Statistics for Analyzing Adversarial Examples: A Framework to Study Adversary Knowledge