The German Protest Registrations Dataset covers protests that have been registered with demonstration authorities in 16 German cities. The data has been compiled from Freedom of Information requests and covers dates, organizers, topics, the number of registered participants, and for some cities the number of observed participants. Covered date ranges vary, with all cities covered in 2022, and 5 cities covered consistently from 2018 to 2022. In comparison to previous datasets that are largely based on newspaper reports, this dataset gives an unprecented level of detail, and is the largest dataset on protest events in Germany to date. This report gives an overview over existing datasets, explains the data retrieval and processing, displays the properties of the dataset, and discusses its limitations.

Data Download

PDF Report

This dataset leverages Freedom of Information laws to collect official protest data from 17 different cities in Germany.

Demonstration authorities. The organizers of demonstrations are legally required in Germany (as well as in many other countries) to notify so-called demonstration authorities before the demonstration takes place (Art. 8 Abs. 2 GG, § 14 VersammlG). Demonstration authorities (Versammlungsbehörden) in Germany are either part of the police or of the municipal administrations, depending on the region. The authorities keep records of all registered demonstrations. Some protests are not registered in advance, especially more disruptive and illegal protest forms, and they are therefore not contained in these records.

Freedom of information laws. Official documents, including protest records, can be obtained via Freedom of Information laws. These exist in more than 100 countries and allow anyone to obtain public documents. The specific requirements, exceptions, and costs vary greatly. In Germany, freedom of information exists on the federal level; but many authorities belong to the regional level, where the extent of freedom of information rights varies greatly; and municipal authorities are not always covered by regional freedom of information laws, sometimes filling the gap with their own legislation.

Freedom of information platforms. Access to public documents has been democratized via platforms that streamline the process of sending requests, escalating the process to oversight authorities or courts if necessary, and making communication and obtained documents available to the public. The Alaveteli network provides software and hosts such platforms in more than 30 countries across the world. Some independent platforms also exist, such as Öffentlichkeitsgestz.ch in Switzerland, and FragDenStaat in Austria and in Germany. These open the possibility of obtaining official protest data at scale.

Multiple approaches have already been used for the creation of datasets on protests in Germany. gives an overview about the existing and new datasets.

Event databases provide disaggregated event data, often on a global scale and with constant updates. The database of choice for protest research is the Armed Conflict Location and Event Dataset [ACLED; Raleigh et al. (2010)]. ACLED is a grand effort that keeps track not only of violent conflicts and riots, but also of ordinary protest events. The data is human-curated based on newspaper reports, and contains coded information on dates, locations, actor groups, police interventions, and more, as well as a short free-text summary for each event, containing an estimate of the size as per the newspaper data source. Data for Germany is available starting from 2020 and is continuously updated. Another relevant event database is the Global Database of Events, Language, and Tone [GDELT; Leetaru and Schrodt (n.d.)]. It relies completely on the automated processing of newspaper articles, and does not have protest-specific information such as organizers, topics, or number of participants.

Academic datasets have been created through the manual coding of newspaper articles and other sources. Protestlandschaft Deutschland provides interactive visualizations for three German datasets:

1. ProDat (Rucht, Hocke, and Ohlemacher 1992) has coded all newspaper articles on Mondays – as well as all days of every 4th week – from 1950 to 2002 in Süddeutsche Zeitung and Frankfurter Rundschau.
2. “Lokale Protestentwicklungen Großstädte” has coded all articles in Sächsische Zeitung (Dresden), Leipziger Volkszeitung, Stuttgarter Zeitung, and Weser-Kurier (Bremen) from 2011 to 2020.
3. “Lokale Protestentwicklungen Mittelstädte” has coded all articles in 12 medium-sized German cities.

Protest events in Germany are also part of international protest datasets: PolDem (Kriesi et al. 2020) provides 3 datasets with different countries and time ranges covered, based on newspaper articles obtained from NexisLexis. And the Mass Mobilization Project (Clark and Regan 2022) records and codes protest events globally from 2014 to 2019.

Official records have been explored only on a small scale: Kanol and Knoesel (2021) compiles government responses to opposition requests about right-wing extremist demonstrations in Germany, starting in 2005. And Demo-Hauptstadt Berlin has used Freedom of Information requests to obtain data on all demonstrations in Berlin from 2018 to 2022.

Retrieval. I send freedom of information requests to German demonstration authorities (depending on the region these are either part of the municipal administrations or of the police) and their supervisory bodies concerning protest data in 32 cities. These cities comprise the political capitals of all 16 regions in Germany, the 17 largest cities by population size, as well as some smaller cities for regions where the request in the regional capital is unsuccessful. 4 requests are not answered, 2 are rejected, 5 state that they do not possess such data, 2 have to be withdrawn due to demanded payments of multiple hundreds of euros, and 19 are partially or completely successful.¹ The requests and responses including the original data files can be found at FragDenStaat.

