Transit
Ridership forecasts miss in predictable directions
Transport models are built on assumptions that fail in known ways, and the failures cluster rather than cancelling out.
This is less a set of instructions about demand forecasting than an argument, and it is worth saying so at the start.
The argument in brief
- Rail forecasts have historically tended to overestimate and road forecasts to underestimate traffic.
- Optimism bias adjustments exist in several national appraisal frameworks.
- Forecast error is rarely audited after schemes open.
Forecasts are inputs to a decision, not predictions
Demand forecasts are produced to support a funding case, and the people producing them work for the party seeking the funding. That does not require dishonesty to bias results; assumption choices within a defensible range accumulate in one direction.
This is a structural feature of appraisal rather than a claim about individuals. It is why several countries now apply explicit optimism bias uplifts to costs and haircuts to benefits.
Errors have historical patterns
Reviews of completed transport schemes across many countries have generally found rail patronage forecasts optimistic and road traffic forecasts conservative more often than the reverse. The sample sizes and comparability of these reviews are limited, so the finding is a tendency rather than a law.
Over a decade, where the pattern holds, it systematically favours road investment over transit in comparative appraisal. Understanding that pattern matters more than any individual forecast figure.
Land use is the largest source of error
Ridership depends on where people live and work, which depends partly on the transport network being forecast. Models that treat land use as fixed input miss the development that a new line induces, understating long-run demand.
Models that assume optimistic development around new stations overstate it when that development does not arrive. Integrated land use and transport models exist, are data-hungry, and are used inconsistently.
Behavioural change is slow and then sudden
Travel habits are sticky, so new services frequently underperform for the first years and then grow as habits reset. Evaluations conducted shortly after opening therefore capture the wrong part of the curve. Conversely, structural shifts such as remote working can invalidate a forecast entirely within a few years.
Forecasts presented as single numbers rather than ranges conceal both of these effects.
Nobody audits the outcome
Post-opening evaluation is required in some appraisal frameworks and conducted inconsistently, and results are rarely fed back into model calibration. Without that feedback loop, systematic errors persist across generations of schemes. A published comparison of forecast against outturn for completed projects would be the single most useful reform available.
Where such comparisons have been published, they have generally been uncomfortable reading.
What to ask about a forecast
Ask what growth in population and employment it assumes, and whether those assumptions come from an independent source. Ask whether demand is allowed to respond to the scheme or held fixed, and what happens to the case under lower demand.
Ask whether the same team produced the case for the alternative options, and how those were specified. A forecast that survives those questions is worth more than a precise number that does not.
The takeaway
Ask what the forecast assumes about the city, not about the line.
Cities are built by a thousand small permissions, not one big plan.
Questions readers ask
Why do new transit lines sometimes carry fewer passengers than forecast?
Common causes include optimistic land use assumptions, slower habit change than expected, fares or service levels differing from those modelled, and connecting services that did not materialise.
Are road traffic forecasts more accurate?
Not necessarily more accurate, but historically they have erred in the opposite direction. Reviews have generally found traffic growth underestimated, which relates directly to induced demand.
