Undergraduate Economic Review Undergraduate Economic Review
Volume 16 Issue 1 Article 10
2019
Reference-Dependent Preferences Among NFL Fans: Evidence Reference-Dependent Preferences Among NFL Fans: Evidence
from Google Trends from Google Trends
Sunjae Lee
Washington University in St. Louis
Follow this and additional works at: https://digitalcommons.iwu.edu/uer
Part of the Behavioral Economics Commons
Recommended Citation
Lee, Sunjae (2019) "Reference-Dependent Preferences Among NFL Fans: Evidence from
Google Trends,"
Undergraduate Economic Review
: Vol. 16 : Iss. 1 , Article 10.
Available at: https://digitalcommons.iwu.edu/uer/vol16/iss1/10
This Article is protected by copyright and/or related rights. It has been brought to you by Digital
Commons @ IWU with permission from the rights-holder(s). You are free to use this material in any
way that is permitted by the copyright and related rights legislation that applies to your use. For
other uses you need to obtain permission from the rights-holder(s) directly, unless additional rights
are indicated by a Creative Commons license in the record and/ or on the work itself. This material
has been accepted for inclusion by faculty at Illinois Wesleyan University. For more information,
please contact [email protected].
©Copyright is owned by the author of this document.
Reference-Dependent Preferences Among NFL Fans: Evidence from Google Reference-Dependent Preferences Among NFL Fans: Evidence from Google
Trends Trends
Abstract Abstract
I look for evidence of reference-dependent preferences in the National Football League (NFL). Under
reference-dependent preferences, sports fans should react more strongly to surprising wins and losses
than expected wins or losses. I use Google Trends to look at the impact of NFL game outcomes on the
use of positive or negative words on Google search. While search activity did respond to NFL games, I did
not 7nd that this response was sensitive to how surprising the outcome was, and so did not 7nd evidence
of reference-dependent preferences.
Keywords Keywords
Reference-Dependent Preferences, Google Trends, Behavioral Economics
Cover Page Footnote Cover Page Footnote
Acknowledgements: I thank Prof. Nick Huntington-Klein for guidance and editing assistance.
This article is available in Undergraduate Economic Review: https://digitalcommons.iwu.edu/uer/vol16/iss1/10
Introduction
Sports fans tend to form expectations about how games will proceed and
who is likely to win. Sports fans also speculate about the result of the game based
on information that is available to them from sports pundits or online betting sites.
However, life is unpredictable, and so are the sports game results. There could be
a case in which the actual result of a game is utterly different from what people
have expected. If people expected a team to win the game by ten points, but in
reality the team loses by ten points, people will be shocked and enraged.
In this paper, I examine the effects of sports game results that completely
go against people’s “reference points on the game outcome” (Ge 2018) on
people’s behaviors. To be more exact, this paper looks for evidence of reference-
dependent preferences, a behavioral economics concept that argues that people
have “reference points” that they use to assess outcomes. People do not have
preferences purely over outcomes, as the rational decision model would suggest,
but evaluate outcomes relative to reference points based on what they expect.
In order to see how people’s behaviors change in response to surprising
sports game outcomes, I connect a data set of anticipated and actual sports
outcomes, specifically from the National Football League (NFL), to a data set
from Google Trends in which sports fans have an opportunity to express their
frustration and loss. Anticipated game outcomes come from an online betting
site, FootballLocks.com, and are linked to actual game scores from the NFL.
Differences between anticipated and actual scores constitute surprises in the
outcome. Negative surprises should be especially frustrating. One place people
might express their frustration is on Google Search, which is not just used to
search for information, but also to express hidden frustrations and opinions that
may not be reported on a survey (Stephens-Davidowitz 2014).
Before examining the data, I hypothesize that an unexpected loss will
result in an increase of searches of negative terms, while an unexpected win will
result in an increase of searches of positive terms on Google, above and beyond
what would occur from an expected loss or win. This would be evidence of
reference-dependent preferences among sports fans.
