Election analysis with natural experiment approach

Category: Innovation (small and large newsrooms)

Country/area: Japan

Organisation: The Chunichi Shimbun

Organisation size: Big

Publication date: 6 Jan 2019

Credit: Isao Matsunami

Project description:

In the national election coverage we used natural experiment approach to the turnout analyses. We found 1) costly early voting stations in the shopping malls do not improve overall turnout, just replacing Sunday vote with early vote, by denying the relationship between the distance to the newly created station and the turnout, 2) pig’s year effect, long cherished by political pundits in Japan, has no ground by comparing turnouts between with-local-election municipalities and without-local-election ones.

Impact reached:

The story reached an above-average level of engagement on social media.

Techniques/technologies used:

To show the relationship between the walking distance to the voting station and voter turnout, we used QGIS network analysis with 100m estimated population mesh and OpenStreetMap. The main graphics of road network is made on d3.js and webGL to explain the intention of the analysis. In pig’s year effect story, data analysis and some charts are done on RStudio. All charts are screen-width responsive.

What was the hardest part of this project?

Precinct map is created manually, with help of reverse geocoding, because it is defined not by geographical area but by address in Japan. Local election data is collected manually by hitting database because it is not collected by local governments. Both data are the most time-consuming tasks but is the starting point of the projects. 

What can others learn from this project?

Natural experiment approach is one of the compelling ways for the reporters to write causal inference stories by analyzing observational and/or public data while avoiding selection bias and statistical cynicism. Although this approach can not always be applied to the questions you want to know, we have still many policy-making arguments that we can check anew and debunk. 

Project links: