Methodology

One headline population, several official counts

A state has several official population figures, and they differ for good reasons. We use the Census Bureau's latest population estimates (Vintage 2025, for July 1, 2025) as the headline figure on every page. The estimates start from the 2020 Census and add births, subtract deaths and add net migration each year, using vital records, tax and Medicare data and other administrative sources.

Each state page also shows the other counts in a small table: the 2020 Census count (April 1, 2020), the estimates base for the same date (the census count adjusted for later corrections and boundary updates), and the American Community Survey totals. The ACS is a survey weighted to the population estimates, so it is the right source for characteristics such as income or education, not for the population total itself.

Which source for which measure

• Population, age, sex, race and Hispanic origin totals: Census Bureau population estimates, Vintage 2025, including the June 2026 release by age, sex, race and Hispanic origin.
• Characteristics (income, poverty, education, housing, language, birthplace, veterans, health insurance, commuting and more): American Community Survey 2025 one-year estimates, released October 29, 2026. Every state has more than 65,000 residents, so one-year estimates are available for all; we use them for state values and trends.
• Small groups: for detailed Asian and Pacific Islander groups, ancestries and languages, we use the one-year estimate where the Census Bureau publishes it with a coefficient of variation of 30% or less, and otherwise the ACS 2020–2024 five-year estimate, labelled with that period. Rankings of states on these groups use the five-year estimates for every state, so all states share one period. American Indian and Alaska Native tribes and groups come from the 2020 Census Detailed DHC-A, the most detailed published count of tribal groups by state.
• Programs: SNAP participation from the USDA Food and Nutrition Service, Medicaid and CHIP enrollment from CMS, homelessness from HUD's point-in-time count and tax returns by income from the IRS. These administrative counts differ from what people report in the ACS, and we label each.
• Births, deaths, fertility and life expectancy: CDC National Center for Health Statistics.
• Income, output and prices: Bureau of Economic Analysis (personal income, GDP, regional price parities).
• Unemployment: Bureau of Labor Statistics, Local Area Unemployment Statistics.
• Migration: Census Bureau components of change, ACS state-to-state flows and IRS migration data, as explained on the migration page.
• Historical population: decennial census counts from 1790, and annual intercensal estimates from 2000.
• Religion: the U.S. Religion Census: Religious Congregations and Membership Study, 2020, distributed by the Association of Religion Data Archives (ARDA). It is the only religion source we publish; see religion below.

When two sources measure the same thing, such as unemployment from the ACS and from the BLS, or veterans from the ACS and from the Department of Veterans Affairs, we name the source each time and explain on the topic pages which to use for what.

Margins of error

ACS figures are survey estimates. Each comes with a margin of error (MOE) at the 90% confidence level, published by the Census Bureau, and we show it next to the value, for example $82,625 (±$670¹). For values we derive, such as a share computed from two published counts, we calculate the margin of error with the approximation formulas in the Census Bureau's ACS handbook. Population estimates, census counts and vital statistics are not samples and have no margin of error.

When a difference is significant

We describe a difference between two survey estimates, between states or between years, as real only if it passes the Census Bureau's test at 90% confidence: the difference divided by the standard error of the difference must exceed 1.645, where the standard error is the square root of the sum of the squared standard errors (each MOE divided by 1.645). Otherwise the text says the values are "not statistically different". Comparisons between years use non-overlapping one-year estimates only.

Ranks, ties and rank ranges

Rank 1 is always the highest value. States with the same value at the displayed precision share a rank, and they share the better position counted from either end: two states tied at the lowest value are both "tied for lowest". For survey measures, we also show a rank range: the best rank a state could hold is one plus the number of states significantly above it, and the worst is the number of states minus those significantly below it. When the range is wide, the exact rank means little; the text then gives the range instead.

Dollars and inflation

Income and housing values are in the dollars of the survey year: ACS 2025 figures are in 2025 dollars. Trends over time are converted to 2025 dollars with the Bureau of Labor Statistics' CPI-U annual averages, and the chart says so. Comparisons with the previous year convert the earlier figure to 2025 dollars with the same CPI-U ratio, the adjustment the Census Bureau describes for comparing ACS dollar values across years.

