Predicting Armed Conflict, 2010–2050

The article predicts changes in global and regional incidences of armed conflict for the 2010–2050 period. The predictions are based on a dynamic multinomial logit model estimation on a 1970–2009 cross-sectional data set of changes between no armed conflict, minor conflict, and major conflict. Core exogenous predictors are population size, infant mortality rates, demographic composition, education levels, oil dependence, ethnic cleavages, and neighborhood characteristics. Predictions are obtained through simulating the behavior of the conflict variable implied by the estimates from this model. We use projections for the 2011–2050 period for the predictors from the UN World Population Prospects and the International Institute for Applied Systems Analysis. We treat conflicts, recent conflict history, and neighboring conflicts as endogenous variables. Out-of-sample validation of predictions for 2007–2009 (based on estimates for the 1970–2000 period) indicates that the model predicts well, with an area under the receiver operator curve of 0.937. Using a p > .30 threshold for positive prediction, the true positive rate 7–9 years into the future is 0.79 and the False Positive Rate 0.085. We predict a continued decline in the proportion of the world's countries that have internal armed conflict, from about 15% in 2009 to 7% in 2050. The decline is particularly strong in the Western Asia and North Africa region and less clear in Africa south of Sahara. The remaining conflict countries will increasingly be concentrated in East, Central, and Southern Africa and in East and South Asia.本文针对2010至2050年全球及各地区武装冲突发生率的变化展开预测。本次预测基于1970—2009年的截面数据集，通过动态多项Logit模型（dynamic multinomial logit model）对无武装冲突、轻度冲突与重度冲突之间的状态转换进行估计。核心外生预测变量包括人口规模、婴儿死亡率、人口结构、受教育水平、石油依赖度、族群分化以及邻域特征。预测过程通过模拟该模型估计结果所隐含的冲突变量变化行为完成。我们采用联合国世界人口展望（UN World Population Prospects）与国际应用系统分析研究所（International Institute for Applied Systems Analysis）发布的2011—2050年预测数据作为预测因子。本文将冲突状况、近期冲突历史以及邻国冲突视为内生变量。针对2007—2009年的样本外预测验证（基于1970—2000年的模型估计结果）显示，模型预测效果优异，受试者工作特征曲线（Receiver Operating Characteristic curve）下面积达0.937。当以p>0.30作为正向预测阈值时，未来7至9年的真阳性率为0.79，假阳性率为0.085。我们预测全球发生内部武装冲突的国家占比将持续下降，从2009年的约15%降至2050年的7%。该下降趋势在西亚与北非地区尤为显著，而在撒哈拉以南非洲地区则相对不明确。剩余存在冲突的国家将日益集中于东非、中非、南部非洲以及东亚与南亚地区。