Social media influences the mainstream media

In April 2022, tech billionaire Elon Musk experimented with to invest in Twitter, declaring the social media corporation demands to be transformed privately. Notably, the founder of PayPal, Tesla, and SpaceX argues that he desires to restore cost-free speech on the system. Quite a few have due to the fact pointed out – rightly – that Musk’s track report with free speech is problematic to say the the very least. Still, there is one more rationale why Musk’s acquisition of Twitter might place democracy at threat: by managing the system, the self-described “free speech absolutist” will also affect the mainstream media agenda.

Many modern papers have proven that social media has transformed culture (e.g. Fujiwara et al. 2021, Levy 2021). But the power of Twitter goes significantly over and above its influence on its users. In a new research venture, relying on approximately two billion tweets and an innovative empirical technique, we quantify what quite a few extensive suspected – that Twitter impacts publishers’ generation and editorial choices (Cagé et al. 2022).

To do so, we commence in three ways. First, we gather a agent sample of all the tweets generated in French concerning August 2018 and July 2019 and blend it with the content revealed on line by all the mainstream media retailers (encompassing newspapers, tv channels, radio stations, pure on line media, and news agencies’ dispatches). Our dataset, which includes around 1.8 billion tweets, encompasses all around 70% of all the tweets in French (which include retweets) for the duration of this time time period. Determine 1 plots the day-to-day distribution of the amount of tweets.

Figure 1 Every day distribution of the selection of tweets in the sample

Notes: The figure plots the every day range of tweets provided in our dataset. The red line plots all the tweets, the blue dotted line displays these tweets when we implement the filter, and the environmentally friendly dashed line plots only the primary tweets. Time period of time is 18 June 2018 – 10 August 2019. The few days without the need of info are because of to unusual occasions when the server collapsed and we ended up so not able to capture the tweets in serious time.

For just about every of these tweets, we collect data on their ‘success’ on Twitter (amount of likes, remarks, and so forth.), as well as data on the attributes of the user at the time of the tweet (e.g. its quantity of followers). To assemble this exclusive dataset, we have put together the Sample and the Filter Twitter Software Programming Interfaces (APIs), and chosen keywords and phrases. Determine 2 summarises our information collection setup.

Determine 2 Diagram of our experimental setup to select the very best tweet collection process

Second, we build novel algorithms to determine all the ‘news stories’ included each on social and standard media. An occasion right here is a cluster of files (tweets and media article content) that talk about the similar news story. So, for instance, all the files (tweets and media article content) speaking about the Hokkaido Jap Iburi earthquake on 6 September 2018 will be classified as aspect of the exact event. Gatherings are detected by our algorithms employing the actuality that the paperwork share adequate semantic similarity. In a nutshell, for Twitter, our solution is made up in modelling the occasion detection trouble as a dynamic clustering problem, working with a ‘first tale detection’ (FSD) algorithm (see Mazoyer et al. 2022 for additional specifics). To detect the news functions among the tales released on line by standard media shops, we abide by Cagé et al. (2020) and describe each news report by a semantic vector (using TF-IDF) and use the cosine distance to evaluate their semantic similarity. Utilized jointly with temporal constraints, we can cluster the posts to kind the activities. Lastly, to generate the intersection concerning social media situations and mainstream media gatherings, we count on the Louvain group detection algorithm (Blondel et al. 2008), as illustrated in Determine 3.

Figure 3 Graphical illustration: Making the joint functions

We recognize 3,992 joint situations, i.e. functions that are protected both on social and on common media, out of which 3,904 originate initial on Twitter.

Third, we depend on the construction of the social media community – and in unique, on the centrality of its users – to isolate ‘exogenous’ shocks to the acceptance of the stories on Twitter (calculated by the range of tweets, retweets, likes, and many others.). In other text, we isolate variations in the reputation of tales on Twitter independent of the intrinsic desire of these stories. To do so, we leverage the enormity of our dataset to propose a novel instrumental variable method: our instrument is the interaction amongst the initial Twitter users’ centrality in the community (calculated computing PageRank centrality just prior to the function) and the news stress in the social media at the time of the 1st tweets on the celebration. Our identification assumption is that, at the time we regulate for the direct impact of centrality and information pressure, the interaction in between users’ centrality and information tension should only have an effect on common news production through its result on the tweets’ visibility on Twitter.

Our conclusions are enlightening. All the things else equivalent – and, in certain, independently of the newsworthiness of a story – a 55% increase in the range of tweets posted in advance of the very first media report on a tale qualified prospects to an maximize in the quantity of news articles or blog posts masking the story corresponding to 17% of the necessarily mean. In other words and phrases, Twitter sets the agenda of media coverage in a quantitatively meaningful way.

Why is this so? Initial, a expanding literature in journalism studies highlights the truth that social media performs an essential part as a information supply. Consistent with this plan, we present that the magnitude of the impact is greater for the media shops that have a significant selection of journalists with a Twitter account, pointing to the job performed by the checking of Twitter by journalists. 

But the use of the platforms as journalistic sources is not the only component at play in this article. In specific, we investigate whether the magnitude of the contagion between social and mainstream media relies upon on the outlets’ small business model. For each individual of the media in our dataset, we accumulate information on no matter whether it uses a paywall (at the time of the info assortment), the properties of this paywall (e.g. delicate versus hard), and the date of introduction of the paywall. This facts is summarised in Determine 4.

