The betting market as a forecast of television ratings for primetime NFL football.
Paul, Rodney J. ; Weinbach, Andrew P.
Introduction
Sports broadcasts offer one of the only remaining items consumers
of televised programs demand to view live. In a media world increasingly
dominated by programs recorded to DVRs and subscription services such as
Netflix, sports offers television advertisers an audience that demands
to watch a program as it unfolds in real time. Other forms of
entertainment, such as television shows and movies, are not as
timedependent as technology allows for consumers to watch these events
at a time that is convenient for them. Sports remains different as the
real-time experience is important to fans who follow their favorite
teams, participate in fantasy leagues, and wager on the outcome of
games.
With this in mind, sports offers a captive audience to advertisers
as fans will not be able to fast-forward through commercials as they
watch the game as it actually happens. Television contracts for popular
sports have skyrocketed in recent years and advertisers are willing to
pay high prices to expose their product to consumers who watch live
sports. In the United States, the top sport for viewers and fans alike
is the National Football League (NFL), whose season runs through the
fall and winter months. Even with its popularity, however, not all NFL
games are created equal. Televised games featuring different teams and
game conditions are likely to attract dissimilar levels of fan demand
for viewing the game.
Understanding the factors that lead to higher ratings for NFL games
is important on a variety of fronts. First, the league has primetime
contracts with NBC and ESPN (Disney/ABC) for Sunday and Monday night
games where only a single NFL game is shown at that time slot. This is
quite different from the normal Sunday afternoon games, where multiple
games are being played simultaneously. The networks would like to choose
the highest-quality games for primetime broadcasts to maximize
viewership and revenues from advertisers.
A second factor that is quite important for NFL broadcasts is that
advertisers are typically guaranteed a certain Nielsen rating or they
will receive future advertising as compensation. This agreement is
called a "make good" and is discussed in the advertising
literature in research such as Rust et al. (1992). If the game does not
meet its ratings goal, the network loses out on future revenues.
Understanding the factors that drive television ratings is pivotal to
choosing games to broadcast, deciding how much advance advertising to
allot to the game, and as a result capture the highest revenues from
advertisers.
Third, understanding the game attributes that lead to higher
television ratings is useful to the league as a whole when they sell
future television packages to the networks. If factors such as game
uncertainty and scoring play pivotal roles in driving television
audiences, structure of the sport itself could potentially be adjusted,
such as rules aimed at increasing scoring, to maximize the value of
future television deals.
This paper addresses these issues by studying the factors that
affect Nielsen ratings for primetime NFL games on Sunday and Monday
nights. Regression models are developed and tested that use expected
game attributes formed in the betting market for National Football
League games as independent variables. The point spread, a measure of
uncertainty of outcome, and the total, a measure of expected scoring,
and other factors are used to explain three different measures of
television viewership available publicly in Variety magazine: the
overall Nielsen rating for the game, the Nielsen rating of the 18-49
year-old viewership bracket, and the share of the rating of the 18-49
year-old group.
The overall Nielsen rating is important to networks and advertisers
and its determinants are the main focus of this study. The 18-49
year-old group, however, is an interesting and timely subset because
attracting and retaining viewers of this age group has recently been
revealed as a problem for the NFL. Although the NFL is a television
ratings giant, the average age of its viewers has been trending older.
Variety magazine, the source of the data for this research, noted that
the average audience size for NFL games has declined by more than 10% in
the last four years and this trend has become a major concern for the
league as a whole (Steinberg, 2014). Understanding the factors that
influence viewership among this group, both as a whole and as a share of
televisions in use, should be helpful to the league in retaining and
improving this segment of the audience, which is often identified as
valuable to advertisers due to their disposable income and consumption
habits (i.e., Storey, 2009).
Additionally, we illustrate how the same factors that influence
television ratings also mirror the amount of interest on a game in the
betting market through a regression model of betting volume taken from
www.sportsinsights.com. Illustrating that bettors and viewers (fans) of
the NFL respond similarly to the game attributes is advantageous as the
betting market for the game is formed before the game is played and may
ultimately be useful in helping to forecast television demand. We note
this by adding the unexplained portion of the betting volume regression,
the residuals, as an independent variable in the Nielsen ratings
regressions. These residuals capture factors that may be difficult to
actually model, but due to the ex-ante betting market being influenced
by the same factors as the ex-post television viewership market, will
help to provide insight into which games are likely to be more or less
popular than originally anticipated. Through this information,
television networks may be better able to plan to meet ratings goals for
advertisers and maximize their profits.
