Faster-recovering night and weekend transit ridership doesn't mean there's no commute peak.
I've rounded so you can compare San Francisco Bay Area regional rail ridership at a glance.
July, 2026
BART, to the nearest 10,000 rides:
200,000 weekdays (fairly stable)
130,000 Saturday
090,000 Sunday
Frequency: Every 10 minutes, same every day except service starts weekdays at 05:00, Saturdays at 06:00, and Sundays at 08:00. Every 20 minutes after 20:00. Last trains start before midnight.
Caltrain, to the nearest 1,000 rides:
47,000 weekdays (rises and falls; 12-month rolling average is only 41,000)
32,000 Saturday
24,000 Sunday
Frequency: Every 30 minutes (66 trains), same every day except 38 extra trains on weekdays during commute hours. Last trains start at or before midnight.
Weekdays are 50% busier than Saturdays and twice as busy as Sundays. A year has 5 times as many weekdays as Saturdays or Sundays, making weekday ridership even more important. This should be obvious.
Days per year, to the nearest 10:
Weekdays: 250
Saturdays: 50
Sundays: 50
Moreover, trains are crowded during commute hours, whereas they don't get crowded on normal weekends. If you think in terms of what matters, people moved (if you ride transit) or minutes of driving delay avoided (whether you ride transit or not), commute service is more efficient and off-peak transit service is less efficient.
Andrew Miller is correct that commute service costs more than all-day service. To add insult to injury, commuters pay the lowest unit price. Most monthly passes are priced at less than 40 single rides, one ride to work and one back home on each of the 20 work days in the month. US commuters buy monthly passes with pre-tax deductions from wages, resulting in even greater discounts and smaller contributions to the public purse.
That's the thing. Transit is a public service whose capital costs are fully subsidized and whose operating costs are heavily subsidized. The Alon Levy cost differential blog post that Andrew Miller cites focuses on capital and depreciation, a little silly in the US, where federal government grants pay most transit capital costs. I'll focus on operating costs, because transit operating money is the scarce resource.
Over the past 50 years, among large US/Canada agencies, only the Toronto Transit Commission and the Bay Area Rapid Transit District covered a large share of operating costs from fares.
The TTC's farebox recovery ratio started to drop with suburban sprawl and the elimination of the suburban fare zone in the early 1970s. You can see remnants of the suburban fare zone on the generously-sized mezzanine level of Warden Station. Municipal- and especially provincial-government austerity pushed the TTC's farebox recovery ratio to between ⅔ and 70% for multiple years in the mid-1990s, before politicians rediscovered the social and political value of transit. The TTC's pass multiplier is still unusually high for the industry, at 47 rides before the monthly price cap kicks in. Capping has just replaced the up-front purchase of TTC monthly passes as of September, 2026, yay!
BART's farebox recovery ratio reached 70% for a couple years in the mid-2010s, due to unprecedented commuter demand. Weekday rides peaked at over 430,000 in the fiscal year ended June 30, 2016. By the time the pandemic struck, demand on a normal weekday had already dropped by more than 30,000 rides. Interestingly, BART has never offered unlimited passes. The San Francisco Muni monthly pass used to cover BART within city limits, that is, 9 closely-spaced train stations out of more than 40, but even this now requires an extra monthly payment.
Unit operating cost -- the cost of running one specific train or providing one specific hour of service -- matters mainly when a politician, or an agency board member, leader, or manager, is intent on cutting transit service.
Unit operating cost is ignored for the special event service that transit agencies boast about and Andrew Miller likes so much. Per Rachel Swan's August 26, 2026 San Francisco Chronicle article, the Santa Clara Valley Transportation Authority, the final link from BART and Caltrain to Levi's Stadium, provided 230,000 World Cup rides at an extra cost of over $21 million, that is, for more than $90 per ride. (It's still more than $45 per ride if we subtract $10 million of one-time US federal government support. FIFA contributed nothing.) If you care about unit operating cost, special event transit is not your friend, because it depends on overtime and unusual positions like docents and crowd control/safety workers.
In the San Francisco Bay Area, unit operating cost is also ignored when regional bureaucrats replace efficient express bus service with trains deep into low-density suburbs.
If anything, the problem facing transit agencies today is that night and weekend demand has risen but post-pandemic commute demand hasn't fallen far enough. Diffuse demand is harder to serve than concentrated demand, and variation -- regardless of the day and time -- is costly. Agencies must either continue providing extra commute service or raise all service to commute levels, resulting in underutilization.
With the debut of electric service in 2024, Caltrain chose the former approach. Caltrain provides baseline service every half hour but still runs extra trains during commute hours, to meet demand.
In 2023, BART chose the latter approach. Before the pandemic, BART ran trains every 10 minutes in the suburbs during commute hours, and less frequently at all other times. Now, BART runs trains in the suburbs every 10 minutes, 7 days a week (except Dublin/Pleasanton, which gets single-line, 20-minute service). BART is underutilized at mid-day and on normal weekends, especially in suburban areas like San Mateo County, Concord, Walnut Creek, Fremont, and San José (where BART touches the northern edge of the city).
Incidentally, delaying the start of BART service on weekdays from 04:00 to 05:00, a change that Andrew Miller mentions, didn't assure on-time completion of the Transbay Tube earthquake retrofit. The project wrapped up a couple years late, in 2024, and the lost hour of service was never restored.
Before the Transbay Tube retrofit project, during, and since, we've had full-weekend maintenance closures far away from the Tube, so it seems the extra hour for maintenance hasn't been used optimally. Meanwhile, BART stubbornly ignores differential demand on Friday and Saturday nights. Final trains roll out before midnight every day. What little overnight bus service remains in the Bay Area is provided by separate agencies, with their own routes, schedules, stops, and fares, so people gravitate to private ride-share services. So much for adjusting transit service to observed human behavior and actual transportation demand!
Excellent analysis. Most transit agencies, especially GO Transit/Metrolinx, is stuck in pre-Pandemic, tidal flow commuter rail thinking, and unwilling to accept that moving people with trains, no matter what day of the week or time, is really important.
The most expensive seat is not the first, rather the unsold seat. The former is finite, the latter is infinite, since it can never be recouped. Finding out ways to fill those unsold seats through a demand driven model (not supply-side) is the key to approaching breakeven or staying relatively solvent as a public good. Unfortunately, not even Alon appears to model this way from what I can tell. We proved this with the PATH. But our work isn't done.
Faster-recovering night and weekend transit ridership doesn't mean there's no commute peak.
I've rounded so you can compare San Francisco Bay Area regional rail ridership at a glance.
July, 2026
BART, to the nearest 10,000 rides:
200,000 weekdays (fairly stable)
130,000 Saturday
090,000 Sunday
Frequency: Every 10 minutes, same every day except service starts weekdays at 05:00, Saturdays at 06:00, and Sundays at 08:00. Every 20 minutes after 20:00. Last trains start before midnight.
Caltrain, to the nearest 1,000 rides:
47,000 weekdays (rises and falls; 12-month rolling average is only 41,000)
32,000 Saturday
24,000 Sunday
Frequency: Every 30 minutes (66 trains), same every day except 38 extra trains on weekdays during commute hours. Last trains start at or before midnight.
Weekdays are 50% busier than Saturdays and twice as busy as Sundays. A year has 5 times as many weekdays as Saturdays or Sundays, making weekday ridership even more important. This should be obvious.
Days per year, to the nearest 10:
Weekdays: 250
Saturdays: 50
Sundays: 50
Moreover, trains are crowded during commute hours, whereas they don't get crowded on normal weekends. If you think in terms of what matters, people moved (if you ride transit) or minutes of driving delay avoided (whether you ride transit or not), commute service is more efficient and off-peak transit service is less efficient.
Andrew Miller is correct that commute service costs more than all-day service. To add insult to injury, commuters pay the lowest unit price. Most monthly passes are priced at less than 40 single rides, one ride to work and one back home on each of the 20 work days in the month. US commuters buy monthly passes with pre-tax deductions from wages, resulting in even greater discounts and smaller contributions to the public purse.
That's the thing. Transit is a public service whose capital costs are fully subsidized and whose operating costs are heavily subsidized. The Alon Levy cost differential blog post that Andrew Miller cites focuses on capital and depreciation, a little silly in the US, where federal government grants pay most transit capital costs. I'll focus on operating costs, because transit operating money is the scarce resource.
Over the past 50 years, among large US/Canada agencies, only the Toronto Transit Commission and the Bay Area Rapid Transit District covered a large share of operating costs from fares.