Trends and gaps

ACS one-year trends start in 2005. The Census Bureau did not release standard one-year estimates for 2020 because of pandemic data-collection problems, so trend lines show a gap there. Population trends use census counts back to 1790 and annual July 1 estimates from 2000.

Religion

Federal statistics do not cover religion: the census does not ask about it. Our religion figures come from one non-federal source, the U.S. Religion Census: Religious Congregations and Membership Study, 2020, designed and carried out by the Association of Statisticians of American Religious Bodies (ASARB) and distributed by the ARDA. We use the ARDA's state file as corrected in September 2026, when the ARDA fixed how missing values were handled in its religious tradition variables. Following the ARDA's terms, every page that uses these figures prints the study's full citation and acknowledges the ARDA and the original collectors.

• What is counted. Religious bodies report, county by county, their congregations and their adherents: full members, their children and others who regularly take part, as each body defines them. It is a count kept by congregations, not a survey of individuals, so it does not measure belief, practice or self-described identity. 216 bodies reported adherents; 156 more reported only congregations, which appear in congregation counts but not in any adherent figure.
• Estimates inside the count. Some bodies did not take part, and several historically African American denominations are under-represented by the study's own account. Catholic totals were estimated diocese by diocese; Muslim figures come from a count of mosques and a survey of mosque leaders; Jewish figures count people affiliated with synagogues; Hindu and Buddhist figures come from counts of temples and centers. Synagogue and mosque figures are not estimates of the Jewish or Muslim population. A person can be counted by more than one body, and people are counted where their congregation meets.
• Shares. We divide adherents by the 2020 Census population of each state and add the states for the United States, which gives 48.6%² nationally. The ARDA file's rate variables are labelled "per 1,000 population" but hold percentages, which differ slightly from adherents divided by population; we do not publish them, and our data tests check that our shares agree with them as percentages.
• Traditions and groups. The six religious traditions (Evangelical Protestant, Mainline Protestant, Black Protestant, Catholic, Orthodox and other) are the ARDA's classification. "Christian" is our sum of every body except the Jewish, Muslim, Hindu, Buddhist and Baha'i groups; it follows the study in counting Latter-day Saints, Jehovah's Witnesses and Unitarian Universalists among the Christian bodies.
• Ranks and change. The counts have no margin of error, so ranks have no range, but they are not precise to the last decimal. Comparisons with 2010 use the ARDA's 2010 state file and the 2010 Census population; because the 2020 study counted more bodies, part of every change reflects coverage.
• Not used. We publish no survey estimates of religion. The Pew Research Center's Religious Landscape Study publishes survey-based state estimates; its public data file has no state identifiers, and we do not copy its published figures, so readers who want survey estimates should consult it directly.

Maps and state outlines

Maps shade the states in five groups of about ten states each, so every shade holds the same number of states. The legend gives the range of each group, and every map has the exact values under "Show the numbers". The District of Columbia is drawn as a dot because its area is too small to see on a national map. A state with no published value for a measure is still drawn, hatched, and named under the map.

The shapes come from the Census Bureau's cartographic boundary file for states (2024 edition, 1:20 million scale), listed on the sources page. They are drawn at build time in an Albers equal-area projection, with Alaska and Hawaii moved below the other states and Alaska shown at a smaller scale, the usual layout for national maps. The outlines are simplified for the web: small islands and fine coastal detail are left out, but neighboring states still share the same border line. On a state's own page the outline leaves out Alaska's Aleutian Islands west of the Alaska Peninsula, so the state is drawn at a readable size.

The colour behind the opening of a state page is the state's Census region: clay for the South, sage for the West, wheat for the Midwest and slate for the Northeast. On topic, ranking and article pages the same four tones only tell the topics apart; they do not refer to a region.

How the pages are written

Every figure in the text is inserted from our data files by an identifier and footnoted automatically, so no number is typed by hand. Statements such as "the highest in the nation" or "not statistically different" are generated from the data with the rules above. Explanations of why a number is what it is are given only when a cited source supports them.

Updates and corrections

We update figures within days of each release on the release calendar. New population estimate vintages revise all years back to 2020, so earlier figures can change. Corrections are listed with their date on the same page.