Literature Review
The role of uncertainty of outcome, team quality, and television
ratings have been studied for the NFL and other sports in the sport
economics literature. In relation to the NFL, Paul and Weinbach (2007)
found that higher quality teams, measured by the sum of win percentages
of the teams, led to higher start-of-game ratings for Monday Night
Football and kept more of the audience at halftime of the game. Closer
games at halftime were also shown to retain a higher rating than games
that were lopsided. Tainsky (2010) also found that teams with higher win
percentages had higher television ratings across all televised games in
the NFL. In addition, Tainsky and McEvoy (2012) showed that better
quality teams improved ratings, while uneven matchups led to lower
overall ratings. Game uncertainty was also shown to play a major role in
television ratings for out-of-market NFL playoff games, but not for
local television markets containing playoff teams (Tainsky, Xu, &
Zhou, 2014). Grimshaw and Burwell (2014) found that fans prefer close
games (more uncertainty of outcome) in NFL markets without local teams
and showed that NFL games on competing networks are substitutes for one
another in viewership.
Television ratings for other sports have also been tested for the
role of uncertainty of outcome and other factors. In college football
bowl games, Salaga and Tainsky (2014) found that less uncertainty of
outcome appeared favorable, pre-game, for BCS bowl games, but
within-game analysis found that viewers respond positively to game
uncertainty. In the NBA, Mongeon and Winfree (2012) found that
television audiences were more sensitive to win percentages of teams
than in-person game attendance. Also for the NBA, Hausman and Leonard
(1997) showed that star players increased television ratings. The role
of race and TV ratings were explored in both the NBA (Kanazawa &
Funk, 2011) and the NFL (Aldrich et al., 2005). For NASCAR racing,
greater uncertainty of outcome was also shown to positively impact
television ratings, in addition to live attendance (Berkowitz et al.,
2011).
The role of uncertainty of outcome was also studied for various
television markets in Europe. Forrest, Simmons, and Buraimo (2005) found
that English Premier League (EPL) ratings were positively influenced by
uncertainty of outcome and also discovered that more uncertainty of
outcome also influenced the selection of games during the flexible
portion of the EPL schedule. When EPL matches were broken down into
minute-by-minute television ratings, uncertainty of outcome was also
shown to positively and significantly affect TV ratings (Alavy et al.,
2010). Greater uncertainty of outcome was also shown to have a positive
effect on television ratings for Spanish Football (Buraimo &
Simmons, 2009).
Expected scoring, proxied by the total (over/under) on sports
contests, have also been shown to have a positive and significant effect
on TV ratings. Paul and Weinbach (2007) showed that expected scoring
played an important role in NFL Nielsen ratings for Monday Night
Football. More recently, Grimshaw and Burwell (2014) showed that
non-local market NFL viewers preferred high scoring games to low scoring
games.
In relation to another key component of this study, betting volume
has recently become available for study to researchers. Paul and
Weinbach (2013) illustrated that betting volume is influenced by
televised games, higher quality team matchups, and higher expected
scoring for the NFL. Similar findings were also shown for betting volume
for other sports in North America (Paul & Weinbach, 2010). In
addition, Paul, Weinbach, and Small (2014) noted the role that betting
volume plays in the study and findings of market efficiency in betting
markets.
This research extends the literature on this subject to examine the
relationship between television ratings and betting volume. This paper
illustrates common factors that influence both television ratings and
betting volume, investigates changes over time in these variables, and
reveals how the unexplained variation in betting volume, known before a
game is played, helps to account for variations in television ratings.
Regression Models
To study this market, we estimate four regression models using
factors from the betting market and elsewhere to explain differences in
different forms of television ratings and betting volume. The dependent
variables in the four regression models consist of three measures of
television viewership and one measure of betting market participation.
