View article

Which variables are most effective in predicting terrorist attacks, and how well do our statistical models forecast transnational terrorism? We seek to address these questions by analyzing data on transnational terrorist incidents from 1968 to 2007. We rely on receiver operating curves (ROC) as a diagnostic tool to assess forecasting ability of various models of terrorist activity. Our analyses yield four central conclusions. First, our model of transnational terrorism provides a fairly strong basis for forecasting attacks–at least at the (relatively broad) level of the country-year. Second, while the overall forecasting capacity of this model is fairly strong, the model is also highly redundant in a predictive sense. That is, many of the variables appear to provide similar information in terms of identifying terrorist attacks. Third, four sets of variables stand out as providing the greatest leverage in predicting future terrorism: distance, wealth, democracy, and a history of terrorist violence. Collectively, these variables perform about as well as a more broadly specified model in forecasting terrorist attacks out of sample. Finally, these results suggest that policy makers would do best to focus on wealth and democracy when thinking broadly about how to reduce the incidence of transnational terrorist attacks.