When is a difference between states real? Reading margins of error
Washington has the highest estimated household broadband subscription rate, 93.6%. Once sampling error is taken into account, though, Washington, Colorado, California and Utah cannot be told apart: any of them could have the highest true rate. This article explains how American Public Data decides when a difference between two survey estimates is real.
Editorial date:
· First published October 7, 2026 Data snapshot apd-2026-10-07-r5, reviewed October 7, 2026
· American Public Data
Every survey estimate is a range
The American Community Survey interviews a sample of households, not every household. Each estimate therefore comes with a margin of error. At the 90% confidence level used by the Census Bureau, intervals built as the estimate plus or minus its margin would contain the true value in about 90% of repeated samples. A margin of ±0.2 percentage points means the data are consistent with a true value a little above or below the published figure.
The test behind “higher” and “lower”
Two estimates are treated as different when the gap between them is larger than the square root of the sum of their squared margins of error. Washington’s rate is 93.6% ±0.1 percentage points and Colorado’s is 93.5% ±0.2 percentage points. The gap between them is not statistically significant at the 90% confidence level. By contrast, Washington and Mississippi (84.1%) differ by far more than their combined margins, so that difference is clearly statistically significant at the 90% confidence level.
Overlapping intervals are not the test
A common shortcut is to check whether two intervals overlap. It is too strict: two intervals can overlap a little while the difference between the estimates is still statistically significant, because the margin of a difference is smaller than the sum of the two margins. The test above uses the correct combined margin. When a profile on this site says one place is “higher,” that statement has passed it.
Explore the figures behind this comparison
Change the measure or switch a percentage to its source count
. The table and download update together. The United States is a separate
benchmark; selected states are examples, rather than a complete ranking.
Households reporting a broadband internet subscription of any source-defined type. Reference period: 2020–2024. Includes cellular data plans. Household subscriptions differ from provider-reported availability, speed, and performance.
Households with broadband subscriptions·
2020–2024· U.S. Census
Bureau
Source counts and periods come from snapshot apd-2026-10-07-r5. Values show 90%
margins of error (published for medians, derived for percentages; pp means
percentage points); population estimates have no sampling margin of error.
Medians have margins too
Median household income is published with its own margin of error. Massachusetts, Maryland and New Jersey have the highest medians, but their estimates are within each other’s margins, so the income ranking shows each with a range of possible positions rather than a single place. Percentages calculated by this site, such as broadband subscriptions, use margins derived from the Census Bureau’s approximation formulas.
Small places need more caution
County estimates rest on smaller samples. Where an estimate’s coefficient of variation exceeds 30%, profiles label it “use with caution” and summaries do not draw conclusions from it. Across all county percentages on the site, 10% carry that label. Those estimates are still published, because hiding them would bias comparisons, but they deserve a wide berth.
What statistical significance does not mean
A statistically significant difference is one unlikely to be explained by sampling alone. It is not necessarily large or important, and it says nothing about cause. The tests on this site are pairwise: when many places are compared at once, some differences will pass by chance, so ranking pages show a range of positions instead of a single rank. The full method is on the methodology page.
Source tables and calculation record
Original producer: U.S. Census Bureau. American Community Survey, 2024 five-year detailed tables, describing 2020–2024.
Website data snapshot: apd-2026-10-07-r5. The article arithmetic uses retained
observations and documented indicator definitions; displayed percentages
are rounded, while exports preserve their underlying precision.
Census comparison guidance
explains why source product and sampling uncertainty belong in a survey
comparison.
Population Estimates Program documentation
explains the estimate framework and vintage revisions. The article
examples do not imply household-level links between separate state
totals.
Colorado leads the state work-from-home point estimates. Explore the worker universe, five-year period and why usual commuting does not measure hybrid schedules.
Adults 19 to 64 are uninsured far more often than children. Compare 2020–2024 ACS coverage by age for every state, with margins of error and source counts.
Florida grew while deaths outnumbered births; California shrank despite more births than deaths. See how natural change and migration shaped every state.