Mobilization and Violence in Syria: Comparing Local and International Text Data

• In Event: New Methods for Studying Political Violence

Abstract

Text-derived data is becoming increasingly important in political science as automated methods to analyze it improve. This paper provides two important contributions to the study of political violence through the use of text-derived data. First, it addresses an open question in the study of civilian violence by examining the effect of pre-war anti-regime mobilization on governments’ retaliatory violence during the war (Balcells 2017) by examining protests and indiscriminate regime violence in Syria. To do so, it draws on novel machine-coded data on the locations of protests in Syria in 2011, using new neural network techniques for classifying text with little training data and novel Arabic-language geolocation algorithms. We train this model on a new dataset we created from Arabic-language reports of protests in news text.

The paper's second contribution comes from generating this data from both English and Arabic-language sources. Many scholars using text data rely solely on English-language sources to understand regional questions. Significant differences in reporting across languages, however, may produce divergent data about political events. We estimate the effect of not using local languages through quantifying the difference in findings resulting from using international and local sources, with major implications for other scholars working with English language text.

Authors

• Andrew Halterman, Massachusetts Institute of Technology

• Jill A. Irvine, University of Oklahoma

• Khaled Jabr, University of Oklahoma