The Recovery Narrative Is Misleading
Transit agencies have spent the past several years measuring "recovery" against 2019 baselines — percent of pre-pandemic ridership restored, month over month. That framing made sense in 2021 and 2022, when the question genuinely was whether people would come back to transit at all. By 2025, continuing to measure against 2019 as the target misses what has actually happened: transit ridership hasn't failed to recover, it has restructured. The distribution of demand across the network, across the week, and across the day looks fundamentally different from 2019 — and treating that as a gap to close rather than a new demand pattern to serve is leading agencies toward the wrong service design decisions.
I want to be clear about what the data actually shows, because the picture is more complicated than either "transit is dead" or "ridership is back." What NTD data and GTFS-RT archives from agencies across the country tell us, as of 2025, is that ridership on most fixed-route systems is within 75–90% of 2019 aggregate levels — but the composition of that ridership has shifted in ways that require different thinking about network design, frequency allocation, and span of service.
Temporal Restructuring: The Collapse of the Traditional Peak
The most analytically significant change in post-pandemic ridership patterns is in the temporal distribution of demand. Pre-2020, most large transit systems showed a pronounced bimodal pattern — AM peak (roughly 7–9 AM) and PM peak (5–7 PM), with a pronounced valley in midday, and significantly lower ridership on weekends than weekdays. That bimodal pattern was a direct reflection of traditional office commute patterns: the majority of transit's workhorse ridership segment was commuting downtown Monday through Friday on a predictable schedule.
What GTFS-RT archives and APC counts show for 2023–2025: the AM peak has compressed and flattened. The sharp 7:30–8:30 AM boarding spike that used to characterize commuter corridors is now a broader, lower, more spread-out morning distribution. The PM peak has shifted later and flattened similarly. On some systems, the weekday–weekend ridership differential has narrowed substantially — in some cases, weekend ridership has recovered to a higher percentage of pre-pandemic levels than weekday peak ridership. Midday ridership — historically the segment associated with healthcare workers, service industry employees, and people without cars — has proven remarkably stable and, on some corridors, grown.
The practical implication is significant: a service planning model that allocates frequency primarily to serve a sharp AM/PM commuter peak is mismatched to a demand distribution that's now flatter and more spread across the day. Agencies that have adjusted headways based on 2019 peak assumptions are running too much service in the peak and too little in the shoulders and midday.
Spatial Restructuring: Corridor Winners and Losers
The temporal shift in demand has a spatial correlate. Corridors that were primarily serving downtown office commuters have generally recovered more slowly than corridors serving healthcare facilities, grocery stores, service employment centers, and community colleges. This is partly a work-from-home effect — office workers with the flexibility to work remotely have reduced their transit commute frequency — and partly a longer-term shift in which trips people choose to make by transit vs. by car.
Consider a demand surface comparison of 2019 APC data vs. 2024 APC data for a mid-size metro that shared both datasets. The downtown-focused radial corridors are running at 70–80% of 2019 ridership. The routes serving hospital campuses, community college corridors, and dense residential areas in the urban periphery are running at 95–110% of 2019 ridership. The aggregate systemwide number (82% of 2019) obscures a bimodal distribution: some corridors have genuinely restructured demand and need to be rethought, while others are essentially at pre-pandemic levels or above.
Agencies that are making uniform service cuts to hit aggregate cost targets are cutting service from corridors with strong demand recovery in order to preserve (underutilized) service on downtown-oriented routes. The data supports a different conclusion: reallocate frequency from underperforming downtown-peak routes to the corridors where demand has proven durable.
The Work-From-Home Variable: Not as Simple as It Looks
The standard narrative attributes post-pandemic ridership changes primarily to work-from-home adoption among office workers. That's a real factor, but it's incomplete. Several other structural changes are operating simultaneously:
- Fare structure changes: Many agencies that suspended fares during the pandemic have not fully restored pre-pandemic fare levels, or have restructured their fare systems in ways that changed rider behavior. Reduced fares tend to increase ridership among price-sensitive segments; free fares attract new trips but also change the trip purpose mix.
- Service reliability: Post-pandemic operator shortages have reduced service reliability on many systems, particularly during peak periods. Passengers who were unreliable service experience are less likely to depend on transit for time-sensitive commute trips — but may continue to use transit for more flexible trips. This creates a selection effect in the ridership data: the transit riders who remain are disproportionately those making non-time-sensitive trips.
- Demographic shifts in the transit-dependent population: Net migration patterns since 2020 have shifted the distribution of transit-dependent populations in ways that vary significantly by metro area. Agencies that are seeing demand growth in outer-ring and suburban areas may be responding to demographic change as much as to mode shift behavior.
We're not saying work-from-home is irrelevant to ridership trends. It clearly is a major factor for downtown-oriented commuter services. But treating it as the only explanation for ridership restructuring leads planners to expect a return to 2019 patterns when office occupancy rates "normalize" — and that expectation may be misguided. The data does not yet support a clear trend toward commuter ridership recovery.
What Good Ridership Analysis Looks Like in 2025
Planning decisions made against 2019 baselines are increasingly unreliable. What the data says agencies should be doing instead:
- Replace 2019 baseline comparisons with trailing-12-month trend analysis. The relevant question isn't whether you're at 85% of 2019 — it's whether demand on a given corridor is stable, growing, or declining over the past year. A corridor at 80% of 2019 but trending upward over 12 months needs different treatment than a corridor at 90% of 2019 but declining.
- Decompose system-level ridership numbers by time-of-day and day-of-week segments. Aggregate numbers obscure distributional shifts that matter for service design. The morning peak, midday, evening, and weekend segments behave differently and need to be analyzed separately.
- Use H3 demand surface mapping to identify spatial concentration changes. Which cell clusters are seeing sustained ridership and which have structurally declined? This is the input to a corridor-level reallocation conversation, not just systemwide headway adjustments.
- Track ridership elasticity with respect to service changes. Post-pandemic rider behavior may be more elastic (more responsive to service improvements) than pre-pandemic behavior, because a larger fraction of current riders made a deliberate mode choice to return to transit. This is a hypothesis worth testing empirically as agencies make service changes.
The Forecasting Challenge
Five years of disrupted ridership data have substantially degraded the reliability of traditional travel demand models that were calibrated on pre-2020 behavior. The mode choice models, trip generation rates, and auto/transit shares that underpin regional travel forecasting were built on observed behavior from a world that no longer exists.
This doesn't mean demand modeling is broken — it means models need to be recalibrated against post-pandemic observed behavior before they can reliably support capital investment decisions. Agencies and MPOs that are using 2018 or 2019 calibration data for their travel demand models should be treating those models' outputs with considerably more uncertainty than they typically do. The range of plausible outcomes is wider than the confidence intervals suggest.
For short-range planning purposes — headway adjustments, span changes, route restructuring — the most reliable input is actual ridership data from the past 12–24 months, analyzed at the granularity of individual route segments and time periods. For long-range capital planning, the demand uncertainty needs to be acknowledged explicitly in project justifications, with sensitivity analysis showing how project benefits change under a range of ridership recovery scenarios.
Transit demand in 2025 isn't recovering toward 2019 — it's stabilizing around a new distribution. Planning against the old baseline means serving where people used to be, not where they are.