This is not the first paper to look for evidence of reference-dependent
preferences in sports. In the literature section of this paper, I look at four different
papers on reference-dependent preferences. All of these papers find evidence of
reference-dependent preferences, either among athletes or fans.
1
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
However, after completing the examination of the data collected through
the Google Trends, I have found, counter to much of the rest of the published
literature, no evidence of reference-dependent preferences. The results imply that
football game results do have a subtle effect on Google search terms, this effect is
not consistent with what reference-dependent preferences would predict.
This study is significant in a sense that it contributes to the understanding of
behavioral economics, and implies that reference-dependent preferences may not
be as well-supported as the prior literature would suggest. The result of the study
suggests a pivotal point for further studies and applications of the reference-
dependent preferences by questioning the real effects of football game results on
football fans’ behavior.
Literature
The primary purpose of this paper is to look for reference-dependent
preferences using Google Trends, a weekly index of the volume of searches for a
particular term on Google. Google Trends is an increasingly popular method for
collecting data for research. In Choi & Varian (Choi and Varian 2012), the
researchers used Google Trends to “forecast near-term values of economic
indicators” such as “automobile sales, unemployment claims, travel destination
planning, and consumer confidence” (Choi & Varian 1). They believe Google
Trends can be useful in predicting near future and present economic situations.
They compared the performance in prediction of an economic model with Google
Trends and without Google Trends. They confirmed that “simple seasonal AR
models that include relevant Google Trends variables tend to outperform models
that exclude these predictors by 5% to 20%” (Choi & Varian 8).
Perhaps the most well-known application of Google Trends data in
economic research is “The Cost of Racial Animus on a Black Candidate:
Evidence using Google Search Data” (Stephens-Davidowitz, 2014). The author
uses Google Trends to “understand the extent of contemporary prejudice” and
“increase our understanding of the determinants of voting” (Stephens-Davidowitz
1). He argues that Google Trends is a “new proxy for an area’s racial animus from
a non-survey source: the percent of Google search queries that include racially
charged language” because of “individuals’ tendency to withhold socially
unacceptable attitudes, such as negative feelings towards blacks, from surveys”
(Stephens-Davidowitz 1). He further claims that “Google search query data can
do more than correlate with existing measure; on socially sensitive topics; they
can give better data and open new research on old questions” (Stephens-
2
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
Davidowitz 2). Unlike surveys, Google search renders more accurate data by
allowing people to be much more honest about their desires and controversial
opinions. Stephens-Davidowitz shows that users don’t simply use Google to
search for information, but also to express emotions or frustrations, which makes
the data valuable as a high-frequency measure of things like frustration or anger,
implying that it can be used to measure these emotions in response to unexpected
sports losses under reference-dependent preferences.
Reference-dependent preferences refers to the behavioral economics
concept that people have “reference points” and evaluate outcomes relative to
those reference points. This means that the way someone feels about an outcome
is relative to what they would have expected it to be. Finding evidence of
reference-dependent outcomes can be difficult because the reference points
cannot be observed. The literature on reference-dependent preferences often uses
sports as a setting where winning probabilities can be calculated ahead of time,
and so unexpected losses and wins can be easily identified. I highlight several
studies of reference-dependent preferences in sports below.
In Bartling, Brandes, & Schunk (2015), the researchers show that
“professional soccer players exhibit reference-dependent behavior during
matches” (Bartling et al. 1). They used data from two soccer leagues to show
evidence that players had reference-dependent preferences. When the flow of the
match did not coincide with players’ expectations (reference points), especially
when their team was losing unexpectedly, the probability that a player would
receive a red card in a given minute increased by more than 20 percent. The same
pattern did not appear when the team was losing but was expected to lose, so this
can be identified as reference-dependent behavior. Reference-dependent behavior
was not diminished by player experience or high-stakes games.
Subsequently Pope & Schweitzer (2011) use golfer performance on the
PGA tour to test for loss aversion, a feature of reference-dependent preferences.
Like Bartling et al. (2015), this paper also concludes that “loss aversion, a
fundamental bias, continues to persist in a highly competitive market” (Pope &
Schweitzer 155), and is not eliminated by competition, large stakes, or
experience.
