)
Pair3 3623.6 *** (0.000) 1996.2 *** (0.000)
Pair4 1521.8 *** (0.000) 1026.7 *** (0.000)
Pair5 2027.4 *** (0.000) 1131.3 *** (0.000)
Pair6 989.4 *** (0.000) 692.4 *** (0.000)
Pair7 2231.4 *** (0.000) 1440.7 *** (0.000)
Pair8 7.19 (0.968) 6.97 (0.962)
Pair10 -19.12 (0.915) 174.4 (0.238)
Pair12 534.1 *** (0.003) 372.4 ** (0.011)
Pair14 -12.11 (0.953) -15.78 (0.921)
Pair15 876.4 *** (0.000) 524.1 *** (0.001)
Pair20 -213.5 (0.195) -82.34 (0.590)
Pair21 918.5 *** (0.000) 614.9 *** (0.000)
Pair28 -224.1 (0.199) -132.03 (0.379)
Holiday 701.7 *** (0.000) 405.06 *** (0.000)
Rainfall -8.03 ** (0.022) -7.14 *** (0.000)
_cons -2460.3 (0.225) -2740.7 * (0.047)
Game Yes Yes Yes Yes
Year Yes Yes Yes Yes
(Pseudo)- 0.524 0.251
[R.sup.2]
Breusch-Pagan test 176.2 (0.00) ***
[H.sub.0]: [CB_SD.sub.25]
= [CB_SD.sub.50]=[CB_SD.sub.75] 3.45 (0.032)**
ATT Quantile
50-th 75-th
Coef. P>t Coef. P>t
CB_SD -2812.4 *** (0.001) -3770.0 ** (0.010)
ADWinr -502.8 ** (0.049) -531.5 (0.252)
DDWinr -99.54 * (0.090) -96.49 (0.339)
SWinr 2077.1 *** (0.000) 2020.1 *** (0.000)
Income 0.001 *** (0.004) 0.001 ** (0.023)
Pop 0 (0.323) 0.000 ** (0.037)
Distance -0.023 (0.621) -0.052 (0.528)
Field 0.021 * (0.052) 0.01 (0.565)
Turf -16.24 (0.881) 75.98 (0.682)
Screen 266.0 ** (0.011) 485.7 *** (0.007)
Pair2 2367.0 *** (0.000) 3058.5 *** (0.000)
Pair3 3003.5 *** (0.000) 5386.6 *** (0.000)
Pair4 1289.4 *** (0.000) 1679.7 *** (0.000)
Pair5 1705.0 *** (0.000) 2608.7 *** (0.000)
Pair6 875.1 *** (0.000) 1119.6 *** (0.000)
Pair7 2281.9 *** (0.000) 2769.8 *** (0.000)
Pair8 106.5 (0.509) 243.2 (0.376)
Pair10 92.98 (0.583) -14.99 (0.959)
Pair12 428.2 ** (0.010) 571.4 ** (0.043)
Pair14 183.4 (0.309) 15.66 (0.960)
Pair15 753.2 *** (0.000) 836.6 *** (0.008)
Pair20 -211.2 (0.228) -305.6 (0.308)
Pair21 764.2 *** (0.000) 959.5 *** (0.001)
Pair28 -200.3 (0.234) -220.1 (0.443)
Holiday 619.3 *** (0.000) 777.5 *** (0.000)
Rainfall -3.228 (0.143) -5.913 * (0.060)
_cons -4502.0 *** (0.004) -2960.1 (0.251)
Game Yes Yes Yes Yes
Year Yes Yes Yes Yes
(Pseudo)- 0.3 0.375
[R.sup.2]
Breusch-Pagan test 176.2 (0.00) ***
[H.sub.0]: [CB_SD.sub.25]
= [CB_SD.sub.50]=[CB_SD.sub.75]
Notes:
(a) *** denotes significance at the 1% level, ** denotes
significance at the 5% level, and * denotes significance
at the 10% level.
(b) Values in parentheses are the p-values.
