How we work with data

Source-led publishing

Imports preserve source files, metadata and checksums. Data move through imported, reviewed and published states. Public pages read a fixed approved snapshot; a failed download cannot replace it. The initial review combines automated checks and source comparison by an AI development assistant. It is not independent expert review.

ACS proportions and margins of error

For current marital status, we sum the male and female cells and divide by the total population aged 15 and over. For household shares, we divide the published couple-household count by all households. We never divide people by two.

Let N and D be the numerator and denominator, p = N/D, and M be a 90% margin of error. For sums of disjoint cells, we approximate the summed MOE by the square root of the sum of squared MOEs. For a subset proportion, the MOE is 100 × √(Mₙ² − p²Mᴅ²) / D. If the radicand is negative, we use the Census ratio approximation with a plus sign. The resulting uncertainty is approximate and includes no invented covariance.

Census guidance for data users · Comparing ACS estimates.

Comparison rules

Measures must share their unit, universe, denominator, survey product and reviewed comparability group. We do not pool CDC crude rates, ONS married-population rates or Eurostat indicators. The original library uses ACS one-year estimates. The research section also includes a separately labelled Data Commons five-year cohort. It does not use overlapping five-year estimates as annual observations or include experimental 2020 one-year estimates.

Change estimates

The original report subtracts the 2021 estimate from the 2024 estimate. Its approximate 90% MOE is √(M₂₀₂₁² + M₂₀₂₄²), treating the separate-year estimates as independent. It is a descriptive comparison, with no adjustment for multiple comparisons and no causal claim.

Reporting and methodological breaks

CDC national reporting areas change across years. Every record retains its coverage and exclusions. Eurostat b and p flags identify breaks and provisional values. ONS notes preserve legal changes and court disruption. A line is not evidence of an unchanged measurement process.

Revisions and rights

Dataset releases are immutable. Corrections create a new release and a public log entry. Attribution and source restrictions follow the data into exports. No source logo or implied endorsement is used.