In addition to the reference-dependent preferences shown in players’
behaviors, there is also research on the effect of surprising sports game losses on
the audience. Card & Dahl (2011) find an effect of unexpected wins and losses by
professional football teams on family violence. They analyze police reports of
3
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
violent incidents on Sundays during football season, and find that “upset losses
lead to a 10% increase in the rate of at-home violence by men against their wives
and girlfriends” which contrasts with the fact that “losses when the game was
expected to be close have small and insignificant effects” (Card et al. 103).
Similar to my research, Card and Dahl also gathered information about reference
points through the NFL betting market. Their finding not only confirms evidence
of loss aversion and reference-dependent preferences, but also that reference
points among football fans are formed rationally and match betting odds.
Lastly, Ge (2018) conducts research on reference-dependent preferences
by analyzing the relationship between sports outcomes and passengers’ tipping
behavior. Ge argues that social norms and consumer sentiment are two main
factors that determine consumers’ tipping behavior. Ge uses data on New York
City taxi fares, tipping, and trip information to show that passengers tend to tip
more when a sports team unexpectedly wins, or win by greater score difference
than the expectation, but they do not pay less tip when there is an upset loss. Ge
explains the absence of loss aversion with the effect of social norms on people. As
a result, Ge’s study “[demonstrates] that while consumers’ reactions can still be
reference-dependent, loss averse behavior may possibly be muted in light of
social norms” (Ge 5).
Evidence from a number of studies finds evidence of reference-dependent
preferences in sports, both among players and fans. Standard approaches include
comparing sports game outcomes to expectations drawn from outside data like
betting markets, and then linking that data to outcomes collected elsewhere, like
from police reports. My study will use Google Trends data, which has been
shown to provide measures of animus and anger in previous work, as an outcome
measure.
Data
The project as a whole performs Google Trends searches on a dictionary
of words, which are coded as positive or negative, and links them to data on
football games. I took a dictionary of 8,223 words available on the University of
Pittsburgh’s Sentiment Lexicon website.
1
The dictionary is from work by Wiebe,
Wilson, and Hoffmann (2005) and Riloff and Wiebe (2003) and contains
information on whether each word is “positive” or “negative” and the strength of
that polarity. For example, “abhorrent” is strongly negative, and “civility” is
1
http://mpqa.cs.pitt.edu/
4
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
strongly positive. I kept only words that were either strongly negative or strongly
positive, removed duplicates, and only used the “non-stemmed” versions of the
words (for example, “angr” might be a stem for both “angry” and “angriest”).
This results in a dictionary of 3,408 words, 1,108 of which are positive and 2,301
of which are negative. I then used the gtrendsR package in R (Massicotte and
Eddelbuettel 2018) to perform Google Trends searches on each of the words in
the dictionary.
I performed the search separately by state, gathering weekly Google
Trends results from January 2015 to December 2018. Google Trends reports an
index score that shows the popularity of that word in that state and how it changes
over time. The score has no absolute meaning, but can be compared to itself and
so provide information on whether a search has gotten more or less popular over
time. Google Trends has previously been used as a measure of sentiment
(Stephens-Davidowitz, 2014). Importantly, Stephens-Davidowitz (2014) finds
that people use Google searches to express frustration, and so searches might be a
way to pick up frustration from sports losses. Choi and Varian (2012) emphasize
that Google Trends can provide relatively accurate predictions of near future and
present economic situations compared to existing surveys because Google Trends
eliminate the effects of the self-serving bias survey participants tend to exhibit.
I gather data on the point spread for games from 2015-2018 from
FootballLocks.com, a football betting site. The spread reports the expected
number of points by which a team will win or lose. I then link the point spread
data to information on the actual score of each game, and the day it was played,
from the NFL website, gathered by the nflscrapR package (Horowitz, Yurko, and
Ventura 2019).
I identify the state that each team plays in by hand, and then merge the
data on football spreads, scores, and dates with the Google Trends data. I assign
Google Trends scores collected on a week that starts 6 or 7 days before game day
as “before game” data, and Google Trends scores collected on a week that starts 0
or 1 days after game day as “after game” data.
