A big data analysis of social media coverage of athlete protests

ABSTRACT

Using a contentious issue in sport – the athlete protests during the playing of the national anthem – this paper examined the relationship between media outlets’ social media coverage of athlete protests and the social media user interest and sentiment. We analysed data sourced from the media outlets’ official Instagram accounts, along with comments on these posts. Using both sentiment lexicons and Random Forrest machine learning models, we derived the sentiment of 496 official Instagram posts and 137,735 user comments. We utilised logit and ordered logit regressions to examine whether media coverage of the athlete protests was responsive to user interest and user sentiment towards the issue. In addition, we employed multinomial logit regressions and two-stage least squared regressions to investigate media’s selection of topics and portrayal of the protests. We found strong evidence that both media’s decisions to cover the protests and how they cover the issue are sensitive to social media user interest and sentiment.

HIGHLIGHTS

• Test the relationship between social media coverage of athlete protests and social media user interest and sentiment.

• Media coverage of the protests was sensitive to social media user interest and sentiment.

• Media outlets were more likely to cover topics at the intersection of sport and politics when user sentiment towards the protests was negative.

• When there was increased social media interest media outlets tend to use more negative tones to cover the protests.

KEYWORDS:

• Athlete activism
• social media
• media bias
• sentiment analysis
• machine learning

Acknowledgments

We would like to thank Professors Rod Ford, Mark Rosentraub, Francine Laffontaine and Matt Shum for their thoughtful comments; our Research Assistants- Brian Johnston and Yinan Zeng for their work; and the participants at the 2019 Marketing Science Conference and the 2019 University of Florida Alumni Conference for their suggestions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Bots are a combination of automated computational algorithms and human manipulations to perform communicative activities on social media with assumed identities (Klinger & Svensson, 2018).

² 71% of Americans between 18 and 24 are reported to use Instagram, compared to the 35% across all adult age groups (Smith & Anderson, 2018).

³ Media outlets refer to news media, sport media and sport organisations.

⁴ Instagram users may utilise other applications such as “Regram” and “Repost” for sharing (or reposting) content. In order to do so, they must first copy and paste the link of the post to the application and use the “Copy to Instagram” feature to share the post on Instagram.

⁵ The selections of newspapers and magazines are based on circulation statistics provided by the Media Intelligence Center. The selection of TV networks is based on the average number of audience information provided by ComScore TV Essentials.

⁶ We remove common words, i.e., “stop words,” such as “the”, “is”, “which” based on three lexicons, Catalan stop words, Romanian stop words and English stop words from the SMART system (Salton, 1971).

⁷ “News report of protests” refers to reports of athletes kneeling or using other protest gestures during the playing of the national anthem. “Reaction to protests” refers to famous people’s comments on the protests. “President Trump” refers to President Trump’s involvement in the athlete protests. “NFL” refers to NFL’s statement and policy on protesting during the national anthem. “Politics” refers to other politicians’ involvement in the protests. Examples of “Other protest-related topics” include high school athletes posing similar gestures to echo the NFL athletes’ protests and weekly recap of sport news where there were athlete protests.

⁸ For “emojis”, we referred to the “genres and explanations” on emojipedia.org/instagram to determine their meanings.

⁹ The actual number of posts per day ranges from 0 to 9. However, there is only a small number of observations where media $i$ posted than more than twice about the protests. The numbers of observations with media $i$ posted 3, 4, 5, 6, 7, 8 and 9 times are 15, 5, 2, 1, 2, 0 and 1, respectively.

Additional information

Funding

This project was funded by a pilot grant provided by the School of Kinesiology at the University of Michigan.