Attribution of polarization to selective exposure versus recommendation algorithms

Ascertain the relative contributions of users’ selective exposure choices compared to platform recommendation algorithms in producing current levels of political polarization observed on social media, by quantifying and disentangling these factors with methodologies that can credibly separate user selection effects from algorithmic amplification.

Background

The authors find extensive polarization and echo chambers in COVID-19 Twitter data, but their analyses are observational and do not establish causal mechanisms. In computational social science, homophily and contagion are often confounded, making it difficult to parse whether users’ own choices (selective exposure) or platform systems (recommendation algorithms) drive observed outcomes.

Disentangling these drivers is essential for designing effective interventions, evaluating platform accountability, and understanding the societal impact of social media architectures. The authors explicitly note that the division of responsibility between user behavior and algorithmic design is unclear.