PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.


Does activism impact consumer behaviour? This note aims to shed light on this issue by using, as a case study, the National Basketball Association's engagement with the Black Lives Matter movement. We show that social action does not affect TV ratings and viewing figures. However, the perceptions of social media users, represented by negatively toned tweets, is shown to have a detrimental impact.
DO NOT shut up and DO dribble:
Media Coverage and TV Consumption
Matteo Pazzona1and Nicola Spagnolo1,2
1Department of Economics and Finance, Brunel University London, UK
2Centre for Applied Macroeconomic Analysis (CAMA), Australian National University,
Does the involvement of firms in controversial social issues have any effect on
consumers? This paper aims to shed light on that question by using, as a case study,
the 2020 National Basketball Association’s engagement with the Black Lives Matter
movement. Using a detailed dataset, we explore the extent to which both social
and traditional media coverage affected the NBA’s TV daily audience. The results
suggest that the latter was not influenced by the intensity of media coverage. Our
findings are robust to an alternative identification approach, difference-in-difference,
with the National Hockey League as control. Despite the absence of an overall effect,
we do find that the tone of social media, as well as the identity of the users, played
a significant role in affecting consumption decisions.
Keywords: Media Coverage,Black Lives Matter,Consumer Behaviour, and Activism.
JEL Classification: D12,D80,L82.
The title of the paper refers to the sentence “Shut Up and Dribble”, used by the journalist Laura
Ingraham to rebuke Lebron James (NBA player) for talking politics. We would like to thank Marco
Barassi, Olivia Cassero, Alejandro Corvalan, Peter Dawson, Roman Gauriot, Giambattista Rossi, Juan de
Dios Tena, and Jan Van Ours. We would also like to acknowledge John Lewis, founder of Sports Media
Watch, for providing us with the TV audience data. Finally, special thanks to Angelo Azzu and Exequiel
Padilla, who helped with the Python and R codes. Corresponding author: Matteo Pazzona, Economics
and Finance, Brunel University London, Middlesex UB8 3PH, UK. Email:
1 Introduction and Background
Until a few years ago, global corporations adopted a low profile concerning controversial
issues, such as immigration or racism (Smith and Korschun, 2018). Recently, companies
have appeared to be more prepared to take a firm position, becoming socially and politically
engaged. At the same time, the extensive role of both social and traditional media amplifies
the stances of the firms and allows more people to be reached. It is natural to ask whether
consumers are affected by the attention devoted by media to such controversial issues. We
aim to shed light on this matter by using, as a case study, the 2020 engagement of the
National Basketball Association (NBA) with the Black Lives Matter (BLM) movement.
More specifically, we focus on the extent to which the media coverage of this engagement
influenced sports consumers’ choices, as measured by TV ratings and viewers.1
BLM is a social movement, which originated in 2013 to protest against police brutality
towards black people in the USA. In June 2020, following the death of George Floyd,
protests erupted across the USA. Floyd’s death prompted one of the biggest civil rights
movements in recent American history. Among the many supporters of the BLM, the
National Basketball Association has been one of the most prominent and outspoken. When
play resumed at the end of July 2020, after the suspension due to Covid-19, the NBA took
a firm but controversial position, allowing athletes to wear a social justice message on the
back of their jerseys instead of their surnames, and to paint the BLM slogan on basketball
courts. The Black Lives Matter protests were endorsed by many leading players, as well
as the NBA commissioner Adam Silver. On June 1, 2020, he declared: “Together with our
teams and players, we will continue our efforts to promote inclusion and bridge divides”.2
Compared to other major sports leagues, the NBA made greater efforts to spread awareness
about social injustice issues and pushed for change. The position taken this time was
certainly stronger and more widely echoed by both traditional and social media platforms.
While the NBA’s BLM activism was intended to increase awareness of social issues, this
endorsement was criticized by some politicians. For example, in September 2020, President
Donald Trump tweeted: “People are tired of watching the highly political NBA. Basketball
ratings are way down, and they won’t be coming back”. President Trump’s statement
implied that consumers punish firms that display too much political engagement. Adam
Silver replied: “No data BLM on-court hurts NBA ratings. There is no doubt there are
1This media has played a central role in the development of sport for most of this century. Baimbridge
et al. (1996) was among the first to examine the above relationship looking at broadcasting and football
using UK data.
2It is worth noting that before the start of the 2020-2021 season, the NBA commissioner announced
that the league would no longer place social justice messages on courts or jerseys. We take advantage of
the fact that the NBA’s political activism was confined to a specific period (June to November 2020) and
the potential effect of political activism on viewers and ratings can therefore be tested.
some people who have become further engaged with the league. They respect their right to
speak out on issues that are important to them”. As we will see shortly, this controversy was
strongly echoed in the traditional media and, especially, social media. It became natural
to ask: did individuals change their NBA TV consumption as a result of the interest
generated by the league’s involvement in the BLM movement? Individuals might decide to
decrease their viewing consumption because they do not approve the NBA being involved
in non-sport businesses. On the other hand, some people might increase their viewership
pattern as a support of the social stance fostered by the league. Some other individuals
might not be affected at all because they have a low elasticity of NBA TV demand with
respect to this social dimension. The latter is a selected group of individuals watching
NBA games regardless of this other dimension. Ultimately, the nature of our research
question is empirical and, as such, we do not have any prior regarding the average impact
on consumers.
To disentangle the role of media coverage from other concurrent factors, we first con-
ducted an event study, employing high-frequency data and including a rich set of control
variables. As a measure of the TV audience, we collected data on the viewers and ratings
for all the NBA matches aired on the national US networks in the 2018-19 and 2019-20 sea-
sons. Although TV audience data does not represent a traditional measure of consumption,
it is closely related to revenues (Kanazawa and Funk, 2001).3
To measure the exposure of individuals to the NBA’s involvement in the BLM move-
ment, we employed three different proxies, based on Twitter, newspapers, and Google
Trends data. These represent sources through which the general public obtains informa-
tion. Twitter - our proxy for social media - is an awareness barometer, which represents
the discussion that the population at large engages in, when actively participating in a
debate (Acemoglu et al., 2018; Austmann and Vigne, 2021; Gorodnichenko et al., 2021).
We calculated the total daily number of original tweets that included words referring to
both the NBA and BLM. On the other hand, the newspapers metric is based on edito-
rial decisions and, thus, it captures the indirect exposure of individuals to a specific topic
(Baker et al., 2016; Caporale et al., 2022, among others). Newspaper coverage is proxied
by story headline counts appearing in the US Today and The New York Times. News
headlines were selected using a string search containing the terms ‘NBA’ and ‘BLM’. Fi-
nally, Google Trends tracks the interest generated by a wider audience, therefore providing
another measure of general interest. The use of Google-sourced data has been heavily used
in the existing literature (Vosen and Schmidt, 2011; Choi and Varian, 2012; Arin et al.,
3Using data on Nielsen ratings for locally televised NBA basketball games, Kanazawa and Funk (2001)
found that higher ratings allowed NBA teams to realize greater advertising revenues.
The baseline empirical exercise consists of regressing NBA match audience data, our
dependent variable, on the social and/or traditional daily media coverage. The results show
that social and political engagement - consistently for all the three measures - did not hurt
TV viewers and ratings. Our results are robust to various identification threats, such as
measurement errors and reverse causality. To take into consideration potentially omitted
variables, we complement our analysis with a difference-in-difference (DiD) approach, using
the National Hockey League (NHL) as a control. The NHL provides an ideal comparison
with the NBA, as both leagues experienced similar negative trends in TV viewers over the
period under consideration. Similarly to the above findings, there is no evidence that the
echo of the NBA involvement contributed to the decline in viewers and ratings. Although
the data do not allow us to infer the change in behavior at the individual level, results
highlight how the aggregate TV demand - with respect to greater involvement of the NBA
in social issues - is relatively inelastic.
