Traffic Engineering and Its Discontents#
SLOP2418 — Urban Design Studio
Week 6 · Idris Fenn · 29 March 2027
direct readout · derived measure · worked hypothetical
A separate widening hypothetical#
Worked hypothetical: these figures illustrate the mechanism; students substitute their own save’s numbers.
| Kerrow corridor at Wickstead | Before (Wk 5, pre-change) | Day 1 after widening | Day 30 after widening |
|---|---|---|---|
| Lanes each way | 2 | 3 | 3 |
| Peak volume, veh/h | 2,040 | 2,040 | 3,050 |
| Volume ÷ capacity | 0.98 | 0.65 | 0.97 |
| Average peak speed | 24 km/h | 41 km/h | 25 km/h |
| Ashgrove back streets, veh/h | 480 | 210 | 140 |
| Bus line passengers, daily | 3,900 | 3,510 | 3,180 |
The argument#
Capacity does not reduce congestion. It relocates the demand that produced it, and then enlarges it.
What the pathfinder is doing#
- Wardrop’s first principle, 1952: at equilibrium, no driver can reduce their own journey time by switching route. All used routes cost the same; unused routes cost more.
- The model’s traffic AI implements exactly this. Agents cost a route and take the cheapest one available, and they re-cost it when the network changes.
- So a widening does not give the existing traffic a faster trip. It lowers the cost of one route until enough traffic has moved onto it that the costs are equal again.
- Equilibrium is not a goal the model is pursuing. It is a property of everybody doing the obvious thing.
Downs’s law#
On urban commuter expressways, peak-hour traffic congestion rises to meet maximum capacity.
— Anthony Downs, “The Law of Peak-Hour Expressway Congestion,” Traffic Quarterly (1962)
Triple convergence#
| Convergence | The driver’s decision | The view that shows it |
|---|---|---|
| Spatial | Take the improved route instead of the back streets | Traffic volume layer, read on Ashgrove |
| Temporal | Leave at 08:00 again instead of 07:20 | Volume by hour across the peak |
| Modal | Drive instead of taking the bus | Transportation Overview, passengers per line |
Downs proposed three. The model supplies a fourth, more slowly.
The arithmetic: capacity#
The Wickstead junction, worked from first principles.
| Term | Value | Source |
|---|---|---|
| Base saturation flow | 1,900 veh/h per lane | standard signalised-arterial figure |
| Effective green ratio, g ÷ C | 0.55 | measured at the Wickstead signal |
| Effective capacity per lane | 1,900 × 0.55 = 1,045 veh/h | |
| Capacity, 2 lanes each way | 2,090 veh/h | before |
| Capacity, 3 lanes each way | 3,135 veh/h | after |
| Capacity increase | +50% |
The arithmetic: demand#
Duranton and Turner (2011) estimate the elasticity of vehicle-kilometres travelled with respect to lane-kilometres at approximately 1.0. The UK Standing Advisory Committee on Trunk Road Assessment (Trunk Roads and the Generation of Traffic, 1994) reached broadly compatible conclusions from different data.
| Elasticity assumed | Volume increase | Long-run peak volume | Volume ÷ capacity | Relief retained |
|---|---|---|---|---|
| 0.0 (the project’s business case) | 0% | 2,040 | 0.65 | all of it |
| 0.5 (short-run, optimistic) | +25% | 2,550 | 0.81 | about half |
| 0.8 | +40% | 2,856 | 0.91 | about a fifth |
| 1.0 (Duranton and Turner) | +50% | 3,060 | 0.98 | none |
The top row is the row every widening is justified on. The bottom row is the row the literature supports.
Where the extra 1,020 vehicles came from#
Decomposition of the +50% observed at day 30, in vehicles per hour at the peak.
| Source | veh/h | Downs’s category | Where you can see it |
|---|---|---|---|
| Re-routed from the Ashgrove streets | +340 | spatial | Ashgrove volume, 480 → 140 |
| Shifted off the Pelling Street bus | +180 | modal | Bus daily passengers, 3,900 → 3,180 |
| Re-timed into the peak from the shoulders | +220 | temporal | 07:00–08:00 volume falling, 08:00–09:00 rising |
| Longer and newly generated trips as households relocate | +280 | land use | Residential demand and new occupancy at the Denton Hill edge |
| Total | +1,020 | = 50% of 2,040 |
Downs–Thomson: the bus set the speed#
- Thomson, 1977, and Mogridge, 1990: on a corridor with a competing transit service, the equilibrium speed of car travel is set by the door-to-door speed of that service. Drivers switch until the two are equally bad.
