Population-scale network embeddings expose educational divides in network structure related to right-wing populist...

Area of Science:

• Social Network Analysis
• Computational Social Science
• Political Science

Background:

• Administrative registry data enables population-scale network construction reflecting shared social contexts.
• Machine learning can encode these networks into numerical representations (embeddings) capturing individual positions.
• Previous research has not fully explored the interpretability and predictive power of such embeddings for social outcomes.

Purpose of the Study:

• To create interpretable, population-scale network embeddings for the Dutch population.
• To assess the predictive power of these embeddings for right-wing populist voting.
• To investigate the link between network structures, particularly educational ties, and voting behavior.

Main Methods:

• Constructed a population-scale social network from administrative data across five contexts (neighborhood, work, family, household, school).
• Generated network embeddings for all individuals using machine learning techniques.
• Utilized embeddings and individual characteristics to predict right-wing populist voting, followed by embedding transformation and analysis.

Main Results:

• Network embeddings alone predicted right-wing populist voting above chance but less effectively than individual characteristics.
• Combining embeddings with individual characteristics yielded only marginal improvements in prediction.
• After transformation, one embedding dimension strongly correlated with voting, revealing that educational ties and attainment differences correspond to distinct network structures associated with this voting pattern.

Conclusions:

• Population-scale network embeddings can be rendered interpretable, offering insights into social phenomena.
• Structural differences in educational networks are significantly associated with right-wing populist voting.
• This study demonstrates a novel methodological approach linking social network structure to political behavior.