# U.S. atlas methodology — version 2.0, September 17, 2026

Coverage: all 3,144 current counties and county equivalents, 50 states and DC. Territories excluded. Source periods remain 2020–2024 and 2015–2019 ACS five-year estimates. Independently retrieved U.S./state values match constituent state/county estimate sums. Map: TIGERweb ACS 2024 layer 82, simplified to 0.025 degrees.

Comparison availability: Connecticut's nine planning regions and Alaska's Chugach/Copper River equivalents have no reported county historical change. Other comparisons match identifiers; minor boundary changes are not harmonized. National correction uses 3,133 comparable county share-change tests at 10% BH FDR. State and U.S. comparisons have separate approximate unadjusted 90% margins. The archive includes full analysis/variable labels, source hashes and reconciliation checks.

Stable study URL: https://familyestateguide.com/estate-planning-data/older-adults-living-alone/
The notes below describe the original two-state calculation definitions. Their formulas and statistical limitations still apply; their old coverage and test-family references are superseded by the nationwide scope above.

# Older Adults Living Alone: California and Florida County Atlas

Edition: September 2026 local preview. Coverage: all 58 California counties and all 67 Florida counties. No national inference is made from this two-state pilot.

## Sources and periods

U.S. Census Bureau, American Community Survey five-year detailed tables:

- [B11010, 2024 release](https://data.census.gov/table/ACSDT5Y2024.B11010): nonfamily households by sex of householder, living alone, and age.
- [B01001, 2024 release](https://data.census.gov/table/ACSDT5Y2024.B01001): sex by age.
- Matching 2019 releases of both tables.
- [Census comparison guidance](https://www.census.gov/programs-surveys/acs/guidance/comparing-acs-data/2024.html).
- [Census worked examples for margins](https://www2.census.gov/programs-surveys/acs/tech_docs/accuracy/2023_ACS_Accuracy_Document_Worked_Examples.pdf).
- [TIGERweb ACS 2024 county boundaries](https://tigerweb.geo.census.gov/arcgis/rest/services/TIGERweb/tigerWMS_ACS2024/MapServer/82), queried for states 06 and 12. Display geometry simplified at 0.015 degrees, coordinates rounded to four decimal places. Maps are for statistical display, not property boundaries.

The 2024 five-year release describes **2020–2024**; the 2019 release describes **2015–2019**. These are non-overlapping survey periods. They are not single-year counts or current 2026 headcounts. In the retrieved data, the selected estimate labels match across periods, and county FIPS sets match. Matching identifiers alone does not establish identical historical polygon boundaries. No historical boundary overlay is presented; the contemporary display map is joined by county identifier.

## Measures

Estimated age-65+ residents living alone:

`N = B11010_005E + B11010_012E`

Those cells represent male and female householders age 65+ living alone. Each one-person household corresponds to one person. People in group quarters are excluded from the numerator.

All age-65+ residents:

`D = sum(B01001_020E ... B01001_025E) + sum(B01001_044E ... B01001_049E)`

These male and female age categories are 65–66, 67–69, 70–74, 75–79, 80–84, and 85+. The denominator includes people in group quarters. That choice answers what share of *all* older county residents live alone, rather than what share of older household residents live alone. It may be affected by changes in institutional residence.

`Share (%) = 100 * N / D`

State figures use independently retrieved state table rows, not averages of county percentages. County means would give small counties the same weight as large counties and do not reproduce statewide shares.

`Change (percentage points) = current share - earlier share`

`Count growth (%) = 100 * (current count / earlier count - 1)`

Count growth is descriptive and is not given a county significance label. Share-change labels follow the procedure below.

