Pathways from Social Media to Problematic Content

The next session at the ICA 2024 conference that I’m attending is presenting articles accepted for a special issue of Political Communication

, and starts with Ryan Moore. Past research has explored the impact of social media on access to mis- and disinformation sources, but remains somewhat inconclusive or very context- and platform-specific. Some of this is drawing on self-reporting; some on browsing data (where it usually focusses on direct referrals from social media platforms); a more indirect link has yet to be explored in full.

Here, social media posts may lead people to other places online that then lead to mis- and disinformation, or simply lead them to adopt ideas that lead to access to problematic content at a later stage. In such scenarios, it is more difficult to assign blame, or even a distinct role, to social media platforms; in this embedded referrers paradigm, some but not all blame probably still should go to social media platforms.

Much past research has also focussed predominantly on Facebook and Twitter; there is a need to add more platforms to these considerations. The present study conducted an ‘ablation study’ where the idea is to remove specific social media platforms from the scene to create a counterfactual world and examine the impact that this has on hypothetical mis- and disinformation uses. It uses Web browsing data as well as participant surveys of users in the US in the last quarter of 2020.

This enables the creation of individual-level transition matrices between sites, and enables the introduction of hypothetical diversion ratios that would divert users along counterfactual browsing paths based on their observed pathways, and this then results in a number of alternative scenarios.

Facebook preceded 9.25% of all visits to misinformation sites in the dataset; in a direct referral model, this is what would disappear if Facebook were to disappear. Under an embedded referrer model, however, some 14% of such referrals would disappear; for Twitter, the effect would increase from 2% to nearly 11% – which indicates the sizeable indirect effect that Twitter has even if it does not necessarily account for so many direct referrals.

The embedded referrals paradigm needs to be further explored, therefore; it can be operationalised using observed media use data, but other approaches to its study may also be possible.