NYC Subway Ridership Analysis (MTA Open Data, 2025–2026)

How New York rides the subway: the busiest stations, rush hours, seasons, unusual days, and an animated map of a day on the subway, built from the MTA's hourly ridership data.

Live demo: mantiereidii.com/mta-ridership, updated daily.

By Mantie Reid II. Free to reuse with credit (CC BY 4.0).

What the data shows

• About 3.6 million subway rides a day. Times Sq-42 St is the busiest station, and 5 PM the busiest hour.
• Ridership in 2026 is up 2.9% on the same days of 2025: Manhattan +3.7%, Brooklyn +3.2%, Queens +1.1%, Bronx −1.1%.
• Storms and holidays empty the subway: a winter storm on Feb 23, 2026 cut ridership to 22% of a normal Monday, and Christmas to 35%.
• Big events show up clearly. Mets-Willets Point jumps to about 14× its normal riders during the US Open. The NYC Marathon, July 4th and Yankees games also stand out.
• The MetroCard faded out: from 33% of entries in January 2025 to under 1% by late 2026.

2026 vs 2025	Fares and the MetroCard

Riders by hour of day	Riders per month

Riders by day of week	Top stations by season

What's in this repo

NYC subway ridership for 2025 and 2026 to date, from MTA's open data. It includes:

• mta_ridership.py: a Python script that writes small JSON/CSV summaries and PNG charts.
• wordpress-plugin/: a WordPress plugin that shows the same trends on a web page and refreshes them daily.

Source: MTA Subway Hourly Ridership: Beginning 2025 (data.ny.gov dataset 5wq4-mkjj). Neither tool downloads the full CSV (40M+ rows). They ask the data.ny.gov API to add up the rows first and download only the totals.

Reproduce the results

Requirements: Python 3.9+ and an internet connection. No account or API key is needed.

git clone https://github.com/MantieReid/mta-data-analysis.git
cd mta-data-analysis
pip install -r requirements.txt
python mta_ridership.py

This writes everything to output/. A first run takes a few minutes. After that, the downloaded totals are cached in output/*_cache.csv (not committed). On later runs the daily station totals are updated incrementally: the last 30 days are downloaded again and replace the cached ones (new days arrive, and any MTA revisions to those days are picked up); use --refresh to download everything again.

Options:

Flag	What it does
--week YYYY-MM-DD	Also writes output/week_<monday>.json: riders per station for that week
--refresh	Ignore the cache and query the API again
--reconcile-days N	How many recent days to download again on a cached run (default 30, 0 = off)
--no-charts	Skip the PNG charts

Tests

pip install pytest pyflakes
python -m pytest tests          # Python tests; the Python/PHP comparison runs if PHP is installed
php tests/php/run_tests.php     # plugin tests (plain PHP, no WordPress or database needed)

The tests use small made-up data in tests/fixtures/ and never call the MTA's API. They cover rounding, dates and seasons, unusual-day detection, holidays, and re-checking recent days (re-running is idempotent, revised days overwrite old values, a revision changes the trend). The plugin tests also cover the run lock, step status, failure alerts and data health. tests/test_parity.py runs the same input through Python and PHP and checks they agree. GitHub Actions runs everything on every push (.github/workflows/ci.yml), on PHP 7.4 and 8.3.

tests/php/live_reconcile_check.php is an optional live check (not run in CI): it runs the re-check against the real API on a copy of a site's stored history and reports any MTA revisions.

Optional: set the SOCRATA_APP_TOKEN environment variable to a free Socrata app token to avoid throttling.

Will I get exactly the same numbers?

The committed output/ files are a snapshot of data through Oct 14, 2026. system_monthly.json records the date the data runs through. If you run the script later, your numbers will differ:

• MTA adds new days, usually 1–2 weeks behind.
• MTA occasionally revises past data.

To compare like for like, look at periods that were complete in both runs. Rows marked "partial": true (months, weeks, seasons and years with missing days) will change.

How the numbers are calculated

• Riders: the ridership column (entries into the station), transit_mode = 'subway', grouped by station_complex_id.
• Weeks run Monday–Sunday. A week belongs to the season of its Thursday.
• Seasons: Winter = Dec–Feb (December counts toward the following winter, so Dec 2025 is in "winter 2025-26"), Spring = Mar–May, Summer = Jun–Aug, Fall = Sep–Nov. Seasonal averages use complete weeks only.
• Partial periods: a month, week, season or year is partial if the data doesn't cover every day in it.
• Hourly averages: riders in each hour of the day ÷ the number of days with data in that year.

