Why Are Some Countries More Politically Fragmented Online Than Others?

TL;DR

The paper tackles cross-country online political fragmentation by introducing a multiscale, structure-based fragmentation score FRAG built on Markov Stability-based community detection and applying it to co-following networks of 18,325 political influencers in Brazil, Spain, and the United States. It demonstrates that online fragmentation varies by country and ideology, with Brazil most fragmented and Spain/US closer in level, and shows that stronger alignment between social identities and ideological positions reduces fragmentation. The method integrates the effective branching factor through the ENC, enabling cross-scale comparisons and linking online fragmentation to offline party-system structure, while revealing cross-national differences in how social sorting shapes fragmentation. The findings highlight the importance of social identity alignment in mitigating fragmentation and provide a principled, scalable approach for cross-national comparisons in online political ecosystems with potential applications for moderation and democratic governance.

Abstract

Online political divisions, such as fragmentation or polarization, are a growing global concern that can foster radicalization and hinder democratic cooperation; however, not all divisions are detrimental, some reflect pluralism and healthy diversity of opinion in a democracy. While prior research has predominantly focused on polarization in the United States, there remains a limited body of research on political divides in multiparty systems, and no universal method for comparing fragmentation across countries. Moreover, cross-country comparison is rare. This study first develops a novel measure of structural political fragmentation built on multi-scale community detection and the effective branching factor. Using a dataset of 18,325 political influencers from Brazil, Spain, and the United States, we assess online fragmentation in their Twitter/X co-following networks. We compare the fragmentation of the three countries, as well as the ideological groups within each. We further investigate factors associated with the level of fragmentation in each country. We find that political fragmentation differs across countries and is asymmetric between ideological groups. Brazil is the most fragmented, with higher fragmentation among the left-wing group, while Spain and the United States exhibit similar overall levels, with the left more fragmented in Spain and the right more fragmented in the United States. Additionally, we find that social identity plays a central role in political fragmentation. A strong alignment between ideological and social identities, with minimal overlap between ideologies, tends to promote greater integration and reduce fragmentation. Our findings provide explanations for cross-national and ideological differences in political fragmentation.

Figures (8)

• Figure 1: (a)–(b) examples of how the effective branching factor (denoted by B) is computed for equal-sized and unequal-sized communities, respectively. (c)-(e) alluvial diagrams of multi-level communities of online political influencers in Brazil, Spain, and the United States, with red indicating that more than half of the self-reported ideological identities are left-leaning, and blue indicating that more than half are right-leaning. No community has a majority of members self-identifying as centrist. The transparency of the colors represents the proportions of the majority ideological identities in each community. The fragmentation score (FRAG) for each level is shown, along with the overall fragmentation score and the scores for left- and right-wing groups (bottom).
• Figure 2: Community merging patterns based on similarity in ideological positions. Heatmaps (left column) show the ideological position similarity between all pairs of communities at each level. Bar plots (right column) compare the average similarity values between communities with shared ancestors versus those with distinct ancestors (where an ancestor refers to a community at the preceding level).
• Figure 3: Community merging patterns based on similarity in social identity. Heatmaps (left column) showing social identity similarity between all pairs of communities at each level. Bar plots (right column) compare the average similarity values between communities with shared ancestors versus those with distinct ancestors (where an ancestor refers to a community at the preceding level).
• Figure 4: Multi-level pie charts showing the proportions of different social identity categories for each community at each level in Brazil, Spain, and the United States. Each circle is color-coded based on the majority ideological position, with color transparency reflecting the proportion. Community size is indicated within each circle. Edges represent shared survey audiences between communities at consecutive levels, weighted by the proportion of the shared audience flowing to lower-level communities relative to the size of the corresponding higher-level community.
• Figure 5: Correlation heatmap between ideological position and social identity in Brazil, Spain, and the United States.