To avoid overlaps where the same week of Trends data is “before” one game but
“after” another, I ignore the impact of a game on searches in a state if that same
state also played a game the week previous or the week following. I also drop
games in which both teams are from the same state. This results in 108 games
examined, one of which could not be linked to betting spread data.
5
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
Table 1: Summary Statistics
Statistic
N
Mean
St. Dev.
Min
Pctl(25)
Pctl(75)
Surprise
107
-8.332
8.571
-33.500
-13.000
-2.000
Expected
Win
107
-1.070
5.671
-16.000
-5.500
3.000
Actual Win
108
-9.370
7.974
-37
-13
-3
Table 1 shows the expected and actual score for each game in the data,
from the point of view of the team that lost.
2
The table shows the spread
(Expected Win), the actual spread (Actual Win, always negative since this table
shows the losing side), and the amount of Surprise, the difference between Actual
Win and Expected Win. Each observation is linked to Google Trends data for
3,408 words, both the week before and the week after the game. After dropping
results for word/state combinations with too few searches to produce results, the
final data set contains 1,360,956 observations, with observations uniquely
identified by the combination of game, state, word, and week-before-game/week-
after-game.
Methods
The primary results perform separate analyses for winning and losing
teams, since the reference-dependent preferences framework suggests the results
should be different for each. In each case, I regress the Google Trends score on:
• Whether the word is positive or negative
• Whether the score is collected before the game or after
• The amount of “surprise” from the results of the game
• And the interactions of all three variables
This gives us a regression equation:
for word at time for game with a team from state , with only one team per
game in the regression because winning and losing teams are estimated
2
All tables were prepared using the stargazer package (Hlavac 2018).
6
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
separately. I also present results using state fixed effects, replacing
with
, to
account for the possibility that some states tend to see larger surprises more often
and to potentially improve the precision of estimates. Regressions use robust
standard errors.
I am interested mostly in the effects of surprise on how the popularity of
words changes from before the game to after, and in particular on how those
effects differ by polarity of the words.
The coefficients of interest are: for losing teams, I focus mostly on
,
which shows how surprise affects the before-to-after change in word popularity
specifically among negative words (for which
). If
is negative,
that means that negative words see larger increases in popularity after a
particularly negatively surprising loss, supporting the reference-dependent
preferences theory. I also look at
, which shows the difference in how surprise
affects the before-to-after change in word popularity between positive and
negative words. If
is zero, then both positive and negative words respond the
same way to surprising games, contrary to the reference-dependent preferences
theory.
For winning teams, I am interested in
. If
is positive, then
particularly positively-surprising games lead to bigger increases in the popularity
of positive words, consistent with reference-dependent preferences.
I run the analysis in a second way. The first regression uses the Google
Trends score directly, but this may give us problems because the scale of the
Google Trends score can’t really be interpreted and may not be comparable across
words and states.
So, I create an indicator variable
equal to 1 if the Google
Trends score for word increased in state from before game to after, and
equal to 0 if it decreased. Words that stayed at the exact same Google Trends
score are dropped.
I then run the analysis
using a linear probability model to easily allow for state fixed effects.
For losing teams, I am interested in
. If
is negative, then big negative
surprises make negative words more popular. I am also interested in
. If
is
7
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
zero, then positive and negative words change in the same way in response to
surprise. Regressions use robust standard errors.
For winning teams, I am interested in
. A positive
shows that larger
positive surprises improve the popularity of positive words more than negative.
