You’ve reached the end of the line.
Transit riders want to know how crowded their vehicles are. One problem: that data didn’t exist in most cities. Now it has arrived.
August 13, 2020
Transit ridership took a sinus-rattling swan dive this winter. (There was this whole thing…) Now, as the pond water slowly drips out of our ears, more and more of us are returning to public transit. Some by choice; some out of necessity. Whether you’re hopping on the bus to rendezvous with that long-lost friend, or because your bike caught a flat, or because work life beckons — making sure you’re safe is Priority #1, #2, and #3.
But we all know there’s a difference between actually “being safe” and that more elusive “feeling safe”. With enforced mask policies, riding the subway or bus carries a low amount of risk. That risk is further reduced when physical distancing between riders is possible. So if we can make sure you know in advance that “this bus is crowded” or “this one is not” you can head to the stop with markedly more pep in your step.
However: if you’re a resource-strapped agency, it’s not as simple as saying “crowding tracker? let’s do it”. You need hard data to justify its implementation. So we did a survey of 6,000 Transit riders to show transit agencies (and their budget overlords) that crowding info really matters.
Before COVID-19, most people weren’t super-bothered by crowding. It was part of life in the big city. Sixty-three percent of people would board a vehicle — even if it was more packed than a mariachi band in a Volkswagen Beetle.
Ask riders how they feel now? No more sardines in the subway:
If only ~25% of urgent travelers will board a packed vehicle — and ~10% of people on non-urgent trips — then transit agencies run the risk of ostracizing tons of customers, forcing many to take pricier private rides, and otherwise leaving the rest marooned at the bus stop.
All because they don’t have crowding info.
But when agencies publish crowding info? Riders say they’re far likelier to take public transit more often.
So let’s assume your transit agency wants crowding info. Great! The ability to share crowding info on apps like Google Maps 🏄♀️, Apple Maps 👯♀️, Transit 🧞♀️, etc. has technically existed since the mid-2010s. But few agencies (or apps) cared much about it. Crowding info was always a “nice to have” feature that permanently hovered around Priority Level Zero.
Ha ha ha… how innocent we were…
Since the beginning of the pandemic, transit agencies have been in a rush to get their dusty crowding stats up to spec. Some of our partner agencies, like the MBTA, were able to jury-rig equipment on 30+ bus lines to detect real-time crowding levels! But most other agencies don’t have that capacity. Instead of real-time crowding levels, those agencies can use predictive crowding measures — done in conjunction with Transit’s engineers.
Many transit vehicles have APCs (Automated Passenger Counters). They’re little invisible laser beam counters, between the bus or train doors. You walk through the lasers, breaking the beam, and depending on your direction, the counter can tell if you’re boarding or disembarking. When there’s lots of crowding, or bags blocking the lasers, you get some counting errors.
Some agencies have overhead cameras that can count passengers with a bit higher accuracy. Other agencies use their fare collection systems, using “bus pass card taps” as a proxy for crowding. All of these are hardware solutions, and necessary if you want to count crowds with precision.
APC hardware was developed for “census-type” reasons. So that agencies knew, generally, which routes were busiest. (Crowding data would get pulled infrequently, and be used to inform, say, a transit network study, or to justify ridership numbers to a government agency overseeing the agency’s budget.) APCs were generally not designed to deliver passenger counts in real time. APCs aren’t 5G devices — they transmit data really slowly. So even those lucky agencies with APCs, have had to hack those same APCs, compress the hell out of their crowding data, and write custom software that can derive real-time, data-rich insights from the bleeps and bloops that the APC transmits over its low-frequency radio waves.
Enter predictive crowding
Assuming you aren’t Neo from the Matrix (and can’t hack your APCs to spit out awesome real-time crowding stats) you can get pretty good results with predictive crowding. Instead of uploading crowding stats in real time, you can upload your vehicle crowding stats at the end of the day, or week, etc. — and feed them into our statistical model. These models can help predict crowding levels in the future. Starting today, this 🔥 crowding collab goes live for our partner agency, LA Metro. The agency uploads their APC data, our team processes it to make smart predictions, and we put those predictions into Transit.
In early tests, our crowding predictions fall within striking distance (88%) of on-the-ground crowding levels — as measured by actual APC data that was checked after we made our prediction.
Designing for easy consumption
When is a train “full”? When is a bus “empty”? Can someone tell the difference between a bus that’s at 20% capacity, vs. 30% capacity? Our friends at the MBTA did some user research to find the best way to communicate crowding levels. They found it was best to divide “crowding” into three buckets:
And that’s how you go from APC laser beams, to a crowding count, to the pretty little crowding indicators you see today in Transit.
More than thirty agencies, be they in Boston or LA or Australia or Ohio, have rolled out crowding info to riders. They’re working with our team at Transit, creating improvised (but working!) solutions for riders, tackling problems in weeks that may have otherwise taken months or years to get around to. They’re actively restoring rider confidence, and putting their organizations in a good place to bring back ridership.
Plus — even outside of plague times — crowding information is terrifically important: for riders with disabilities, parents with strollers, folks transporting big items like bicycles or banjos or flat screen TVs. And there’s plenty of aces up our sleeves yet. Next up: a solution for our agency friends without any APC hardware whatsoever… because what could be better at measuring a crowd… than the crowd itself…
Evaluating GBFS feed quality for shared mobility services
Eva, a Montreal ridehail company, joins the STM, BIXI, Communauto, and Uber as the newest option in the Transit app for riders across the city
Riders can use one app to purchase a BCycle pass, unlock a bike and get multimodal directions in Madison, Greenville, Broward County and Santa Barbara