National Transit Equity Research Audit

Wait gap = median(wait in lower-income group) − median(wait in higher-income group), in minutes. Positive means scheduled waits are longer in lower-income neighborhoods (worse). Negative means shorter (better).

Density gap = median(trips/km² in lower-income group) − median(trips/km² in higher-income group). Negative means lower-income neighborhoods have lower scheduled service density (worse).

This site uses GTFS Schedule (timetables). It does not measure delays or cancellations. That requires GTFS‑Realtime.

1. National pattern (schedule-based): Lower-income neighborhoods do not consistently have worse planned service in this audit.
2. Where equity risk concentrates: The clearest risk signal is transit deserts and coverage gaps, especially in the lowest-income quintile.
3. How to use this: Prioritize “desert reduction” and coverage improvements. Treat “significant city” lists cautiously (multiple testing).

Many systems show equal or stronger scheduled service supply in lower-income neighborhoods (often reflecting dense corridors and core areas). Equity concerns still exist, but may concentrate in deserts, in specific outliers, or in reliability (not measured by static schedules).

Deserts are the most communicable equity signal: “places with very weak or absent service.” Use this as the primary KPI for public reporting.

When many systems are tested, some will appear significant by chance. For public-facing lists, apply multiple-testing control and report confidence intervals.

• Transit: GTFS Schedule (planned trips, stop times, calendars). GTFS Schedule reference.
• Income: Statistics Canada Census Profile (2021). Census Profile / WDS.

• Scheduled wait: derived from scheduled headways in a defined time window (weekday/weekend).
• Scheduled density: scheduled supply normalized by neighborhood area (trips/km²).
• Transit deserts: rule-based flag for very weak/absent service.

• Primary: Bottom50 vs Top50 (median split within system).
• Robustness: Q1 vs Q5 (quintiles within system).
• Effect: median gap (low − high). Wait gap > 0 = worse waits in low-income; density gap < 0 = worse density in low-income.
• Inference: Mann–Whitney U p-values (two-sided) and bootstrap confidence intervals.

This is a planned-service audit. It does not capture delays/cancellations/crowding. For that, add GTFS‑Realtime where available.

• Minimum sample sizes per system and per income group to compute robust statistics.
• “Report tables” prefer complete finite statistics (gap + CI + p-value) to avoid NaN outputs.
• For “significant city” lists, apply multiple-testing control (FDR) and require practical effect size thresholds.

Track desert rate overall and by income group; prioritize where the lowest-income neighborhoods have the highest desert burden.

Where there are zero scheduled trips in the time window, wait is undefined—treat this as a coverage gap and report it explicitly.

When GTFS‑Realtime is available, measure delay/cancellation patterns to capture reliability inequity.

Apply FDR and require minimum practical effect sizes before listing systems publicly.

• Schedule ≠ reliability: planned service only.
• Wait assumptions: most interpretable for frequent service.
• Geography: CT and DA differ in size and coverage.
• System vs city: GTFS feeds may span multiple municipalities.

• Privacy: aggregated neighborhood statistics only; no personal data.
• Avoid stigma: frame as service gaps and investment opportunities.
• Accessibility: plain language + readable tables; avoid color-only meaning.
• Licensing: respect open-data licenses and provide attribution.

• GTFS Schedule feeds (planned service): stops, trips, stop times, calendars. Reference: GTFS Schedule.
• Statistics Canada — Census Profile (2021) for neighborhood income and geography-level characteristics. Terms: Statistics Canada Open Licence.

“This analysis uses GTFS Schedule feeds published by Canadian transit agencies (see agency portals for license details) and Statistics Canada Census Profile data (2021) licensed under the Statistics Canada Open Licence. Analysis and visualizations: NTERA (ntera.ca).”

GTFS feeds are usually distributed via municipal or provincial open-data portals. License terms can differ by agency. If you publish a downstream dataset, include a short “Data & Licensing” section and link to each portal’s license page when possible.

Income values come from the 2021 Census Profile (income year 2020). GTFS feeds represent the schedules available at the time they were collected. If you refresh GTFS inputs, regenerate the outputs and replace the files in /data.

1. Regenerate outputs and export updated CSVs.
2. Replace the CSVs in /data (same filenames).
3. Commit and push — GitHub Pages updates automatically.

https://github.com/jasonpark26/ntera

Median scheduled wait (low-income) − median scheduled wait (high-income), minutes. Positive = worse waits in low-income areas.

Median trips/km² (low-income) − median trips/km² (high-income). Negative = worse density in low-income areas.

A rule-based flag for neighborhoods with very weak or absent planned service in the analysis window.

GTFS Schedule measures planned service. Reliability (delays/cancellations) requires GTFS‑Realtime or agency performance data.

Most commonly: no scheduled trips serve that neighborhood in the analysis window. Treat it as a coverage gap.

No. This is planned-service (GTFS Schedule). Reliability requires GTFS‑Realtime where available.

Many higher-income neighborhoods are low-density suburbs with lower frequency and fewer routes.

• v5: Added Data Quality/Coverage, Last Updated timestamp, Reproducibility, Glossary, SEO tags, and a reliability disclaimer banner.
• v4: Added Sources/Licensing and “How to read” definitions.