Abstract
This study proposes a research framework for analyzing legislative voting activities based on member-level voting similarity, using the 18th National Assembly of South Korea as a case study. We introduce a bill contentiousness measure derived from information entropy to identify bills where voting is diversified across both conservative and progressive parties. Additionally, we propose a member-level voting similarity measure that adjusts for the high frequency of 'Aye' votes. Analysis using Multi-Dimensional Scaling (MDS) and network analysis reveals that while party cohesion is generally high, controversial bills trigger complex cross-party alignments. Results show that the 'Hanara' party maintains higher cohesion than the 'Minju' party as contentiousness increases, and factors such as election type (local vs. proportional) significantly influence individual voting behavior.
Introduction: Decoding the Floor
Traditional metrics for lawmaker performance often focus on quantity—attendance rates or the number of bills proposed. However, these measures fail to capture the "quality" of political alignment and dissent. To better understand the underlying structures of the 18th National Assembly (2008–2012), this research employs information theory and network analysis to map legislative behavior beyond simple party lines.
The Challenge of Voting Sparsity
In many parliamentary systems, including South Korea’s, the majority of bills reaching the floor pass with near-unanimity. This creates a "sparsity" problem where "Yes" votes are so prevalent that they obscure meaningful political disagreement.
To address this, the study filters the dataset to focus on "contentious" bills—those that actually divide the assembly and provide a clearer signal of political alignment.
Methodology: Measuring Contentiousness and Similarity
The researchers developed two primary metrics to extract signal from the noise:
- Bill Contentiousness Measure: Using the concept of information entropy, bills are scored based on how divided the vote is. A bill receives a high score if "Ayes" and "Noes" are balanced and if that division exists within both the conservative and progressive camps, rather than just along party lines.
- Member-level Voting Similarity: Standard similarity measures fail when one category (the "Yes" vote) is overwhelming. The authors adapted the F1-score to create a similarity index that gives equal weight to "No" votes, preventing them from being statistically ignored.
Key Insights and Results
Asymmetric Cohesion and the Contentiousness Filter
By analyzing the top 5%, 10%, and 20% of contentious bills, the study found that the conservative 'Hanara' party exhibited higher internal cohesion on controversial matters compared to the progressive 'Minju' party. As bills became more controversial, Hanara members tended to align more closely with their party leadership, whereas Minju members showed more diverse voting patterns.
The Proportionality Gap
The data revealed that lawmakers elected via proportional representation (party lists) often voted differently than those representing local constituencies. This suggests that representatives beholden to a local base are more likely to break party ranks when a bill impacts their specific region or local interests.
Vanishing Party Lines
Using Multi-Dimensional Scaling (MDS) to project lawmakers onto a 2D map, the researchers observed that while parties are clearly separated on routine bills, these boundaries blur significantly on contentious ones. In high-stakes votes, individual ideology or local interests often override the party whip, leading to the formation of "mixed-party" communities in network analysis.
Conclusion
This framework provides a sophisticated lens for evaluating legislative activity. By filtering for contentiousness and adjusting for voting sparsity, it becomes possible to identify the true ideological fault lines within a national assembly. Future work may integrate temporal analysis to track how these voting networks evolve over time and incorporate natural language processing to link voting patterns directly to bill topics.
Reference: Kang, P., Park, Y., Cho, S., & Kim, S. B. (2014). Analysis on the Voting Activities of the 18th National Assembly of South Korea based on the Member-level Similarity. Journal of the Korean Institute of Industrial Engineers, 40(1), 60-83.
Summary updated