POLLING

Since January 2014, Fabio Wasserfallen and I have been running surveys in Switzerland for various newspapers, companies, foundations, and academic projects. We work on projects that are interesting to us and are related to the res publica — we are only interested in carrying out surveys that are eventually published and available to the broader public. We combine survey statistics with modern data science tools and leverage vast amounts of data to produce high quality results.

We run two kinds of surveys. With our media partners we survey their readers and rely on a number of algorithms to process the raw data before we employ sophisticated models to create our estimates. Part of this has also been published in the leading academic journals (Paper 1, Paper 2). Since we started in 2014 we have been at least as good as polls based on traditional techniques. Whether looking at the last poll or a linear extrapolation, we are closer to the actual outcome in more than 50% of the cases.

We have gathered one of the largest online panels in the country and field surveys over this panel for various projects. In the past, we have worked with industry partners on relevant topics such as digitization or new technologies such as 5G. We have also collaborated with academic partners from various institutions and civil society organizations.

Before each national ballot vote we also publish a forecast and on the voting day itself we provide projections that often deliver precise results about 5 min after the first data is made available. Both of these projects are carried together with Tamedia and rely heavily on machine learning. For more information visit leewas.ch.