D'Orazio Vito, Yonamine James E
Institute for Quantitative Social Science, Harvard University, Cambridge, MA, USA.
RPX Corporation, San Francisco, CA, USA.
PLoS One. 2015 May 7;10(5):e0122472. doi: 10.1371/journal.pone.0122472. eCollection 2015.
While many studies have suggested or assumed that the periods preceding the onset of intra-state conflict are similar across time and space, few have empirically tested this proposition. Using the Integrated Crisis Early Warning System's domestic event data in Asia from 1998-2010, we subject this proposition to empirical analysis. We code the similarity of government-rebel interactions in sequences preceding the onset of intra-state conflict to those preceding further periods of peace using three different metrics: Euclidean, Levenshtein, and mutual information. These scores are then used as predictors in a bivariate logistic regression to forecast whether we are likely to observe conflict in neither, one, or both of the states. We find that our model accurately classifies cases where both sequences precede peace, but struggles to distinguish between cases in which one sequence escalates to conflict and where both sequences escalate to conflict. These findings empirically suggest that generalizable patterns exist between event sequences that precede peace.
虽然许多研究表明或假设国内冲突爆发前的时期在时间和空间上是相似的，但很少有研究对这一命题进行实证检验。利用1998 - 2010年亚洲综合危机早期预警系统的国内事件数据，我们对这一命题进行了实证分析。我们使用三种不同的度量标准（欧几里得距离、莱文斯坦距离和互信息），将国内冲突爆发前序列中政府与反叛组织互动的相似性与随后更长和平时期前的互动相似性进行编码。然后，这些得分被用作二元逻辑回归中的预测变量，以预测我们是否可能在其中一个或两个国家都观察到冲突。我们发现，我们的模型能够准确地对两个序列都先于和平的情况进行分类，但难以区分其中一个序列升级为冲突和两个序列都升级为冲突的情况。这些发现从实证上表明，和平之前的事件序列之间存在可推广的模式。
