Travel demand forecasting, and how to read a forecast
Every number in a transportation plan comes out of a model. Models are useful and they are wrong, and the useful part depends on understanding the second half of that sentence.
A travel demand model estimates how many trips will be made, where they will go, by what mode and along which route. It is built from household surveys, employment data, counts and census figures, then calibrated until it reproduces today reasonably well. Having matched the present, it is run forward against assumptions about population, jobs and land use.
Where the uncertainty actually lives
The arithmetic is not the weak point. The assumptions are. A forecast that says traffic grows twenty percent by 2045 is really saying: if population grows as projected, if employment concentrates where the land use plan says, if nobody's travel behaviour changes much. Each of those has been wrong before, sometimes badly.
| Input | How confident | Why it moves |
|---|---|---|
| Population totals | Reasonably high | Demography is slow and well observed |
| Where growth lands | Moderate | Depends on zoning decisions not yet made |
| Employment by sector | Low to moderate | Sensitive to events nobody models |
| Mode choice | Low | Behaviour shifts faster than models update |
| Trip generation rates | Moderate | Remote work moved these sharply and permanently |
Probability is the honest framing
A single forecast line invites false precision. Ranges and scenarios are more truthful: a high growth case, a low growth case, and a central case, each with its own implications. When a plan presents one number without a range, that is a presentational choice, not a modelling result.
The same caution applies anywhere outcomes are quoted as odds. A probability is a statement about a distribution of outcomes, not a prediction of the next one, and confidence intervals get dropped in summary long before they get dropped in the analysis. Reading any forecast well means asking what was assumed, how wide the range was, and who chose which number to print.
Common questions
Why do forecasts keep over-predicting traffic?
Historically most have, because growth in vehicle miles slowed while models continued to extrapolate earlier trends. The correction has been gradual and is still working through.
Can a model justify any conclusion?
Not quite, but assumptions have enough room that two defensible runs can support different projects. This is why the assumptions, not the outputs, deserve the scrutiny.