Ghost Dollars

An interactive map that makes visible what New York City's capital-spending data leaves out: the investment immigrant neighborhoods don't get, and the spending the city can't place on a map.

The Two Framings

Ghost dollars: the equity gap. For each of NYC's 59 community districts, ghost dollars represent the capital investment a neighborhood did not receive relative to an equal-per-resident baseline. A positive ghost-dollar figure means the district received less than its population-proportional share of visible capital. The tool reports this as a percentile rank (not a false-precision dollar entitlement) to make the comparative tilt legible without overstating precision.

Dark dollars: two transparency gaps that compound. There is no single dark-dollars number. There are two separate, stacking gaps, and the tool is careful never to conflate them:

1. Unplaceable share (computed here, ~74%): Of the capital the city does publish in the Capital Projects Dashboard, roughly $116B of $158B (the current commitment plan) is tagged Citywide, borough-only, or lump-sum, so it names no community district. This is our computed darkSharePct. These dollars exist in the public record but cannot be assigned to any neighborhood.

2. Missing share (NYC Comptroller, "Flying Blind on Billions," Dec 2025, ~47%): The dashboard contains only about 47% of project IDs and ~58% of planned dollars present in the city's underlying financial system. Roughly half of projects never appear in the public-facing data at all.

These stack: of all capital, roughly half never reaches the dashboard, and of what does, roughly three-quarters can't be placed in a neighborhood. The tool presents #1 as its own measured number and cites #2 as a separate, compounding fact.

Methodology & Caveats

Reporting period: The pipeline uses only the latest reporting period (202601 at build time, discovered dynamically from the dataset). The Capital Projects Dashboard stores one row per project budget-line per reporting period and republishes ~3× per year, so naively summing all rows overstates capital ~10×. We deduplicate within the latest period by fms_id | budget_line.

Deduplication: One budget-line per project is counted once per the latest reporting period.

Multi-CD projects: A small fraction (~3%) of rows name more than one community district (e.g. "Manhattan 01, Manhattan 12"). These projects split total_budget equally across the listed districts. Borough-only and Citywide rows are classified as dark.

Parity baseline: A flat equal-per-resident dollar figure: (sum of all placeable visible capital) ÷ (sum of all placeable-CD populations). It is the most conservative baseline, and the simplest to read and to defend. A need-weighted baseline is a planned future toggle, never the default.

Percentile reporting: Per-resident visible capital is ranked across all 59 CDs and reported as a percentile. Ghost dollars are shown for orientation, not as a precise entitlement claim.

PUMA approximation: ACS 2019–2023 5-year population data is sourced via a PUMA-to-CD crosswalk. For 8 community districts with population under 100,000, the PUMA boundaries do not align cleanly with CD boundaries, so population figures are approximations for those districts.

Dark-dollar consistency: A dollar excluded as dark is excluded from both the numerator (visible capital per CD) and the parity pool. It is never silently dropped from one side.

Data Sources

Layer	Source	Dataset ID	Notes
Capital spending	NYC Capital Projects Dashboard (NYC Open Data / Socrata)	fb86-vt7u	Fields: community_board, total_budget, spend_to_date. Updated ~3×/yr.
Demographics	ACS 2019–2023 5-year, via PUMA-to-CD crosswalk	—	Population, race/ethnicity, median household income by CD. PUMA approximation disclosed above.
CD geometry	NYC Open Data	5crt-au7u	GeoJSON export, plain download. (Was yfnk-k7r4, retired 2024.)

Refreshing Data

1. Fetch raw capital data:

   npx tsx src/lib/pipeline/fetchRaw.ts
   

2. Run the full pipeline (aggregate, deduplicate, join demographics, compute metrics):

   npm run pipeline
   

   This overwrites public/data/cd-metrics.json and the merged GeoJSON.

Running Locally

npm install
npm run dev

Open http://localhost:3000.

Tests:

npx vitest run