Do different types of information disseminate differently online? In recent years, studies have sought answers to these questions by comparing the statistical properties of network paths taken by different types of content delivered online. Here, we demonstrate the importance of controlling the correlations between the compared properties. In particular, we show that the previously reported structural differences between fake and real news dissemination paths on Twitter disappear when only comparing stunts of the same size; differences between the channels for disseminating images, videos, news and petitions persist. Coupled with a theoretical analysis of dissemination processes, our results suggest that, in order to limit the spread of fake news, it may be sufficient to focus on reducing the average ‘contagiousness’ of information.
Do some types of information spread faster, more widely, or further than others? To understand how news broadcasts differ, researchers compare the structural properties of the paths content takes as it propagates through a network, studying what are known as waterfalls. Commonly studied cascade properties include range, depth, width, and velocity of propagation. Drawing conclusions from statistical differences in these properties can be difficult, as many properties are dependent. In this work, we demonstrate that it is essential to control the size of waterfalls when studying structural differences between collections of waterfalls. We first revisit two datasets of notable recent online streaming studies that reported content-specific differences in the waterfall topology: an exhaustive corpus of Twitter cascades for content verified by Vosoughi as true or fake news. et al. [S. Vosoughi, D. Roy, S. Aral. Science 359, 1146–1151 (2018)] and a comparison of Twitter stunts of videos, images, news and petitions by Goel et al. [S. Goel, A. Anderson, J. Hofman, D. J. Watts. Manage. Sci. 62, 180–196 (2016)]. Using methods that control joint stunt statistics, we find that for fake and real news stunts, the reported structural differences can almost entirely be explained by the fact that the fake news stunts are larger. For videos, images, news and petitions, structural differences persist when controlling size. By studying classical diffusion models, we then give the conditions under which the differences in structural properties under different models reduce or not to differences in size. Our results are consistent with the fact that the mechanisms underlying the dissemination of true and fake news are quite similar, differing primarily in the basic infectivity of their propagation process.
- Accepted August 30, 2021.
Author contributions: research designed by JLJ and JU; JLJ and JU carried out research; JLJ and JU contributed new reagents / analytical tools; JLJ analyzed the data; and JLJ and JU wrote the paper.
The authors declare no competing interests.
This article is a direct PNAS submission. DW is a guest editor invited by the Editorial Board.
This article contains additional information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.2100786118/-/DCSupplemental.
Previously published data (10, 11) were used for this work.