A late bus costs more than a slow one

The theory of service reliability is well covered elsewhere. This is about the version you meet in practice.

People budget for the bad day

Somebody who must arrive at a fixed time leaves early enough to survive a delayed trip, and that margin is spent on every journey. A service that is slow but consistent therefore costs less total time than a faster one that occasionally fails badly.

This is why average journey time is a poor description of what a network delivers to the people using it. It also explains why passengers describe as unreliable a route that performs well on most days and collapses on a predictable few. The margin never appears in any operator statistic, because it is time spent waiting somewhere at the destination rather than on the vehicle.

Bunching is a feedback loop

A bus that falls slightly behind meets a larger crowd at the next stop, spends longer loading and falls further behind. The bus following it meets a smaller crowd, loads quickly and catches up, until the two arrive together with a long gap behind them. The process is self-reinforcing rather than random, which is why bunching appears reliably on frequent routes and rarely on infrequent ones.

No amount of driver instruction prevents it, because the mechanism operates through passenger numbers rather than through driving behaviour. Holding a leading vehicle at a controlled point, or having a supervisor manage headways in real time, interrupts the loop directly.

Padding is real time that somebody spends

Timetables include recovery time so a late vehicle can return to schedule, and that padding sits mostly at the ends of routes. Too little padding and delays propagate through the whole day, too much and vehicles wait at termini while passengers wait for them elsewhere. Padding is also expensive, because a vehicle standing at a terminus is a vehicle and a driver not producing service.

In practice, operators facing worsening traffic often add running time rather than fixing the road conditions, which makes the timetable honest and the service slower. A route whose scheduled journey time has crept upward across successive timetables is telling you about the street rather than the operator.

Reliability is won on the street

The largest sources of variation are traffic congestion, signal delay and the time spent stationary at stops while people board. Bus priority at signals, continuous bus lanes and off-vehicle ticketing each attack one of those sources directly. Measures that only cover part of a route deliver part of the benefit, since the vehicle simply loses the time somewhere else.

Where a route crosses several administrative boundaries, priority is often implemented in some sections and refused in others, which is why corridors perform unevenly.

The unglamorous conclusion is that reliability is a highway engineering outcome as much as an operational one.

Measuring the right thing

Punctuality against a timetable is the traditional measure and it suits infrequent services where people consult a departure time. On frequent routes, excess waiting time is the better measure, because it captures what passengers actually experience while standing at a stop.

Over a decade, averages hide the tail, so a percentile measure showing the worst journeys reveals the variability that drives passenger behaviour. Automatic vehicle location data makes all of this measurable continuously, and publishing it is a political decision rather than a technical one. What gets measured tends to get managed, so the choice of indicator quietly determines which improvements an operator pursues.

The data here is patchy — most cities do not publish it consistently.

What a passenger can observe

Stand at a stop through several headways and note the gaps, because unevenness rather than length is usually the source of complaint. Watch how people board, since queuing to pay at the front door is often the largest controllable delay at a busy stop. Look at where the bus lane ends, as the final approach to a junction is where most of the priority benefit is won or lost.

Measured properly, note whether real-time predictions drift, because predictions that jump backwards indicate the vehicle is stuck rather than the system being broken. These observations are available to anybody with ten minutes and they describe a route more honestly than an annual performance figure.

The takeaway

Judge a route by its worst trips. That is the one passengers plan their day around.

Cities are built by a thousand small permissions, not one big plan.