ZIP code economic data: sources and error margins
ZIP code economic data comes from four or five federal sources that each measure something different and each lag by a different amount. Put them on one page without labels and you get confident-looking numbers that disagree with each other. Read them with their vintage and margin of error attached and they become a useful picture of an area's economy, for territory planning, market sizing, or context on where a business operates.
Key takeaways
- A Census "ZIP" figure is really a ZCTA figure, an approximation of the USPS delivery area.
- Each source has its own vintage: ACS 5-year estimates span five years, IRS ZIP data trails by about three tax years.
- Always read an ACS median with its margin of error; treat high-error estimates as rough.
- Area statistics describe places, never people, and must not drive decisions about any individual.
The four sources behind most ZIP-level numbers
| Source | What it measures | Vintage on a typical page today | Watch out for |
|---|---|---|---|
| Census American Community Survey, 5-year | Income, poverty, education, housing, rent | 2020 to 2024 pooled | Margins of error; small areas are noisy |
| IRS Statistics of Income, ZIP data | Adjusted gross income, high-income returns | Tax year 2022 | Counts tax returns, not households |
| FHFA House Price Index | Repeat-sales home price change | Through 2025 | Developmental series; not every ZIP has its own index |
| Census ZIP Business Patterns | Establishments, employees, payroll, sector mix | 2023 | Employer establishments only |
The ACS number is not a 2024 snapshot. A 5-year estimate pools responses collected over all five years, which buys sample size for small areas at the cost of timeliness. When you compare it to a 2025 home price change, you are comparing different years on purpose, and you should say so.
ZCTA is not ZIP
USPS ZIP codes are mail routes, not areas, and they change. The Census Bureau publishes data for ZIP Code Tabulation Areas, built from census blocks to approximate those routes. Most of the time the match is close. It breaks down for PO-box-only ZIPs and single-building ZIPs, which have no ZCTA at all, and along ZIP boundaries that cut through blocks. CoverFi's public ZIP Code profiles are indexed only where a ZIP has its own ZCTA estimate, a population of at least 1,000, and a published median household income, which is about 22,900 ZIPs.
Read every median with its margin of error
The ACS publishes a 90 percent margin of error with every estimate. A quick reliability check is the coefficient of variation:
standard error = margin of error / 1.645
CV = standard error / estimate
CV above 30% → treat the estimate as low reliability
Worked example from a live profile: ZIP Code 33130 in Miami reports a median household income of $61,009 with a margin of error of plus or minus $7,485. The standard error is about $4,550, so the CV is about 7.5 percent. That is solid. A small rural ZIP with a $45,000 median and a $20,000 margin of error has a CV near 27 percent; its true median could plausibly be anywhere from $25,000 to $65,000, and ranking it against neighbors means little. CoverFi stamps estimates above the 30 percent line as low reliability.
Percentiles: compared with whom?
A percentile is only as meaningful as its comparison group. CoverFi's ZIP profiles rank each measure within the county, the state, and the United States, and show how many ZIP codes sit in each group. For ZIP Code 33130, the county comparison covers 78 ZIPs, the state comparison 912, and the national comparison about 22,900. A ZIP can look middling nationally and unusual locally, or the reverse, because the groups differ. County and state percentiles are left blank when a group has fewer than five ZIPs.
Two rules keep percentiles honest. A higher percentile means a larger value, not a better or worse area; a high poverty rate and a high income can both rank high. And percentiles describe where a ZIP sits among peers on one measure, which says nothing about the people who live there.
Home prices: three tiers of fallback
Home-price change comes from the FHFA index, which is not available for every ZIP. A profile falls back through a chain: first the ZIP's own index; then an index built from the census tracts inside it; and finally the county index, which is labelled by county name and never ranked against other ZIPs. The Miami example above reads off the county index. FHFA describes its series as developmental and not seasonally adjusted, so use it to compare direction and magnitude, not to price a house. Note also that median home value comes from the ACS and is a resident-reported estimate, while the price index tracks repeat sales; the two can disagree.
Business mix: reading the sector table
The business section draws on ZIP Business Patterns, which counts employer establishments, employees, and payroll by sector. A ZIP profile lists the top sectors with each sector's share and its relative concentration against the national share, so a sector that is more common here than in the country overall stands out. The same Miami ZIP shows about 47.7 establishments per 1,000 residents, at the 76th percentile within its county. Employer establishments exclude most sole proprietors and the self-employed, so the figure undercounts one-person businesses.
One number is never the picture
The same Miami ZIP shows why. Its median household income sits at the 27th percentile of Florida ZIPs, while its per-capita income sits at the 80th and its average adjusted gross income per IRS return at the 73rd. Its Gini index, a measure of income inequality, is 0.576, at the 96th percentile in the state. A median describes the middle household; a per-capita figure or an average gets pulled up by high earners. In a ZIP with wide income spread, those measures diverge sharply, and picking one tells half the story.
A short checklist for any ZIP you are sizing up:
- Pair a median with a mean (median household income with average AGI per return).
- Pair a level with a change (median home value with the 5-year price index change).
- Check the margin of error on every ACS median before comparing ZIPs.
- Compare within the county as well as the state, since state percentiles mix very different labor markets.
- Note which year each number describes.
Comparing two ZIPs in practice
When you set one ZIP against another, work through the same steps each time:
- Confirm both have their own ZCTA estimate; if one is a PO box or building ZIP, the comparison is invalid.
- Note the vintage of each measure and make sure you are comparing like with like.
- Read the margin of error and compute the coefficient of variation for each median.
- Rank within the same county first, then the state, and only then nationally.
- Look at a companion measure for every headline figure.
- Write down what the comparison cannot tell you, such as anything about a specific household.
A comparison that survives all six steps is worth putting in front of a client. One that fails at step 1 or 3 was never a comparison.
What area data must not be used for
ZIP statistics describe places. They say nothing reliable about any single household in the ZIP, and using them to decide whether a person gets credit, insurance, a job, or housing is off the table: CoverFi is not a consumer reporting agency, and its area pages say so on every page. The FCRA vs non-FCRA guide explains why that line sits where it does. For the business side of an area, such as which companies are registered there and how fast new ones are forming, see the business entity search guide.
Frequently asked questions
Why do two websites show different median incomes for the same ZIP?
They usually use different ACS vintages, or one uses a ZCTA estimate and the other an approximation built from census tracts. Check the year range and the geography.
How often does ZIP-level economic data update?
ACS, IRS, FHFA, and ZIP Business Patterns each release once a year, on different schedules, so a ZIP page mixes vintages by design.
What is a good margin of error for ACS data?
There is no single cutoff, but a coefficient of variation under 15 percent is generally reliable and one above 30 percent should be treated as rough.