Table 3: Estimation Results of OLS and Quantile Regressions (CR_R)
ATT OLS
25-th
Coef. P>t Coef. P>t
CB_R -1422.6 *** (0.001) -521.6 * (0.053)
ADWinr -548.4 * (0.094) -350.9 * (0.075)
DDWinr -108 (0.166) -64.72 (0.178)
SWinr 2336.0 *** (0.000) 1719.8 *** (0.000)
Income 0.001 * (0.064) 0.001 *** (0.004)
Pop 0.000 * (0.015) 0.000 ** (0.027)
Distance -0.068 (0.343) -0.038 (0.306)
Field 0.004 (0.758) 0.008 (0.338)
Turf -11.32 (0.931) 31.97 (0.716)
Screen 514.9 *** (0.001) 189.4 ** (0.025)
Pair2 2517.2 *** (0.000) 1667.3 *** (0.000)
Pair3 3632.0 *** (0.000) 2004.1 *** (0.000)
Pair4 1525.6 *** (0.000) 1037.6 *** (0.000)
Pair5 2018.4 *** (0.000) 1131.5 *** (0.000)
Pair6 991.1 *** (0.000) 693.8 *** (0.000)
Pair7 2229.9 *** (0.000) 1436.2 *** (0.000)
Pair8 0.751 (0.997) 0.289 (0.998)
Pair10 -12.38 (0.945) 159.4 (0.245)
Pair12 537.2 *** (0.003) 378.5 *** (0.005)
Pair14 -18.78 (0.928) -17.35 (0.905)
Pair15 874.1 *** (0.000) 546.9 *** (0.000)
Pair20 -218.3 (0.185) -83.86 (0.552)
Pair21 918.5 *** (0.000) 606.1 *** (0.000)
Pair28 -213.6 (0.220) -126.7 (0.352)
Holiday 699.9 *** (0.000) 407.7 *** (0.000)
Rainfall -8.01 ** (0.022) -5.87 *** (0.000)
_cons -2579.4 (0.201) -2732.2 ** (0.030)
Game Yes Yes Yes Yes
Year Yes Yes Yes Yes
(Pseudo)- 0.524 0.252
[R.sup.2]
Breusch-Pagan test177.4(0.000) ***
[H.sub.0]: [CB_R.sub.25] 5.66
= [CB_R.sub.50]=[CB_R.sub.75] (0.003) ***
ATT Quantile
50-th 75-th
Coef. P>t Coef. P>t
CB_R -991.0 *** (0.002) -2010.8 *** (0.000)
ADWinr -547.7 ** (0.023) -452.7 (0.206)
DDWinr -98.87 * (0.075) -94.19 (0.227)
SWinr 2120.5 *** (0.000) 2147.3 *** (0.000)
Income 0.001 *** (0.003) 0.001 ** (0.014)
Pop 0.000 *** (0.281) 0.000 ** (0.018)
Distance -0.026 (0.559) -0.073 (0.270)
Field 0.020 ** (0.047) 0.01 (0.462)
Turf -31.75 (0.758) 37.22 (0.797)
Screen 278.9 *** (0.005) 522.5 *** (0.000)
Pair2 2385.2 *** (0.000) 2988.3 *** (0.000)
Pair3 3050.7 *** (0.000) 5328.0 *** (0.000)
Pair4 1329.4 *** (0.000) 1637.7 *** (0.000)
Pair5 1756.9 *** (0.000) 2532.3 *** (0.000)
Pair6 888.6 *** (0.000) 1149.5 *** (0.000)
Pair7 2332.7 *** (0.000) 2703.3 *** (0.000)
Pair8 134.9 (0.374) 181.2 (0.395)
Pair10 101.5 (0.525) -23.42 (0.919)
Pair12 470.6 *** (0.003) 524.4 ** (0.017)
Pair14 224 (0.190) -17.62 (0.942)
Pair15 779.8 *** (0.000) 860.3 *** (0.000)
Pair20 -183.7 (0.265) -275.1 (0.228)
Pair21 787.9 *** (0.000) 904.4 *** (0.000)
Pair28 -152.5 (0.338) -233.7 (0.297)
Holiday 602.3 *** (0.000) 732.7 *** (0.000)
Rainfall -3.29 (0.114) -4.93 * (0.064)
_cons -4496.6 *** (0.002) -2741.1 (0.171)
Game Yes Yes Yes Yes
Year Yes Yes Yes Yes
(Pseudo)- 0.3 0.376
[R.sup.2]
Breusch-Pagan test177.4(0.000) ***
[H.sub.0]: [CB_R.sub.25]
= [CB_R.sub.50]=[CB_R.sub.75]
Notes:
(a) *** denotes significance at the 1% level, ** denotes
significance at the 5% level, and * denotes significance
at the 10% level.
(b) Values in parentheses are the p-values.
