Individual Submission Summary
Share...

Direct link:

A Supervised Learning Approach to Counting Conflict Events: Evidence from Egypt

Sat, October 2, 4:00 to 5:30pm PDT (4:00 to 5:30pm PDT), TBA

Abstract

Although conflict events such as protests, riots and armed clashes have long been an important subject of study in political science, research in this field remains plagued by measurement issues. Most existing datasets are hand coded from English-language media sources, however, there is evidence to suggest that these sources underreport conflict events in non-English speaking countries. An alternative has been to hand code from local media sources, a resource- and time-intensive endeavor. As a result, the study of conflict events has often relied on either poor-quality data or temporally and spatially circumscribed data. This paper presents an alternative approach to measuring conflict events in non-English speaking countries, one which captures locally reported events in an efficient manner. To demonstrate this approach, we build an original dataset of protest events in Egypt. We use hand-coded data from an independent Egyptian newspaper, Al-Masry Al-Youm, from 2010 to 2011 as a training dataset. We then use supervised modelling on hundreds of thousands of Al-Masry Al-Youm articles to predict all protest events in Egypt between 2005 and 2013. This approach has the potential to considerably improve the measurement of conflict events in non-English speaking countries. We compare this dataset to commonly used existing datasets to explore biases in their coverage and consider the substantive implications for the study of protest.

Authors