Results
Results Section A: Main Results
Table 2: Main Regression Results
Dependent variable:
Raw Google Index
Winning
Teams
Losing
Teams
Winning
Teams (State
FE)
Losing
Teams (State
FE)
(1)
(2)
(3)
(4)
After the game
0.089
*
0.070
0.089
*
0.070
(0.054)
(0.060)
(0.053)
(0.060)
Positive
3.137
***
3.284
***
3.129
***
3.269
***
(0.071)
(0.080)
(0.071)
(0.079)
Surprise
0.011
***
0.0003
-0.007
**
0.009
**
(0.003)
(0.004)
(0.003)
(0.004)
After the
game*Positive
0.150
0.049
0.150
0.049
(0.101)
(0.113)
(0.100)
(0.112)
After the
game*Surprise
0.003
-0.003
0.003
-0.003
(0.004)
(0.005)
(0.004)
(0.005)
Positive*Surprise
0.015
***
-0.002
0.014
**
-0.002
8
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
(0.006)
(0.007)
(0.006)
(0.007)
After the game*
-0.009
-0.003
-0.009
-0.003
Positive*Surprise
(0.008)
(0.009)
(0.008)
(0.009)
Constant
10.642
***
10.920
***
9.107
***
9.121
***
(0.038)
(0.043)
(0.074)
(0.090)
Observations
706,246
654,710
706,246
654,710
R
2
0.012
0.011
0.023
0.022
Adjusted R
2
0.011
0.011
0.023
0.022
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
Table 2 shows the results of the regressions described in the Methods
section, run separately for winning (Columns 1 and 3) and losing (Columns 2 and
4) teams, both without (Columns 1 and 2) and with state fixed effects (Columns 3
and 4). State fixed effects are included to account for potential state-level
differences in Google search activity.
The coefficient on Positive is positive, indicating that on average
positively-coded words have higher indices. However, since the trends score is
not necessarily meant to be comparable across words, this is not a result of
interest.
The coefficients on Surprise and Positive*Surprise are both often
significant, which is interesting because it suggests that more-surprising games
are related to more popular searches, especially for positive words, but both
before and after the actual game. This may have something to do with excitement,
but that is a speculative interpretation.
The coefficient on Surprise*After is insignificant for losing teams. This is
from our regression equation. The lack of significance here indicates that more
surprising losses are not related to increasing popularity of negative words. This
result fails to support reference-dependent preferences.
There is also no significance on Surprise*After*Positive,
, for either
winning or losing teams. The lack of significance here indicates that there is no
9
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
difference between Positive and Negative words in how surprising outcomes
affect popularity. This again fails to support reference-dependent preferences
State fixed effects do not change much, which is not too surprising as the
Google Trends scores are within-state.
Table 3: Before-After Increase Regression Results
Dependent variable:
Increase
Winning
Teams
Losing
Teams
Winning Teams
(State FE)
Losing Teams
(State FE)
(1)
(2)
(3)
(4)
Positive
0.002
0.004
*
0.002
0.004
*
(0.002)
(0.002)
(0.002)
(0.002)
Surprise
-0.0001
0.0001
0.0002
**
0.001
***
(0.0001)
(0.0001)
(0.0001)
(0.0001)
Positive*Surprise
-0.0001
-0.0002
-0.0001
-0.0002
(0.0002)
(0.0002)
(0.0002)
(0.0002)
Constant
0.505
***
0.501
***
0.501
***
0.492
***
(0.001)
(0.001)
(0.003)
(0.004)
Observations
353,123
327,355
353,123
327,355
R
2
0.00001
0.00004
0.001
0.002
Adjusted R
2
0.00000
0.00003
0.001
0.002
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
Table 3 shows regression results as described in the Methods section,
where the dependent variable is a binary indicator equal to 1 if the word increased
10
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
in popularity from before the game to after the game. The dependent variable is
equal to 0 if the word decreased in popularity. Words with no change are dropped.
When I include state fixed effects, precision increases and Surprise is now
significant. However, both values are very tiny and positive. This result implies
that, for winning teams, more-positive (better) surprises make negative words
more popular, counter to what is expected, since reference-dependent preferences
implies Surprise should improve popularity of positive words for winning teams,
not negative. Moreover, for losing teams, more-positive (i.e., less-negative, better)
surprises increase the popularity of negative words, when I would expect those
smaller surprises to have less of an effect.
Also, unlike what I have expected, there was no effect of
Positive*Surprise, which indicates that any impact of Surprise affects positive and
negative equally.
