Karlsson, E. (2026). Divide and conquer? The impact of media narratives on digital election interference on political and perceived polarisation. New Media & Society, 28, 2812–2837. https://doi.org/10.1177/14614448251410509
Summary
This paper asks whether news coverage about digital election interference — rather than the interference itself — can polarise a democratic public. Using a pre-registered UK survey experiment (N=3,779) that presented fictitious articles about coordinated fake account activity in the 2019 general election, Karlsson tests effects across perceived, affective, and ideological polarisation. Aggregate exposure produced no significant polarising effect, but when respondents saw the opposing party portrayed as the beneficiary, they reported greater perceived societal division and stronger affective hostility toward ideological out-groups. The paper argues that interference narratives can advance the polarising goals of the campaigns they describe, even when the campaigns themselves fail to persuade voters.
Key Contributions
- Provides causal experimental evidence on polarisation as an outcome of interference discourse, extending literature that has largely examined trust and democratic satisfaction.
- Disaggregates polarisation into perceived, affective, and ideological dimensions and shows they react differently to the same stimulus.
- Identifies partisan-alignment asymmetry: effects are conditional on whose side is portrayed as benefitting.
- Suggests ideological (left–right) identities are more emotionally reactive than partisan identities in a multi-party UK context, complementing US-centric work.
- Raises normative implications for how journalists should cover interference without amplifying its intended harm.
Methods
A pre-registered Qualtrics/Prolific online experiment fielded in March 2024 with ten conditions (nine treatments plus control, ~378 per cell) crossing interference source (foreign/domestic/unspecified) with beneficiary party (Labour/Conservative/unspecified). Sampling was balanced between Labour and Conservative partisans, with manipulation checks used to filter inattentive treated respondents. Outcomes captured perceived polarisation (societal division, perceived extremism), affective polarisation (in/out feeling thermometers for party and ideology), and ideological polarisation (party-support strength, left–right self-placement). Linear regression with robust standard errors compared treatment vs. control and partisan “match” vs. “no-match” conditions, with Benjamini–Hochberg q-values for multiple testing.
Findings
- No significant aggregate treatment effect on perceived political division or perceived extremism (H1a/H1b rejected).
- “No-match” respondents (opposing party portrayed as beneficiary) showed higher perceived societal division (~0.026 on a 0–1 scale, p=.01, q=.09), but not higher perceived extremism.
- No aggregate effect on the affective polarisation index (H2a rejected).
- “No-match” respondents showed elevated affective polarisation toward ideological out-groups (~0.026, p=.016, q=.094); no equivalent effect for partisan out-groups.
- Exploratory patterns suggest respondents whose preferred party benefits report lower affective polarisation, consistent with motivated reasoning or dissonance reduction.
- No treatment effects on ideological polarisation (H3a/H3b rejected).
Connections
This paper sits at the intersection of election-interference discourse and polarisation research, connecting to work on how disinformation narratives themselves shape democratic attitudes and trust — see DeVerna2025-dl on perceived exposure effects, Nenno2025-xa and Bennett2025-xs on election-interference framing, and Kalsnes2025-zb on media coverage of information threats. Its focus on affective polarisation via ideological rather than partisan identities links to broader hybrid-media polarisation studies such as Tornberg2025-ir, Green2025-ap, and Arceneaux2026-xk, while the asymmetric partisan-alignment mechanism resonates with motivated-reasoning work like Lyons2026-ca and van-der-Linden2026-jt.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify