Why is the mean often not a good measure of central tendency for a skewed distribution?