Replication data for: Learning to do Better: Using Boosting Neural Networks and Leader Experience Measures to Improve the Accuracy of Conflict Prediction Models

It has become almost cliche to bemoan the sorry state of quantitative conflict research. This paper will be no exception. Indeed, most of the assumptions necessary in order to analyze conflict using statistical methods are rather egregiously violated by the data. The purpose of this paper is to propose and test (a) the use of boosted, single hidden layer, feed-forward perceptron neural networks to predict conflict likelihood,(b) the addition of an instrumentalization of state learning into the predictive model. I find that while standard neural network regression does offer an improvement over logistic regression methods, the advantage of neural networks can be significantly extended by using boosting methods to transform the “weak” neural net into a “strong” learner. The predictive advantage of boosted neural networks is further enhanced by the inclusion of measures of leader tenure as proxy measures of state learning.