Electoral Reform and Pork Barrel in Parliamentary Questions

Measuring legislator behaviours and tendencies towards constituencies under different electoral systems is important. This paper quantitatively investigates this topic using the case of Taiwan Legislative Yuan and data on written parliamentary questions through an electoral reform from multi-member districts (MMD) to single-member districts (SMD). With existing labelled pork legislation, I train deep learning models using convolutional neural networks with an embedding layer extracted from Transformer BERT to detect pork-barrel features in parliamentary questions over time. Evidence exists to show that legislators under MMD are more likely to express political intention about pork-barrel projects in written parliamentary questions. The reform subsequently demonstrates heterogeneous effects on major parties vis-à-vis minority parties, respectively.