Despite the absence of an overall statistically significant effect, we find interesting results
by exploiting the granularity of Twitter data. For example, Tweets written by popular
authors have a positive, and significant, effect on ratings. We show that Tweets sent by
NBA stakeholders - a potential measure of activism - lead to an increase in the TV audience.
Finally, we find some evidence that the tone of the messages plays a role: positive tweets
are associated with higher ratings, although the evidence is weaker for negative ones.
This study relates to various strands of the literature. Firstly, it contributes to the
strand that analyzes whether media coverage, and its tone, affects consumer behavior
(Carroll, 2003; Lerner et al., 2007; Lamla and Lein, 2014; Biolsi and Lebedinsky, 2021).4
The last of these studies analyzed how consumer spending changed in response to news
coverage of the investigation into collusion between Donald Trump’s presidential campaign
and Russia. This work rests on the idea that when individuals receive news favorable to
their preferred political party, they will feel more optimistic and thus spend more. We use
a similar argument investigating whether individuals sharing (or not sharing) the view that
the NBA should engage in social activism would make them more inclined to watch (or not
watch) televised NBA matches.
Our work also contributes to the literature looking at beliefs and consumption patterns
(Angeletos and La’O, 2013; Gillitzer and Prasad, 2018; Benhabib and Spiegel, 2019). These
studies recognize the influence of non-fundamental factors such as beliefs and opinions
on consumers’ spending decisions. Within this context, recent literature has found that
social media is increasingly influencing beliefs and fostering activism among consumers
(Bovitz et al., 2002; Hendel et al., 2017; Enikolopov et al., 2020; Zhuravskaya et al., 2020;
4Relatedly, Ananyev et al. (2021) showed how exposure to Fox News Channels affected physiscal dis-
tancing during COVID-19. Finally, there is also a strand of the literature investigating the impact of news
on media users (Mastrorocco and Minale, 2018).
Gorodnichenko et al., 2021).5In this study, we explore the consequences of the NBA’s
activism, with the media functioning as an echo chamber.
This paper also contributes to the large literature that analyzed the demand for sports,
especially for TV broadcasting (Buraimo et al., 2022; Hausman and Leonard, 1997; and
Forrest et al., 2005). This literature focused on aspects such as outcome uncertainty,
team identification, or quality of players. Very few studies analyzed the role of other
determinants. Buraimo et al. (2016) - the closest to our work - documented the negative
effect of the (in)famous scandal - Calciopoli - on stadium attendance for the convicted
teams. However, that scandal had a negative connotation - being related to collusion with
the referees - whereas in our setting, the involvement in the BLM could be welcomed, or
not, by the fans.
Finally, this work is also linked to the recent marketing literature investigating whether
social and political activism affects a firm’s performance (Scherer et al., 2014; Smith and
Korschun, 2018). We contribute to this literature by separating the role of firms’ involve-
ment from other co-founders.
The layout of the paper is as follows. Section 2 explains the data. Section 3 presents
the results of the event study design alongside the robustness checks. Section 4 analyzes
further questions, using data from Twitter. Section 5 provides a summary, and offers some
concluding remarks.
2 Data
In this section, we describe the variables employed in the empirical analysis presented in
Section 3 and Section 4.
2.1 TV Audience
Major US sports leagues are not public companies and are, therefore, not listed on the
stock market. As such, we cannot use the share value, as a measure of the firm’s success.
Therefore, we opted to estimate the performance of the NBA by using the following two
indicators: TV viewers and ratings.6
5Gorodnichenko et al. (2021) showed that social media diffusion had a profound influence on opinions
related to the Brexit referendum and the 2016 U.S. Presidential Election. Larson et al. (2019) highlighted
how individuals involved in the 2015 Charlie Hebdo incident were much more connected than those not
involved. Similar findings have been obtained by Halberstam and Knight (2016). Enikolopov et al. (2020)
found a robust link between the penetration of social media and the protests in Russia in 2011. There is
abundant literature concerned with the role of traditional media (DellaVigna and La Ferrara, 2015).
6TV viewership and ratings are different but related measures of TV audience (Nielsen, 2021). Ratings
refer to the percentage of TV homes in the U.S. tuned in to a specific program. A similar concept is ‘TV
share’, which is a percentage based on the number of households watching television. ‘Viewers’ and ‘viewing
Broadcasting of NBA games in the USA operated in a similarly across major sports
leagues, such as the National Football League or National Hockey League. Some games are
aired on national channels, whereas the rest is available only on regional networks. Our
analysis focuses on the former due to the lack of detailed data for the latter. Nationally
broadcasted games are available on cable TV channels, such as ABC and TNT. Consumers
usually subscribe - by paying a fee - to a package that includes a bundle of other channels,
offering news, movies, etc. Once users have subscribed, they have unlimited access to
all the programs broadcasted by those networks.8From an economic perspective, the fee
represents a sunk cost, with an almost zero marginal cost of consumption.9This is ideal
for our empirical analysis because consumers can instantly, without additional costs, adapt
their viewership behavior. We can also think about the decision to pay the subscription
as the extensive margin, whereas the actual consumption represents the intensive margin.
Our data, and settings, are more suited to investigate the latter.
We gathered data on the former for all the 506 NBA matches aired by the national
networks in the 2018-19 and 2019-20 seasons. The data for ratings refers to 490 matches.
For each of those, we obtained detailed information about the date and time, the broad-
casting network, and the type of game whether regular season or play-off. It should be
noted that, on some days, more than one match was televised. Table 1 shows that the
average number of viewers was 2.35 million, whereas the average rating was 1.47 million.
About 68% of the matches were aired on prime time (after 8 pm ET), 33% were played
during the playoff, and 20% at the weekend. The majority of the games were broadcasted
by ESPN and TNT.10 Table 1 also reports some key summary statistics for the NHL,
which we will use as a comparison to the NBA in the difference-in-difference setting. We
collected viewing-figure data for 414 NHL matches and ratings data for 278 matches. All
the TV audience data is at the national level, as data for lower administrative levels were
not available.
[Please insert Table 1 about here]
figures’ refer to the total number of people that watch a program. TV ratings and viewing figures do not
necessarily capture the entirety of media consumption, as in recent years there has been a shift in the way
that people watch TV, with programs increasingly offered on internet-related platforms. Unfortunately,
the latter data is not available. The revenue generated by national TV deals represents the main source
of income for professional sports leagues. As of 2021, such deals were worth 2.7 U.S
billions, more
than a third of total NBA revenues. Furthermore, there is a clear positive link between TV audience and
advertising revenues Kanazawa and Funk (2001). 7
8That is the biggest difference with pay-per-view.
9It is not exactly zero because of the presence of ads in TV programs.
10These figures are not shown in Table 1 to save space.
2.2 NBA-BLM News Coverage
To capture the echo chamber of the NBA/BLM topic, we employ three different measures.
The first is based on social media, which has been shown to affect consumption and in-
vestments.11 We consider Twitter rather than other social media because messages on
this platform are more likely to indicate the general patterns of opinions and provide a
valid measure of public attention (Soroka et al., 2018; Shen et al., 2019). More specifically,
we searched all the original Twitter messages that contained the words ‘NBA’ and ‘BLM’
simultaneously, either as abbreviations or in their fully spelled-out forms. For example, we
searched for ‘BLM’, ‘#BLM’, ‘Black Lives Matter’, ‘Black Live Matters’, etc.12 We decided
to focus on original tweets to follow the existing literature (Hatte et al., 2021). Informa-
tion on Tweets was all retrieved from Twitter Streaming API, and we only considered only
messages written in English - or with an undefined language only if hashtags were included.
We thus created the variable Tweets NBABLM, to represent the total daily number of
original tweets. As the top panel of Figure 1 shows, the values were consistently negligible
until the end of May 2020, but then spiked dramatically. There were two clear peaks: one
after the resumption of the NBA, at the end of July 2020, while the other was linked to
the boycott following the police shooting of a black man in Wisconsin, in August of the
same year.
