How the analysis is computed

Figures are sourced from organizations' IRS Form 990, 990-EZ, and 990-PF returns. This prototype loads them via the ProPublica Nonprofit Explorer API (summary financials extracted from the filings), joined to IRS Business Master File registry details for NTEE classification and subsection.

For detail beyond the summary, we resolve each organization's return in the IRS year index and load the full e-file XML from the GivingTuesday Data Lake, normalized across schema years via a field concordance. That XML carries the functional-expense split (→ program-expense ratio), per-officer compensation, grant line-items, and mission text — so most orgs carry full-detail analysis, with the summary feed as the fallback.

A transparent weighted composite of four components, each scored 0–100:

• Program-expense ratio — weight 34% (program ÷ total expenses; 50%→0, 85%→100)
• Operating margin — weight 24% ((revenue − expenses) ÷ revenue; −10%→0, +10%→100)
• Months of cash — weight 24% (net assets ÷ monthly expenses; 0→0, 6 mo→100)
• Liabilities-to-assets — weight 18% (lower is stronger; 0→100, 1.0→0)

Available components are re-weighted when data is missing, and the result is graded A+ to F. The score is a heuristic for comparison, not a verdict.

The program-expense ratio comes from the functional-expense breakdown in the full e-file XML; the few orgs without an available XML filing are scored on the remaining components, re-weighted to sum to 100%.

Cohorts are defined by NTEE major group × revenue band × geography. We compute the distribution of a metric across the cohort and place an org by percentile rank (ties use the midpoint method). Distributions are pre-computed per cohort so comparisons read instantly.

We total each officer's reported compensation (base + related-org + other) from the full e-file XML and benchmark the highest-paid against peers by sector and size. Orgs without an available XML filing fall back to the reported aggregate. An “outlier” flag means the figure sits in the top few percent or several times the peer median — a prompt to read the filing, nothing more.

We flag values more than ~2 standard deviations from the peer mean, and year-over-year swings beyond 50%. These are framed as “unusual versus peers,” never as an accusation. Always read the underlying filing before drawing conclusions.

Found a figure that looks wrong? It may reflect the filing itself, or our parsing. Every figure links to its source return so you can verify, and we welcome corrections.