The three measures of television viewership are overall Nielsen rating,
18-49 year-old Nielsen rating, and 18-49 year-old Nielsen share of
rating. A Nielsen rating point is equal to one percent of the total
number of television households in the United States. For the year 2014,
this is approximately equal to 1,156,000 households. Nielsen share of
ratings is calculated for a particular time slot as the number of
household televisions tuned to the program divided by the number of
televisions in use, expressed as a percentage. As stated in the
introduction, the overall ratings are the prime focus of this research,
but the 18-49 year-old audience is also studied due to estimated
declining interest in NFL viewership among this demographic and their
appeal to advertisers due to their disposable income and consumption
habits.
Each of these measures are taken from Variety magazine for the
2009-2013 NFL seasons. We use each game reported by Variety magazine
during this timeframe as our sample. The measure of betting market
activity is betting volume taken from www.sportsinsights.com daily
during the timeframe of this sample. The betting volume is taken from
the number of bets for each game offered on the sports betting rotation.
Sportsinsights presents an aggregated total of bets from three popular
on-line sportsbooks, BetUS.com, CaribSports.com, and SportBet.com. The
average number of bets per game was nearly 90,000 for the sample. This
figure is not the number of dollars bet, as that information is not
given by Sportsinsights, but research into the relationship between the
number of bets and dollars bet revealed an extremely high degree of
correlation between these numbers (Humphreys et al., 2013).
The independent variables included in each of the regressions are
structured the same across all four models. The first set of independent
variables includes NFL football game attributes. The first independent
variable is the sum of the win percentages of the teams participating in
the nationally televised game. The win percentages used in this
summation are the win percentages going into the game for each team.
This variable represents the overall quality of the teams participating
in the game. Assuming fans prefer games between the best teams, this
variable should have a positive impact on television ratings.
The second independent variable included in the model is the
absolute value of the point spread on the game taken from
www.sportsinsights.com. This serves as the measure of uncertainty of
outcome of the game. The absolute value of the point spread is used due
to the home field advantage not being important in the setting of
understanding how uncertainty of outcome impacts television ratings.
Point spreads closer to zero indicate games that are expected to be more
competitive; if uncertainty of outcome is important, this variable
should have a negative and significant effect on Nielsen ratings.
The betting market total, also gathered from
www.sportsinsights.com, is another independent variable that represents
on-field expectations of game play; in this case the total (over/under)
represents expected scoring in the game. Given previous research results
(Paul & Weinbach, 2007), the expected sign on this variable is
positive as higher scoring games are expected to be more popular with
fans of the game.
Given that the sample contains both primetime NFL games on Sunday
and Monday nights, a dummy variable is included in the regression model
for Sunday Night Football games on NBC. Sunday night games are available
for free over-the-air (in addition to cable and satellite subscribers),
while ESPN's Monday Night Football is only available through a paid
cable or satellite subscription. Therefore, given a wider potential
audience for primetime football on Sunday night, the dummy variable for
Sunday Night Football should have a positive and significant effect on
ratings and volume.
The regression model also includes dummies for the months of the
season and years within the sample. The months of the season should
account for changing interest in football throughout the season due to
the playoff push, some teams being eliminated from playoff contentions,
starting or ending of other sports, and holiday effects. The yearly
dummies are included to take into account changing technology throughout
the sample as more households have obtained alternative ways to watch
television programming (mobile devices, computers, etc.) and other
entertainment viewing options have multiplied (i.e., Netflix). The
yearly dummies are also important for the betting volume regression, as
government rules were introduced that prohibited the funding of offshore
betting accounts from U.S. banks and credit cards.
A dummy variable is also included in the regression model to
account for primetime NFL games that directly competed with Major League
Baseball World Series games. On the days where the two sports went
head-to-head, the value of this dummy variable is one. Past research has
shown that the two products are substitutes for viewers (Paul &
Weinbach, 2007). If this has continued in recent years, this dummy
variable should have a negative and significant effect on NFL TV
ratings.
Summary statistics of the data used in the regression model are
presented in Table 1. As can be seen in the table, Sunday Night Football
ratings and betting volumes are higher overall than Monday Night
Football. Sunday Night Football games also have greater uncertainty of
outcome (lower absolute value of the points spread) and scoring (total).
Table 2 presents the regression results for each of the four
models, showing results for overall Nielsen ratings, 18-49 year-old
Nielsen ratings, 18-49 year-old Nielsen share, and Betting Volume.
