Home > News > In Tokyo elections, Twitter reveals voter preferences in a way surveys do not
193 views 6 min 0 Comment

In Tokyo elections, Twitter reveals voter preferences in a way surveys do not

- February 5, 2014

Joshua Tucker: The following is a guest post by political scientists Steve Pickering (Kobe University), Seiki Tanaka (Syracuse University), and Kyohei Yamada (International University of Japan).
*****

On Feb. 9, a new governor of Tokyo will be elected. Following the resignation of the former governor Inose Naoki due to a bribery scandal, voters in Tokyo will elect a new leader to manage what is one of the largest administrative areas in the world, home to more than 13 million people and with a Gross Regional Product of US$90 billion (almost as much as South Korea’s Gross Domestic Product).

In many elections, voters choose their candidate along party lines. But in the case of the Tokyo governor, issues are believed to be more important than the party. We want to understand which issues are important, and how these issues vary across Tokyo. The Japanese newspaper Mainichi Shinbun has been looking at the issues voters care about in this election by running telephone surveys in Tokyo. But there may be problems with telephone surveys, such as survey respondents being older or more conservative than the social media generation. To address this gap, we use Twitter to analyze candidates’ and voters’ tweets to draw inferences about voter preferences regarding the election.

To date, there has been very little research on the effect that the Internet has had on Japanese elections. This is hardly surprising; prior to July 2013, political candidates were banned from using the Internet as part of their campaigns! A highly connected society such as Japan, where Internet campaigning has only recently been legalized, is a perfect place to use social media to study the issues that voters care about. Specifically, we picked three salient issues that candidates have been reported to emphasize in their election platforms:

  • Nuclear power plants
  • The Tokyo Olympics
  • Social welfare

For each issue, we checked the frequency of tweets at the municipality or ward level (both are equivalent to boroughs in the United States) within Tokyo. First, we simply examined the number of tweets per day for each of the three key words between Jan. 25 and 28, without geographical information. Figure 1 suggests that the most frequently tweeted phrase is “nuclear power plants,” while fewer people are interested in the Tokyo Olympics and the least number of people tweeted about social welfare. Another thing we notice is that while the number of tweets for nuclear plants and social welfare is relatively constant over time, the number of tweets about the Tokyo Olympics varies. (As an aside, we also find that like television, there is a “prime time” for tweets – people in Japan used Twitter in the evening most frequently, and the clear peak was at 9 p.m.)

Frequency of tweets by subject area in Tokyo (Figure and Data: Steve Pickering, Seiki Tanaka, and Kyohei Yamada)

The next figure shows the geographic distribution of tweets within Tokyo’s municipalities and wards. We can see that each of the key words is most frequently tweeted by users registered in the seven wards that comprise downtown Tokyo (popular areas such as Shinjuku and Suginami). However, “nuclear power plants” is more widely spread out across space, with a large number of tweets in the western suburbs such as Chofu, Fuchu, and Kunitachi. On the other hand, the “Tokyo Olympics” and “social welfare” are not as frequently tweeted in those outer areas.

Maps represent the natural log of tweets sent within each municipality for the three key issues on Jan 28 2014. (Figure and Data: Steve Pickering, Seiki Tanaka, and Kyohei Yamada)

Maps represent the natural log of tweets sent within each municipality for the three key issues on Jan 28 2014. (Figure and Data: Steve Pickering, Seiki Tanaka, and Kyohei Yamada)

Now here’s the interesting bit: when we compare our findings with the telephone survey by Mainichi Shinbun on Jan. 27 (see this Japanese language survey), we find that the Twitter data are not consistent with the survey reported in the newspaper. The research team there conducted a telephone-based survey and visualized the issue that the respondents most cared about by using a similar municipality map. Their findings show that “nuclear power plants” was the most salient issue in only a few wards, whereas “social welfare” was salient in most of the municipalities and wards.

We can speculate several potential reasons for the discrepancy, including: i) Twitter users are different from the average voter; and ii) Twitter users are different from the average telephone survey respondent. While respondents in a telephone-based survey are likely to be older and have party affiliations, those who use Twitter are probably younger and less organized than the average voter and the average telephone survey respondent (as in the United States, young people are much less likely to have landline phones). If the assumption holds, we can reasonably argue that i) telephone surveys are useful, but only give part of the picture; ii) Twitter analyses are an important complement to telephone surveys in order to reveal voters’ preferences. Candidates with a clearer understanding of what voters care about may be in a better position to get out the vote — and win the election.