Results Section B: Robustness Checks
There are two concerns I have about our results: one is a possibility that
there is no response not because I am failing to replicate reference-dependent
preferences, but rather because the results of sports games have no effect on the
Google Searches at all. The other is that there is no effect because I am using a
linear term for Surprise. To check the first concern, I bring in the Table 4, which
pools together both winning and losing teams and repeats the analysis from
Tables 2 and 3.
Table 4: Pooled Analysis
Dependent variable:
Google Trends Index
Increase
Raw Score
Raw Score
(State FE)
Increase
Increase
(State FE)
(1)
(2)
(3)
(4)
After the game
0.090
**
0.090
**
(0.042)
(0.042)
11
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
Positive
3.295
***
3.283
***
0.007
***
0.007
***
(0.056)
(0.056)
(0.002)
(0.002)
Won
-0.204
***
-0.171
***
0.006
***
0.001
(0.041)
(0.042)
(0.002)
(0.002)
After the game*Positive
0.072
0.072
(0.080)
(0.079)
After the game*Won
0.027
0.027
(0.059)
(0.058)
Positive*Won
-0.038
-0.038
-0.006
**
-0.006
**
(0.078)
(0.077)
(0.003)
(0.003)
After the game*Positive*Won
0.003
0.003
(0.110)
(0.110)
Constant
10.938
***
9.154
***
0.501
***
0.495
***
(0.030)
(0.058)
(0.001)
(0.004)
Observations
1,366,966
1,366,966
462,692
462,692
R
2
0.011
0.023
0.00004
0.001
Adjusted R
2
0.011
0.023
0.00003
0.001
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
The first two columns use raw Google Trends scores as the dependent
variable. Here, search activity for both kinds of words increases after a game
relative to before. Winning the game is related to search behavior, but none of the
interactions are significant. These results imply that sports results have influences
on the search activity, even though the previous section showed that the
12
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
relationship does not appear to be consistent with reference-dependent
preferences.
The second two columns use the Increase from before variable as the
binary dependent variable. They show that positive words are about .7% more
likely than negative words to increase in popularity from before the game to after.
Moreover, one major discovery is that winning affects positive and negative
words differently, with positive words .6% less likely to increase after the game
than negative words.
These results imply that Google searches do respond in some way to
football results, but not in the way that reference-dependent preferences would
expect. These effects are tiny but nonzero. Connecting this result to another
literature on the reference-dependent preferences, Ge (2018) conducts a research
on reference-dependent preferences by analyzing the relationship between sports
outcomes and passengers’ tipping behavior. Ge argues that there are two main
factors that determine consumers’ tipping behavior: “social norms and consumer
sentiment” (Ge 3). Data used in the study came from a dataset of the New York
City Taxi and Limousine Commission which contains” fare, tipping and trip
information for taxi rides in New York City” (Ge 3). Ge found out that passengers
tend to tip more when a sports team unexpectedly wins, or win by greater score
difference than the expectation, but they do not pay less tip when there is an upset
loss. Ge explains the absence of loss aversion with the effect of social norms on
people. As a result, Ge’s study “[demonstrates] that while consumers’ reactions
can still be reference-dependent, loss averse behavior may possibly be muted in
light of social norms” (Ge, p. 5).
To check the second concern about the potential nonlinearity of the effect
of Surprise, I check the Increase for each word across each value of Surprise, non-
parametrically, in four graphs. I have a separate graph for positive and negative
words, and for winning and losing teams.
13
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
Figures: Nonlinear Effects of Surprise
Figure 1: Positive Words for Losing
Teams
Figure 2: Negative Words for Losing
Teams
Figure 3: Positive Words for Winning
Teams
Figure 4: Negative Words for Winning
Teams
In all four graphs, the LOESS curve is never significantly different from
the overall mean of .5, indicating that Surprise has no real relationship with
Increase for either positive or negative words, linear or otherwise. As a result, I
can conclude that our null result is not simply because of linearity.
