Relative error vs. sample size