If your data have noises/randomness, it most likely does, and you use your statistic to reduce the variance of the noises then it would be sharper would it not?