Figure 1

(A) Distribution of the Untrustworthiness index of Twitter accounts in TWITIMM, i.e., accounts involved in the immigration debate. The numbers, weighted by the activity of the users, tend to be very low; in logarithmic scale we can see that the vast majority of the users have low U, but the right tail of the distribution decreases slowly, highlighting a core of high-U users at the very end of the distribution; (B) Distribution of U disaggregated by clusters and tested against a randomly reshuffled community partitioning. All the distributions are statistically non-random according to the Mann-Whitney test (\(p \leq 10^{-4}\)). We see how RT2, identified as an anti-immigration cluster, clearly exceeds the average number of high-U users by far, showing a higher and longer right tail than the other communities, suggesting that the prevalence of low credibility media outlets in this cluster is much higher. (C) Shares of URLs disaggregated by community and by type of URL. Links to reliable and unreliable outlets are those that concur with the calculation of the Untrustworthiness score (i.e., users showing a striking preference for one type or the other can be classified, respectively, as trustworthy and untrustworthy), while the category other includes all URLs that do not belong to our lists, thus not contributing to the calculation of U. We observe very different sharing patterns between the communities, with users in some sharing a significantly higher number of reliable URLs, while others show smaller differences between reliable and unreliable media outlets. Differences have been statistically tested as shown in Tab. 2