"Here's Your Evidence": False Consensus in Public Twitter Discussions of COVID-19 Science

TL;DR

The paper introduces a mixed-methods framework to quantify and compare scientific consensus on COVID-19 with public consensus expressed on Twitter, using messages linked to medRxiv/bioRxiv preprints and a retweet-based stance classifier. It defines consensus metrics Cs, Cp, and Cf to measure alignment and exposes a robust false consensus where contrarian discourse dominates public discussion despite strong scientific backing, especially for vaccines. The authors combine large-scale quantitative analyses (topic modeling via BERTopic, network-based node classification with a GCN, regression analyses) with qualitative case studies to reveal tactics like cherry-picking, insinuations of conflicts of interest, and selective reporting that drive misrepresentation. Temporal dynamics show a surge of contrarian activity and new accounts during the pandemic, with highly mentioned contrarian papers disproportionately shaping discourse. The findings highlight the need for perceptible, transparent scientific consensus to counter misinformation and inform policy design in crisis contexts.

Abstract

The COVID-19 pandemic brought about an extraordinary rate of scientific papers on the topic that were discussed among the general public, although often in biased or misinformed ways. In this paper, we present a mixed-methods analysis aimed at examining whether public discussions were commensurate with the scientific consensus on several COVID-19 issues. We estimate scientific consensus based on samples of abstracts from preprint servers and compare against the volume of public discussions on Twitter mentioning these papers. We find that anti-consensus posts and users, though overall less numerous than pro-consensus ones, are vastly over-represented on Twitter, thus producing a false consensus effect. This transpires with favorable papers being disproportionately amplified, along with an influx of new anti-consensus user sign-ups. Finally, our content analysis highlights that anti-consensus users misrepresent scientific findings or question scientists' integrity in their efforts to substantiate their claims.

Figures (11)

• Figure 1: Data collection pipeline.
• Figure 2: 2D histograms of pairwise correlations between log-transformed metrics. ***p$<$ 0.001, *p$<$ 0.5.
• Figure 3: Different types of networks colored by node label.
• Figure 4: CDF of paper share distribution between conformists and contrarians.
• Figure 5: Topics and allocated papers shown as 2D UMAP embeddings. The first topic is a generic one for outliers. High-interest topics are numbered, and their respective positions are indicated in the plot.