Yoichi Masuzoe celebrates his gubernatorial election victory in Tokyo. (Kyodo News/Associated Press)
Joshua Tucker: We are pleased to welcome back political scientists Steve Pickering (Kobe University), Seiki Tanaka (Syracuse University), and Kyohei Yamada (International University of Japan) with a follow up to their post last week regarding the ways in which Twitter helped identify public opinion related to the Tokyo mayoral election. Here, they assess the extent to which this same Twitter data could predict voting patterns in the election.
*****
On Feb. 9, Tokyo’s voters elected a new governor. Masuzoe Yoichi (see photo above), a former Diet member obtained the highest vote share (43 percent). In our post last week on this election, we presented the frequency of tweets on three key issues: nuclear power plants, social welfare and the Tokyo Olympics. Now that we have voting data at the municipality and ward levels, we want to check if these tweets had any relationship to the vote outcome.
Here, we focus on nuclear power plants to see if the frequency of tweets on this issue is related to the vote shares of the two candidates who had a clear position on this issue – Hosokawa Morihiro, a former prime minister, and Utsunomiya Kenji, a lawyer supported by the Japan Communist Party and the Social Democratic Party. Since these two candidates had a similar position, many speculated that they would split the left-leaning votes, and they did: Hosokawa received 19.6 percent of the vote and Utsunomiya 20.2 percent. We combine their vote shares, assuming that voters who prioritized the nuclear issue and were against nuclear plants would have likely voted for one of these two candidates.
Here’s our first map, which shows the levels of support for the winning candidate Masuzoe as compared to the two anti-nuclear plants candidates.
Here’s our map from last week’s post.
Comparing the two maps tells us that vote share of the anti-nuclear candidates tended to be higher in municipalities and wards that had more tweets on nuclear power plants. This is just a correlation, and more rigorous analysis is under way, but it indicates that looking at tweets can provide information that surveys alone may not be able to reveal.