14
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
Conclusion
The primary goal of this research paper is to find out whether the
surprising results of football games affect the terms people search on Google. I
was mainly interested in the effects of surprise on how the popularity of words
changes from before the game to after, and how that change is related to whether
those words are positive or negative. In order to do so, I utilized Google Trends
data. Before the analysis, I predicted that an unexpected loss would result in
increased searches for negative words, and an unexpected win would result in
increased searches for positive words.
Contrary to my expectation, I found no evidence in favor of reference-
dependent preferences. The results implied that Google searches do respond in
some ways to football results, but not in the way that reference-dependent
preferences would expect. These effects were tiny but nonzero.
This finding that the effects of football game results are insignificant in
terms of influencing football fans’ behavior provides a crucial point to consider in
further application of the reference-dependent preferences.
A number of studies, which include studies I have discussed in the
literature section, suggest that both players and fans behave according to
reference-dependent preferences in sports games. However, my paper has
discovered a potential weakness to these findings by demonstrating a failure to
replicate, questioning the credibility of studies on the reference-dependent
preferences. Consequently, governments and institutions should consider
carefully before taking reference-dependent preferences into account when
making decisions or establishing policies.
Bibliography
Bartling, Björn, Leif Brandes, and Daniel Schunk. 2015. “Expectations as
Reference Points: Field Evidence from Professional Soccer.” Management
Science 61 (11): 2646–61. https://doi.org/10.1287/mnsc.2014.2048.
Card, David, and Gordon B. Dahl. 2011. “Family Violence and Football: The
Effect of Unexpected Emotional Cues on Violent Behavior*.” The
Quarterly Journal of Economics 126 (1): 103–43.
https://doi.org/10.1093/qje/qjr001.
15
Lee: Reference-Dependent Preferences Among NFL Fans
Published by Digital Commons @ IWU, 2019
Choi, Hyunyoung, and Hal Varian. 2012. “Predicting the Present with Google
Trends.” Economic Record 88 (SUPPL.1): 2–9.
https://doi.org/10.1111/j.1475-4932.2012.00809.x.
Ge, Qi. 2018. “Sports Sentiment and Tipping Behavior.” Journal of Economic
Behavior & Organization 145 (January): 95–113.
https://doi.org/10.1016/j.jebo.2017.10.016.
Hlavac, Marek. 2018. Stargazer: Well-Formatted Regression and Summary
Statistics Tables. (version 5.2.1). https://CRAN.R-
project.org/package=stargazer.
Horowitz, Maksim, Ron Yurko, and Samuel Ventura. 2019. NflscrapR:
Compiling the NFL Play-by-Play API for Easy Use in R (version 1.8.1).
https://github.com/maksimhorowitz/nflscrapR.
Massicotte, Philippe, and Dirk Eddelbuettel. 2018. GtrendsR: Perform and
Display Google Trends Queries (version 1.4.2). https://CRAN.R-
project.org/package=gtrendsR.
Pope, Devin G, and Maurice E Schweitzer. 2011. “Is Tiger Woods Loss Averse?
Persistent Bias in the Face of Experience, Competition, and High Stakes.”
American Economic Review 101 (1): 129–57.
https://doi.org/10.1257/aer.101.1.129.
Riloff, Ellen, and Janyce Wiebe. 2003. “Learning Extraction Patterns for
Subjective Expressions.” In Proceedings of the 2003 Conference on
Empirical Methods in Natural Language Processing -, 10:105–12. Not
Known: Association for Computational Linguistics.
https://doi.org/10.3115/1119355.1119369.
Stephens-Davidowitz, Seth. 2014. “The Cost of Racial Animus on a Black
Candidate: Evidence Using Google Search Data.” Journal of Public
Economics 118: 26–40. https://doi.org/10.1016/j.jpubeco.2014.04.010.
Wiebe, Janyce, Theresa Wilson, and Claire Cardie. 2005. “Annotating
Expressions of Opinions and Emotions in Language.” Language
Resources and Evaluation 39 (2): 165–210.
https://doi.org/10.1007/s10579-005-7880-9.
16
Undergraduate Economic Review, Vol. 16 [2019], Iss. 1, Art. 10
https://digitalcommons.iwu.edu/uer/vol16/iss1/10
