Cluster-Based Early Warning Indicators for Political Change in the Contemporary Levant

Abstract

We use cluster analysis to develop a model of political change in the Levant as reflected in the World Event Interaction Survey coded event data generated from Reuters between 1979 and 1998. A new statistical algorithm that uses the correlation between dyadic behaviors at two times identifies clusters of political activity. The transition to a new cluster occurs when a point is closer in distance to subsequent points than to preceding ones. These clusters begin to “stretch” before breaking apart, which serves as an early warning indicator. The clusters correspond well with phases of political behavior identified a priori. A Monte Carlo analysis shows that the clustering and early warning measures are not random; they perform very differently in simulated data sets with similar statistical characteristics. Our study demonstrates that the statistical analysis of newswire reports can yield systematic early warning indicators, and it provides empirical support for the theoretical concept of distinct behavioral phases in political activity.