A wider road fills up and the mechanism is not mysterious

These are listed in the order worth acting on, which with induced demand is not the order they are usually presented in.

Capacity lowers the price of a trip

Time is the main cost of driving, and adding lanes reduces it, which makes the trip more attractive at the margin. Trips that were not worth making, or were made by another mode, become worth making by car.

This is not a special property of roads; it is what happens whenever the cost of something falls and demand is not fixed. The confusion arises because traffic engineering historically treated demand as a fixed quantity to be accommodated.

The response arrives in stages

In the first weeks, drivers switch routes to the faster one, which makes the new road look busy and the parallel roads look better. Over months, departure times spread back toward the peak because the peak is no longer as painful.

Over years, people take jobs further away, developers build at the edge, and the land use pattern adapts to the new travel time. Each stage adds traffic, which is why a widened corridor often returns to its previous congestion level only after several years.

The evidence is unusually consistent

Studies across different countries and decades have repeatedly found that vehicle distance travelled rises with lane capacity over the medium term. Estimates of the size of the effect vary considerably with context, and corridors with no viable alternatives behave differently from congested urban networks.

On the ground, the direction of the effect is not seriously contested; the magnitude in any specific case is. Presenting it as a universal one-for-one cancellation overstates a finding that is strong enough without exaggeration.

Removing capacity works the same way in reverse

When road space is taken away for a plaza, a bus lane or a cycle track, some traffic disappears rather than relocating. Reviews of road closures and reallocations have found this repeatedly, with the effect varying by how good the alternatives are. The mechanism is the same one running backwards: the trip got more expensive and some people stopped making it.

Modelling that assumes fixed demand predicts gridlock in these schemes and is regularly wrong in the same direction.

Models encode the assumption

A traffic model with a fixed demand matrix will always show that widening reduces congestion, because it cannot represent the trips that widening creates. Variable demand models exist and are more demanding of data, calibration and time.

Appraisal frameworks in a number of countries now require them for large schemes, which changed several business cases materially. When a scheme is justified by a fixed-demand model, that is worth knowing before reading the benefit figure.

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

What capacity is still good for

Induced demand is not an argument against all infrastructure; it is an argument about what infrastructure delivers. Capacity given to a mode that occupies little space per person — rail, bus lanes, cycle tracks — induces demand for that mode instead.

Pricing rather than expansion manages road demand directly, which is why congestion charging and road pricing keep returning to the agenda. The honest case for a road is usually access to somewhere, not the permanent elimination of congestion.

Everything above, in order of what to do first

1. Capacity lowers the price of a trip. Time is the main cost of driving, and adding lanes reduces it, which makes the trip more attractive at the margin.
2. The response arrives in stages. In the first weeks, drivers switch routes to the faster one, which makes the new road look busy and the parallel roads look better.
3. The evidence is unusually consistent. Studies across different countries and decades have repeatedly found that vehicle distance travelled rises with lane capacity over the medium term.
4. Removing capacity works the same way in reverse. When road space is taken away for a plaza, a bus lane or a cycle track, some traffic disappears rather than relocating.
5. Models encode the assumption. A traffic model with a fixed demand matrix will always show that widening reduces congestion, because it cannot represent the trips that widening creates.
6. What capacity is still good for. Induced demand is not an argument against all infrastructure; it is an argument about what infrastructure delivers.

The takeaway

Ask what the road is for. If the answer is less congestion, the answer is probably wrong.

The design decision is visible long after the people who made it have gone.