The Interplay between Distribution Parameters and the Accuracy-Robustness Tradeoff in Classification