JMIR Publications

ABSTRACT

Background:

Echo chambers are the engines behind various phenomenon. When individuals segregate with like-minded people (chambering) and these ideas impact individuals in a non-rational way (echoing), this polarisation affects downstream behaviour. No less is this true for vaccine hesitancy, whereby echo chambers are thought to push vaccine hesitancy movements or spread of misinformation. Further, certain influential users are responsible for this movement.

Objective:

The objective of this study is to describe the growth and levels of chambering between different vaccine stances. In addition, it aims to describe how influential nodes temporally contribute to the chambering process. Assessing the temporal development of echo chambers, and the relation that influential users have on their growth, gives insight on effective communication for preventing rise in vaccine hesitancy.

Methods:

In this study, a Taiwanese forum dataset is used, and a multilayer network model is constructed to assess the existence of echo chambers. Each layer represents either pro-vaccination, vaccine hesitant, or anti-vaccination board posts, and their cross-commenting and comment-receiving behaviour is used to measure echo chambering. For understanding the behaviour of influential users – or key nodes – in the network, node degree and PageRank is used, in addition to the conceptualisations of these measurements on a multilevel network. Layer-level (number of nodes, number of edges, average degrees) and node-level (indegree, outdegree, PageRank, cross-indegree, cross-outdegree, Multiplex PageRank) attributes that could be validated through bootstrapping were done.

Results:

In total, the pro-vaccination layer (G^P) had n=1,283 board posts, 11,087 average nodes, and 23,504 edges; vaccine-hesitant layer (G^H) n=1,322 boards, 14,037 average nodes, and 33,520 edges; and anti-vaccination layer (G^A) had n=387 boards, 5,477 nodes, and 8,696 edges. The decreasing homophily between layers over time suggests that there is no existence of echo chambers. Instead, there are increasing cross-cutting exchanges between layers, supporting a “battleground” engagement from both polarised layers on vaccine hesitant boards. When looking at behaviour of hardliners – a type of influential node mostly commenting within their layer – anti-vaccination hardliners are more active in engaging with the vaccine hesitant layer than the pro-vaccination equivalent. When looking at behaviour of high PageRank nodes in a layer, they also mostly engage with the vaccine hesitant layer.

Conclusions:

The findings suggest that the pro-vaccination layer should increase its efforts to engage with vaccine hesitant individuals. In addition, efforts should be made to target hardliner and influential nodes in the anti-vaccination layer to reduce engagement with the vaccine hesitant layer, preventing sway in the battleground.