One measure, one source release, one universe
A ranking starts with a defined observation: a household median, a share of a particular population, a resident count, or a change between two dates. We store the unit and period with the value. Every default table ranks the 50 states. DC and Puerto Rico are reference rows where source coverage permits. Other territories are omitted because the ingested files do not provide equivalent observations for them.
The DC toggle changes the eligible universe to 51 jurisdictions and recalculates ranks. Puerto Rico remains a reference. A state filter only hides rows; it does not recalculate their rank among the full universe. Sorting lowest first reverses the order while retaining the declared “rank 1 = highest” convention.
Precision and true ties
For descending rankings, rank equals 1 plus the number of eligible observations strictly greater than that value. Equal source values share a rank: 1, 2, 2, 4. Values are ranked before display rounding. The source files themselves have finite precision; we never invent hidden digits to break a source tie.
Our CSV exports retain the source’s available numeric precision, along with units and source identifiers. The displayed percentage of population growth is rounded for readability; the ranking uses the full calculated ratio.
Survey uncertainty
ACS comparison files publish 90% margins of error. For independent estimates from different states, the standard error of a difference is the square root of the sum of squared standard errors. At the same 90% confidence level, this simplifies to comparing the absolute difference against √(MOE₁² + MOE₂²). We flag a difference only when it exceeds that threshold.
The tests follow the Census statistical testing guidance. They are not adjusted for multiple comparisons, and are not simultaneous rank confidence intervals. Non-significance is not proof of equality. Pairwise non-significance is not transitive, so we avoid artificial disjoint “tie groups.” Published-value ties and statistical uncertainty are different concepts.
National benchmarks are shown for context. We do not use a naive independent-estimates test between a state and the U.S. total, because the state contributes to that total. Population Estimates Program files do not supply ACS-style margins of error; they are labeled “not published.”
Change and history
Population growth is 100 × (population in 2025 − population in 2024) ÷ population in 2024. Numeric change is the numerator alone. Both observations come from Vintage 2025. Population history on report cards uses annual July 1 estimates from that same vintage, which revises past years. It does not show the history of what earlier releases originally reported.
Selected ACS measures include 2021–2024 annual values, ranks and margins of error. Nominal dollar histories are not inflation-adjusted. We do not connect series across a known definition break or compare overlapping five-year periods. A missing baseline stays unavailable.
Custom scores
The custom tool uses min-max normalization among the 50 states. Higher-preferred values receive 100 × (value − minimum) ÷ (maximum − minimum); lower-preferred values receive its complement to 100. Slider weights are divided by their sum and multiplied by those normalized scores. Contributions sum to the final score. A constant measure receives 50 for all states.
Example weights are illustrative choices, not an objective “best states” model. Source uncertainty is not propagated to the composite. Correlated inputs can count similar patterns twice; pairs with absolute Pearson correlation above 0.8 are flagged. Published code: ranking and composite calculation.
Publication gate
Every included measure must have valid observations for all 50 states, a producer file, a period, units, and reviewed rights. Source bytes are archived with SHA-256 checksums. A reviewed build publishes pages, maps, downloads, report cards, and search together. No third-party proprietary ranking values are in this data store. See Sources for the exact release registry.