Bounding distributional errors via density ratios