Coefficients are presented for each independent variable, t-stats are
shown in parentheses, and statistical significance is noted with
*-notation. Due to autocorrelation and heteroskedasticity issues with
the data, Newey-West HAC standard errors and covariances are used and
shown in the regression results below.
Across the four regression model results, there are clear
similarities in the factors that influence television ratings and
betting volume in the NFL. First, in terms of game attributes, it was
found that the sum of win percentages of the teams was found to have a
positive and significant effect on TV ratings and betting volume in each
regression model. More successful teams increase fan and bettor demand
to watch and wager on primetime games.
Our measure of uncertainty of outcome of the game, the absolute
value of the point spread, was not found to have a statistically
significant effect on television ratings or betting volume. It appears
the overall quality of the teams participating in the game and other
factors are more important to viewers and bettors than uncertainty of
outcome when there is only one game being broadcast in primetime. In
addition, due to the selection process by NBC and ESPN, these games are
often chosen with the anticipation of a close football game. When only
studying primetime games, the variation in uncertainty of outcome may
not be large enough, especially in a limited sample, to fully capture
fan preferences for uncertainty of outcome.
Another possibility to explain the lack of statistical significance
of the absolute value of the point spread is that this measure of
outcome uncertainty may be related to other factors that affect demand
for live games, such as loss aversion, home win preference, and a
preference for watching upsets (Coates et al., 2014). Although studied
in the context of attendance, these same factors may also impact
television demand. In related work, Humphreys and Zhou (2014) note that
as many as four different factors can be captured in the observed
relationship between outcome uncertainty and consumer demand. Therefore,
an identification problem may lead to an imprecise parameter estimate on
outcome uncertainty. More research into this area may help to clarify
these results as it relates to demand for sports on television.
Expected scoring, proxied by the betting market total on the game,
was found to have a positive and significant effect on television
ratings, similar to the findings of Paul and Weinbach (2007). All else
equal, having an anticipated higher scoring game led to increases in
overall television ratings, ratings among 18-49 year olds, and the share
of the ratings for 18-49 year olds.
The dummy variable for Sunday Night Football games on NBC, as
opposed to Monday Night Football games on ESPN, was shown to have a
positive impact on television ratings and betting volume. Due to Sunday
Night Football being available over-the-air and Monday Night Football
being only available on cable and satellite, the greater potential
audience leads to more viewers overall. Under the assumption that there
is a sizeable consumption component to sports betting, having a game
available to a larger audience on NBC likely leads to more bettors as
they can easily watch the game and enjoy following the successes and
failures of the teams (and their bets) throughout the telecast.
The regression model results show that there were monthly effects
on the ratings share of 18-49 year olds, as ratings share of NFL games
fell in October, November, and December (compared to September) for this
age group. Perhaps after the start of the season, as some teams drop
from contention, some of the younger audience changes their viewing to
other forms of entertainment or simply does not watch television during
these telecasts. Statistically significant monthly effects were not
evident in the other regression results.
Yearly dummies played an important role in betting volume for
Sunday and Monday night NFL games, as betting volume has declined over
time. This is likely due to tighter internet restrictions on banks,
credit cards, and other funding mechanisms for offshore sports books
under U.S. law. These restrictions have led U.S. bettors to move away
from offshore gambling, potentially into other similar markets such as
one-day fantasy sports leagues, which were deemed legal under U.S. law.
Yearly dummies were generally shown to not have an impact on television
ratings, with the exception being the 2013 season for the ratings share
of 18-49 year olds. This declining younger audience, specifically males,
was the subject of much discussion in the media during the early 2014
NFL season. Although declining ratings for the NFL within this age group
is evident, this may not reflect fewer fans in this demographic watching
football games, but could indicate they are watching games through
mobile devices or the internet, which the Nielsen ratings may not
capture.
The dummy variable representing when the NFL was up against Major
League Baseball's World Series during primetime broadcasts was
shown to have a negative and significant effect on overall Nielsen
ratings, 18-49 ratings, and 18-49 ratings share. The final series of the
Major League Baseball playoffs appears to be a substitute for NFL games,
as baseball's champion is determined during the second month of the
NFL season. In contrast, betting volume for NFL primetime games was not
impacted in a statistically significant matter by the World Series.
