How to Read the ACS5 Margin of Error (and When to Ignore the Number)

The American Community Survey is a sample, not a census of every household. Every published estimate is paired with a margin of error (MOE) at the 90% confidence level. For populous places — entire states, large counties, big cities — MOEs are tiny relative to the estimates and can be ignored. For small places they can swallow the number.

A practical rule of thumb

Compute the coefficient of variation: MOE divided by 1.645, then divided by the estimate. If the result is under 12%, the estimate is reliable. Between 12% and 40%, treat it as approximate. Above 40%, treat it as suggestive at best.

On City Zip Compare we surface the underlying ACS5 number on every page, but for ZIPs with very small populations (under ~500) you should expect noisier numbers, especially for income and housing-value medians.

Why ZCTAs are not exactly ZIP codes

The Census does not publish data by USPS ZIP code; it publishes by ZIP Code Tabulation Area (ZCTA), which is a Census-defined geographic approximation of a USPS ZIP. About 95% of residential ZIPs map cleanly to a ZCTA. P.O. Box-only ZIPs and unique-recipient ZIPs (large companies with their own code) usually do not.

When City Zip Compare cannot find a ZCTA for a ZIP query, that's why.

A worked example

Suppose a small ZCTA reports a median household income of $52,000 with a published margin of error of $18,000. The coefficient of variation is $18,000 ÷ 1.645 ÷ $52,000, or roughly 21% — solidly in the 'treat as approximate' range. That doesn't mean the number is wrong; it means the true value could plausibly fall anywhere from roughly $41,000 to $63,000, and small year-over-year swings within that range shouldn't be read as a meaningful local trend.

Compare that to a large city reporting the same $52,000 median with a $1,500 margin of error — a coefficient of variation under 2%, comfortably reliable. Same headline number, very different confidence behind it.

When margin of error matters most in practice

This isn't an abstract statistics lesson — it has real implications for how you use place data. If you're comparing two small towns and one shows a $3,000 higher median income than the other, check whether that gap is larger than the combined margins of error before treating it as a meaningful difference. For large cities and states, gaps of a few thousand dollars are almost always statistically real. For small ZIPs, they often aren't.