Transportation Demand Analysis

The goals of transportation demand analysis are to describe travel in

meaningful terms, to explain travel behavior, and on the basis of an
understanding of travel behavior, to predict demand for various types of
transportation services.

Travel Behavior

In most cases, the trip is considered to be the basic unit of travel behavior.
Trips involve movement from a single origin to a single destination, and are
usually described in terms of their origins, destinations, purposes, and times
of occurrence, travel modes, and routes. In some cases, more complex units
of travel known as trip chains or trip patterns are studied. These are sets of
trips, usually beginning at the traveler’s home and proceeding to several
destinations in sequence before returning.

Transportation demand analysis attempts to explain travel behavior in

primarily economic terms. In economics, the theory of consumer behavior is
based on the concept of utility, which is expressed in mathematical terms by
means of a utility function. Reduced to its simplest terms, the utility function
expresses the consumer’s indifference between various alternative choices or
the attributes of these choices. The utility function

u = f(x, y)

implies that the individual is indifferent in a choice between x and y for all
combinations which produce the same value of u. Furthermore, the
individual will prefer all combinations u1 of x and y to all combinations u2 of
x and y if

u1 = f1(x, y)>u2(x, y)

In the analysis of travel choices, it usually applies to mutually exclusive

alternatives, since it is impossible to make two trips at the same time. In this
case, the idea is that an individual will make one trip only if the utility of
traveling exceeds that of not traveling and that where different trips are
possible, the one with the greatest utility will be chosen.

1
It is also generally assumed that travel is not undertaken for its own sake.
Rather, it is considered to be a secondary good and its utility is said to be
derived utility. This means that the utility of any particular trip is partially
dependent on the activity it makes possible, and the way in which the
activity is valued by the individual. Thus a work trip is said to be motivated
by the thought of money to be earned, and to be worth no more than the
amount that would be lost if it were not made.

On the other hand, trips also involve costs, or disutilities, which must be
subtracted from their values. One of these is obviously the monetary cost of
the trip; another which is usually considered is the amount of time it takes.
Since some research has indicated that time is valued differently in different
uses, the time cost of a trip is sometimes broken into components such as
time spent riding and time spent walking or waiting for service.

Travel Demand Modeling

The most common approach has been to structure the choice sequence by
first stratifying trips by purpose and time of day. Such stratification is
usually quite simple, involving only a few categories. For instance, trip
purpose may be stratified only into work and nonwork trips. A slightly more
elaborate scheme would recognize work, shopping, personal business,
educational/religious, and social/recreational trips. As can be seen, such
schemes are based primarily on the frequency, regularity, and time of day
that the different types of trips are made. Time-of-day stratification has often
been between work-trip peaks and off-peak periods. In addition, trips are
sometimes classified according to whether they take place on weekdays or
weekends.

For each combination of trip purpose and time of day, separate models are
usually developed for the number of trips made (trip generation), their
origins and destinations (trip distribution), the mode used (mode choice) and
the route followed (trip assignment). Such models require information about
socioeconomic conditions and land use, and where they are intended to
predict travel demands very far into the future, it may also be necessary to
model these conditions.

A second major simplification has to do with the uniqueness of individual

preferences. In theory, every individual has a unique utility function for
travel. Not only would it be virtually impossible to quantify this utility

2
function, but the results would be of little interest. Instead, transportation
demand modeling is based on structuring the population into more or less
homogeneous groups of people and quantifying utility functions for
representative individuals.

Two types of grouping have traditionally been used. In the first, the
population is grouped into geographical zones, and the average
characteristics of the residents of the zone are used to describe it. In the
second, an attempt is made to identify subgroups, which may be present in
all the zones, whose socioeconomic and trip-making characteristics are
similar. Such groups are known as market segments.

For purposes of establishing market segments, the population is classified

according to characteristics likely to influence travel choices, activity
patterns, and relative preferences for various trip characteristics.
Classification schemes have involved categories such as automobile
availability, family size and structure, number of employed persons in the
household, income, and ethnicity. Since it is necessary to know the
geographical distribution of the various market segments, market
segmentation is usually based on characteristics recorded in census data.

The reasoning behind market segmentation is that the available travel

options, activity patterns, and preferences of individuals with similar
socioeconomic characteristics are apt to be similar. For instance, individuals
without access to automobiles face a very different set of travel choices and
travel costs than those who have access to them. To give another example,
individuals with high incomes are more likely to have automobiles than
those with low incomes, and are apt to place a higher value on time relative
to money.

Where the population is divided into market segments, demand models

usually attempt to predict the probability that a representative individual
from a given market segment will make a particular travel choice. Such
models are known as disaggregate models, and usually involve an explicit
utility function. It is sometimes said that they are models of individual
behavior, but this is not strictly true, since they assume that all individuals
within a particular market segment face the same choice set and have the
same preferences for whatever travel characteristics are included in the
model.

3
On other hand, aggregate models attempt to predict numbers of trips from
the average characteristics of the trips and of the population of a
geographical zone.

Since the goal of demand modeling is to predict the number of trips made,
the results of disaggregate models must be aggregated. The expected number
of trips from any market segment is the number of individuals in the market
segment times the probability any individual makes a trip. The total number
of trips expected is the sum of the number expected from each market
segment.

The development of predictive travel demand models involves statistical

analysis and is based on the assumption that the future will in some way
resemble the past. The general strategy in each case is to construct a
mathematical model of current travel behavior; such models relate current
trip-making behavior to a number of independent variables by means of
some sort of regression analysis. Determination of regression coefficients
through analysis of data about present travel demand patterns is called
calibration of the model. In order to predict future trips, the relationships
expressed by the model are assumed to remain unchanged over time;
changes in the independent variables are predicted; and, from, these a value
of the dependent variable is calculated.