Overall, the regression results show that television viewers and
bettors both value many of the same attributes when making the decision
to watch/wager on NFL games. They prefer games between good teams,
individual game uncertainty of outcome does not play a major role in the
sample, while expected scoring leads to higher ratings and generates
more betting volume.
Betting Market Volume as a Preview of Television Ratings
Given that betting market volume and television ratings are shown
to be influenced by many of the same determinants, it is possible that
betting volume in some form may be helpful in explaining television
ratings. Since betting market volume is known before the game is
actually played and develops in a market during the course of the week
leading up to the nationally televised game, this information may help
to provide a glimpse inside the preferences of television audiences in
advance of the game.
Most television contracts offer some assurances of ratings for
advertisers. If the contracted ratings are not met, the advertiser will
receive some form of compensation from the network. The network may give
them additional advertising at no cost through "make-backs."
Given that Nielsen ratings are important to the bottom line of the
network in this contract cycle (and in those to come in the future), any
insight that may help to project if the game is going to meet the
contracted ratings number or not can be valuable to the network. For
instance, if the game is not expected to meet its viewership target,
additional advertising can be put in place on the network and its family
of associated networks or money may actually be spent to advertise the
game on other networks to attempt to stimulate demand to watch the game.
On the other hand, if the game is expected to exceed the contracted
ratings, normal commercial time dedicated to advertise the upcoming game
can be switched into actual commercials for products, which will
increase the revenues of the network overall.
Knowing that both the betting market and television ratings respond
to game quality factors (in addition to other known factors), simple
aggregated betting volume may not provide additional insight into
television ratings beyond wagering market-formed prices. However, the
residuals of the betting volume regression shown in the previous section
may be helpful in explaining future television ratings for the game.
When the residuals are greater than zero, it implies that more betting
is occurring on the game than would be anticipated due to game quality
expectations, the month of the season, the network airing the game, etc.
If the preferences of the bettors translate into preferences of fans,
television ratings may be influenced by the same factors.
The residuals of the betting market volume regression may capture
factors that are difficult to model outside of the point spread and
total, such as effects of star players, fantasy sports implications, or
very popular teams with the public (which will likely overlap with star
players and being a good team to begin with), and may play a key role in
explaining higher or lower television ratings than predicted based upon
game characteristics.
Therefore, using the same regression model for overall Nielsen
ratings, ratings of the 18-49 year-old demographic, and the ratings
share of the 18-49 year-old marketplace, we now add the residuals of the
betting volume regression as an additional independent variable in the
model. If games that are more (less) popular with bettors are also more
(less) popular with television viewers of the NFL, this variable will
have a positive and statistically significant effect on ratings. The
results are shown in Table 3.
When taking the residuals of the betting volume regression and
using them as an independent variable to explain television ratings, the
residuals were shown to have a positive and significant effect on
overall television ratings for prime time football. When betting volume
was higher than expected due to game attributes, month and year effects,
etc., television ratings were also shown to be positively impacted.
Unexplained factors not included in the betting volume regression
model, such as potential impact of star players, popular teams (beyond
team success and uncertainty of outcome), and others are shown to
positively impact overall Nielsen ratings for Sunday Night Football and
Monday Night Football games. It was not shown to be statistically
significant, at the 5% level or below, for the 18-49 year-old
demographic. The interest of the younger demographic does not appear to
be influenced as heavily by the unexplained activity in the betting
market.
Given that betting market volume is known before the game is
actually played, this variable could be useful in forecasting television
ratings. Games that are receiving more betting volume than expected are
likely to have greater television rating success. Games that are lagging
in betting volume, compared to expectations, are likely to have lower
television ratings. Further research into the progression of betting
volume during the week, its volatility, and whether it is forecastable
during the window that would be useful to networks and advertisers may
be fruitful avenues to pursue.
If possible, monitoring betting market volume could be helpful to
the television networks broadcasting NFL games. If betting market volume
is lower than anticipated, given the point spread and total on the game,
it may be worth further advertising on its network (or the family of
networks owned by Disney/ABC/ESPN or NBC) to attempt to boost ratings to
meet ratings expectations of advertisers. If betting volume is beyond
what is expected, those advertising spots could be transferred to actual
product advertisements, eliminating the need to forego ad revenue at the
expense of promoting the upcoming NFL game on these networks.
