Question 1 of 7: Land Use, Mode Choice and Traveller Information

Question 1: Land Use, Mode Choice and Traveller Information (20 marks)

Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.

(a) The land use – transportation feedback cycle

Travel is a derived demand: almost nobody travels for its own sake, so the pattern of trips in a region is a direct consequence of where activities are located. Land use development therefore drives the transportation system in a very literal sense. A new subdivision on the edge of a city creates a fixed number of households, each generating a predictable number of daily trips; a regional shopping centre or a hospital creates a concentrated set of attractions. The quantity, the origin and destination pattern, the time of day and the mode of those trips all follow from the type, the intensity and the spatial arrangement of the development. Density, land use mix and street connectivity matter as much as raw floor area, because a dense mixed-use district generates far more walk and transit trips per dwelling than a low-density single-use suburb generating essentially only automobile trips.

The causality also runs the other way, and this is the half that planners most often underestimate. Improved transportation facilities reduce the generalised cost of reaching a parcel of land, which raises that parcel's accessibility, and accessibility is one of the principal determinants of land value and of development pressure. A new interchange, a rapid-transit station or a widened arterial makes previously marginal land economically viable for development, so residences, retail and employment relocate toward it. The additional development then generates new trips on the very facility that was built to relieve congestion. This is the mechanism behind induced demand: part of the traffic growth that appears after a capacity improvement is not diverted from elsewhere but genuinely new, created by the land use response.

The two directions together form a closed feedback loop, usually drawn as the land use – transportation cycle: land use determines travel demand, travel demand and the supplied network determine accessibility, accessibility drives land value and location decisions, and location decisions change land use. The loop closes over a period of years to decades, far longer than the annual cycle over which traffic is normally observed.

The consequence for travel demand forecasting is decisive. The conventional four-step model treats the land use inputs — zonal households, employment and income — as exogenous, fixed by a separate land use forecast, and then predicts trips on a fixed network. If the network is being changed by the very project under evaluation, that assumption is internally inconsistent: it holds land use constant when the project itself will move it. The forecast then systematically understates future traffic on an improved facility, overstates the durability of the congestion relief, and consequently overstates the benefit-cost ratio of road widening while understating the ridership benefit of a transit investment that would concentrate development around stations. Sound long-range practice therefore either iterates between an integrated land use model and the travel model until the two converge, or at minimum tests alternative land use scenarios so that the sensitivity of the forecast to the induced development is visible to the decision maker.

(b) The IIA property of the multinomial logit model

The multinomial logit model assigns to alternative $i$ the probability

from which the odds between any two alternatives are

The ratio depends only on the utilities of those two alternatives. It is completely unaffected by whether a third alternative exists, by how attractive it is, or by how similar it is to either of them. That is the independence of irrelevant alternatives, and it follows directly from the assumption that the random error terms of the utility functions are independently and identically distributed Gumbel variates.

IIA is unrealistic whenever two alternatives share unobserved attributes, because their errors are then correlated rather than independent. The textbook illustration is the red bus – blue bus paradox. Suppose travellers split evenly between automobile and a bus service, so each carries half the market. Now paint half the fleet a different colour and treat the two colours as separate alternatives. Since colour does not enter the utility function, the three alternatives have equal utility, and the logit model predicts a one-third share for each — the automobile share collapses from 50 per cent to 33 per cent simply because a bus was repainted. The true answer is obviously that the automobile keeps its half and the two bus alternatives share the other half, because the two buses are near-perfect substitutes for one another and poor substitutes for the car. The same failure appears in real work in a less comic form: adding a light-rail line to a network that already has a bus service draws proportionately from every existing mode, whereas in reality it should draw disproportionately from bus, whose users share unobserved tastes for transit. Any policy forecast for a new mode that resembles an existing one is therefore biased, usually optimistically.

The limitation can be overcome in several ways. The most common is the nested logit model, in which alternatives that share unobserved attributes are grouped into a nest — a transit nest containing bus and rail, competing at the upper level against the automobile nest. The logsum of the lower nest carries the composite utility of the group upward, and the nest's dissimilarity parameter measures how substitutable its members are, restoring IIA within a nest but not between nests. Where the nesting structure is not obvious, the cross-nested or generalised extreme value family allows an alternative to belong partly to several nests. Alternatively the error correlation can be modelled directly: the multinomial probit model admits an unrestricted covariance matrix, and the mixed logit or random-parameters logit represents correlation and taste heterogeneity by integrating logit probabilities over a distribution of coefficients, at the price of simulation-based estimation. A cheaper partial remedy is to enrich the deterministic utility with mode-specific constants and with attributes that capture the shared characteristics, so that less of the similarity is left in the unobserved term. Finally, the IIA assumption should be tested rather than assumed — the Hausman-McFadden specification test compares coefficients estimated on the full choice set with those estimated on a restricted set, and a significant difference is direct evidence that IIA has been violated.

(c) Advanced traveller information systems and demand management

An advanced traveller information system is the branch of intelligent transportation systems that collects network conditions in real time — from loop detectors, probe vehicles, transit automatic vehicle location and incident reports — and delivers them to travellers through variable message signs, transit arrival displays, broadcast advisories and, above all, smartphone applications. It manages demand not by adding capacity but by changing the choices travellers make with the capacity that exists, and it acts on every dimension of the trip decision.

The most immediate effect is on route choice. Travellers who learn of a queue ahead divert to alternative paths, which shifts flow from the saturated link toward under-used ones and moves the network from the poorly informed state toward a genuine user equilibrium. Because delay grows sharply and non-linearly once a link approaches capacity, quite modest diversion can produce disproportionate reductions in total network delay. The second effect is on departure time: a traveller who is told that the peak has not yet cleared can delay or advance a trip by fifteen minutes, flattening the peak and raising the utilisation of the shoulder periods without any physical works. The third and most valuable effect for sustainability is on mode choice. Real-time transit arrival information is consistently shown to reduce the perceived burden of waiting, which the mode-choice models of Question 6 weight more heavily per minute than in-vehicle time; combined with journey planners that present a transit or cycling itinerary beside the driving one, and with real-time parking availability that exposes the true cost of driving, this shifts trips to lower-emission modes. A fourth effect is trip suppression or consolidation: a traveller who sees that a corridor is closed may cancel a discretionary trip, chain it with another, or substitute a telephone call.

These outcomes serve sustainability on all three of its usual dimensions. Environmentally, reduced congestion and shorter queues cut fuel consumption and greenhouse-gas and criteria-air-contaminant emissions, which are highest in stop-and-go conditions, and any mode shift toward transit and active modes reduces emissions per passenger-kilometre further. Economically, deferring or avoiding capacity expansion by extracting more service from existing infrastructure is far cheaper per unit of benefit than construction, and travel-time reliability improves, which for freight and commercial vehicles is worth more than average travel time. Socially, better information reduces the anxiety and the schedule buffer that travellers build into unfamiliar or infrequent trips, and it improves access for those who do not own a car. Advanced traveller information is also the natural complement to more direct demand-management measures, since the responsiveness of travellers to congestion pricing or to high-occupancy-toll lanes depends on their being told, in real time, what the price and the saving actually are.

Two limitations deserve mention in an honest answer. Information changes behaviour only if it is credible, timely and specific, so a system that issues stale or vague advisories is quickly ignored. And where a large share of drivers act on the same advice simultaneously, the diversion can overload the alternative route and oscillate — the reason modern systems increasingly recommend individualised rather than uniform routings.