Methodology

The composite is a 0–100 benchmark where higher means a stronger overall fire posture relative to conditions. Rank 1 is the strongest national position. It is built from three published sub-scores:

• Fire Burden Score
  How hard is fire hitting this jurisdiction relative to its size?
  Formula. Equal-weighted mean of the standardized rates: fire incidents per 1,000 residents, fire deaths per 100k, fire injuries per 100k, and property loss per capita.
  fires_per_1kfire_deaths_per_100kfire_injuries_per_100kproperty_loss_per_capita
• Fire Vulnerability Score
  How exposed is this community to fire loss before any response?
  Formula. Weighted mean of standardized demographic, housing, and environmental risk indicators, anchored to the FEMA National Risk Index social vulnerability layer.
  pct_povertypct_age_65pct_pre1970_housingpct_ruralsocial_vuln_index
• Fire Readiness Score
  How well-resourced is this jurisdiction for the risk it carries?
  Formula. Weighted mean of standardized funding and capacity indicators per capita, with coverage-flagged fiscal inputs down-weighted.
  grant_per_capitafire_expenditure_per_capitafire_tax_per_capitafirefighters_per_1k

All counts are first converted to rates (per capita, per 1,000 residents, or per 100,000 residents) so population never dominates the score. Rates are then standardized across the 51 jurisdictions using a z-score and rescaled to a 0–100 range for display. Indicators where 'lower is better' (all Burden and Vulnerability inputs) are oriented so the sub-score direction is consistent.

Each metric's percentile is its rank among all 51 jurisdictions for that metric and year, expressed 0–100. Percentiles are shown to the public because they are more intelligible than z-scores; z-scores are retained internally for computation.

A regression model estimates each jurisdiction's expected fire burden from its vulnerability and demographic profile. The residual (actual minus expected) classifies the jurisdiction as Outperforming (actual burden materially below expected), Near expected, or Underperforming (actual burden materially above expected). In the production platform this model is replaced by the Fireground Analytics risk-adjusted engine.

Every metric is traceable to a public source. The full catalog lives on the sources page.

• National Fire Incident Reporting System (NFIRS)
  U.S. Fire Administration (USFA / FEMA)
• National Emergency Response Information System (NERIS)
  U.S. Fire Administration (USFA / FEMA)
• USFA / NFPA National Fire Statistics
  U.S. Fire Administration; National Fire Protection Association
• CDC WONDER — Mortality (NVSS)
  Centers for Disease Control and Prevention
• American Community Survey (ACS) 5-Year
  U.S. Census Bureau
• OpenFEMA — Assistance to Firefighters & Mitigation Grants
  Federal Emergency Management Agency
• USAspending.gov
  U.S. Department of the Treasury
• Census of Governments — State & Local Finance
  U.S. Census Bureau
• FEMA National Risk Index
  Federal Emergency Management Agency
• National Interagency Fire Center (NIFC) / USFS
  National Interagency Fire Center; U.S. Forest Service

A jurisdiction is included only if it publishes fire incident data meeting a documented completeness threshold. Jurisdictions below the threshold are excluded and listed transparently rather than imputed into the rankings.

Missing values are never silently imputed into a public score. A metric with insufficient coverage is flagged, may be down-weighted within its sub-score, and is surfaced with a data-quality indicator. Jurisdictions are labeled High, Medium, or Low coverage.

• NFIRS participation is voluntary and incomplete; incident counts are normalized to rates and cross-checked against CDC mortality, but coverage still varies by state.
• Fire tax and local fiscal line items are sparse and inconsistently reported; Readiness inputs derived from them are flagged and may be modeled.
• Scores describe relative position, not causation. A funding–outcomes relationship is a benchmark, not a finding of cause.
• Low-population jurisdictions can swing on a small number of incidents; multi-year smoothing is applied where data allows.
• This is a Phase 1 sample build: all values shown are illustrative and generated, not real public data.

1. 1.0.0-sample2026-06-01

   Initial public methodology for the state-level sample scorecard. Defines the three sub-scores, the composite weighting (40/25/35), normalization, percentile method, and the expected-vs-actual model.

2. 0.9.0-draft2026-05-10

   Internal draft. Established the inclusion rule, source catalog, and the jurisdiction-agnostic data model.