[Please insert Figure 1 about here]
Our second measure of news coverage is based on traditional media outlets (Baker et al.,
2016; Caporale et al., 2022). Following the similar criteria adopted for social media, we
gathered data on the number of daily articles published in USA Today and The New York
Times, two of the most read daily newspapers in the USA. We searched for online articles
that included the key terms ‘NBA’ and ‘BLM’, and created the variable USA & NYT,
which combines the daily articles for the two newspapers. Tweets NBABLM and USA &
NYT are highly related, with a correlation coefficient of 0.82. This can be seen in the top
panel of Figure 1, which reports the two daily series one next to each other.
Our last measure of news exposure is based on Google Trends, which reflects the Google
search intensity. The results are scaled on a range of 0 to 100, where 100 represents the
maximum number. Google Trends is a proxy of the level of interest in a particular topic
and has been used frequently in the literature, especially to predict consumption patterns
(Vosen and Schmidt, 2011; Choi and Varian, 2012). The correlation coefficient of Google
11As pointed out by Austmann and Vigne (2021), the discussions on Twitter are most likely to extend
outside its users, and reach a much wider audience.
12The results are consistent with more restrictive criteria, such as including only Tweets with the NBA
and BLM hashtags, as shown in the appendix.
Trends with Tweets NBABLM is 0.61 and with USA & NYT is 0.69. The two highest
values for all our measures are found during the boycott period, at the end of August 2020.
In the bottom panel of Figure 1, we report the trends for Tweets NBA and Tweets
BLM. The former represents the number of tweets with the hashtag #NBA (excluding
#BLM), and the latter is the number of tweets with the hashtag #BLM (without #NBA).
We calculated these two variables to capture all the news coverage related to the NBA and
the BLM separately, i.e. independent of the one on the NBA-BLM together. The number
of Tweets NBA follows a cyclical path, in which more tweets were released at the beginning
of the regular season and during the play-off. On the other hand, Tweets BLM peaked
immediately after the death of George Floyd. Comparing the bottom and top panels, we
notice that the involvement of the NBA in the BLM movement did not coincide perfectly
with the trend of Tweets NBA and Tweets BLM. For example, Tweets NBABLM,USA
& NYT and Google Trends did not peak in the aftermath of Floyd’s death, which shows
how the NBA’s involvement in the movement manifested itself at a later period. This is
reassuring, as the NBA/BLM-related news coverage captures a specific trend.
3 Empirical Analyses: Baseline & Robustness
In this section, we study whether the coverage of news related to the NBA’s involvement
in the BLM had any significant impact on TV consumption. We start by analyzing the
baseline model, followed by two sets of robustness.
3.1 Baseline Model
Our baseline model is the following:
T V Consumptioni,t =α+β M edia E xposurei,t1+γXi,t +εi,t,(1)
,where istands for matches and tfor day. TV Consumption refers to the number of
viewers (in logs) and ratings, taken in turns.13 Media Exposure captures the NBA/BLM
news coverage, which is a proxy of the interest in the topic. As detailed in the previous
section we have three measures: Tweets NBABLM,USA & NYT, and Google Trends.
For all these measures, we consider the value of the day before the match - t1 - to limit
potential simultaneity issues. Xis an extensive set of control variables, which are related
to both news coverage and TV audience. As a measure of news coverage for both NBA
and BLM, we consider Tweets BLM and Tweets NBA separately. We also include Key
Dates, a binary variable for days referring to the restart of the NBA in July 2020, and the
13Results with viewers in levels are consistent and available upon request.
boycott in August 2020. After takes a value of one for any matches played following the
resumption of the NBA at the end of July 2020 (see Figure 1, top panel). Several binary
variables capturing the round of play-offs - keeping regular season matches as the excluded
category - are controlled for. Two additional variables indicate whether the match was
played in prime time, or during the weekend. The regression also controls for the three
TV networks which broadcasted the highest number of matches: ESPN,ABC, and TNT.
Finally, we include a linear daily trend and season-fixed effects.
[Please insert Table 2 about here]
The results - presented in Table 2 - start with Viewers. Column (1) reports the findings
with Tweets NBABLM, column (2) with USA & NYT and (3) with Google Trends. The
coefficient is not statistically significant, at conventional levels, in all three specifications
proposed. In columns (4) through (6) we repeat the exercise using Ratings as dependent
variable. Again, we find limited evidence of an impact of media intensity on TV Consump-
tion, except in column (4) where the coefficient for Media Coverage is significant at the 0.1
Turning to the other variables, Tweets NBA and Tweets BLM are never significant. The
variable After shows the loss in TV consumption that occurred when the NBA resumed
playing, but the effect is significant only for Ratings. Matches during the play-off period
attracted a bigger audience, with the Finals having the highest value. As expected, Prime
Time is positive and significant. Overall, the results point to a non-effect - or a relatively
small one - of political engagement on the demand for NBA entertainment.
3.2 Robustness I: Match FE & 2SLS
By including a rich set of variables and using high-frequency data, as well as lagged values,
we argue to offer a convincing identification strategy.14 However, it is still plausible that
the media coverage related to the NBA/BLM is endogenous to the TV audience. A possible
concern regards the omission of co-founders factors that could affect both media coverage
and TV audience. For example, a match between the Chicago Bulls and the Los Ange-
les Lakers might attract more interest - ceteris paribus - to another less prestigious one.
Similarly, games played between teams in urban, and multicultural, areas might attract
different audiences compared to those in more rural and homogeneous ones. To control for
such specific characteristics, in Table 3, we replicate the baseline results adding match-up
14However, as noted by Reed (2015), the use of lagged values does not necessarily solve the endogeneity
fixed effects.15 As done previously, we report the results first for Viewers and then for Rat-
ings. These confirm, and somehow reinforce, the absence of any significant effect of Media
Coverage on TV Consumption. To further control for the presence of time-varying omitted
factors, in Section 3.3, we will conduct a difference-in-difference analysis, comparing the
National Basketball Association(NBA) with the National Hockey League(NHL).
[Please insert Table 3 about here]
There are still two potential sources of endogeneity that need to be addressed: mea-
surement error and reverse causality. We deal with the latter in Section 4 using detailed
data from Twitter. The former might be an issue if our measure of media exposure does
not capture correctly the level of interest generated by the involvement of the NBA in the
BLM movement. The use of three alternative measures - all pointing to the same results -
should shield us from this potential identification threat. However, as a further check, we
run a 2SLS regressions model following a procedure used by Chalfin and McCrary (2018)
on the effect of police on crime in the USA. The authors showed that when two variables
are noisy measures that aim at capturing the same thing, it is possible to use one as an
instrument for the other. In our case, we have three measures capturing NBA/BLM media
exposure, i.e. we can use two as instruments. We opted to instrument Tweets NBABLM
with USA & NYT and Google Trends. The results - shown in columns (4) and (8) - confirm
the absence of an effect on TV Consumption. As anticipated, the instruments are highly
correlated with the Tweets NBABLM, and the F-statistics is well above the rule of thumb
suggested by Stock et al. (2002).
3.3 Robustness II: Difference-in-Difference
The results presented so far suggest that days with high public interest in the involvement
of the NBA in the BLM movement were not followed by significant changes in media con-
sumption. To further control for the presence of omitted time-varying factors, in this section
we run a difference-in-difference model using the National Hockey Association (NHL) as a
comparison.16 The NHL is the best counterfactual for the NBA, as it shares many similar-
ities. Firstly, the NHL runs from October to June in non-pandemic years, like the NBA.
Secondly, both leagues were forced to stop playing in March 2020 and then resumed in
July 2020, as shown in the top panel of Figure 2. Thirdly, a decrease in the number of TV
15Match-up refers to a game played by the same two teams. In our data we have a total of 248 match-ups,
which implies a severe reduction of the degrees of freedom and explains why we do not have them in the
baseline model.
