It would not be a random fair sample, as the sample categories each should be 30 minimum for such sample tests to be effective. There are 20 cities so the maximum any category group can be sampled is 3 per city and 2 from other cities as 50/2 = 25 = 2.5 each. Therefore by having data from other cities this creates 20 other categories as the samples are based on geographic wider scale such survey would require 30 from each city to show effectiveness and it would require reliable data of 30 from from each city instead of 2-3 each city.