Trip Generation Models

Trip generation models are intended to predict the total number of trips
produced or attracted by a zone. Trips are usually thought of as being two-
way excursions originating at the trip-maker’s home. They are said to be
produced by residential development and attracted by economic or other
activity. Trips are normally stratified by purpose; for each trip type, the
number produced in a particular zone is assumed to depend on the size and
the characteristics of the zone’s resident population. The number of trips
attracted to the zone is assumed to depend on the characteristics of the
economic activity. In most cases, trip generation models are of the general
form

Pi = f (z1, z2, z3……)

Ai = f (z’1, z’2, z’3…)

4
where Pi = trip productions from zone i
Ai = trip attractions to zone i
zi = predictive factors for productions
z’i = predictive factors for attractions

The most common form of trip generation model is a linear function of the
form

Pi = ∑j αjzij

Where the αj’s are coefficients to be determined through regression analysis

Factors which have commonly been used to predict trip productions include
zone population, the average number of workers in households in the zone,
automobile availability, income, and the fractions of the zone population
with different occupations. Factors which have commonly been used to
predict trip attractions include floor space, total employment in the zone, or
even the amount of developed land.

An alternative technique for modeling trip generation is use of cross-

classification or category analysis. In this technique, the population is
broken down into a set of classes known to be correlated with trip-making
behavior. For instance, households might be broken down according to the
number of persons in the household and the number of vehicles available.
For each combination of classes, a trip rate is calculated.

This results in a Table similar to Table 13.1. The resulting cross-

classification table can then be used as a disaggregate model for predicting
trip generation from zones with different distributions of household
characteristics. For instance, suppose the trip rates given in Table 13.1 apply
to a travel zone whose population breakdown is given in Table 13.2. The
average number of trips per household may be estimated by multiplying the
trip generation rate for each category by the fraction of the population in that
category. That is,

Pi = ∑j ∑k rjkpjk

Where rjk = trip rate for cross-classification table cell jk

pjk = fraction of population for cross-classification table cell jk

5
Table 13.1

Cross-classification table giving trip rates in trips per household per day

Vehicles available per household

Persons per household 0 1 2 or more

1 1.02 1.90 2.10

2 2.12 3.25 3.70
3 2.15 3.75 3.90
4 or more 3.96 5.00 6.54

Table 13.2

Fraction of population of hypothetical zone in various household size

and vehicle availability categories

Vehicles available per household

Persons per household 0 1 2 or more

1 0.05 0.21 0.02

2 0.03 0.16 0.13
3 0.02 0.10 0.13
4 or more 0.01 0.08 0.06

For the hypothetical zone represented by Table 13.2, the average number of
trips per household would be estimated as

P = 0.05(1.02) + 0.21(1.90) + 0.02(2.10) + 0.03(2.12) + 0.16(3.25) +

0.13(3.70) + 0.02(2.15) + 0.10(3.75) + 0.13(3.90) + 0.01(3.96) + 0.08(5.00)
+ 0.06(6.54) = 3.31 trips/day

6
Trip Distribution Models

Trip distribution models are intended to predict zone-to-zone trip

interchanges. The final product is a projected origin-destination matrix of the
form shown in Table 13.3, where T ij refers to the number of trips with
origins in zone i and destinations in zone j.

There are several types of trip distribution models but the most important is
the gravity model. Of these, the gravity model is by far the most popular.
The gravity model has the following alternative forms:

Tij = Pi (AjFij)/Σj (AjFij) (13.9)

Tij = Aj (PiFij)/Σi (PiFij) (13.10)

Where Tij = trips from zone i to zone j

Aj = trip attractions in zone j
Pi = trip productions in zone i
Fij = impedance of travel from zone i to zone j

The impedance of travel from zone I to zone j is usually some function of

the travel time or generalized cost of travel between the two zones. One
common form is

Fij = Cij-α

Where Cij = generalized cost function for travel from zone i to zone j and α =
model parameter to be determined to be determined by calibration.
Originally, α was assumed to be 2.0, by analogy to the inverse square law of
gravitation-hence the name gravity model.

The generalized cost term Cij most commonly represents the travel time
between zone I and zone j; it can also represent a money cost, some weighed
combination of time and money, or possibly other trip characteristics such as
comfort.

Example: Using a gravity model with an impedance term of the term C -α,
estimate the number of trips from zone I to all the other zones, α = 1.90.

7
 Travel time to
Zone 1, min Productions Attractions
Zone___________________________________________________________________

1 20 000 10 000
2 10 15 000 30 000
3 20 30 000 18 000
4 15 25 000 10 000
5 30 18 000 40 000

Since the problem is to find trips from zone 1 to all other zones, distribute
productions from zone 1. Use

Tij = P1 (Aj/Cijα)/∑j (Aj/Cijα)

Calculate T1j Procedure:

1. Calculate Aj/Cijα for all j.

2. Sum over all j.
3. Calculate Aj/Cijα/∑j (Aj/Cijα) for all j; check that it sums to 1.00.
4. Calculate T1j; check that it sums to P1.

Example calculation (for zone 2):

Aj/Cijα = 30 000/101.9 = 377.68

Aj/Cijα/∑j (Aj/Cijα) = 377.68/559.12 = 0.675

T1j = P1 (Aj/Cijα)/∑j (Aj/Cijα) = 20 000×0.675 = 13 500

Zone Aj Cij Aj/Cijα Aj/Cijα/∑j(Aj/Cijα) T1j

1 10 000
2 30 000 10 377.68 0.675 13 500
3 18 000 20 60.72 0.109 2 180
4 10 000 15 58.27 0.104 2 080
5 40 000 30 62.45 0.112 2 240
∑ 559.12 1.000 20 000

8
9