16It needs to be stressed that the event study analysis and the DiD are conceptually different. In the
former, the treatment variable represents the media intensity. In the latter, it assumes a constant effect
throughout the after period.
viewers had been experienced by both leagues over the previous few seasons. Despite these
similarities, however, the two leagues differed greatly in their engagement with the BLM
movement. The NHL was less involved, as shown in the summary statistics in Table 1.17
We are aware that other major leagues might be slightly more similar to the NBA in terms
of demographic including race - and political affiliation, compared to the NHL.18 The
National Football League was not considered given the limited number of matches played,
and the absence of overlap in the regular season.
We first provide a visual inspection of the evolution of Viewers for both seasons and
leagues. Figure 2 is divided into three sub-samples, where the first represents the entire
2018-19 season; the second shows the earlier part of the 2019-20 season until the break
caused by the Covid 19 pandemic; the last sample covers from the end of July until October
2020, i.e. when the seasons resumed. The figure reveals how the average number of viewers
is higher for the NBA, compared to the NHL. However, both leagues share a similar pre-
event trend, supporting the use of a difference-in-difference technique.19 To corroborate
the visual inspection, we estimate the following model:
T V Consumptioni,t =αNBAi+βAf tert+θNBAiAf tert+γXi,t +εi,t,(2)
the variables have been already discussed in Section 3. In addition, we control for
Tweets NHL, which measures the number of tweets with NHL without BLM. θis the
coefficient of interest. We report the results with two specifications: the first includes
season’s fixed effects only, whereas the second adds the whole set of control variables, Xi.
The reason for doing so is to investigate how the results are affected by the inclusion of
relevant TV consumption determinants. The residuals are clustered at the league level,
and P-values are calculated by means of wild bootstrap (Cameron and Miller, 2015). In
total, we estimate four models, two for each audience measure. Table 4 shows that θis
never statistically significant. These results suggest that the NBA did not lose viewers
(or experienced reduced ratings) as a result of its engagement with the BLM movement.
Overall, the latter results appear to provide further evidence in support of findings reported
in Section 3.
[Please insert Table 4 and Figure 2 about here]
17The NHL did not make any clear statements about the movement and rarely addressed racial issues
directly. This is despite some players showing support for the BLM movement on their social media
platforms. Moreover, players decided to boycott four play-off games in November, following the killing of
Jacob Blake. Similar to the NBA, though, two games in the 2020 Stanley Cup finals were postponed, on
August 27 and 28.
18There are four major professional leagues in USA/Canada: the NFL, NHL, MLB and NBA.
19Results are robust to the use of monthly leads and lags, and are available on request.
We are quite confident about the absence of an effect of media exposure on TV audience.
However, the zero average effect might be driven by a substantial degree of heterogeneity.
In the next section, we investigate such a possibility by exploiting the granularity of the
social media data.
4 Social Media Echo Chamber & Sentiment Analysis
In this section, we exploit the quality, and granularity, of the social media data to answer
additional research questions.
4.1 Social Media
Our main regressor in Table 2 - Tweets NBABLM - represents the number of original
tweets. As such, it does not necessarily inform on a) how well these were received by
Twitter users, and b) the role of influential authors.20 To take the first issue into account,
in columns (1) through (3) of Table 5 for Viewers - and (5) to (7) for Ratings - we employ
alternative measures based on the popularity of the messages. In columns (1) and (5) we use
Tweets NBABLM Pop I, which sums tweets, retweets, and likes. This variable is 38 times
bigger than Tweets NBABLM, with a very high standard deviation.21 In columns (2) and
(6) we add the quotes, and replies, for each original tweet, along with retweets and likes.
To take into account the identity of the authors of the messages, in columns (3) and (7) we
multiply the daily number of original tweets by the average number of followers of Twitter’s
authors that posted that day. Results reveal a positive and significant effect only for the
latter measure, both for Viewers and Ratings. This suggests that the popularity of authors
- measured by followers - could be important in explaining changes in viewers’ behaviors. In
columns (4) and (8), we consider the role of the identity of the authors of the Tweets. More
specifically, we create the variables T weetsN BA BLM, Leag +P la and T weetsN BA
BLM , F ans. The former represents the total number of tweets sent by accounts belonging
to NBA stakeholders, from the league, teams, and players respectively. This measure is
the closest proxy to the activism of the NBA because it represents the unfiltered support
of the league to the BLM movement. T w eetsNB A BLM, F ans represent the Tweets
sent by fans, i.e. those who are not the stakeholders. This exercise reveals a positive,
and significant, effect of T weetsN BA BLM, Leag +P la on both Viewers and Ratings,
while no effect is found on T weetsN BA B LM, F ans. If anything, activism was able to
20The same issue applies to the other measures based on traditional media, and Google Trends. However,
social media data allow us to identify precisely the impact on consumers.
21The correlation between the two measures is 0.75.
mobilize viewers.22
[Please insert Table 5 about here]
As mentioned in the previous section, a threat to identification is reverse causality if
social media activity is influenced by TV audience. It is fair to suspect that the NBA
and its players adjust their twitting behavior according to TV audience performance. For
example, the NBA media department might refrain to send BLM-related Tweets if ratings
are decreasing to avoid upsetting potential viewers. To take this possibility into account,
in column (9), we regress the number of Tweets by the NBA and players (T weetsN BA
BLM, Leag +P la) on a binary variable which takes a value of one if the day before
the release of the Tweets, the number of viewers was in a decreasing trend. The results
reveal an insignificant coefficient that points to the absence of strategic behavior by the
4.2 Sentiment Analysis
In this section we explore whether a potential factor explaining sports consumption behavior
is not the number of tweets but rather their tone (Dawson et al., 2014; Lamla and Lein,
2014). The following question arises: are TV consumers more affected if the general tone
towards the NBA is negative or positive? Therefore, we classify messages according to
their sentiment. We first grouped each tweet by its constituent words and then removed
all link tokens, punctuation, links, etc.23 All the remaining words were classified using
popular sentiment lexicons (Salvatore et al., 2021). In Table 6, we report the results
using two popular lexicons Bing and Afinn. The former classifies 6,788 English words
as either positive or negative. The score for positive words is tagged with 1, while for
negative words it is -1. Afinn classifies 2,476 English words using a scale from -5 (the most
negative sentiment) to +5 (the most positive). For each lexicon, we calculated a measure
representing the daily number of tweets associated with a particular sentiment.24 The
results for Bing are reported in columns (1) and (3). In columns (2) and (4), we employ
Afinn, considering all words with a score between -5 and 0 as negative, and from 1 to 5 as
positive.25 A pattern now appears to emerge, with the results suggesting that the tone of
the message matters. Days with a greater number of positively toned tweets are followed
22This result could also be linked to the literature on the impact of role models on individuals’ behavior
(Farina and Pathania, 2020).
23This process is referred to as tokenization.
24We first calculated the average daily score of the daily tweets. We then multiplied such measures by
the total number of tweets, i.e. T weetsN BA BLM .
25The results are consistent with other classifications, such as considering -5, -4, and -3 as negative and
+3, +4 and +5 as positive.
by an increase in the number of Viewers and Ratings, whereas negatively toned tweets by
a decrease. However, the results for positive are more significant compared to negative
ones. This is consistent with the view that response to positive and negative messages are
asymmetrical (Soroka, 2006; Akhtar et al., 2011).
[Please insert Table 6 and Table 7 about here]
In Table 7, we consider another popular lexicon, NRC, which classifies words into eight
basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) and two
sentiments (positive and negative). The results for Viewers and Ratings - in columns (1)
and (5) - show a similar pattern to those shown in Table 6. However, individual sentiments
do not explain much of the variation in TV audience. In the remaining columns, we provide
some robustness exercises to the findings reported in Table 6. In columns (2), (3), (6), and
(7) we multiplied the daily average score using Bing, or Afinn, by the total number of
tweets with at least a word matched by that lexicon.26 Results are consistent with previous
findings. Finally, in (4) and (8), we regress TV Audience on the standard deviation of the
Afinn Score, a measure of tones’ polarization. This analysis does not reveal a significant
5 Conclusions
This paper investigates whether the media exposure of the NBA to the BLM movement had
any effect on TV ratings and number of viewers. As a measure of media exposure, we have
considered data from Twitter, newspapers and Google Trends. Our analysis shows that
media exposure is not associated with any statistically significant change in the number
of viewers or ratings. These findings are robust to various model specifications, such as a
difference-in-difference model, using the NHL as a control group.
