Characterizing the Robustness of Science