As my battery runs out for today, I’m in a final session on partisanship at the 2026 International Communication Association conference in Cape Town, which begins with a paper by Harry Yan. His focus is on the alignment between news headlines and images; such alignment is especially important now that headlines and header images frequently circulate together as news is shared on social media platforms.
To the extent that these align, they may represent a form of multimodal media bias, and such bias might also result on differences in social media engagement. The project tested this for a dataset of 48 news sources over five years (2017-2021), focussing on 8 US politicians; it used computational techniques to confirm the identity of politicians depicted, and identify the stance shown in headlines and images.
Headline favourability followed partisan biases in US media; for images, this was not the case (with slightly more positive images used throughout for Democrat than Republican politicians). Headline and visual favourability were therefore only weakly aligned; there were no obvious partisan biases in this.
Negative headlines produced higher engagement than neutral and positive headlines, but positive headlines also generated more engagement than neutral headlines. Alignment between headlines and images assisted by generating slightly more engagement than non-alignment. Partisan bias on social media is thus multimodal, but mostly focussed on text headlines; positive or (especially) negative bias in the media is more important, though.