Exploiting the granularity of Twitter data, we show that messages posted by NBA
stakeholders such as individual players and the league itself increase ratings. It can
be argued that the league’s activism might have impacted consumer decisions. We also
show that the popularity of the authors of tweets is relevant. As a final exercise, we have
studied whether consumer sentiment plays a role, finding robust evidence that positively
toned tweets increase consumption. On the other hand, the role of negatively toned tweets
is less clear.
Our results have some important implications. In the context of increasing demands
for firms to advocate for social causes, we find that media exposure is not costly, at least
26As such, the final number of tweets associated with sentiment is lower than the analysis presented in
Table 2. Further differences in the number of tweets employing the two lexicons are due to the heterogeneity
in the number of classified words in each lexicon.
in the short term. Our analysis offers further insights into the elasticity of demand for
sports with respect to non-economic factors. Continuing, we show that the role played
by social media in supporting (or opposing) the activism of firms appears to be relevant.
Further research could explore the implications of political engagement using micro level
data, therefore controlling for individual consumer characteristics and preferences, as well
as for a wider range of mainstream media news and social media platforms.
Acemoglu, D., T. A. Hassan and A. Tahoun, “The power of the street: Evidence
from Egypt’s Arab Spring,” The Review of Financial Studies 31 (2018), 1–42.
Akhtar, S., R. Faff, B. Oliver and A. Subrahmanyam, “The power of bad: The
negativity bias in Australian consumer sentiment announcements on stock returns,”
Journal of Banking & Finance 35 (2011), 1239–1249.
Ananyev, M., M. Poyker and Y. Tian, “The safest time to fly: pandemic response
in the era of Fox News,” Journal of Population Economics 34 (2021), 775–802.
Angeletos, G.-M. and J. La’O, “Sentiments,” Econometrica 81 (2013), 739–779.
Arin, P., K. Akoz, S. Sarangi and N. Sayour, “Do Social Movements Affect Ed-
ucational Outcomes? Theory and BLM Evidence,” Technical Report, Working Paper,
Austmann, L. M. and S. A. Vigne, “Does environmental awareness fuel the electric
vehicle market? A Twitter keyword analysis,” Energy Economics 101 (2021), 105337.
Baimbridge, M., S. Cameron and P. Dawson, “Satellite television and the demand
for football: A whole new ball game?,” Scottish Journal of Political Economy 43 (1996),
Baker, S. R., N. Bloom and S. J. Davis, “Measuring economic policy uncertainty,”
the Quarterly Journal of Economics 131 (2016), 1593–1636.
Benhabib, J. and M. M. Spiegel, “Sentiments and economic activity: Evidence from
US states,” The Economic Journal 129 (2019), 715–733.
Biolsi, C. and A. Lebedinsky, “Can changes in sentiments influence consumer be-
havior? Evidence from the Trump-Russia investigation,” Economic Inquiry 59 (2021),
Bovitz, G. L., J. N. Druckman and A. Lupia, “When can a news organization lead
public opinion?–Ideology versus market forces in decisions to make news,” Public Choice
113 (2002), 127–155.
Buraimo, B., D. Forrest, I. G. McHale and J. d. D. Tena, “Armchair fans:
Modelling audience size for televised football matches,” European Journal of Operational
Research 298 (2022), 644–655.
Buraimo, B., G. Migali and R. Simmons, “An analysis of consumer response to
corruption: Italy’s Calciopoli scandal,” Oxford Bulletin of Economics and Statistics 78
(2016), 22–41.
Buraimo, B. and R. Simmons, “Uncertainty of outcome or star quality? Television
audience demand for English Premier League football,” International Journal of the
Economics of Business 22 (2015), 449–469.
Cameron, A. C. and D. L. Miller, “A practitioners guide to cluster-robust inference,”
Journal of Human Resources 50 (2015), 317–372.
Caporale, G. M., F. M. Ali, F. Spagnolo and N. Spagnolo, “Cross-border port-
folio flows and news media coverage,” Journal of International Money and Finance 126
(2022), 102638.
Carroll, C. D., “Macroeconomic expectations of households and professional forecast-
ers,” the Quarterly Journal of Economics 118 (2003), 269–298.
Chalfin, A. and J. McCrary, “Are US cities underpoliced? Theory and evidence,”
Review of Economics and Statistics 100 (2018), 167–186.
Choi, H. and H. Varian, “Predicting the present with Google Trends,” Economic Record
88 (2012), 2–9.
Dawson, P., P. Downward and T. C. Mills, “Olympic news and attitudes to-
wards the Olympics: a compositional time-series analysis of how sentiment is affected by
events,” Journal of Applied Statistics 41 (2014), 1307–1314.
DellaVigna, S. and E. La Ferrara, “Economic and social impacts of the media,” in
Handbook of Media Economicsvolume 1 (Elsevier, 2015), 723–768.
Enikolopov, R., A. Makarin and M. Petrova, “Social media and protest partici-
pation: Evidence from Russia,” Econometrica 88 (2020), 1479–1514.
Farina, E. and V. Pathania, “Papal visits and abortions: evidence from Italy,” Journal
of Population Economics 33 (2020), 795–837.
Forrest, D., R. Simmons and B. Buraimo, “Outcome uncertainty and the couch
potato audience,” Scottish Journal of Political Economy 52 (2005), 641–661.
Gillitzer, C. and N. Prasad, “The effect of consumer sentiment on consumption:
cross-sectional evidence from elections,” American Economic Journal: Macroeconomics
10 (2018), 234–69.
Gorodnichenko, Y., T. Pham and O. Talavera, “Social media, sentiment and public
opinions: Evidence from# Brexit and# USElection,” European Economic Review (2021),
Halberstam, Y. and B. Knight, “Homophily, group size, and the diffusion of political
information in social networks: Evidence from Twitter,” Journal of Public Economics
143 (2016), 73–88.
Hatte, S., E. Madinier and E. Zhuravskaya, “Reading Twitter in the Newsroom:
Web 2.0 and Traditional-Media Reporting of Conflicts,” Technical Report, 2021.
Hausman, J. A. and G. K. Leonard, “Superstars in the National Basketball Associa-
tion: Economic value and policy,” Journal of Labor Economics 15 (1997), 586–624.
Hendel, I., S. Lach and Y. Spiegel, “Consumers’ activism: the cottage cheese boy-
cott,” The RAND Journal of Economics 48 (2017), 972–1003.
Kanazawa, M. T. and J. P. Funk, “Racial discrimination in professional basketball:
Evidence from Nielsen ratings,” Economic Inquiry 39 (2001), 599–608.
Lamla, M. J. and S. M. Lein, “The role of media for consumers? inflation expectation
formation,” Journal of Economic Behavior & Organization 106 (2014), 62–77.
Larson, J. M., J. Nagler, J. Ronen and J. A. Tucker, “Social networks and
protest participation: Evidence from 130 million Twitter users,” American Journal of
Political Science 63 (2019), 690–705.
Lerner, J. S., S. Han and D. Keltner, “Feelings and consumer decision making: Ex-
tending the appraisal-tendency framework,” Journal of Consumer Psychology 17 (2007),
Mastrorocco, N. and L. Minale, “News media and crime perceptions: Evidence from
a natural experiment,” Journal of Public Economics 165 (2018), 230–255.
Nielsen, “Nielsen,” (2021), accessed: 2021-09-30.
