I think his point was that it's more likely taking the total income and total property costs and computing the average that way. But if you instead compute the same per-city and then look at the middle of those it can give you a drastically different value.