Census data methodology and statistical limits
How source artifacts become versioned observations, calculations, charts and reusable evidence.
From original files to one observation
The launch collection uses original U.S. Census bulk files. Selected ACS geographic rows and original PEP CSVs are retained with source URLs, acquisition dates and SHA256 checksums. The 2024 ACS geographic directory and original table shells supply identifiers, labels and universes.
Acquisition is separate from templates. Validation checks exact identifiers, schema cells, duplicate entities, units, nonnegative usable values, valid percentage denominators and source-defined geographic relationships. PEP county/state/national control totals reconcile for all six annual years, and every component of population change reconciles with the annual estimates. Normalized observations feed the page, chart, table and export, avoiding separately maintained figures.
Different periods remain different
ACS five-year estimates describe 2020–2024, not a point-in-time observation for 2024. Publication was January 29, 2026. PEP population history uses annual July 1 observations from Vintage 2025, with national/state publication January 27 and county publication March 26, 2026. Source files were acquired September 30, 2026; the current snapshot was published October 7, 2026 from those same files.
Exact source IDs permit a geographic join but do not make product definitions, boundary vintages or universes identical. ACS records retain 2024 geographic definitions; PEP records retain Vintage 2025. Territory, tribal, city, metro and institution comparisons have not been integrated.
Direct and calculated measures
Counts, medians and means from source tables remain direct observations. Calculated percentages preserve their numerator, denominator and source cells. Rent burden uses B25070_007E through B25070_010E divided by B25070_001E minus B25070_011E. The threshold is at least 30%; the not-computed group is excluded.
Medians cannot be added or averaged to make a regional median. Nonadditive rates and adjusted measures require approved methods. This release offers no custom aggregation, unsupported subgroups or ACS historical trends. State ranking pages order point estimates and show the range of positions that pairwise significance tests cannot rule out. Annual population history and its components of change use one consistent PEP vintage.
Uncertainty, significance and reliability
ACS figures are survey estimates. Direct estimates (counts, medians and means) carry the Census Bureau’s published margin of error (MOE) at the 90% confidence level. Every percentage calculated by this site carries a derived 90% MOE computed from the published MOEs of its component cells with the Census Bureau’s documented approximations: the MOE of a sum is the square root of the summed squared component MOEs (keeping only the largest MOE among zero-estimate cells), and the MOE of a proportion p = N/D is √(MOEN² − p²·MOED²)/D. Where that quantity is zero or undefined, the more conservative ratio formula √(MOEN² + p²·MOED²)/D is used. Exports record which formula applied.
These approximations ignore correlation between cells, so they can differ slightly from MOEs in Census subject tables built with replicate weights. A statistically controlled source MOE is treated as zero; any other unavailable component MOE leaves the derived MOE unavailable rather than invented.
Two estimates are described as statistically different only when the difference exceeds √(MOE1² + MOE2²), a 90% test. Profile summaries, comparison tables and article statements use this rule; otherwise they say no significant difference was found. Comparing a county with its own state or with the nation, or a state with the nation, ignores the overlap between the two, which makes the test conservative. PEP population estimates have no sampling MOE, so they are compared descriptively.
A coefficient of variation (CV) is the standard error (MOE ÷ 1.645) divided by the estimate. Estimates with a CV above 30%, or a zero estimate with sampling error, are marked “use with caution.” They are still shown, because hiding them would bias comparisons, but summaries do not draw conclusions from them.
Positions among states or counties count down from the highest point estimate. Because neighbors are often statistically indistinguishable, profiles and rankings also show a supported range: from 1 plus the number of areas significantly higher to the total minus the number significantly lower. The tests are pairwise and not adjusted for multiple comparisons, so the range is a guide to precision, not a guaranteed interval.
Missing values
Source negative sentinels remain visible in the raw source cells and are excluded from arithmetic. Missing, inapplicable and source-unavailable observations remain separate from a measured zero. Percentages with no valid denominator remain unavailable.
Capability and privacy rules
Each indicator specifies supported comparisons, trend suitability, subgroups and aggregation. Only annual PEP history and its components of change support time-series tables and charts in this release. PLACES records are catalog-only and explicitly disallow local trends, overall-health rankings and intervention-effect claims. Education cohorts, fiscal accounting and administrative universes require independent source rules.
Public outputs are aggregate. No suppressed cells are recovered by subtraction or cross-dataset joins. County observations do not establish household or individual circumstances, immigration status, personal risk or the effectiveness of a policy.
Frozen release and update review
The current release is apd-2026-10-07-r5; earlier releases apd-2026-10-01-r3 and apd-2026-09-30-r1 remain available at their original paths. Exports include the source version, transformation and MOE method. Future updates require a new immutable directory, schema/rights review, validation and review of material methodological changes. Source failure leaves the approved release available; it does not authorize substituting unreviewed numbers. The statistical review record documents the current methods and checks.
The publisher is responsible for the planned monthly source and link checks, with annual ACS/PEP release review. No unattended update scheduler is active. Original producer files remain linked; source discovery records are not presented as completed connectors. See the update log and frozen downloads.