Mapping the Flemish election landscape : An alternative visualization
In october 2018, local elections were held in the Flemish municipalities. Usually, the results of elections are visualized on a geographical map. Inspired by an analysis of German election data by Moritz Stefaner and a similar project of Jan Willim Tulp (*) based on the parliament elections in the Netherlands, I was looking for an alternative way to visualize the results of the municipality elections in Flanders (14 october 2018).
The workflow used is as follows.
After collecting the election results for each municipality, a data normalization was performed in order to make elections results (available as percentages of votes for political parties) comparable between parties. We used the Z-score normalization that transforms each percentage of votes for a political party into standard deviations from the average. This allows to determine which municipalities voted more of less than on average for a political party.
After normalization, cosine similarities ware calculated between municipalities.
The matrix of cosine similarities is then used to perform a multidimensional reduction and visualization analysis. We used the tsne-package to reduce dimensionality of the election data to 2 dimensions. The visual representation of the data can be used to check for the presence of clusters. Municipalities which vote in a similar way are located nearby.
The plot of the election data clearly shows that municipalities where the winning party is a local list are quite well separated from all other municipalities. Only a small number of municipalities with a local list as winner are situated in the lower part of the plot and the percentage of votes that can be attributed to the winning local list is lower than is the case for municipalities in the upper left quadrant of the diagram.
Apart from local lists, municipalities with other winning political parties also cluster quite well. However, independent from geographical location, the picture is more scattered for these political parties (especially for CD&V).
Although geographically separated, the visualization of vote similarities allows us to see the flemish election landscape as grouped by lifestyles and preferences of municipalities.
(*) See M. Stefaner, Mapping electionland, http://well-formed-data.net/archives/955/mapping-electionland (september 23, 2013) and TULP interactive, Close votes. Which cities vote like yours ?, http://tulpinteractive.com/close-votes/ (20 july 2015).