Reed, W. R., “On the practice of lagging variables to avoid simultaneity,” Oxford Bulletin
of Economics and Statistics 77 (2015), 897–905.
Salvatore, C., S. Biffignandi and A. Bianchi, “Social media and twitter data
quality for new social indicators,” Social Indicators Research 156 (2021), 601–630.
Scherer, A. G., G. Palazzo and D. Matten, “The business firm as a political actor:
A new theory of the firm for a globalized world,” Business & Society 53 (2014), 143–156.
Shen, D., A. Urquhart and P. Wang, “Does twitter predict Bitcoin?,” Economics
Letters 174 (2019), 118–122.
Smith, C. and D. Korschun, “Companies can’t avoid politics-and shouldn’t try to,”
Harvard Business Review (2018).
Soroka, S., M. Daku, D. Hiaeshutter-Rice, L. Guggenheim and J. Pasek,
“Negativity and positivity biases in economic news coverage: Traditional versus social
media,” Communication Research 45 (2018), 1078–1098.
Soroka, S. N., “Good news and bad news: Asymmetric responses to economic informa-
tion,” The Journal of Politics 68 (2006), 372–385.
Sport Media Watch, “NBA playoff drop in line with broader trend,”
low-sports-viewership/ (2021), accessed: 2021-09-30.
Stock, J. H., J. H. Wright and M. Yogo, “A survey of weak instruments and
weak identification in generalized method of moments,” Journal of Business & Economic
Statistics 20 (2002), 518–529.
Vosen, S. and T. Schmidt, “Forecasting private consumption: survey-based indicators
vs. Google trends,” Journal of Forecasting 30 (2011), 565–578.
Zhuravskaya, E., M. Petrova and R. Enikolopov, “Political effects of the internet
and social media,” Annual Review of Economics 12 (2020), 415–438.
Table 1: Summary Statistics
Observations Mean SD Min Max
Viewers (millions) 506 2.35 2.14 0.25 18.76
Ratings 490 1.47 1.23 0.17 10.70
Tweets NBABLM(-1d) 506 51.04 165.17 0 1436
USA&NYT(-1d) 506 1.05 2.30 0 19
Google Trends(-1d) 506 15.09 27.72 0 100
Tweets NBA(-1d) 506 4011.46 1474.05 1267 10738
Tweets BLM(-1d) 506 892.49 1499.32 17 8101
Tweets(-1d) Positive Bing 506 16.38 53.33 0 474.04
Tweets(-1d) Negative Bing 506 16.79 49.68 0 405.62
Tweets(-1d) Positive Afinn 506 22.67 84.54 0 820
Tweets(-1d) Negative Afinn 506 18.30 56.68 0 490
Tweets NBABLM(-1d), Pop I 506 1942.76 8824.26 0 69465
Tweets NBABLM(-1d), Pop II 506 2159.59 9701.55 0 78055
Tweets NBABLM(-1d), Pop III (xK) 506 7068 44400 0 4.78E+05
Tweets NBABLM, Leag+Pla(-1d) 506 0.40 1.11 0 7
Tweets NBABLM, Fans(-1d) 506 50.63 164.48 0 1432
After 506 0.24 0.43 0 1
Play-off 506 0.33 0.47 0 1
Prime Time 506 0.68 0.47 0 1
Weekend 506 0.20 0.40 0 1
Viewers (millions) 414 0.72 0.85 0.12 0.87
Ratings 278 0.56 0.54 0.14 4.90
Tweets NBABLM(-1d) 414 19.75 65.48 0 349
Notes: This table shows the summary statistics for the most relevant variables employed in this study. TV audience-related
data are sourced by Sport Media Watch (2021), whereas news-related data are retrieved from Twitter API. Data on newspaper
and Google Trends have been web-scraped from the USA Today,The New York Times and Google Trends websites.
Table 2: Media Exposure and TV Consumption: Baseline Model
Viewers(Log) Ratings
(1) (2) (3) (4) (5) (6)
Tweets Mainstream
Trends Tweets Mainstream
Tweets NBABLM(-1d) 0.296 0.574
[0.196] [0.324]
USA&NYT(-1d) -0.001 -0.020
[0.013] [0.028]
Google Trends(-1d) 0.002 0.015
[0.003] [0.009]
Tweets NBA(-1d) -0.011 -0.005 -0.005 -0.009 0.005 0.002
[0.025] [0.023] [0.022] [0.037] [0.035] [0.034]
Tweets BLM(-1d) 0.010 0.026 0.005 0.084 0.149 -0.015
[0.032] [0.041] [0.037] [0.077] [0.103] [0.083]
After -0.115 -0.142 -0.214 -0.882∗∗ -1.001∗∗ -1.457∗∗∗
[0.167] [0.174] [0.189] [0.356] [0.387] [0.557]
Play-Off:1st Rd 0.549∗∗∗ 0.544∗∗∗ 0.534∗∗∗ 0.695∗∗∗ 0.687∗∗∗ 0.613∗∗∗
[0.065] [0.067] [0.069] [0.119] [0.119] [0.127]
Play-Off:Semi 1.117∗∗∗ 1.112∗∗∗ 1.095∗∗∗ 1.586∗∗∗ 1.574∗∗∗ 1.455∗∗∗
[0.071] [0.072] [0.078] [0.133] [0.135] [0.160]
Play-Off:Conf Finals 1.488∗∗∗ 1.479∗∗∗ 1.477∗∗∗ 2.405∗∗∗ 2.389∗∗∗ 2.369∗∗∗
[0.071] [0.071] [0.071] [0.167] [0.166] [0.170]
Play-Off:Finals 1.537∗∗∗ 1.552∗∗∗ 1.568∗∗∗ 4.646∗∗∗ 4.698∗∗∗ 4.794∗∗∗
[0.126] [0.127] [0.125] [0.640] [0.644] [0.626]
ABC 1.734∗∗∗ 1.719∗∗∗ 1.715∗∗∗ 1.295∗∗ 1.2571.237
[0.426] [0.429] [0.429] [0.646] [0.649] [0.643]
ESPN 1.056∗∗ 1.054∗∗ 1.052∗∗ 0.212 0.208 0.194
[0.496] [0.498] [0.497] [0.732] [0.734] [0.729]
TNT 1.024∗∗ 1.023∗∗ 1.023∗∗ 0.208 0.207 0.206
[0.496] [0.497] [0.497] [0.724] [0.726] [0.722]
Prime Time 0.234∗∗∗ 0.233∗∗∗ 0.233∗∗∗ 0.211∗∗∗ 0.211∗∗∗ 0.213∗∗∗
[0.038] [0.039] [0.039] [0.053] [0.054] [0.054]
Weekend 0.035 0.048 0.053 -0.182 -0.150 -0.122
[0.065] [0.063] [0.063] [0.131] [0.127] [0.125]
Key Dates -0.010 0.160 0.147 -0.031 0.361 0.236
[0.159] [0.130] [0.135] [0.278] [0.227] [0.245]
Season FE, Trend Yes Yes Yes Yes Yes Yes
Observations 506 506 506 490 490 490
Adj. R-sq 0.656 0.655 0.655 0.724 0.723 0.726
Notes: This table presents the baseline analysis regarding the impact of the NBA/BLM daily media exposure on Viewers
and Ratings for NBA matches (the unit of analysis). We employ three measures of media exposure, all referring to the day
before the matches were played(-1d). Tweets NBABLM represents the number of original tweets obtained using as ‘NBA’
and ‘BLM’ key words. USA& NYT is the total number of news reports or articles which appeared in the USA Today and
The New York Times with the keywords ‘NBA’ and ‘BLM’. Google Trends is the daily value of Google searches. Tweets
NBA represents the counts of tweets for the day before the match using the hashtag ‘NBA’ but excluding ‘BLM’. Tweets
BLM considers tweets with the hashtag ‘BLM’ but excluding ‘NBA’. The other variables are described in the text. *,**, ***
represent significance at the 10%, 5% and 1% levels, respectively.
