Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Affective Polarization Amongst Swedish Politicians

François t'Serstevens, Roberto Cerina, Gustav Peper

TL;DR

This study investigates affective polarization among Swedish politicians on Twitter (2021–2023), focusing on positive versus negative partisanship and how definitions of the in-group shape online discourse. By combining 25,129 tweets with multi-model sentiment analysis and three ideology measures (DW-NOMINATE, CHES, and LLM-based ratings), the authors reveal that negative partisanship dominates when in-groups are defined at the party level, and this negativity yields substantially higher online engagement (roughly 3.18 more likes and 1.69 more retweets). The results also show that extremist actors are more prone to NP under a party-level in-group, though findings vary with the ideology metric used; DW-NOMINATE performs poorly in this multiparty, EU context. The work highlights the value of LLMs for analyzing non-English political data at scale and provides new insights into how multiparty dynamics shape polarizing discourse and online visibility.

Abstract

This study investigates affective polarization among Swedish politicians on Twitter from 2021 to 2023, including the September 2022 parliamentary election. Analyzing over 25,000 tweets and employing large language models (LLMs) for sentiment and political classification, we distinguish between positive partisanship (support of allies) and negative partisanship (criticism of opponents). Our findings are contingent on the definition of the in-group. When political in-groups are defined at the ideological bloc level, negative and positive partisanship occur at similar rates. However, when the in-group is defined at the party level, negative partisanship becomes significantly more dominant and is 1.51 times more likely (1.45, 1.58). This effect is even stronger among extreme politicians, who engage in negativity more than their moderate counterparts. Negative partisanship also proves to be a strategic choice for online visibility, attracting 3.18 more likes and 1.69 more retweets on average. By adapting methods developed for two-party systems and leveraging LLMs for Swedish-language analysis, we provide novel insights into how multiparty politics shapes polarizing discourse. Our results underscore both the strategic appeal of negativity in digital spaces and the growing potential of LLMs for large-scale, non-English political research.

Affective Polarization Amongst Swedish Politicians

TL;DR

This study investigates affective polarization among Swedish politicians on Twitter (2021–2023), focusing on positive versus negative partisanship and how definitions of the in-group shape online discourse. By combining 25,129 tweets with multi-model sentiment analysis and three ideology measures (DW-NOMINATE, CHES, and LLM-based ratings), the authors reveal that negative partisanship dominates when in-groups are defined at the party level, and this negativity yields substantially higher online engagement (roughly 3.18 more likes and 1.69 more retweets). The results also show that extremist actors are more prone to NP under a party-level in-group, though findings vary with the ideology metric used; DW-NOMINATE performs poorly in this multiparty, EU context. The work highlights the value of LLMs for analyzing non-English political data at scale and provides new insights into how multiparty dynamics shape polarizing discourse and online visibility.

Abstract

This study investigates affective polarization among Swedish politicians on Twitter from 2021 to 2023, including the September 2022 parliamentary election. Analyzing over 25,000 tweets and employing large language models (LLMs) for sentiment and political classification, we distinguish between positive partisanship (support of allies) and negative partisanship (criticism of opponents). Our findings are contingent on the definition of the in-group. When political in-groups are defined at the ideological bloc level, negative and positive partisanship occur at similar rates. However, when the in-group is defined at the party level, negative partisanship becomes significantly more dominant and is 1.51 times more likely (1.45, 1.58). This effect is even stronger among extreme politicians, who engage in negativity more than their moderate counterparts. Negative partisanship also proves to be a strategic choice for online visibility, attracting 3.18 more likes and 1.69 more retweets on average. By adapting methods developed for two-party systems and leveraging LLMs for Swedish-language analysis, we provide novel insights into how multiparty politics shapes polarizing discourse. Our results underscore both the strategic appeal of negativity in digital spaces and the growing potential of LLMs for large-scale, non-English political research.

Paper Structure

This paper contains 10 sections, 2 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Theoretical Framework
  3. Methods and Materials
  4. Tweet Data
  5. Tweet Sentiment
  6. Estimating Politician Ideology
  7. Results and Analysis
  8. Discussion
  9. Partisanship Online
  10. LLMs for Data Labelling

Figures (5)

  • Figure 1: Amount of tweets posted by date.
  • Figure 2: Negative and Positive Tweets for the In- and Out-Groups
  • Figure 3: Tweet Weighted Network
  • Figure 4: Posterior distribution of the ratio of negative to positive partisanship reactions
  • Figure 5: Estimated probability of posting a negative partisan tweet by party and ideological rating