Processing. I convert all files to CSV using the pandas library (team 2020), Adobe PDF Converter, and PdfTables.com. Since the column names and formats vary widely, I create a separate table reader script for each city. For the challenge of I use the dateparser library (parsing about 90% of dates), hand-written regex rules based on the most common patterns in the data (parsing about 9% of the data), and manual coding for the remaining data points. Multi-day events are reduced to their start date. When a single data point describes multiple events, it is reduced to the first day of the first event. Data points without date information are dropped. Participant numbers are also parsed using a combination of regex rules and manual coding; specifiers such as “max.”, “min.”, “bis zu” are ignored, and for number ranges the rounded average is applied. The dataset for Augsburg cannot be converted from PDF to CSV, the dataset for Mannheim is poorly structured, and the dataset for Dortmund does not have topic information, so I exclude them.

The resulting dataset contains 57.010 events from 16 cities. For 13 cities the ex-ante number of expected participants are given, and for 2 of them (Berlin and Magdeburg) the ex-post extimates by the police are also included. For all cities the topic of the protest is given, as specified by the organizers themselves; and for 4 cities (Erfurt, München, Wiesbaden, Wuppertal) the name of the organizing group is also known.

Further statistics about the dataset can be seen in table Table 1. Figure 1 shows the regional coverage of the dataset.

As apparent from Table 1, the covered years vary between cities. Using the whole dataset unfiltered would therefore lead to inconsistencies in developments over time. Therefore the dataset is provided in three subsets (see also the bottom of ):

1. The 2018-2022 dataset contains only cities that have coverage throughout 2018-2022, and that also have the number of registered participants available. This results in a large dataset (about 40.000 entries) that strikes a balance between regional diversity (5 cities) and time coverage (4 years).
2. The 2022 dataset contains all cities that have data available in 2022, and also have the number of registered participants available. With 13 cities covered but only 12.500 overall entries, it is useful especially for the study of geographic variations.
3. The unfiltered dataset contains all data without restrictions on included cities and time ranges, overall 57.000 entries. It also contains cities where the number of registered participants is not available. This dataset should not be used for direct analysis; but it may be used for the creation of alternative consistent sub-datasets similar to the two ones above.

The main benefits of authority data are:

1. Level of detail: Many more events (per timeframe, and overall) are included than in any existing dataset based on newspaper analyses.
2. Avoidance of selection bias of the media. This is feared to be an immense problem associated with newspaper-based data (Hutter 2014; James Ozden and Sam Glover 2022). This dataset may also help examining the actual amount of media selection bias.

The two main downsides of using data from demonstration authorities are:

1. Unreliability: Since the data is only about registrations, we do not know whether the events have actually taken place, and how many people have actually participated. (Only for some cities – Berlin and Magdeburg – we can know it via the numbers of observerd participants, and for other cities – München – via indications of cancelled events.)
2. Incompleteness:
   • Not all protests are registered. Some disruptive groups such as Letzte Generation deliberately do not register many of their protest actions.
   • Only a handful of municipalities are included, focusing on large and politically relevant cities and ignoring the places of residence of more than 85% of the population.

Criticisms of authority data. Protest research has identified two main sources of protest event data – newspaper data and data from authorities – and overwhelmingly relies on the first source, that is newspaper data. Objections are raised against the use of data from demonstration authorities, as being biased, uninformative about the motives and organizers, uncomparable across regions, often unavailable or unobtainable, and because it is restricted to only registered demonstrations Wiedemann et al. (2022).

I agree with these objections only in part:

• With Freedom of Information laws and platforms such as FragDenStaat it is now possible to obtain this data easily, cheaply, and at scale – however with the exception of regions with weak Freedom of Information laws or poor data management (see Figure 1).
• In 17 out of 19 cities the data contains information about the topics of the demonstrations, as specified by the organizers; this may well give a better perspective on the organizers and their motives than subsequent newspaper coverage may give.
• The data from demonstration authorities avoids both media bias in judging the goals of the organizers, as well as media bias in the selection of events worth reporting, and is thus arguably less biased than media-derived data. Only the numbers of participants are judged by the police and may be subject to bias, just as the estimates by journalists may also be subject to bias; researchers will be aware of topics where this may be especially problematic, such as anti-police protests.
• Regarding the issue of comparability, the current datasets are specifially designed to be comparable across regions, by focusing on key information that exists in all regions in similar formats.

The German Protest Registrations dataset presents opportunities for more detailed and less biased protest event analysis. Researchers should be aware of reliability problems with this type of data, and of the exclusion of non-registered protests. While the unfiltered dataset comes with inconsistent data over time, two consistent subsets are provided and researchers are encouraged to create their own consistent subsets. The dataset may drive insights internal to protest research – such as the relation between registration and observation data –, as well as to applications such as protest impact evaluation.

Please contact me via email to [email protected], or using the discussion and issue features on Github.