Margin of Error vs. Geography in the ACS

You may want to mention that the main two determinants of the MoE are the sample size, as the square root of the n (lower for a tract than for a county, in most situations), roughly (there are technicalities), and the mean income itself. I think of income as approximately constant coefficient of variation variable. Coefficient of variation = standard deviation in the population / mean. For a ZIP code with mean income of $15K, you can expect the standard deviation to be may be $20K, and for a ZIP code with mean income of $100K, the standard deviation could be $130K. When you divide these standard deviations by the sqrt(n) for the same n, you get bigger MoEs in the second case.

Stas,

Thank you so much for this excellent comment. You clearly know a lot more about this than me. If you have time, can I ask you to also leave this comment on the blog post itself? I think that most of my readers would be interested in reading this as well.

Ari