Read the population before the percentage.

Every comparison keeps the marital-status definition, age universe, sex, geography, survey product and reference period together.

Five categories, without double counting

ACS B12001 describes residents aged 15 and over. We use never married; married except separated; separated; widowed; and divorced. The married subtotal in the original table includes separated people. We remove that overlap for this five-category presentation by combining “spouse present” and “other spouse absent” cells.

The existing general marital-status article uses a now-married subtotal that includes separated people. Its label differs intentionally. Do not compare that subtotal with “married, except separated” here.

Counts, denominators and uncertainty

A share is 100 × the status count ÷ the matching population aged 15+. Male and female breakdowns use their own population denominators. Estimates are people, never fabricated couples. Original counts, numerator cells, denominator cells and 90% margins of error remain in the downloadable records.

For summed Census cells, we combine margins of error by the square root of their squared sum. For a subset share p = numerator ÷ denominator, its margin is 100 × √(MOE(numerator)² − p² × MOE(denominator)²) ÷ denominator. If the radicand is negative, we use the Census ratio approximation with addition. These are approximate survey-error calculations, not a proof that nearby estimates differ.

The selected Data Commons facet does not return margins of error. They remain unavailable; the interface does not invent them or display a confidence ranking. Cross-checking the 2020–2024 counts against original Census tables does not make the absent uncertainty available in the Data Commons series.

One year and five years answer different questions

The original Census cohort uses one-year 2021–2024 estimates. The Data Commons cohort uses five-year estimates labelled by the entire period. A date of 2024 in that API means the 2020–2024 estimate in this selected facet. Adjacent five-year periods overlap by four years; they are not independent annual observations.

We do not merge products or silently switch facets for a missing value. The comparison tool offers common periods within one product. It does not compare experimental one-year 2020 estimates. Earlier and later Census methods and geographic boundaries can differ; our descriptive history is not a causal time-series model.

Compare areas carefully

These are resident characteristics, not the location where weddings or divorces occurred. National, state, county and city boundaries represent different populations. Los Angeles city is inside Los Angeles County; New York County is only part of New York City. Do not sum overlapping places. Differences may reflect age composition, migration and many other factors.

What marital status does not measure

Unmarried, never married, unpartnered and seeking a partner are different. Current divorced share is neither an annual divorce rate nor lifetime divorce incidence. Dividing annual divorces by annual marriages combines different cohorts and cannot estimate the probability that a marriage ends in divorce. We do not infer first-marriage timing from current married age distributions.

Release and correction process

Imports are cached and staged. Exact source hashes, explicit facets, five-category partitions and original Census checks are validated before an approved immutable snapshot is published. A failed request cannot replace the published snapshot. Review dates describe this publication’s checks; they do not change survey dates.

Read the coverage and rights audit · Inspect the published release · Census comparison guidance