Conclusions
This research illustrated that television viewers and bettors of
the National Football League enjoy many of the same characteristics.
Using data from primetime NFL broadcasts of Sunday Night Football (NBC)
and Monday Night Football (ESPN), television viewers and bettors were
shown to enjoy the same on-field game features. Betting volume and
television ratings, both overall and within the coveted 18-49 yearold
bracket, were shown to positively be influenced by the quality of the
two teams playing in the game (sum of win percentages) and the expected
amount of scoring (betting market total). Controlling for these
characteristics, expectations of uncertainty of outcome (point spread on
game) was not shown to have a significant impact on either market.
Given that fans and bettors alike are shown to enjoy anticipated
high-scoring matchups between two high-quality teams, the ex-ante prices
formed and the subsequent volume seen in the betting market can be
useful in projecting television ratings. Betting volume is known before
the game is played and may be useful in forecasting ratings. Using the
residuals of the betting volume regression as an independent variable in
the television ratings regression models, it was shown that the
residuals of the volume regression have a positive and significant
effect on overall television ratings. Higher-than-expected or
lower-than-expected betting volume is shown to transfer its effects to
Nielsen ratings as the betting market serves as an indicator of future
television viewership.
Following the likely progression of betting volume during the week
leading up to the nationally televised game may provide insight about
Nielsen ratings to television network executives. Given that television
contracts usually have ratings targets, which may be costly to the
network if the contracted ratings are not reached, betting market volume
may help in choosing the frequency and viability of advertisements for
the upcoming NFL game on the network and its associated family of
networks. If betting volume is not reaching expected levels, given the
quality of the teams and the expected amount of scoring, more
advertising time may be useful in spurring viewership to meet
contractual levels. On the other hand, if betting volume is exceeding
expectations, advertising spots that would typically advertise the game
can be substituted with actual revenue-generating product
advertisements. In other words, market-based prices (point spreads and
totals) and consumer response (wagering volume) from betting markets may
serve a useful purpose to television networks when forecasting Nielsen
ratings.
Rodney J. Paul is a professor in the Sport Management Department in
the David B. Falk College of Sport and Human Dynamics. His research
interests include studies of market efficiency, prediction markets,
behavioral biases, and the economics and finance of sports.
Andrew P. Weinbach is an associate professor of economics and the
Colonel Lindsey H. Vereen Endowed Business Professor at the E. Craig
Wall Sr. College of Business Administration. His research interests
include the economics and finance of sports, consumer behavior, and the
economics of lotteries and gambling.
References
Alavy, K., Gaskell, A., Leach, S., & Szymanski, S. (2010). On
the edge of your seat: Demand for football on television and the
uncertainty of outcome hypothesis. International Journal of Sport
Finance, 5, 75-95.
Aldrich, E., Arcidiancono, P., & Vigdor, J. (2005). Do people
value racial diversity? Evidence from Nielsen ratings. The B.E. Journal
of Economic Analysis & Policy: Topics in Economics Analysis and
Policy, 5, Article 4.
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, 326-338.
Berkowitz, J. P., Depken, C. A. II, & Wilson, D. P. (2011).
When going in circles is going backward: Outcome uncertainty in NASCAR.
Journal of Sports Economics, 12, 253-283.
Coates, D., Humphreys, B. R., & Zhou, L. (2014).
Reference-dependent preferences, loss aversion, and live game
attendance. Economic Inquiry, 52, 959-973.
Forrest, D., Simmons, R., & Buraimo, B. (2005). Outcome
uncertainty and the couch potato audience. Scottish Journal of Political
Economy, 52, 641-661.
Grimshaw, S. D., & Burwell, S. J. (2014). Choosing the most
popular NFL games in a local TV market. Journal of Quantitative Analysis
in Sports, 10, 329-343.
Hausman, J., & Leonard, G. (1997). Superstars in the National
Basketball Association: Economic value and policy. Journal of Labor
Economics, 15, 586-624.
Humphreys, B. R., Paul, R., & Weinbach, A. (2013). Consumption
benefits and gambling: Evidence from the NCAA basketball betting market.
Journal of Economic Psychology, 39, 376-386.
