讲座人：山本铁平，日本早稻田大学

讲座时间：2026年10月21日上午10:30-12:00

讲座地点：北京大学国家发展研究院承泽园校区344教室

讲座题目：Using Conjoint Tasks for Voting Advice Applications

内容简介：Voting advice applications (VAAs) are widely used across multi-party democracies to aid voters' decision-making about their electoral choices among parties or candidates. A typical VAA features dozens of policy issues and asks a voter about their preferred position on each of the issues, one by one. The application then calculates the voter's overall score indicating how closely their position matches each of the parties or candidates. This common approach, however, suffers from several important limitations.

First, the calculation of the overall match scores involves adding up scores on individual issues with arbitrary (typically uniform) weights, implicitly making a strong assumption about how much voters care about each issue. Second, the calculation also fails to incorporate possible interactions between different issues, ignoring the possibility that some voters might, for example, take into account the consistency of multiple policies as a package when they make a vote choice. Third, responses on politically sensitive policy issues, such as gender and ethnicity, might be prone to social desirability bias.

In this research, we propose a novel type of VAA based on conjoint tasks. Instead of asking about individual policy issues, we instead present packages of issue positions and ask voters to choose which hypothetical party or candidate they would vote for in an election. This approach overcomes the limitations of existing VAAs by way of measuring both direction and intensity of voters' preferences about policy issues via conjoint tasks (Bansak et al. 2023). Moreover, the responses collected through this application can be analyzed with a variety of statistical techniques for standard conjoint survey experiments, which have been rapidly developed over the recent years.Our application uses a new algorithm to calculate a voter's overall proximity to the target policy bundles on the basis of their own conjoint responses. The algorithm is simple, fast, and scalable to large-scale deployment for public use in actual elections. We demonstrate the performance of the proposed algorithm via simulations and implement it for the two Japanese National Elections in July 2025 and February 2026. The empirical application involved collaboration with a major Japanese national newspaper, the Nikkei, and yielded conjoint responses from more than 900,000 unique users in total, enabiling highly granular analyses of voters’ policy preferences.

主讲人简介：山本铁平教授是早稻田大学政治与经济学部Faculty of Political Science and Economics, Waseda University教授，长期聚焦于因果机制识别与实验设计创新，在PNAS，APSR，AJPS，JASA，JRSS-A，Political Analysis上发表多篇论文。此前，他曾在MIT政治学系担任正教授、政治学方法实验室（Political Methodology Lab）主任，现任顶尖期刊AJPS副主编、政治学方法论学会（Society for Political Methodology）执行委员会委员。

组织者：

（国家发展研究院 ）李力行、席天扬、王轩、徐化愚、刘诗尧、黄清扬

（经济学院）刘冲、吴群峰、曹光宇、年永威、施新政

（光华管理学院）张晓波、仇心诚