A Study on Quantitative Quality Measurement Factors for Passed Bills

Abstract

This study differs from prior legislative research, which has primarily focused on identifying factors influencing a bill’s likelihood of passage, by instead examining the factors that affect the quality of bills that have already been enacted. To overcome the cost and consistency limitations of traditional expert-driven qualitative evaluations, the study adopts the LLM-as- Judges methodology, employing large language models (LLMs) as proxy evaluators. The empirical scope covers 2,785 member-sponsored bills passed during the 20th and 21st Korean National Assembly. From the quantitative characteristics available for these bills, the study identifies and analyzes which indicators meaningfully influence legislative quality. Pre- dictive performance was evaluated using five-fold cross-validation across four machine learning classification models, including logistic regression, and statistical significance was evaluated through permutation testing. The results show that sentence length exerts a significant effect on the clarity and readability dimensions of legislative quality. This study carries methodological and policy implications by shifting the analytical paradigm from predicting “factors influencing passage” to predicting “factors determining high-quality legislation,” and by empirically demonstrating that quantitative assessment of legislative qual- ity is feasible through LLM-based evaluation.