In 2019, Canada went through a general election. As the digital innovation team of CBC/Radio-Canada, the Canadian public broadcaster, we decided to cover this very important democratic process with a data-driven perspective and in innovative visual forms.
It was a big challenge since the electoral campaign lasted just six weeks. You will find in this submission the result of our very intense journalistic sprint: five projects, all covering different aspects of the election with different technics, including data analysis, data visualizations, machine learning, cartography, and simulations.
Canada is a bilingual country. Some stories were published just in French, while others were published in both French and English.
We kicked off the electoral campaign with the project Set up to fail: Why women still don’t win elections as often as men in Canada. By analyzing 15 years of electoral data and 10 years of financial data, we demonstrated that women candidates had fewer chances to be elected for two main reasons: they are more often placed in less winnable districts and they receive less money from their party to finance their campaign. We illustrated this issue with data visualizations in scrolly-telling, including one with small multiples.
Then we published Laissez-nous deviner pour qui vous allez voter which allows you to build a character (gender, age, province, income, religion, etc.) and visualize which political party is more likely to get its vote. To calculate this probability, we fed a machine learning algorithm with the answers of 387 671 voters, who were asked about their demographics and which party they would vote for. We also identified the biggest factors influencing that decision.
For the project Ce qu’il faut retenir du débat des chefs, we worked all night after the leaders’ debate to publish a data analysis with visualizations of each leaders’ interventions.
Finally, during the vote night, we published two projects. We presented an analysis of the results in Voici où s’est jouée l’élection en carte, with maps animated in scrolly-telling. And we also made a simulation of the results with different electoral systems. This is the project called How would proportional representation have shaped this election’s results?
In the end, we provided hundreds of thousands of Canadians with crucial information, published in innovative formats.
When a data analysis was needed, we did it in Python, with the Pandas module.
For the machine learning algorithm, we partnered with the non-profit Vox Pop Labs. The algorithm itself was based on three models: one multinomial logistic regression, one support vector machine, and one extreme gradient boosting.
What was the hardest part of this project?
The electoral campaign is very short in Canada. The last one was only six weeks long. Therefore, it was a real challenge to publish many customized projects, with many different technics. We standardized as many things as possible. For example, our React template helped us to be more efficient. But the real key is a great multidisciplinary team, with members trusting each other.
What can others learn from this project?
We think this project showcases very different ways to adopt a data-driven approach, but always with a journalistic purpose. All the projects are about the electoral process, but with different angles, different visual forms and different experiences for the user.
We hope other reporters, designers, and developers will be encouraged to be more creative after reading our stories. Part of our job is to find and publish quality information, but it’s also to constantly find the best way to communicate this information.