- So the widening’s long-run car speed was never a property of the road. It was a property of the bus.
- And the widening pulled 180 veh/h off the bus, which cut the line’s revenue, which — for most of the studio — was met by cutting the fleet from the four vehicles issued in Week 5 to three.
- A 6-minute headway became an 8-minute headway. Average wait rose from 3 minutes to 4. The competing trip got worse, so the equilibrium car speed the corridor settles at got worse with it.
- The widening made itself less effective by the mechanism it triggered.
Braess, 1968#
The tempting extra link can make every selfish route slower.
Buchanan, 1963: the other capacity#
- Buchanan’s Traffic in Towns argued that a street has two capacities, and that planning had only ever measured one.
- Vehicular capacity: the volume above which the traffic stops moving. On Pelling Street at three lanes, 3,135 veh/h.
- Environmental capacity: the volume above which the street’s other functions stop working. On Pelling Street, reached at roughly 1,600.
- The second number is not in the Highway Capacity Manual, has no standardised method, and is the number the district is actually built around.
What the third lane cost the frontage#
Pelling Street frontage, before the widening and one simulated year after.
| Measure | 2 lanes each way | 3 lanes each way | Working |
|---|---|---|---|
| Carriageway width | 14.0 m | 20.5 m | one 3.25 m lane each way |
| Pedestrian crossing time at 1.2 m/s | 11.7 s | 17.1 s | +5.4 s of exposure |
| Traffic noise at 10 m | 68 dB(A) | 69.8 dB(A) | 10 log₁₀(1.5) = +1.8 dB |
| Ground-floor commercial holding level 3 or above | 71% | 44% | Land Value info view |
| Frontage land value index | 100 | 87 | Land Value info view |
1.8 dB is barely audible and entirely sufficient.
The tradeoff of the week#
Mobility performance against livability.
The lane raised peak speed by 17 km/h on day one and 6.5 metres of carriageway for good.
What to do instead#
The honest menu, with what each costs.
| Instrument | Effect on the corridor | What it costs |
|---|---|---|
| Retime the Wickstead signal; raise g ÷ C from 0.55 to 0.62 | +13% effective capacity | Almost nothing fiscally; the same elasticity applies, so it buys less and lasts as briefly |
| Reallocate one existing lane to transit | Vehicular capacity down a third; corridor person-throughput up, if the headway is held | Politically the hardest move in the studio, and the one Downs–Thomson says works |
| Shorten the trips: mix uses in the station areas, per Week 4 | Removes demand rather than serving it | Slow — a simulated year or more before it is visible — and it is a zoning project, not a road project |
| Price the peak | The only instrument that addresses demand directly | Not available. The model has no congestion-pricing mechanism, so nothing you argue from this save file can rest on one |
A working week and its evidence record#
- No weighted assessment is due in Week 6. Upload the save, statement and evidence by 21:00; preserve the before-and-after record for the remaining two assessments.
- The next brief is Assessment 2 — Public Realm and Infrastructure Plan, due Week 9 and worth 25%.
- A plan that adds capacity is not penalised. A plan that adds capacity and describes a short-term observation as a permanent congestion improvement has exceeded its evidence.
Before the break#
- Two weeks off. The studio reconvenes Monday 19 April.
- Leave the save file in a state you can return to. Save under a name with the week in it, and write your three peak figures into the process log. You will not remember them.
- Week 7 — Infrastructure You Don’t See, Nadia Ilkhom. Water, sewage, power and the capacity curve. Its deck follows the same evidence discipline as the other eleven.
- Nadia’s framing, which is worth arriving with: every network in this city has a capacity curve, and you have spent two weeks learning what happens at the top of one. The next four are about the networks where failure is not a queue but an outage.