Twitter in the News: An Analysis of Embedded Tweets in Political News Coverage

Abstract

This study examines the growing journalistic practice of embedding full tweets in online political news coverage. Against the background of a hybrid media system, we pursue three research goals. First, we evaluate the scope of the Twitter-in-the-news phenomenon relative to news coverage as a whole. Second, we examine the functions of embedded tweets. Third, we identify characteristics that increase the likelihood that tweets will be selected for publication in a news article. We combine computational methods with a manual content analysis and analyze political news coverage of one month outside election periods in four German online news outlets. Our results show that embedding tweets in the news is not a niche phenomenon but has been established as a routine journalistic practice to a moderate extent. In the majority of news articles, the function of tweets is to illustrate an argument or information provided in the text. Geographical proximity and – with some dependencies – the number of popularity cues increased the likelihood of a tweet to be picked up for reporting. The paper shows how transformation processes currently observed in journalism manifest themselves in reporting.

Keywords:

• Political journalism
• hybrid media system
• news values
• popularity cues
• Twitter
• automated and manuel content analysis

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Notes

1 As some news articles embedded more than one tweet, the N for the following analyses is 298 tweets (which are embedded in 222 news articles).

2 As a likelihood ratio test showed that there is no overdispersion, Poisson models are preferable over negative binomial regressions. The main results are robust when using OLS or negative binomial regressions.

3 To test whether ‘sentiment’ makes a difference independent of its direction, we collapsed positivity and negativity into a dummy variable indicating that one of the two is present. We ran an additional model including this new variable instead of the two individual predictors (otherwise, the models do not converge due to multicollinearity). While the sentiment variable is not significant (b = −.28, SE = .25, p = .26), the main findings remain robust. The p-value of proximity is even smaller (p < .001) than in the original models.