Humphreys, B. R., & Zhou, L. (2014). The Louis-Schmeuing
paradox and the league standing effect. Economics Working Papers, 14-22,
College of Business & Economics, West Virginia University,
Morgantown, WV.
Kanazawa, M., & Funk, J. (2001). Racial discrimination in
professional basketball: Evidence from Nielsen ratings. Economic
Inquiry, 39, 599-608.
Mongeon, K., & Winfree, J. (2012). Comparison of television and
gate demand in the National Basketball Association. Sport Management
Review, 15, 72-79.
Paul, R., & Weinbach, A. (2007). The uncertainty of outcome and
scoring effects on Nielsen ratings for Monday Night Football. Journal of
Economics and Business, 59, 199-211.
Paul, R., & Weinbach, A. (2010). The determinants of betting
volume for sports in North America. International Journal of Sport
Finance, 5, 128-140.
Paul, R., & Weinbach, A. (2013). Wagering preferences of NFL
bettors: Determinants of betting volume. Journal of Prediction Markets,
6, 42-55.
Paul, R., Weinbach, A., & Small, K. (2014). The relationship
between sportsbook volume and market efficiency in football betting
markets. Journal of Prediction Markets, 8(2), 29-42.
Rust, R., Kamakura, W., & Alpert, M. (1992). Viewer preference
segmentation and viewing choice models for network television. Journal
of Advertising, 21, 1-18.
Salaga, S., & Tainsky, S. (2015). The effects of outcome
uncertainty, scoring, and pregame expectations on Nielsen ratings for
Bowl Championship Series games. Journal of Sports Economics, 16,
439-459.
Steinberg, B. (2014, September 17). The NFL's biggest
challenge? Keeping younger viewers. Variety. Retrieved from
http://variety.com/2014/tv/news/the-nfls-greatest-test-not-ray-ricebut-young-crowds-who- tune- out-games-1201307613/
Storey, M. (2009, April 23). THE TV COLUMN: Not in 18-49 age group?
TV execs write you off. Arkansas Democrat Gazette. Retrieved from
http://www.arkansasonline.com/news/2009/apr/23/tv-column-not-18-49-age-group-tvexecs-wr-20090423/
Tainsky, S. (2010). Television broadcast demand for National
Football League contests. Journal of Sports Economics, 11, 629-640.
Tainsky, S., & McEvoy, C. (2012). Television broadcast demand
in markets without local teams. Journal of Sports Economics, 13,
250-265.
Tainsky, S., Xu, J., & Zhou, Y. (2014). Qualifying the
uncertainty of outcome effect: A game-level analysis of NFL postseason
broadcast ratings. Journal of Sports Economics, 15, 219-236.
Rodney J. Paul [1] and Andrew P. Weinbach [2]
[1] Syracuse University
[2] Coastal Carolina University
Table 1: Summary Statistics--2009-2013 Sunday Night Football and
Monday Night Football
18-49 18-49 Overall Betting Point Total
Rating Share Rating Volume Spread
n = 133 Overall Sample
Mean 6.85 18.58 17.75 88,276.11 5.19 45.12
Median 6.80 18.00 17.31 87,149.00 4.00 45.00
Standard 1.65 10.73 4.54 18,178.67 3.39 4.74
Deviation
n = 73 Sunday Night Football
Mean 7.88 19.73 20.65 91,089.95 4.59 45.84
Median 7.80 19.00 20.86 87,149.00 3.50 46.50
Standard 1.29 3.40 3.59 18,648.88 2.99 4.74
Deviation
n = 60 Monday Night Football
Mean 5.61 17.18 14.23 84,852.60 5.92 44.26
Median 5.60 15.00 13.99 87,149.00 4.75 44.50
Standard 1.08 15.48 2.72 17.125.95 3.72 4.64
