# Housing infrastructure and solar methods

These two HousingStabilityIQ analyses use the 2025 American Community Survey one-year release, published October 29, 2026. Estimates describe 2025, not conditions measured in October 2026. They are household-level descriptive comparisons, not causal models or national rankings.

## Geography and universes

The Raleigh-Cary, North Carolina metropolitan area uses full metro GEO_ID 310M700US39580. The component counties are Wake 0500000US37183, Johnston 0500000US37101 and Franklin 0500000US37069. We use each official geography's estimates rather than deriving metro margins of error by adding counties. Selected county estimate totals also reconcile exactly to the official metro totals. The metro is not Raleigh city, Cary town or the larger Research Triangle combined area.

The solar study is San Diego County, California, GEO_ID 0500000US06073, not San Diego city. All geography labels are verified in the retained official geography file. Component 00 is required.

B25149 and B25145 both cover occupied housing units. A household corresponds to an occupied housing unit, but estimated units are not sample sizes. Allocation tables B992528 and B992526 cover all housing units, including vacant units: their percentages must not be substituted for the response or allocation rates of occupied owner/renter subgroups.

## Formulas

Summary-file column order differs from the API variable naming order. For example B25149_E004 in the DAT file is B25149_004E in API metadata. The full names used are recorded in estimates.csv.

- Septic share of all occupied homes: (B25149_E004 + B25149_E008) / B25149_E001 × 100. These are owner and renter septic counts; the other-system category is not included.
- Metro owner and renter septic supplementary measures: E004/E002 and E008/E006, respectively.
- Solar share among owner-occupied homes: B25145_E003/B25145_E002 × 100.
- Solar share among renter-occupied homes: B25145_E012/B25145_E011 × 100.
- Solar share among owner-occupied one-unit homes: B25145_E004/(B25145_E004+B25145_E008) × 100.
- Solar share among renter-occupied one-unit homes: B25145_E013/(B25145_E013+B25145_E017) × 100.
- Supplemental solar share among renters in structures with two or more units: B25145_E014/(B25145_E014+B25145_E018) × 100.
- Allocation shares: E002/E001 × 100 in the relevant allocation table. Those tables publish estimates without margins of error.

One-unit housing includes attached and detached homes. This is a restricted-category descriptive comparison, not statistical adjustment for income, housing age, location, race, financing or other characteristics. It does not identify a causal effect of tenure.

## Uncertainty and annotations

All published margins of error are 90% margins. For a sum, M_N = sqrt(sum(M_i²)). For a subset proportion p=N/D, M_p in percentage points = 100×sqrt(M_N² − p² M_D²)/D. If the radical is negative, the script uses the Census ratio fallback with a plus sign. No selected metric required that fallback. Approximate differences use sqrt(M_A² + M_B²). Covariance is unavailable, so these are approximations that can overstate or understate uncertainty. See the [2025 Census worked examples](https://www2.census.gov/programs-surveys/acs/tech_docs/accuracy/2025_ACS_Accuracy_Document_Worked_Examples.pdf).

The reader-facing percentages and MOEs are rounded once from unrounded calculations. The tables preserve full precision and raw inputs. `cell-audit.csv` logs each selected estimate and MOE cell. All selected cells are finite nonnegative numbers; no negative jam value, blank, suppression marker or nonnumeric annotation is converted to zero. Table-based files provide E/M values, not separate data-query EA/MA annotation fields. No independent API annotation fetch is claimed. The chosen allocation cells are also valid nonnegative numeric values; no MOE is fabricated where none is published.

Checks verify 19 table-partition identities across selected geographies, the county-to-metro estimate sums, and valid selected cells. These checks validate arithmetic and table interpretation, not survey fieldwork or nonsampling error. Allocation reflects editing/imputation; its prevalence does not by itself quantify bias. MOEs do not capture every potential reporting, processing, coverage or nonresponse error.

## Selection and interpretation limits

The author explored counties and metros in the newly released tables, then selected these two geography/topic pairs for local reporting value. This was not a preregistered test. Multiple exploratory comparisons can generate false discoveries. The published results are descriptive baselines, not claims that these are the largest gaps nationwide; no simultaneous confidence guarantee, multiple-testing-adjusted discovery or causal conclusion is asserted. Large selected contrasts and their approximate intervals are available in validation.json for transparent inspection. Comparing a county with its containing metro as if independent is prohibited.

The sewer and solar questions are new to the ACS in 2025. The [2025 subject definitions](https://www2.census.gov/programs-surveys/acs/tech_docs/subject_definitions/2025_ACSSubjectDefinitions.pdf), printed pages42–43, explicitly say no prior ACS comparison exists. The solar measure concerns panels generating electricity for household use or sale and is distinct from solar as the principal heating fuel. It is not limited to rooftop systems. The sewer categories describe connection type, not tank condition, compliance, safety, repair cost or contamination. ACS occupied-unit counts are not system-permit inventories. Neither story joins parcel, utility account, risk, program-beneficiary or income microdata.

## Reproduce

Keep this folder and its raw subfolder together. Run `python reproduce.py` to regenerate the estimates, selected-cell audit, allocation table and validation report using Python's standard library. Add `--charts` with matplotlib installed to regenerate the SVG/PNG charts. No API key, authentication, network request or credential is required for this offline calculation. `source-manifest.json` identifies retained raw files, URLs and SHA-256 hashes. No send or publication action is performed.