Table 3: Media Exposure and TV Consumption - Robustness I
Viewers(Log) Ratings Tweets
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Tweets Mainstream
Trends 2SLS Tweets Mainstream
Trend 2SLS First
Tweets NBABLM(-1d) 0.366 0.114 0.746 0.428
[0.365] [0.584] [0.573] [0.968]
USA&NYT(-1d) -0.014 -0.012 0.028∗∗∗
[0.021] [0.032] [0.003]
Google Trends(-1d) 0.000 -0.001 0.003∗∗∗
[0.006] [0.010] [0.001]
Tweets BLM(-1d) -0.052 -0.003 -0.033 0.018 -0.090 -0.023 -0.040 0.091 -0.034∗∗∗
[0.060] [0.064] [0.061] [0.049] [0.091] [0.104] [0.079] [0.082] [0.011]
Tweets NBA(-1d) -0.010 -0.006 -0.007 -0.007 -0.033 -0.025 -0.026 -0.006 0.014∗∗∗
[0.030] [0.029] [0.029] [0.018] [0.046] [0.045] [0.046] [0.030] [0.002]
Season FE, Trend Yes Yes Yes Yes Yes Yes Yes Yes Yes
Match-Ups FE Yes Yes Yes No Yes Yes Yes No No
Wald F statistic 47.15
Observations 506 506 506 506 490 490 490 490 506
Adj. R-sq 0.747 0.746 0.745 0.655 0.770 0.768 0.768 0.724 0.833
Notes: This table presents the robustness analysis and analyses the impact of the NBA/BLM daily media exposure on Viewers
and Ratings for NBA matches (the unit of analysis). We employ three measures of media exposure, all referring to the day
before the match(-1d). Tweets NBABLM represents the number of original tweets obtained using as ‘NBA’ and ‘BLM’ key
words. USA& NYT is the total number of news reports or articles which appeared in the USA Today and The New York
Times with the keywords ‘NBA’ and ‘BLM’. Google Trends is the daily value of Google searches. Tweets NBA represents the
counts of tweets for the day before the match using the hashtag ‘NBA’ but excluding ‘BLM’. Tweets BLM considers tweets
with the hashtag ‘BLM’ but excluding ‘NBA’. Columns (4) and (8) present the results for the 2SLS regressions, using USA&
NYT and Google Trends as instrument for Tweets NBABLM. Column (9) refers to the first stage of the 2SLS for model in
column (4). All models include: Key Dates,After, binary variables for the playoff series, networks dummies, Prime Time,
Weekend, trends and season effects. The other variables are described in the text. All but columns (4),(8) and (9) include
match fixed effects. *,**, *** represent significance at the 10%, 5% and 1% levels, respectively.
Table 4: Difference-in-Difference NBA vs NHL
Viewers(Log) Ratings
(1) (2) (3) (4)
Only Season FE All Controls Only Season FE All Controls
NBA 1.6750.389 1.443∗∗ 0.876∗∗
[0.158] [0.071] [0.025] [0.034]
After 0.613∗∗ -0.048 0.792 -0.361
[0.034] [0.015] [0.353] [0.248]
NBA X After -0.055 -0.290 0.233 -0.148
[0.071] [0.116] [0.040] [0.113]
Tweets BLM(-1d) 0.000 0.001 -0.107 0.029
[0.035] [0.052] [0.073] [0.068]
Tweets NBA and NHL(-1d) -0.145 0.011 -0.234-0.021
[0.073] [0.016] [0.020] [0.020]
Prob>Cluster-Adjusted test .052 .273 .162 .250
Prob>Wild Bootstrap test .274 .778 .223 .167
Other Controls No Yes No Yes
Observations 920 920 768 768
Adj. R-sq 0.508 0.773 0.256 0.656
Notes: This table presents the results for the difference-in-difference model. The treated group is the NBA whereas the
control is the NHL. After is a dummy taking value one from when the NBA and NHL resumed- at the end of July-
afterwards. Models (1) and (3) include only season fixed effects. The ones in (2) and (4) include all controls, which are
Tweets NBA(-1d),Tweets BLM(-1d),Key Dates,After, binary variables for the playoff series, networks dummies, Prime
Time,Weekend, trends and season effects. -1d means that the variable refers to the day before the match. *,**, ***
represent significance at the 10%, 5% and 1% levels, respectively.
Table 5: Media Exposure and TV Consumption - Robustness Only with Tweets
Viewers(Log) Ratings Tweets, Leag+Pla
(1) (2) (3) (4) (5) (6) (7) (8) (9)
and Likes
Tweets, Retweets
Likes, Quotes
and Replies
by Followers
+ League
vs. Fans
and Likes
Tweets, Retweets
Likes, Quotes
and Replies
by Followers
+ League
vs. Fans Reverse
Tweets NBABLM(-1d), Pop I 0.002 0.004
[0.003] [0.006]
Tweets NBABLM(-1d), Pop II 0.002 0.004
[0.002] [0.005]
Tweets NBABLM(-1d), Pop III 0.0100.022∗∗
[0.005] [0.009]
Tweets NBABLM, Leag+Pla(-1d) 0.509∗∗ 1.303∗∗∗
[0.208] [0.405]
Tweets NBABLM, Fans(-1d) 0.228 0.400
[0.177] [0.295]
Decrease Audience(-1d) 0.060
Tweets NBA(-1d) -0.009 -0.009 -0.013 -0.016 -0.004 -0.004 -0.016 -0.023 0.048
[0.024] [0.024] [0.025] [0.025] [0.037] [0.037] [0.037] [0.037] [0.039]
Tweets BLM(-1d) 0.027 0.027 0.037 -0.016 0.118 0.118 0.1420.019 0.305
[0.028] [0.028] [0.030] [0.033] [0.075] [0.075] [0.080] [0.076] [0.259]
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 506 506 506 506 490 490 490 490 506
Adj. R-sq 0.655 0.655 0.657 0.658 0.723 0.723 0.726 0.730 0.689
Notes: This table shows the results for several models using Viewers and Ratings as dependent variables.Tweets NBABLM,
Pop I represents the total number of original tweets plus retweets and likes. Tweets NBABLM, Pop II is the sum of the
original tweets, retweets, likes, quotes and replies. Tweets NBABLM, Pop III multiplies the daily number of original tweets
by the average number of followers of the Twitters’ authors that posted that day. Tweets NBABLM, Leag+Pla,Tweets
NBABLM, Fans represent the number of Tweets sent by NBA stakeholders and causal fans. Tweets NBA represents the
counts of tweets for the day before the match using hashtag NBA excluding BLM. Tweets BLM considers tweets with hashtags
BLM excluding BLM. -1d means that the variable refers to the day before the match. All models include: Key Dates,After,
binary variables for the playoff series, networks dummies, Prime Time,Weekend, trends and season effects. *,**, *** represent
significance at the 10%, 5% and 1% levels, respectively.
Table 6: Tweets’ Sentiments and TV Consumption I
Viewers(Log) Ratings
(1) (2) (3) (4)
Bing Afinn Bing Afinn
Tweets(-1d) Positive 1.612 1.2222.7122.748∗∗∗
[1.032] [0.657] [1.587] [1.034]
Tweets(-1d) Negative -1.041 -1.010 -1.779 -3.738
[1.453] [1.344] [2.391] [2.155]
Tweets BLM(-1d) 0.040 0.056 0.1390.212∗∗
[0.037] [0.045] [0.082] [0.102]
Tweets NBA(-1d) -0.005 -0.007 0.003 0.003
[0.023] [0.022] [0.035] [0.034]
Controls Yes Yes Yes Yes
Observations 506 506 490 490
Adj. R-sq 0.660 0.660 0.727 0.729
Notes: This table shows the several exercises that capture the impact of the tone of
the tweets on Viewers and Ratings as dependent variables. Tweets Positive and Tweets
Negative represents the number of original tweets with positive and negative tone using
the Bing and Afinn lexicons. (-1d) means that the variable refers to the day before
the match. All models include: Tweets NBA,Tweets BLM,Key Dates,After, binary
variables for the playoff series, networks dummies, Prime Time,Weekend, trends and
season effects. *,**, *** represent significance at the 10%, 5% and 1% levels, respectively.
