Larsson, A. O. (2026). “Meet the new boss – Same as the old boss”: A longitudinal study of political post sentiment and Facebook engagement. First Monday. https://doi.org/10.5210/fm.v31i1.14448
Summary
This paper offers a decade-long (2013–2023) longitudinal study of Facebook communication by Norway’s populist Progress Party and its two consecutive leaders, Siv Jensen and Sylvi Listhaug. Larsson asks how the sentiment of political posts relates to audience engagement over time, combining a supply-side analysis of what parties post with a demand-side analysis of how users respond. Working with 11,160 posts, the author documents a growing reliance on negative campaigning—especially after Listhaug’s 2021 leadership takeover—and shows that negativity reliably drives shares and comments, while likes follow a distinct, more positively-valenced logic. The paper doubles as a methodological demonstration of LLM-assisted content analysis in a non-English (Norwegian) setting.
Key Contributions
- Rare 10-year longitudinal evidence on sentiment and engagement in populist political Facebook communication outside the Anglo-American context.
- A reproducible hybrid workflow pairing GPT-4 zero-shot annotation with human validation for sentiment coding in Norwegian, achieving high reliability.
- Demonstrates that likes operate on a different logic from shares and comments with respect to sentiment.
- Integrates content-production (supply) and audience-engagement (demand) analyses within a single design.
- Speaks to debates on negative campaigning, populism, permanent campaigning, and political professionalization.
Methods
- Longitudinal collection via CrowdTangle of 11,160 public Facebook posts from three pages: the Progress Party (5,941), Sylvi Listhaug (2,791), and Siv Jensen (2,428).
- Hybrid content analysis: zero-shot classification by GPT-4 (gpt-4-0314) using a Norwegian-language prompt to label posts as negative, positive, or neutral, validated against human coding.
- Reliability: LLM re-coding of a 20% sample (Krippendorff’s α = .91); human coding of a 5% sample (α = .82).
- Engagement analysis using medians, Kruskal-Wallis rank sum tests, and Dunn’s post-hoc tests (given non-normal distributions), plus visualizations across account, sentiment, and year.
Findings
- The share of negative posts rose over the decade, sharply so after Listhaug became leader in 2021; her own content shifted from mostly positive (2013–2016) to predominantly negative from 2017.
- The Progress Party’s page stayed positive until 2021, then turned negative—suggesting leaders shape party-level communication style.
- Negative posts consistently earned the highest median shares across accounts and years, and drove more comments in most years (2021 being an exception).
- Likes diverged: positive content became the dominant driver of likes for Jensen (2020–2023) and Listhaug (2021–2022).
- Election years no longer reliably produced engagement surges, pointing toward more permanent campaigning rhythms.
Connections
This paper’s methodological core—LLM zero-shot classification with human validation in a non-English language—connects to broader work on the reliability and use of LLMs for computational content analysis, such as Le-Mens2025-qz and Balluff2026-if. Its substantive focus on affective/negative content and engagement dynamics resonates with research on emotional amplification and outrage online like Brady2026-ln, while its populist political-communication framing links loosely to studies of partisan online behavior in this register.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts