A Conditional Logistic Regression Predictive Model of World Conflict considering Neighboring Conflict and Environmental Security
Abstract
Forecasts of conflict are of utmost importance for assisting combatant commanders in developing strategic and operational campaign and country plans that consider the dynamic changes that evolve within their area of responsibility. This research formulates and constructs five suites of statistical models to better understand the collinearity of environmental factors affecting conflict and compares the classification accuracy between forcing these factors into logistic regression models. A total of thirty-nine predictor variables are tested and evaluated for inclusion in a six region, two conflict state combination suite. The five suites of twelve models calculate the probability of whether a country will transition to either an In Conflict or Not In Conflict state for the following year. Handpicking the best models proposed in this study from each suite achieves modeling classification accuracies of 92.0 with 82.6 prediction accuracies. Through exploring new variables and selection methods, the models demonstrate that leveraging the collinearity of environmental factors help provide strategic insight in developing Department of Defense Theater Campaign Plans to effect the stability of national security.