Original Study

Flamino, J., Galeazzi, A., Feldman, S., ..., Szymanski, B. K. (2021). Political polarization of news media and influencers on Twitter in the 2016 and 2020 US presidential elections. Nature Human Behavior, 7, 904–916. https://doi.org/10.1038/s41562-023-01550-8

Abstract. Social media has been transforming political communication dynamics for over a decade. Here using nearly a billion tweets, we analyse the change in Twitter’s news media landscape between the 2016 and 2020 US presidential elections. Using political bias and fact-checking tools, we measure the volume of politically biased content and the number of users propagating such information. We then identify influencers—users with the greatest ability to spread news in the Twitter network. We observe that the fraction of fake and extremely biased content declined between 2016 and 2020. However, results show increasing echo chamber behaviours and latent ideological polarization across the two elections at the user and influencer levels.

Open original study

Replication

Research project, Philipp Knöpfle, 2025

Replicators: Knöpfle, Philipp; Haim, Mario; Breuer, Johannes

Abstract. Flamino et al. (2023) estimate the levels of ideological polarization and echo chamber behavior for Twitter (now X) users during the 2016 and 2020 U.S. presidential elections using political bias classification and network analysis methods. Using 873 million tweets, they find a decline in the proportion of fake and extremely biased content but identify an increase in echo chamber behaviors and latent ideological polarization among both users and influencers over the investigated period. Using the Twitter data and analysis code provided in the complementary OSF.io repository, we were able to reproduce the results of their analysis successfully with only minor deviations due to small technical adjustments. In general, social media analyses frequently blur the distinction between reproduction and replication due to the dynamic nature of platform data and changing access policies resulting in difficulties when retrieving consistent datasets over time. Hence, to assess the replicability of this study, we conducted a robustness check by querying the Twitter/X Batch Compliance API with a subsample from the original analysis to evaluate how many tweets from the initial dataset remain accessible today. Our "rehydration" attempts exposed substantial limitations in the Twitter/X API, as data retrieval issues arose across both free and paid access tiers, preventing us from re-collecting the original dataset or obtaining reliable estimates of tweet accessibility from the original study. While the study was largely reproducible with the intermediary and aggregated data provided, its full reproducibility (e.g. including data recollection and reproducing the data preprocessing) and replicability is constrained by restrictive social media platform data access policies.