Logistic Regression for Extremely Rare Events: The Case of School Shootings