Table 7: Tweets’ Sentiments and TV Consumption II
Viewers(Log) Ratings
(1) (2) (3) (4) (5) (6) (7) (8)
Measure II Afinn
Measure II Std. Dev.
Afinn NRC
Measure II Afinn
Measure II Std. Dev.
Tweets(-1d) Positve 1.523∗∗∗ 0.991 0.843∗∗ 1.993 3.555∗∗ 2.458∗∗
[0.581] [0.649] [0.387] [1.274] [1.670] [1.029]
Tweets(-1d) Negative -0.564 -0.526 -0.787 -1.506 -3.052 -3.221
[0.811] [0.986] [0.779] [1.737] [2.379] [2.077]
Tweets(-1d) Disgust 2.864 13.303
[3.705] [10.614]
Tweets(-1d) Fear -0.179 4.826
[2.741] [4.889]
Tweets(-1d) Joy 0.376 3.148
[2.571] [5.919]
Tweets(-1d) Sadness 1.6873.380
[0.890] [2.202]
Tweets(-1d) Surprise -0.015 -0.180
[0.327] [0.719]
Tweets(-1d) Trust -1.532 -1.350
[0.970] [2.663]
Tweets(-1d) Anger -6.451 -13.371
[4.392] [14.175]
Tweets(-1d) Anticipation 0.234 0.685
[0.567] [1.077]
Afinn Score Std. Dev. -0.013 -0.064
[0.049] [0.071]
Tweets BLM(-1d) 0.0590.037 0.067 0.024 0.1910.198 0.2770.115
[0.034] [0.044] [0.050] [0.029] [0.112] [0.126] [0.158] [0.076]
Tweets NBA(-1d) -0.017 -0.009 -0.012 -0.005 -0.018 -0.009 -0.016 0.003
[0.026] [0.024] [0.025] [0.022] [0.039] [0.036] [0.037] [0.034]
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Observations 506 506 506 506 490 490 490 490
Adj. R-sq 0.655 0.655 0.656 0.655 0.724 0.725 0.726 0.723
Notes: This table shows the impact of the tone of the tweets on Viewers and Ratings as dependent variables. Columns (1)
and (5) show the results with the lexicon NRC, which considers 8 emotions and two sentiments. Columns (2),(3),(6) and
(7) represents the count of positively and negatively toned tweets using the Bing and Afinn lexicons. Columns (4) and (8)
consider the standard deviation of the Afinn score per day. All models include: Tweets NBA,Tweets BLM,Key Dates,After,
binary variables for the playoff series, networks dummies, Prime Time,Weekend, trends and season effects. *,**, *** represent
significance at the 10%, 5% and 1% levels, respectively.
Figure 1: Tweets and Newspapers Trends
Notes: Top Panel: Number of Tweets NBABLM (left y axis) and USA & NYT (right panel). Bottom Panel: Number of
Tweets NBA and Tweets BLM.
Figure 2: Trend in Viewers and Tweets for the NBA and NHL
SEASON 2018-19 SEASON 2019-20
NBA/NHL stop for CoronavirusSummer Break
Playoff 2020Playoff 2019
.121 18.763
10/1/2018 4/1/2019 10/1/2019 4/1/2020 10/1/2020
Notes: Trend in Viewers (in millions) for the NBA(grey) and NHL(black) for the whole period under consideration. After
refers to all matches played after the NBA and NHL resumed from the Covid-19 related break (30th of July 2020).
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Social media represent an excellent opportunity for the construction of timely socio-economic indicators. Despite the many advantages of investigating social media for this purpose, however, there are also relevant statistical and quality issues. Data quality is an especially critical topic. Depending on the characteristics of the social media a researcher is using, the problems that arise related to errors are different. Thus, no one unique quality evaluation framework is suitable. In this paper, the quality of social media data is discussed considering Twitter as the reference social media. An original quality framework for Twitter data is introduced. A reformulation of the traditional quality dimensions is proposed, and the new quality aspects are discussed. The main sources of errors are identified, and examples are provided to show the process of finding evidence of these errors. The conclusion affirms the importance of using a mixed methods approach, which involves incorporating both qualitative and quantitative evaluations to assess data quality. A collection of good practices and proposed indicators for quality evaluation is provided.
Full-text available
This paper examines the ways that social movement organizations affiliated with the Black Lives Matter (BLM) movement use Twitter through three content analysis studies. The main finding presented in the paper is that the modal tweet generated between December 1, 2015 and October 31, 2016 was an emotional response—an expression of sadness, outrage, or despair—to police brutality and the killings of African Americans. The second key finding is that BLM organizations generated more tweets that framed the movement as a struggle for individual rights than ones that utilized frames about gender, racial, and LGBTQ identities. Finally, the paper shows that BLM activists urge their followers to pursue disruptive repertoires of contention less frequently than they encourage other political behaviors. These findings suggest that the BLM movement is intelligible through both the resource mobilization and new social movement paradigms within social movement studies.
This paper studies information diffusion in social media and the potential role of bots in influencing public opinions. Using Twitter data on the 2016 E.U. Referendum (“Brexit”) and the 2016 U.S. Presidential Election, we find that diffusion of information on Twitter is largely complete within 1-2 hours. Stronger diffusion between agents with similar beliefs is consistent with the “echo chambers” view of social media. Our results are consistent the notion that bots could have a tangible effect on the tweeting activity of humans and that the degree of bots’ influence depends on whether bots provide information consistent with humans’ priors. Overall, our results suggest that the aggressive use of Twitter bots, coupled with the fragmentation of social media and the role of sentiment, could enhance political polarization.
Pinning down the role of social ties in the decision to protest has been notoriously elusive, largely due to data limitations. Social media and their global use by protesters offer an unprecedented opportunity to observe real‐time social ties and online behavior, though often without an attendant measure of real‐world behavior. We collect data on Twitter activity during the 2015 Charlie Hebdo protest in Paris, which, unusually, record real‐world protest attendance and network structure measured beyond egocentric networks. We devise a test of social theories of protest that hold that participation depends on exposure to others' intentions and network position determines exposure. Our findings are strongly consistent with these theories, showing that protesters are significantly more connected to one another via direct, indirect, triadic, and reciprocated ties than comparable nonprotesters. These results offer the first large‐scale empirical support for the claim that social network structure has consequences for protest participation.
This paper adds to the growing literature of Bitcoin by examining the link between investor attention and Bitcoin returns, trading volume and realized volatility. Unlike previous studies, we employ the number of tweets from Twitter as a measure of attention rather than Google trends as we argue this is a better measure of attention from more informed investors. We find that the number of tweets is a significant driver of next day trading volume and realized volatility which is supported by linear and nonlinear Granger causality tests.
We study a consumer boycott on cottage cheese, organized in Israel on Facebook in the summer of 2011 following a steep price increase since 2006. The boycott led to an immediate decline in prices, which remain low even six years later. We find that (i) demand at the start of the boycott would have been 30% higher but for the boycott, (ii) own- and especially cross-price elasticities increased substantially after the boycott, and (iii) post-boycott prices are substantially below the levels implied by the post-boycott demand elasticities, suggesting that firms were concerned with public backlash due to high prices.
Past work suggests that the priorities for information propagation in social media may be markedly different from the priorities for news selection in traditional media outlets. We explore this possibility here, focusing on the tone of both newspaper and Twitter content following changes in the U.S. unemployment rate, from 2008 to 2014. Results strongly support the expectation that while the tone of newspaper content exhibits stronger reactions to negative information, the tone of Twitter content reacts more strongly to positive economic shifts.