Table 2: TV Ratings and Betting Volume Regression Results--NFL
2009-2013 Sunday Night and Monday Night Football
Dependent Overall 18-49 18-49
Variable: Nielsen Nielsen Nielsen
Rating Rating Share of
Rating
Intercept 4.63 2.23 ** 7.08 ***
(1.62) (2.10) (2.87)
Sum of Win 2.11 *** 0.75 *** 1.67 ***
[Percentages.sub.t-1] (2.82) (3.76) (3.07)
Absolute Value of -0.07 -0.01 0.01
Point Spread (-0.77) (-0.61) (0.02)
Betting Market Total 0.18 *** 0.07 *** 0.19 ***
(2.95) (3.02) (3.51)
Sunday Night Football ill 6.16 *** 2.18 *** 4.32 ***
(11.19) (8.06) (6.56)
October -0.98 -0.62 ** -2.54 ***
(-1.29) (-2.36) (-3.48)
November -0.66 -0.63 *** -2.67 ***
(-0.88) (-2.78) (-4.01)
December 0.12 -0.59 -2.26 ***
(0.15) (-2.09) (-3.02)
January 1.44 -0.22 -1.97
(0.63) (-0.20) (-0.71)
2010 1.04 0.26 0.86
(1.20) (-0.96) (1.15)
2011 0.18 0.06 -0.59
(0.21) (0.17) (-0.65)
2012 -0.15 -0.07 -0.23
(-0.18) (-0.21) (-0.27)
2013 -1.09 -0.70 ** -2.00 **
(-0.18) (-2.13) (-2.39)
World Series -2.70 ** -1.29 *** -3.56 ***
Dummy Variable (-2.00) (-3.96) (-5.58)
R-squared 0.61 0.64 0.59
Dependent Betting
Variable: Volume
Intercept 54768.55 ***
(3.52)
Sum of Win 7156.887 **
[Percentages.sub.t-1] (2.38)
Absolute Value of -105.65
Point Spread (-0.27)
Betting Market Total 587.33
(1.84)
Sunday Night Football 5620.04 **
(1.99)
October 5185.64
(1.15)
November 1465.95
(0.45)
December -6184.37
(-1.58)
January 3394.97
(0.38)
2010 12912.68 **
(2.20)
2011 -12393.10 ***
(-2.99)
2012 -6692.44
(-1.66)
2013 -17201.47 ***
(-3.50)
World Series -4361.20
Dummy Variable (-0.67)
R-squared 0.40
*-notation notes statistical significance of the t-test that the
coefficient is different from zero. ** represents statistical
significance at the 5% level and *** represents statistical
significance at the 1% level.
Table 3: Impact of Residuals from Betting Volume Regression on
Television Ratings--NFL Sunday Night Football and Monday Night
Football 2009-2013
Dependent Variable: Overall 18-49 18-49
Nielsen Nielsen Nielsen
Rating Rating Share of
Rating
Intercept 4.63 2.23 ** 7.08 ***
(1.44) (2.13) (2.75)
Sum of Win 2.11 *** 0.74 *** 1.67 ***
[Percentages.sub.t-1] (3.49) (3.08) (2.93)
Absolute Value of Point -0.07 -0.01 0.01
Spread (-0.77) (-0.58) (0.02)
Betting Market Total 0.18 *** 0.07 *** 0.19 ***
(2.87) (3.31) (3.49)
Sunday Night Football 6.16 *** 2.18 *** 4.32 ***
(12.75) (8.48) (6.64)
October -0.98 -0.62 -2.54 ***
(-1.32) (-1.87) (-3.27)
November -0.66 -0.63 ** -2.67 ***
(-1.20) (-2.22) (-3.83)
December 0.12 -0.59 -2.26 ***
(0.13) (-1.77) (-2.92)
January 1.44 -0.22 -1.97
(0.42) (-0.21) (-0.77)
2010 1.04 0.26 0.85
(1.11) (0.93) (1.16)
2011 0.18 0.06 -0.59
(0.17) (0.18) (-0.67)
2012 -0.15 -0.07 -0.23
(-0.15) (-0.22) (-0.28)
2013 -1.09 -0.70 ** -2.00 **
(-0.88) (-2.42) (-2.51)
World Series Dummy -2.71 -1.29 *** -3.56 ***
Variable (-1.93) (-4.32) (-5.26)
Residuals of Betting 0.00004 ** 0.000009 0.00003
Volume I Regression (2.26) (1.45) (1.75)
R-squared 0.63 0.64 0.60
*-notation notes statistical significance of the t-test that the
coefficient is different from zero. ** represents statistical
significance at the 5% level and *** represents statistical
significance at the 1% level.