Table 4: Estimation Results of the OLS and Quantile
Regressions (HHI)
ATT OLS
25-th
Coef. P>t Coef. P>t
HHI -0.006 (0.618) 0.004 (0.513)
ADWinr -876.4 *** (0.006) -491.6 *** (0.004)
DDWinr -102.8 (0.191) -75.04 * (0.082)
SWinr 2313.1 *** (0.000) 1651.4 *** (0.000)
Income 0.001 ** (0.049) 0.001 *** (0.001)
Pop 0.000 *** (0.009) 0.000 ** (0.011)
Distance -0.059 (0.414) -0.031 (0.350)
Field 0.003 (0.832) 0.004 (0.586)
Turf -5.27 (0.968) 62.53 (0.417)
Screen 514.1 *** (0.001) 175.7 ** (0.018)
Pair2 2514.2 *** (0.000) 1696.8 *** (0.000)
Pair3 3620.9 *** (0.000) 1988.6 *** (0.000)
Pair4 1511.8 *** (0.000) 1005.0 *** (0.000)
Pair5 2035.8 *** (0.000) 1146.5 *** (0.000)
Pair6 985.1 *** (0.000) 721.0 *** (0.000)
Pair7 2260.2 *** (0.000) 1497.7 *** (0.000)
Pair8 2.99 (0.987) 6.5 (0.956)
Pair10 -13.24 (0.941) 169.3 (0.163)
Pair12 534.9 *** (0.003) 364.5 *** (0.002)
Pair14 -8.03 (0.969) -70.43 (0.586)
Pair15 893.2 *** (0.000) 571.7 *** (0.000)
Pair20 -196.4 (0.230) -57 (0.650)
Pair21 950.2 *** (0.000) 570.7 *** (0.000)
Pair28 -234 (0.182) -119.5 (0.318)
Holiday 698.7 *** (0.000) 411.8 *** (0.000)
Rainfall -7.84 ** (0.028) -7.44 *** (0.000)
_cons -2847.1 (0.156) -2336.5 ** (0.037)
Game Yes Yes Yes Yes
Year Yes Yes Yes Yes
(Pseudo)- 0.521 0.252
[R.sup.2]
Breusch-Pagan test 177.4 (0.00) *** 0.730
[H.sub.0]: (0.483)
= [HHI.sub.50] = [HHI.sub.75]
ATT Quantile
50-th 75-th
Coef. P>t Coef. P>t
HHI 0.005 (0.542) -0.02 (0.108)
ADWinr -647.9 *** (0.007) -665.6 * (0.063)
DDWinr -102.2 * (0.078) -114 (0.173)
SWinr 2035.5 *** (0.000) 2090.8 *** (0.000)
Income 0.001 *** (0.000) 0.001 ** (0.029)
Pop 0 (0.330) 0.000 *** (0.007)
Distance -0.001 (0.977) -0.018 (0.787)
Field 0.021 ** (0.041) 0.007 (0.633)
Turf -89.02 (0.405) 21.96 (0.885)
Screen 295.7 *** (0.004) 475.7 *** (0.001)
Pair2 2226.0 *** (0.000) 3020.6 *** (0.000)
Pair3 3132.1 *** (0.000) 5475.6 *** (0.000)
Pair4 1299.6 *** (0.000) 1745.5 *** (0.000)
Pair5 1784.4 *** (0.000) 2655.3 *** (0.000)
Pair6 857.7 *** (0.000) 1131.6 *** (0.000)
Pair7 2368.1 *** (0.000) 3055.7 *** (0.000)
Pair8 77.12 (0.625) 325.8 (0.151)
Pair10 62.07 (0.709) -83.04 (0.734)
Pair12 530.1 *** (0.001) 616.8 *** (0.008)
Pair14 196.4 (0.269) 11.94 (0.963)
Pair15 779.5 *** (0.000) 959.8 *** (0.000)
Pair20 -149.7 (0.382) -290.2 (0.235)
Pair21 839.9 *** (0.000) 1045.3 *** (0.000)
Pair28 -168.3 (0.310) -260.7 (0.271)
Holiday 593.6 *** (0.000) 729.2 *** (0.000)
Rainfall -2.65 (0.221) -5.45 ** (0.029)
_cons -5205.4 *** (0.001) -2813.2 (0.188)
Game Yes Yes Yes Yes
Year Yes Yes Yes Yes
(Pseudo)- 0.298 0.374
[R.sup.2]
Breusch-Pagan test 177.4 (0.00) ***
[H.sub.0]:
= [HHI.sub.50] = [HHI.sub.75]
Notes:
(a) *** denotes significance at the 1% level, ** denotes
significance at the 5% level, and * denotes significance
at the 10% level.
(b) The values in parentheses are the p-values.