[Retric]

My point is it's a high variance statistic which is hard to control for.

If population A = 95%, Population B = 30%, and Population C = 1% then it becomes vary important to get the correct ratio of populations A, B, and C. And correctly sampling culturally diverse populations in the same geographic area is vary difficult. Something as simple as a phone screen is unlikely to work due to things like % of population in prison or number of phone numbers in a household. Even block by block analysis can fail with transient populations as it's easy to introduce bias by counting someone twice or not at all.