Station trends (the script and the WordPress plugin use the same method and get identical results):

• Unusually busy/quiet days: a station's riders that day compared with its normal level, which is the median of the same weekday over the previous 26 weeks (at least 10 of them). The comparison then divides by the whole system's own ratio that day, so holidays and storms that affect every station don't count. "Busy" means at least 3× normal. "Quiet" means 35% of normal or less. Stations normally under 1,500 riders/day (3,000 for quiet days) are skipped, and back-to-back days are grouped into one event.
• Weekend vs weekday: average daily riders over the last 364 days, skipping days a station was nearly closed (under 20% of its median).
• Late-night share: riders between midnight and 5 AM ÷ all riders, last 365 days.
• Transfers: the dataset's transfers column, last 365 days.
• Map ("a day on the subway"): riders per station per hour over the last 56 days ÷ the number of days with data. The API times out when grouping a longer range by station and hour, so it's queried one week at a time and added up.
• Holidays and storms calendar: system riders per day ÷ the median of the same weekday over the previous 26 weeks. US holidays are labeled automatically; storms and events come from a short hand-kept list in mta_ridership.py (EVENTS) and the plugin (includes/calendar.php).
• Year over year: Jan 1 to the latest day of data this year vs the same calendar days last year. Station changes only include stations with 2,000+ riders a day last year.
• Find your station: each station's riders over the last 365 days ÷ the number of days with data, ranked busiest first. Its change vs last year uses the same calendar days as above, shown only for stations with 100+ riders a day last year. Stations that average under 1 rider a day (St George on the Staten Island Railway, which records almost no entries) are left out.

Outputs

File	Contents
station_monthly.csv	Riders per station per month, with a partial flag
station_month_avg.csv	Average riders per month per station (complete months only)
system_monthly.json	System-wide monthly totals and the average month
weekly.csv / weekly.json	Weekly totals and average riders per station
seasons.json	Average weekly riders per season
station_yearly.csv, yearly.json	Yearly totals per station, top 10 per year
day_of_week.json	Average daily riders per weekday, per year
station_hourly_avg.csv, system_hourly_avg.json, month_hourly_avg.json	Average riders per hour of day
station_seasonal.csv, seasonal_top5.json	Season totals per station, top 5 per season
top_stations_latest.json	Top 5 stations on the latest day of data and over the last 7 days
unusual_days.json	Unusually busy and quiet days per station: the last 14 days and the biggest per year, each with its daily riders around the day
weekend_vs_weekday.json	Stations busiest on weekends vs mostly weekday, last 364 days
late_night.json	Stations with the largest share of riders between midnight and 5 AM, last 365 days
transfers.json	Stations with the most transfers, last 365 days
calendar.json	Riders per day vs a normal day for that weekday (median of the 26 weeks before), with holidays, storms and events labeled
extras.json	Map data (each station's location and average riders per hour, last 8 weeks), this year vs last year (total, by borough, biggest station changes), the station lookup (every station's average riders a day over the last 365 days, rank and change vs last year), fare types (last 365 days) and OMNY vs MetroCard by month
charts/*.png

Congestion pricing page

The plugin also has a congestion pricing page, [mta_congestion] (live), built from MTA Congestion Relief Zone Vehicle Entries and the zone outline. It shows the tolls with a calculator (E-ZPass rates from mta.info), the zone map, vehicles entering per hour, this year vs last year on the same days (Jan 5 to the latest day, since tolling began Jan 5, 2025), every bridge and tunnel, entries by vehicle type, and subway ridership at stations inside the zone vs outside (a station is inside if its location falls within the outline).

Subway line report cards

[mta_lines] (live) ranks every subway line by on-time rate and gives each line a report card at /subway-lines/<line>/ (for example the B train). It combines five MTA monthly datasets: Terminal On-Time Performance, Customer Journey-Focused Metrics, Trains Delayed, Wait Assessment and Major Incidents.

• On time: trips reaching the last stop within 5 minutes of schedule ÷ scheduled trips, weekdays and weekends, over the last 12 months of data. The change is vs the 12 months before.
• 5+ minutes late, extra wait, extra ride time: from the customer journey metrics, weighted by passengers.
• The W line is left out: the MTA publishes no on-time data for it.

Bus route report cards

[mta_buses] (live) ranks NYC bus routes (busiest, slowest, least reliable, most bus-lane camera violations) and gives each route a report card at /bus-routes/<route>/ (for example the M15 Select Bus Service).

• Riders a day: Bus Hourly Ridership over the last 12 months ÷ days, queried a month at a time. Riders per hour use the last 8 weeks of weekdays, queried a week at a time.
• Speed: Bus Speeds, total miles ÷ total operating time on weekdays (includes time at stops and lights). Rankings of speed and spacing leave out express buses.
• Spacing (Wait Assessment), service run (Service Delivered), extra time (Customer Journey Metrics, weighted by customers) and camera violations (ACE Violations), last 12 months.
• Busiest stops: Stop Ridership Estimates, latest month, weekday rows ÷ the number of weekdays in that month (each row adds up every Monday, Tuesday… of the month). These come from passenger counters, so they can run higher than fare-based ridership.
• Temporary shuttles (e.g. T321) and routes with under 50 riders a day are left out.

WordPress plugin

See wordpress-plugin/README.md. In short:

1. Upload wordpress-plugin/mta-ridership.zip under Plugins → Add New → Upload Plugin.
2. Activate it.
3. Add [mta_ridership] to a page.

License

Everything in this repo (code, output files and charts) is licensed under CC BY 4.0 (full text). You can use, share and change it for any purpose, including commercially, as long as you give credit.

How to credit, for example under a chart or in an article:

NYC subway ridership analysis by Mantie Reid II, https://github.com/MantieReid/mta-data-analysis (CC BY 4.0)

Say if you changed anything. GitHub's "Cite this repository" button (from CITATION.cff) gives the same credit in APA and BibTeX formats.

The underlying ridership data is published by the MTA on data.ny.gov and isn't covered by this license.