Due to a unique data-related complexity, Israeli election results have never been mapped accurately beyond municipality-level. Data analysts have produced approximations and journalists have explored specific neighborhood-level stories, but only a combination of journalistic research into the election mechanisms and a painstaking analysis of the datasets – partly inaccessible to the publich – enabled the creation of a detailed, granular, and accurate results map. This project entails both the map, which was updated live as results were released, and a map-based story that introduced these results to international readers.
The main impact of the project was the proof that such an analysis is possible. Journalists have used the map for their reporting or reported on the map. In the long run, this project will hopefully set a new standard for election mapping in the Israeli media, where this practice is still immature. More importantly, repeating it in three election rounds in 2018-2019, during which the authorities recieved the same inquiries over and over again, could lead to the understanding that the necessary datasets should be released to the public.
The core of the project was data analysis using Python in Jupyter Notebooks. For the production of the resulting map I used Plotly’s Dash and deployed it as an independent Heroku site. For the story that presents the map and the political crisis that unfolded after the election I used Mapbox and its scrollytelling templates.
What was the hardest part of this project?
The hardest part was processing the data. In fact, the simple result – in many countries banal – of an election map by neighborhood, hides a complex system completely unknown to anybody outside a small circle of officials managing the elections (who, interestingly, did not think it was possible to conduct this analysis).
In this system, votes can be mapped to geographical units, polling stations, and polling places (which often include several stations). Results are provided by polling station. Shortly before election day, in cases where there were big differences between the number of votes allocated to different stations within the same polling place, all voters’ lists in that polling place were joined together and re-divided equally between the stations. (The purpose was to minimize queues.) This process is called “balancing.” Since different stations within the same polling place can serve different geographical areas, these must merge on the map – because there is no way to determine where the votes in the joined list stemmed from.
This complicated process of reversing the “balancing” must be conducted recursively, since balancing occured in many polling places and led to the merging of several areas that were themselves already a result of merging, and so on.
Perhaps the only thing that was harder was to grapple with the results themselves, which left the country in a hopeless political deadlock…
What can others learn from this project?
1. To continue when they hear “impossible.”
2. That a lot can be done by implementing projects that are common in one place and rare in another, like an election map.
3. That the strength of data journalism is precisely at the intersection between being a journalist who can get data and understand its meaning and being a data anlyst who’s able to process it.