Are Metro service cancellations finally declining?

Photo via Busologist (Flickr)

Following the COVID lockdowns, many transit agencies had to contend not only with recovering lost riders but also ensuring service reliability and delivery. In addition to the pandemic itself, subsequent crises in staffing and supply chains hit transit systems hard. Drivers and maintenance staff were hard to find and spare parts needed for vehicle maintenance were often held up in other parts of the world.

Even as riders were slowly returning to transit, this double whammy meant that bus trips were often getting cancelled either due to a lack of an operator or lack of a bus. Anecdotally, I experienced this a lot myself throughout 2021-2023. Early on, it was hard to find information about what was cancelled and it wasn't uncommon to find a scheduled trip never end up going live" on real-time apps like OneBusAway.

It wasn't until the tail end of 2023 that I started observing trips being restored and less uncertainty around arrivals, which presumably was due to some stabilization in both staffing and supply chain flows. I don't think these issues have fully resolved but a general improvement is certainly noticeable.

Earlier this year, I requested a log of all service cancellations in 2022 and 2023. Coincidentally, Alex has also been collecting data on trip cancellations via GTFS as part of his personal project portfolio. Although his work requires more backend work to neatly publish and visualize results, our findings were fairly consistent.

As part of my public records request, Metro returned a huge collection of data that is broadly referred to as coordinator service records (CSRs). CSRs are a sort of record-keeping system for Metro to log service loss for any number of reasons. Ultimately, they were able to send me a breakdown of all work cancelled" by base, route/run, and reason for cancellation.

A quick footnote about the data: on the operational end, it's easier for Metro to track cancellations at the unit of a work piece," which is more or less analogous to a run." A run can encompass multiple trips because it represents a driver's own discrete assignment: deadheading from base to a route terminus, driving any number of that route's trips back and forth, deadheading to another route, deadheading back to base, etc. So when a base is short an operator with no extraboard (i.e., spare operator) to fill in, then that entire piece of work is cancelled along with its corresponding trips.

The data I received is at the level of cancelled workpieces. The implication here is that a single piece of cancelled work translates into multiple trips, which is orders of magnitude for impactful for riders. For example, one cancelled full shift of around 8 hours may consist of several trips lost.

At any rate, my main objective was to uncover if cancellations were generally decreasing over time. My first exploration of the data was to plot an aggregate count of cancelations as a monthly time series, broken out by operating base.

All cancellations by operating base

A few noteworthy observations:

• Cancellations across bases largely trended together, suggesting generally systematic factors driving the cancellations
• Although I don't have 2021 data in hand, there was a very noticeable precipitous drop in the early half 2022
• There is some evidence of stability through the middle of 2022, followed by more erratic spikes toward the end of 2022 and throughout 2023

Although there is some evidence of peaking in December and the summer months (July and August), there don't seem to be very obvious seasonal effects. As such, I found it worthwhile to break out the data further by cancellation reason. By and large, the two most common reasons for cancelled work were a lack of coach (NoCo") and lack of operator (NoOp"), which is where I focused my analysis.

Cancellations by lack of coach or operator

Breaking things out by NoCo" and NoOp" yield a pretty stark picture. Operator-related cancellations plummeted from the beginning of 2022 and have remained more or less stable since mid-2022. Coach-related cancellations, on the other hand, were largely insignificant until a fairly extreme spike toward the end of 2022 into early 2023.

This suggests a kind of duality: Metro's reduced service in 2021 and lack of operators likely meant relatively low vehicle maintenance requirements. But once service was restored and operator staffing stabilized, maintenance needs appear to have skyrocketed. Around that same time (early 2023), I coincidentally had reached out to a friend who drives out of East Base, who offered the following observations:

On Wednesday, there were twenty pieces of work canceled with plenty of operators available. The [coordinator responsible for routing coaches in the yard] didn't want me to write up a minor maintenance issue since they are short coaches. So that's one data point to indicate Metro is struggling to hire enough vehicle maintenance personnel...

...Metro has been low on vehicle maintenance staffing for quite a while now. During the pandemic, it wasn't as much of an issue since they weren't trying to run more service.

Based on this anecdote and the data, it looks like Metro's lack of coaches was simply due to its inability to maintain them. When I broke out NoCo" cancellations by base, it was evident that this was a system-wide problem, although some bases were more heavily inflicted than others.

Very lastly, I wanted to explore how NoOp" (lack of operator) cancellations trended when broken out by operating base.

When plotting operator-related cancellations against time, the picture is promising and buttresses the narrative that service reliability has been improving because there are now finally enough operators to fill the seats.

Across all bases, there was a steep decline from early 2022 and, with the exception of Central Base in July 2022, a steady trend downward since then. There is some evidence of a very modest increase at the end of 2023, which could be holiday-related but I don't have enough to data to detect regular seasonal effects.

To corroborate what I gleaned visually, I also ran a few simple statistical tests to detect a downward trend across the 2022-2023 monthly time series. Both tests - a linear model and Mann-Kendall test - affirmed that operator-related cancellations have dropped over the entire time period in a statistically significant way, but not coach-related cancellations, presumably due to that spike in early 2023.

I think it's fair to presume that the NoCo" spike was a one-time phenomenon as a kind of lagging indicator of service delivery. In other words, the more buses you put on the road, the more maintenance you need for those buses. In a sense, it's a good problem to have and it appears Metro was about to make adjustments by the second half of 2023 to support its service delivery targets.

Overall, the data tells a positive story around the recovery of Metro service. It validates my own experience in the past year, where I've had much less occasion to deal with missing trips and unreliable service. Although we still have a ways to go on the road to transit recovery, steps are being taken in the right direction.