Analyzing Tripartite Split-Ticketing: A Machine Learning Approach

• In Event: Explaining the 2024 Taiwan Presidential Election

Abstract

This study aims to examine complex patterns of straight- and split-ticket voting in concurrent presidential and legislative elections under the mixed-member majoritarian (MMM) system. The findings can answer both theoretically and substantively interesting questions. For example, do small party supporters allocate their three ballots strategically when their identified party also fields a presidential candidate?
Under the MMM electoral system, there is a lack of linkage between the single member district (SMD) and the proportional list (PR) tiers. Therefore, the advantage of SMD for bigger parties would be unlikely to be offset by PR. If the president is elected through plurality and both presidential and legislative elections are held concurrently, the number of congressional parties would be small. This makes the survival of small parties in the MMM system considerably difficult. Yet small parties that pass the seat-allocation threshold can still be “pivotal” if no major party wins majority seats in the congress.
The literature is abundant concerning strategic and split ticket voting in mixed electoral systems, but almost focusing exclusively on the SMD and PR tiers of legislative branch. For mixed-member proportional (MMP) systems, this is mainly because parliamentary systems do not hold presidential elections. Even for MMM with presidential elections, the executive and congressional elections are not necessarily held concurrently. Taiwan’s concurrent presidential and legislative elections under the MMM system provide a rare opportunity to measure tripartite split-ticketing and analyze the determinants of vote allocation decisions.
Taiwan’s concurrent presidential and legislative elections held on January 13 of 2024 is an interesting case. The vote shares of the top three parties are as below:
DPP KMT TPP others
presidential 40.05% 33.49% 26.46% --
MMM-SMD 45.17% 40.42% 2.97% 11.44%
MMM-PR 36.16% 34.58% 22.07% 7.19%

For the first time since 2012 that a third party TPP garnered more than 22% in both presidential and PR votes, albeit a meager 3% vote share in the SMD tier. It is thus intriguing to see through these aggregate results and examine how individual voters decide their voting choices. This task requires survey data beyond the aggregate results released by the Central Election Commission.
The micro-level data of pre- and post-election telephone surveys collected by the Taiwan’s Election and Democratization Study (TEDS2024-T) Project are employed to analyze patterns of straight- and split-ticket voting in Taiwan’s 2024 concurrent presidential and legislative elections. Pre-election surveys consists of three waves of rolling cross-sections, while the post-election survey traces all the pre-election respondents. This paper will use the reported voting choices of the post-election survey and all the demographic and attitudinal variables recorded in the pre-election polls to avoid the simultaneity problem.
Related literature indicates that cross-pressures behind ticket-splitting involve friction between partisan, social, demographic, and attitudinal elements. Such complexities make it difficult to analyze using standard regression-based approaches such as multinomial logit/probit models. This study instead will divide the TEDS2024-T dataset into training and test data and use the training data to train a tree-based random forest ensemble, and then evaluate the performance of trained model out-of-sample on the hold-out test data. This supervised machine learning approach allows us to identify which conceivable variables matter empirically and discern how they interact to shape voters’ choices.

Author

• Chi Huang, National Chengchi University