Mega-protests Meet ‘Big Data’: A Behavioral Approach to Protest Participation

• In Event: Big Data and Machine Learning

Abstract

The unexpected outburst around 2011 of massive protests in diverse contexts poses both theoretical and methodological challenges. Movements like the Arab Spring and the Spanish and Greek Indignados were exceptional in their scale, and correspondingly in the way they crossed traditional social boundaries. Research on the encompassingness of mega-protests and the diversity of their participants is underdeveloped and constrained by the types of data currently available. We introduce a new type of behavioral data that offers unique opportunities for studying large-scale contentious political behavior, and use Israel’s 2011 “social justice” movement for illustration.
The standard data sources for studying protest participation, namely retrospective self-reports or on-site interviews, suffer from limitations when addressing the diversity of participants due to their limited reliability, coverage and representativeness. In this paper big data on geographic positioning and population movements is exploited for the first time to study participation in street demonstrations. Information automatically recorded in real time on the location of anonymous cellular phone users provided unique and reliable behavioral data for 38 large-scale protest events that took place in 18 different locations. The data are aggregated by participants’ residential localities, at the level of census tracts.
The paper details the properties and advantages of this complex dataset and the processing procedures required to ready it for analysis. We then illustrate some novel applications to the study of protest participation. A major strength of this data set is its high resolution in terms of the timing and place of participation, which allows investigating the temporal trends and spatial patterns of participation in a protest campaign. Adding background information on the political and socio-demographic characteristics of protesters’ residential localities enables analysis of their social composition, making it possible to identify differences between distinct social sectors in when, where and with whom they joined demonstrations. The data also make it possible to examine the political geography of protest participation, including analysis of ‘political commuting’ and the effects of the distance from the location of the demonstrations on the level of participation. Finally, localities’ participation rates can serve as an independent variable that helps explain other dependent variables. This can be applied at the aggregate level, as in the analysis of the impact of protest engagement on voter turnout and changes in the distribution of votes in the subsequent elections. Participation rates can also serve as contextual factor for explaining individual attitudes (e.g. support for the protest) or actions (e.g. protest participation, or voting behavior).
While emphasizing the benefits to researchers of this type of big data, we also point to difficulties and limitations. The aggregated nature of the dataset secures the privacy of individual users, but at the same time limits the ability to characterize participants by properties like age and gender. Similarly, collecting behavioral data avoids reliance on self-reports, but at the same time limits the research to focus solely on behavior without any information on the personal motivations and interpretations of the protesters. Further, since these data are limited to only one form of political participation, different sources are required to make linkages to other types of political behavior. More broadly, the newness of the data raises a number of issues, including our inability as yet to make comparisons to similar research elsewhere, the absence of standardized procedures ensuring comparability and replicability, and the lack of clear ethical standards (a problem shared by all users of big data). That said, we stress the substantial advantages of our data and encourage other researchers to explore the potential for obtaining similar data in their own countries.

Authors

• Michael Shalev, University of California

• Assaf Rotman, Tel Aviv University