Elepost – a crowdsourced analysis of electoral posters

Even in our digital times, printed campaign posters are still one of the most important means of electoral propaganda. For the Bavarian regional elections that took place this fall, BR Data started a crowdsourced investigation on which election posters the parties had plastered on the streets of Bavaria, and where.

The elections for the regional parliament of Bavaria are one of the most important political events for our mothership Bayerischer Rundfunk, the Bavarian public broadcaster. On 14th of October, the Bavarian electorate not only decided on their future parliament, their vote was also considered a benchmark for Angela Merkel’s grand coalition in Berlin. Of course, BR Data couldn’t go without a data story covering the election campaign.

The idea: Born in Buenos Aires

About three months before election day, the first election posters popped up in the center of Munich and I remembered my two-month visit to the Argentinean newsroom La Nación data in 2017. La Nación’s data journalists had experimented many times with crowdsourced investigations, a method we at BR Data hadn’t used before. In 2014, La Nación collected pictures and GPS coordinates of giant election posters in the Buenos Aires region in a story about campaign spending. The crowdsourcing part played only a minor part in that story, but the idea for our project was born. If we collected enough pictures, as well as location information about election posters in Bavaria, we could investigate where the parties and candidates tried to get which kinds of messages across. Was it possible to microtarget audiences on the streets?

Codename: Elepost 🐘

The team consisted of our newest team member Maxi Richt who had just joined BR Data as a developer, along with Marco Lehner, an aspiring journalism/coding student who did an internship with us over the summer. Our plan was to build a prototype and collect around 100 pictures and coordinates of posters as quickly as possible. Then we wanted to evaluate whether the idea was worth pushing further. Codename: Elepost. Election Posters.

Testing the Elepost WhatsApp bot with some pictures of our favorite Thai food: Up and running!

We knew, when we wanted the project to succeed, we needed to make submitting photos and coordinates as easy as possible for our users. Maxi hacked together a quick WhatsApp integration for Airtable, a cloud based database service with spreadsheet view for easy editing and a powerful API. After two days of coding – and adding a profile picture of a cute little elephant – the first Elepost prototype was up and running on Maxi’s laptop. The photo safari began! In just one week, BR Data team members, friends and family submitted around 200 pictures from all over Munich. We challenged our team to submit more and more pictures to rise on the leaderboard. Shout-out to Steffen, who even changed his bicycle commute to get more photos.

Internal ranking after day one of BR Data’s election poster photo safari.

In the meantime, we had contacted all political parties that took part in the elections and asked for a list of their campaign posters and designs. Only one party sent us pictures of their posters, the others were not able or willing to provide us with a complete list of their poster designs, let alone locations. Although we didn’t find any direct hints for microtargeting, we knew we had a unique opportunity to get an almost complete overview of all the different election posters the parties put out on the street. So we pushed forward.

Technical overview: Infrastructure and analytics

Maxi expanded our technical infrastructure by building a Telegram bot and a web upload form capable of automatically extracting GPS coordinates from the pictures’ EXIF data. Pictures and coordinates were managed in Airtable. We pseudonymized the phone numbers and other user data in order to comply with the famous European data protection regulation. In order to analyze the images on a thematic level, we used Google’s cloud vision API for optical character recognition. Marco refined his R skills by creating an analytics dashboard for the data so we could oversee the number of submitted pictures, regional distribution and most frequent topics. I was very busy managing the project, promoting the story via radio and a dedicated website and getting BR’s regional correspondents involved.

One photo instance of the famous DIE PARTEI headscarf poster, manually categorized as poster design no. 271.

When we went live, the pictures started rolling in. A big chunk of my time went into manually categorizing thousands of pictures. In the future, this might be a task worth automating. Data-wise, we had one table for all the pictures and their GPS coordinates, and another one for all the individual poster designs, which we manually matched and linked in Airtable.

Results: 3000 pictures and some surprises

Within about three weeks, our users submitted more than 3000 pictures from all regions of Bavaria. We identified more than 1000 unique poster designs in the data.

Final map view of submitted campaign posters in Munich’s city center.

One of the most surprising insights was the fact that most parties in some way or another used the topic of “Heimat”, or “homeland” in their poster campaigns, enticing patriotic sentiments in their voters. Even traditionally left-wing parties like ÖDP or even the Green party used “Heimat” vibes in their posters. We also found some regional differences: some parties advertised for bigger investments in broadband infrastructure in rural areas, while focusing on rents and housing politics in the bigger cities.


Maximilian Zierer

Maximilian Zierer is a data and investigative reporter for BR Data, the data journalism unit of Bayerischer Rundfunk. He studied journalism at Deutsche Journalistenschule, loves radio reporting and still hasn’t updated to Python 3.
Biggest career achievement: the successful implementation of the Döner Kebab emoji into Unicode 9.

Runs on:

How many times per week do you have to explain what "data journalism" is?
"So you’re working in data protection, right?"

How many items are on your desktop?

Swear words per day?
At BR Data, we never use bad words.

How big was your biggest data set?
For our investigation on the Madeira tax haven, we used some 10.000 pages of pdf documents.

Your funniest file name?
doener-emoji-unicode-anschreiben-100 (2).pdf

snow flake
© 2018 Journocode