The TTC's farebox recovery ratio started to drop with suburban sprawl and the elimination of the suburban fare zone in the early 1970s. You can see remnants of the suburban fare zone on the generously-sized mezzanine level of Warden Station. Municipal- and especially provincial-government austerity pushed the TTC's farebox recovery ratio to between ⅔ and 70% for multiple years in the mid-1990s, before politicians rediscovered the social and political value of transit. The TTC's pass multiplier is still unusually high for the industry, at 47 rides before the monthly price cap kicks in. Capping has just replaced the up-front purchase of TTC monthly passes as of September, 2026, yay!
BART's farebox recovery ratio reached 70% for a couple years in the mid-2010s, due to unprecedented commuter demand. Weekday rides peaked at over 430,000 in the fiscal year ended June 30, 2016. By the time the pandemic struck, demand on a normal weekday had already dropped by more than 30,000 rides. Interestingly, BART has never offered unlimited passes. The San Francisco Muni monthly pass used to cover BART within city limits, that is, 9 closely-spaced train stations out of more than 40, but even this now requires an extra monthly payment.
Unit operating cost -- the cost of running one specific train or providing one specific hour of service -- matters mainly when a politician, or an agency board member, leader, or manager, is intent on cutting transit service.
Unit operating cost is ignored for the special event service that transit agencies boast about and Andrew Miller likes so much. Per Rachel Swan's August 26, 2026 San Francisco Chronicle article, the Santa Clara Valley Transportation Authority, the final link from BART and Caltrain to Levi's Stadium, provided 230,000 World Cup rides at an extra cost of over $21 million, that is, for more than $90 per ride. (It's still more than $45 per ride if we subtract $10 million of one-time US federal government support. FIFA contributed nothing.) If you care about unit operating cost, special event transit is not your friend, because it depends on overtime and unusual positions like docents and crowd control/safety workers.
In the San Francisco Bay Area, unit operating cost is also ignored when regional bureaucrats replace efficient express bus service with trains deep into low-density suburbs.
If anything, the problem facing transit agencies today is that night and weekend demand has risen but post-pandemic commute demand hasn't fallen far enough. Diffuse demand is harder to serve than concentrated demand, and variation -- regardless of the day and time -- is costly. Agencies must either continue providing extra commute service or raise all service to commute levels, resulting in underutilization.
With the debut of electric service in 2024, Caltrain chose the former approach. Caltrain provides baseline service every half hour but still runs extra trains during commute hours, to meet demand.
In 2023, BART chose the latter approach. Before the pandemic, BART ran trains every 10 minutes in the suburbs during commute hours, and less frequently at all other times. Now, BART runs trains in the suburbs every 10 minutes, 7 days a week (except Dublin/Pleasanton, which gets single-line, 20-minute service). BART is underutilized at mid-day and on normal weekends, especially in suburban areas like San Mateo County, Concord, Walnut Creek, Fremont, and San José (where BART touches the northern edge of the city).
Incidentally, delaying the start of BART service on weekdays from 04:00 to 05:00, a change that Andrew Miller mentions, didn't assure on-time completion of the Transbay Tube earthquake retrofit. The project wrapped up a couple years late, in 2024, and the lost hour of service was never restored.
Before the Transbay Tube retrofit project, during, and since, we've had full-weekend maintenance closures far away from the Tube, so it seems the extra hour for maintenance hasn't been used optimally. Meanwhile, BART stubbornly ignores differential demand on Friday and Saturday nights. Final trains roll out before midnight every day. What little overnight bus service remains in the Bay Area is provided by separate agencies, with their own routes, schedules, stops, and fares, so people gravitate to private ride-share services. So much for adjusting transit service to observed human behavior and actual transportation demand!
This made me feel surprisingly optimistic on a Friday, GO transit fumbles aside
Excellent analysis. Most transit agencies, especially GO Transit/Metrolinx, is stuck in pre-Pandemic, tidal flow commuter rail thinking, and unwilling to accept that moving people with trains, no matter what day of the week or time, is really important.
The most expensive seat is not the first, rather the unsold seat. The former is finite, the latter is infinite, since it can never be recouped. Finding out ways to fill those unsold seats through a demand driven model (not supply-side) is the key to approaching breakeven or staying relatively solvent as a public good. Unfortunately, not even Alon appears to model this way from what I can tell. We proved this with the PATH. But our work isn't done.