Transit Peers

A benchmarking tool for comparing US public transit agency performance using National Transit Database (NTD) data from 2019–2025.

Live at transitpeers.net

What It Does

Transit Peers helps transit professionals identify peer agencies and compare performance over time. The workflow has two steps:

1. Peer Selection — Pick a home agency, then filter and rank ~1,000+ agencies by similarity across criteria like population, ridership, operating costs, and more.
2. Performance Comparison — View trend charts, bar charts, and summary tables comparing your agency against selected peers across 8 metrics. Export data as CSV.

Tech Stack

• Frontend: React 19, TypeScript, Recharts, Vite
• Data Processing: Python 3.12 (pandas), managed with uv
• Data Source: NTD annual metrics CSVs (2019–2025)

Project Structure

transit_benchmark/
├── preprocess.py              # CSV → JSON data pipeline
├── metrics/                   # Raw NTD CSV files (2019–2025)
│   └── data_dictionary.csv
└── app/                       # React frontend
    ├── src/
    │   ├── App.tsx            # Main app with state management
    │   ├── types.ts           # TypeScript interfaces
    │   ├── data.ts            # Data loading utilities
    │   └── components/
    │       ├── FilterStep.tsx  # Peer selection & similarity ranking
    │       └── ExploreStep.tsx # Charts, tables, CSV export
    └── public/data/           # Generated JSON (from preprocess.py)

Setup

Prerequisites

• Python 3.12+ with uv
• Node.js with npm

Data Processing

source .venv/bin/activate
python preprocess.py   # Processes metrics/*.csv → app/public/data/*.json

This generates 5 JSON files consumed by the frontend: agencies.json, agency_yearly.json, agency_modes.json, yearly_mode_totals.json, and metadata.json.

Frontend

cd app
npm install
npm run dev       # Start dev server at http://localhost:5173
npm run build     # Production build → app/dist/
npm run lint      # Run ESLint
npm run preview   # Preview production build

Data Pipeline

The preprocessing script handles several NTD data quirks:

• NTD ID normalization — 2019–2021 data uses longer IDs (e.g., 7R03-70133); the script extracts the last 5 characters for consistency with 2022+ data.
• Currency parsing — Older CSVs have formatted values like $5,206,727,193 that get stripped to numeric.
• Population imputation — Missing population values are backfilled from the nearest available year, preferring 2022+ data.
• Derived metrics — Calculates cost per trip, farebox recovery, rides per capita, fare per trip, and trips per hour.
• Filtering — Removes agencies with zero ridership.

Similarity Ranking

Agencies are ranked by similarity to the home agency using a configurable set of criteria:

1. Log-transform values (handles wide ranges across small and large agencies)
2. Normalize each criterion to 0–1
3. Sum the absolute differences across selected criteria
4. Lower score = more similar

Available criteria: population, ridership, fare per trip, cost per trip, operating expenses, vehicle revenue hours, vehicle revenue miles, rides per capita.