Figure 4 News editors’ company design

Notes: The Figure studies the share of the media retailers in our sample relying on their on the net business enterprise design. 52% of the media in our sample do not have a paywall (“no paywall”), and 4.3% affliction the reading through of the paid posts on the actuality of looking at an advertisement (“paid article content can be accessed by viewing an ad”). Of the shops that do have a paywall, we distinguish between a few designs: difficult paywall, metered paywall, and gentle paywall (“some articles or blog posts locked driving paywall”). 

We clearly show that the magnitude of our consequences is a lot increased for the media retailers that rely fully or strongly on advertising revenues than for those people whose on-line material is driving a paywall (and as a result largely count on subscriptions). For the former, a 50% raise in recognition leads to an maximize in information coverage corresponding to 22.% (no paywall), 20.3% (delicate paywall) and 21.1% (‘watch-an-ad’ paywall) of the necessarily mean, as opposed to 6.2% of the imply for the shops employing a metered paywall, a coefficient that is additionally not statistically considerable. In other text, Twitter influences mainstream media simply because of short-term concerns produced by promotion earnings-bearing clicks. 

Though there are widespread fears that new technologies are worsening editorial high-quality – in distinct because they have led to discounts in the newsroom, which in transform have lowered the quality of news provision and the creation of initial material (Cagé et al. 2017) – our findings thus suggest that they are disproportionately worsening the excellent for people who simply cannot afford to pay for or are unwilling to pay back for information. Set an additional way, because media outlets whose material is obtainable online for absolutely free are inclined to be far more affected by the attractiveness of stories on Twitter than those people using a paywall, the platform generates an boost in facts inequality, earning disadvantage voters further more vulnerable to manipulation (Kennedy and Prat 2019).

In addition to, our findings – which capture the outcomes of a variation in popularity that is uncorrelated with a story’s underlying newsworthiness – suggest that social media may possibly offer a biased sign of what visitors want, which may perhaps in change explain why, as highlighted by survey facts, a significant share of the populace is not intrigued in the information created by the media (and may so determine not to consume news). Twitter customers are without a doubt not agent of the common information-reading through population. This factors to a unfavorable effect of social media driven by the output facet, steady with modern improvements in the two The Guardian and The New York Situations social media suggestions, which highlight the actuality that journalists tend to depend too significantly on Twitter as each a reporting and comments tool1 and that it may distort their perspective of who their audience is.

Turning to the demand for news and making use of audience data, we at last clearly show that the news content articles covering situations that are much more common on Twitter do not get extra views in comparison to the other article content, additional reflecting the truth that the journalists’ reliance on Twitter may well distort the info they develop when compared to what citizens essentially choose.

Whether Elon Musk will in fact get Twitter stays an open dilemma. Irrespective of whether the new European regulations these as the Electronic Markets Act (Crémer et al. 2022) and the Electronic Companies Act will be productive at regulating written content on social networks has nonetheless to be verified, even if the DSA is a phase in the ideal path. In the meantime, it is very important to maintain in thoughts that social media matters for democracy outside of what anyone could have expected. Indeed, not only does it effects the end users who devote time on the platforms, but also there is a contagion from social to mainstream media. This contagion casts question on the organization design of the legacy media, as properly as the welfare outcomes of the platforms. In distinct, our results phone into dilemma no matter if citizens would be superior informed in the absence of Twitter, and whether social media may well be damaging to the two journalism and democracy.


Blondel, V D, J-L Guillaume, R Lambiotte, and E Lefebvre (2008), “Fast Unfolding of Communities in Substantial Networks”, Journal of Statistical Mechanics: Idea and Experiment 2008 (10): P10008.

Cagé, J, N Hervé and M-L Viaud  (2017), “The industrial worth of news in the web period”,, 19 June.

Cagé, J, Nicolas H, and M-L Viaud (2020), “The Output of Details in an On the net World”, The Critique of Economic Reports 87(5): 2126–64.

Cagé, J, N Hervé, and B Mazoyer (2022), “Social Media Influence Mainstream Media: Evidence from Two Billion Tweets”, CEPR Discussion Paper No. 17358.

Crémer, J, D Dinielli, A Fletcher, P Heidhues, M Schnitzer and F Scott Morton (2022), “The Digital Markets Act: An financial viewpoint on the final negotiations”,, 11 February.

Fujiwara, T, K Muller, and C Schwarz (2021), “The Impact of Social Media on Elections: Evidence from the United States”, NBER Performing Paper No. 28849.

Kennedy, P J, and A Prat (2019), “Where Do Individuals Get Their News?”, Financial Policy 34(97): 5–47.

Levy, R (2021), “Social Media, Information Intake, and Polarization: Proof from a Discipline Experiment”, American Financial Evaluate 111(3): 831–70.

Mazoyer, B, N Hervé, C Hudelot, and J Cagé (2022), “Short-Text Embeddings for Unsupervised Event Detection in a Stream of Tweets”, Advancements in Awareness Discovery and Management 10, forthcoming.



About the Author: AKDSEO

You May Also Like