Keywords: Exit Poll, Election, Methodology, Japan

Exit polls in Japanese elections have been conducted since around 1990. The Asahi Shimbun began experimenting with exit polls in the late 1990s and launched its first full-scale implementation during the 1998 House of Councillors election. The purposes of exit polling in Japan are twofold. First, exit polls are used to predict the election outcomes in each electoral district immediately after voting closes at 8:00 PM. Accurately forecasting which candidates are likely to win is particularly important for television broadcasters, as it allows them to provide timely and reliable reporting. Second, exit polls serve as a tool for analyzing voter behavior, such as understanding which political party received the most support from unaffiliated voters.
For the October 2024 House of Representatives election, six organizations, including the Asahi Shimbun, collaborated to conduct a joint exit poll. This survey was carried out at 8,335 polling stations across 289 electoral districts, yielding valid responses from 376,787 voters. On average, 1,304 responses were collected per electoral district.
The survey methodology involved stationing interviewers at designated polling station exits. Voters who had completed voting were handed a tablet and asked to answer questions by interacting with the touchscreen. These questions included their voting preferences and party affiliations. The collected data were transmitted in real-time to a central server, allowing clients to download raw data as needed.
Given that there are approximately 46,000 polling stations across Japan, selecting which stations to include in the exit poll is critical. The accuracy of an exit poll largely depends on the representativeness of the selected polling stations. Stations must statistically reflect the demographic and voting patterns of their respective electoral districts.
In addition to statistical considerations, logistical factors must also be addressed. In the Asahi Shimbun's exit polls, interviewers often conduct surveys at multiple locations in a single day. Consequently, the time and cost of transportation, whether by public transit or automobile, must be taken into account.
Determining polling locations is akin to solving a complex optimization problem, balancing the following parameters:
Number of surveyors: Limited by cost constraints.
Number of polling locations: A higher number increases representativeness.
Number of valid responses: A higher response count enhances reliability.
Geographic distribution of locations: A dispersed distribution is ideal, but proximity is preferred to minimize travel costs and time.
The selection process begins by stratifying polling stations based on key election-related factors such as voter turnout, the incumbent's vote share in the previous election, and party support rates. Once stratified, polling stations are sampled using systematic random sampling, starting with a randomly chosen number and selecting stations at fixed intervals.
In the presentation, we will analyze the discrepancies between the vote shares estimated by last year’s exit polls and the actual vote shares for each candidate. Additionally, we will evaluate whether the candidates ranked first in the exit polls ultimately secured victory. This analysis will provide insights into the accuracy and predictive power of exit polling in Japan.