Three reasons why we at SVT open source our algorithm for the Election Compass (VAA)
Firstly we want to be truly transparent in our methodology and how our matching algorithm works, secondly we welcome suggestions on how we can do even better and finally we want other actors to be able to use what we have in order to improve overall VAA quality. Those are the main reasons why we release the VAA matching algorithm as open source.
Voting Advice Applications (VAA) – or Election Compasses/Valkompasser as they are called in Sweden – have been around for a long time and with every election their importance for voters rises along with actual usage numbers. At SVT we have made VAA’s since 2002 and in the 2010 parliament election we leveled up our ambitions by introducing the possibility to match not only against parties but also all participating candidates. In the latest national election in 2018 we went even further by also offering local VAA’s for municipalities and regions as well as for the national parliament – since all three of those elections are held simultaneously in Sweden. The latest VAA we’ve developed was the one for the EU election in 2019.
The usage of our VAA’s is huge. In a randomized evaluation survey (Novus for SVT) after the 2018 election, four out of ten of all Swedish voters stated they had used the SVT VAA before casting their vote and almost 80% of those users said it provided them with good or very good guidance in making their choice. Hence the responsibility for us to do a good job with the Election Compass is very high.
The work methodology and quality in the development of VAA’s is therefore crucial, and the importance of transparency made SVT adopt the “The Lausanne Declaration on Voting Advice Applications” before the 2018 election. This declaration is the work of the ECPR Research Network and aims to “recommend certain standards and minimal requirements that should be respected by all the makers of VAA’s”. By adopting this declaration we now have a more standardized way of being transparent in our work.
The declaration states that every maker of VAA’s needs to be transparent regarding who’s behind the application and that it’s free of charge for all voters to use. It also points out the importance of explaining the construction and methodology so that everyone interested can scrutinize the work by looking into the details.
At SVT we have always been open with how we work with the VAA’s. In 2018 we published an extensive FAQ as well as a methodology description, but then we never released the matching algorithm. After the most recent election in Sweden – the EU Election in May 2019 – we decided to package the matching algorithm as open source (https://github.com/SVT/election-compass-match). Our algorithm has basically been the same since 2010, with minor improvements in 2014, 2018 and 2019. We have for example added different ways of answering the questions and fine-tuned the algorithm after comprehensive testing.
Then why are we doing this? As previously mentioned the transparency reason is the most obvious one, but since there are a lot of great minds working with code and VAA’s we also welcome error corrections and your suggestions on how we could improve. The best solutions are usually found when a lot of people work together.
Finally, we hope this example can work as a good way to improve both openness and the quality of all VAA’s in future elections. In Sweden the next general election will be held in 2022, and we hope all VAA makers by then will follow our example and release their matching algorithm as open source or at least be more transparent regarding how it works. Together we can make all the VAA’s better.
SVT (Eva Landahl, Kristofer Sjöholm, Rickard Andersson, Fredrik Stålnacke, Hans G Andersson)
P.S. There are many VAA’s in Europe that work very openly and our Public Service colleagues at NRK in Norway also open sourced their algorithm from their most recent election in 2019 (https://github.com/nrkno/valg-valgomat-algoritme).