## Approximate uncertainty

Source margins are at 90% confidence. For sums, approximate margin is the root sum of squares of the component margins:

`MOE_sum = sqrt(sum(component_MOE^2))`

For a subset proportion, with `p = N/D`:

`MOE_share_pp = 100 * sqrt(MOE_N^2 - p^2 * MOE_D^2) / D`

If the expression under the square root is negative, use the Census ratio fallback with a plus sign:

`MOE_share_pp = 100 * sqrt(MOE_N^2 + p^2 * MOE_D^2) / D`

Across non-overlapping periods:

`MOE_change_pp = sqrt(MOE_share_current_pp^2 + MOE_share_earlier_pp^2)`

Approximate standard error is the 90% margin divided by 1.645. Normal two-sided p-values are computed for county share changes and adjusted across **all 125 county tests** using Benjamini–Hochberg at a **10% false discovery rate**. This is an exploratory multiple-comparison procedure, not a probability that an individual labeled change is true. Approximation and test-dependence limitations remain; the procedure does not promise a realized error rate for this particular dataset.

States are assessed separately by whether the absolute share change exceeds its approximate 90% change margin. Those two state labels are unadjusted. County labels use the FDR criterion. The exported data retains both unadjusted and adjusted results.

We do not use variance replicate estimates, so covariance is approximated rather than calculated. We do not test all pairs of county shares or claim statistically distinct county ranks. Chart and sort order reflect point estimates only. The displayed count coefficient of variation is `MOE_N / (1.645 * N)`; a value of at least 0.30, or undefined precision, is flagged as wide count uncertainty. This is our reporting flag, not an official Census certification. Unflagged estimates still have uncertainty.

Missing values and negative Census sentinel codes are unavailable, never silently converted to zero. Missing input makes the corresponding derived value unavailable. Zero denominators do not produce a percentage. The current pilot has valid nonnegative inputs for its rendered measures.

## Download dictionary

The CSV has one county per row; the JSON includes statewide summaries and exact source-variable labels.

| Field | Definition |
|---|---|
| `geoid` | Five-character county FIPS identifier; preserve as text to retain California's leading zero |
| `county`, `state` | County and state names |
| `2019_...`, `2024_...` | Fields belonging to that five-year release, not a single-year headcount |
| `period` | Complete survey period |
| `living_alone_65_plus` | Estimated number of older people in one-person households |
| `living_alone_moe90` | Approximate 90% count margin |
| `population_65_plus` | Estimated older-resident denominator |
| `population_65_plus_moe90` | Approximate denominator margin |
| `total_population` | B01001 total-resident estimate, for aggregation checks |
| `share_pct` | Older residents living alone as a percent of all older residents |
| `share_moe90_pp` | Approximate 90% share margin, in percentage points |
| `count_cv` | Approximate count standard error divided by count |
| `wide_count_uncertainty` | Reporting flag described above |
| `share_change_pp` | Current minus earlier share, in percentage points |
| `share_change_moe90_pp` | Approximate 90% change margin |
| `share_change_p_value` | Unadjusted normal two-sided p-value |
| `share_change_detected90` | Absolute difference exceeds the unadjusted 90% margin |
| `share_change_q_value` | Benjamini–Hochberg adjusted p-value |
| `share_change_supported_fdr10` | Adjusted value at most 0.10 |

Numeric exports are rounded to six decimal places for measures and eight for test statistics; full computed precision is retained in data.json. Filtered downloads are a compact subset of these fields. Use the full CSV/JSON for research.

## Validation and maintenance

Raw response bodies and table metadata are cached under `raw/`. No credential-bearing request URL is stored. `validation.json` includes their SHA-256 hashes, label checks, county counts, map-join checks, and county-sum comparisons with independently retrieved state values. The build rejects missing counties, mismatched geography sets, or large county/state discrepancies. The current build's county/state differences are recorded rather than assumed to be zero.

The reproducible build reads CENSUS_API_KEY from the workspace .env. API access is server-side during generation; the atlas contains only public data. It works as a standalone HTML page without remote scripts or a live data API. Update it on relevant source releases, retain prior editions, and review release notes before changing periods or variables.

Our original charts and analysis are offered under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/), with attribution to Family Estate Guide and the U.S. Census Bureau as the original statistical source. The publication URL is not yet assigned. No Census endorsement is implied.
