# Retirement population analysis methods

## Scope and vintage

Five distinct descriptive analyses use the 2025 ACS 1-year detailed tables, retrieved November 2, 2026. These are state-level current-resident/household estimates and one national employment comparison. No rankings, causal inference, annual trend, investment recommendation, personalized risk score, city extrapolation or current-year projection is produced. The 2025 ACS 1-year release was October 29, 2026. Primary raw tables and machine-readable API group shells are frozen in raw/. The group shells give exact labels and cell IDs; Geos20251YR.txt gives geographic labels. Source-manifest.json contains URLs, retrieval times, SHA256 and byte sizes. Neither scripts nor published numeric outputs require a private contact file or network access.

The Census October 2026 collection note says a six-week funding-lapse suspension affected August–November panels, annual sample size, response rates and margins of error. Preserve this source caveat. No ad hoc correction or assertion of unaffected survey quality is made.

## Reproduce

Run `python reproduce.py` from this folder (Python3 with matplotlib) to regenerate public/metrics.json, public/claim-ledger.json, public/claims.csv and five charts/*.png. No source acquisition is needed. `acquire.py` records the original table acquisition pattern; it is optional and a later run can fetch revised sources, so do not use it for frozen replay. SHA256 checks should run first via `python verify_sources.py`. Charts are original matplotlib plots, not imagegen.

## Cell definitions and uncertainty

The claim ledger records each numerator and denominator, table, geography, cell list, estimate and MOE. `B23001` male+female employed cells are combined separately for 65–69, 70–74, 75+. Their denominators are the matching total age-band cells. These age groups do not establish retirement status, working hours, occupation or motivation.

`B09020` cells15+18 count people 65+ living alone. All people 65+ denominator is cell1. `B11007` cell3 is the same estimated living-alone count; cell2 counts households with at least one person 65+. Direct published versus aggregated MOEs can differ even where point estimates coincide. Group quarters are not equated with nursing homes, living alone with loneliness, or co-residence with care.

`B07001` sums65–69, 70–74, 75+ cells separately for each prior-residence category. The movers denominator sums the four moving categories, excluding same home. It is a current-Florida-resident view, not out-migration or net migration; the survey compares home now and one year earlier, not every intervening move. Interstate origin is not retirement motivation.

`B10056` cells5+10 count grandparents 60+ responsible for their co-resident grandchildren; cells3+8 count responsible grandparents across ages 30+. The universe is co-resident grandparents, not all grandparents. The questionnaire asks about responsibility for most basic needs; not legal custody, occasional childcare, unique households, children, retirees or institutional service caseloads. No statistically secure older-majority claim is made. Approximate complement MOEs are not necessarily equal due to published component covariance limitations.

For counts aggregated across disjoint cells, MOE is the square root of the sum of squared component MOEs. For a subset proportion p=N/D, approximate MOE in percentage points is100*sqrt(MN²−p²*MD²)/D; if the radicand is negative, use100*sqrt(MN²+p²*MD²)/D. These are standard ACS approximation formulas, not replicate-weight estimates. Derived uncertainty does not capture nonsampling error or all covariance. Published and derived MOEs use 90% confidence. No multiple-comparison rankings or exhaustive statistical-significance screen is conducted. Chosen states are purposive reporting examples, not claimed extremes. Point-estimate statements are qualified where differences are uncertain.

Colorado uses separate household receipt marginal tables. The intersection point-estimate lower bound is max(0,B19051_E002−B19055_E003)=187067, equivalent to earnings+Social Security−total. Upper bound is min(earnings,Social Security)=670082. This algebraic range is NOT a confidence interval or directly measured overlap. A sensitivity lowers earnings by its published MOE and raises no-Social Security by its published MOE: 187067−14832−13271=158964. These marginal adjustments are NOT a simultaneous confidence statement. Label `summed_input_moes90` is descriptive only. Social Security includes Railroad Retirement and survivor/disability benefits; earnings may belong to another household member and need not arrive simultaneously in the past 12-month reference period. No individual working-retiree count or income sufficiency inference follows.

## Selection and disclosure

These mechanisms were selected for interpretive utility: age composition, a cross-universe denominator bridge, movement origin, caregiving responsibility, and an algebraic overlap bound. State examples were selected purposively with local reporting fit; they are not an exhaustive search for extremes. Earlier network articles about mortgage-free housing, hazard counts, heating/facilities/internet and location risk scores address different questions. No exclusive discovery claim is made. Private newsroom novelty/contact research is separate and not a dependency of the public numerical analysis.

AI assisted code, drafting and independent-agent checks. Independent AI review is not human editorial approval. Raw data, formulas, uncertainty and source-year limits must remain with this publication.

## Primary technical references

- [2025 ACS subject definitions](https://www2.census.gov/programs-surveys/acs/tech_docs/subject_definitions/2025_ACSSubjectDefinitions.pdf), frozen raw PDF. Employment definitions at printed pages 76–77; income and grandparent definitions are also included.
- [2025 ACS worked accuracy examples](https://www2.census.gov/programs-surveys/acs/tech_docs/accuracy/2025_ACS_Accuracy_Document_Worked_Examples.pdf), frozen raw PDF, aggregation and proportion approximations.
- [2025 ACS accuracy of data](https://www2.census.gov/programs-surveys/acs/tech_docs/accuracy/ACS_Accuracy_of_Data_2025.pdf), frozen raw PDF, sampling/non-sampling errors.
- [2025 questionnaire](https://www2.census.gov/programs-surveys/acs/methodology/questionnaires/2025/quest25.pdf), frozen raw PDF, responsibility and income questions.
- [2025 collection interruption note](https://www.census.gov/programs-surveys/acs/technical-documentation/user-notes/2026-04.html), frozen official excerpt, revised October 28, 2026.
