Institute of Transport Studies
Graduate School of Business
The University of Sydney
NSW 2006
Working Paper
ITS-WP-93-6
STATED PREFERENCE ANALYSIS OF
TRAVEL CHOICES: THE STATE OF
PRACTICE
David A. Hensher
March 20, 1993
Prepared for a Special Issue of Transportation on the Practice of Stated Preference
Methods, guest edited by David Hensher. An earlier modified version was presented as
Stated Preference Analysis David A. Hensher
2
the keynote address to the 1993 National Conference on Tourism Research, held at the
University of Sydney, 19 March 1993. The comments of Jordan Louviere, Lester Johnson, Paul
Hooper, W.G. Waters II and Mark Bradley are appreciated.
NUMBER: Working Paper ITS-WP-93-6
TITLE: Stated Preference Analysis of Travel Choices: The State of
Practice
ABSTRACT: Stated preference (SP) methods are widely used in travel
behaviour research and practice to identify behavioural responses
to choice situations which are not revealed in the market, and
where the attribute levels offered by existing choices are modified
to such an extent that the reliability of revealed preference models
as predictors of response is brought into question. This paper
reviews recent developments in the application of SP models
which add to their growing relevance in demand modelling and
prediction. The main themes addressed include a comparative
assessment of choice models and preference models, the
importance of scaling when pooling different types of data,
especially the appeal of SP data as an enriching strategy in the
context of revealed preference models, hierarchical designs when
the number of attributes make single experiments too complex for
the respondent, and ways of accommodating dynamics (i.e. serial
correlation and state dependence) in SP modelling.
AUTHOR: David A. Hensher
CONTACT: Institute of Transport Studies
Graduate School of Business
University of Sydney NSW 2006
Australia
Telephone: + 61 2 550 8631
Facsimile: + 61 2 550 4013
Stated Preference Analysis David A. Hensher
3
DATE: March, 1993
1. Introduction
It is twenty years since the seminal papers by Davidson (1973) and Louviere et al (1973)
in transportation were published which alerted us to the appeal of methods for evaluating
an individual's response to combinations of levels of attributes of modes of transport
which are not observed in the market, but which represent achievable levels of service.
Widespread interest in this "new" approach to travel behaviour modelling, however, was
slow in developing, in part due to the high agenda interest in the development of
discrete-choice models and activity approaches to the study of the continuous sequences
of human actions over a period of time (see Hensher and Stopher 1979)
1
. Indeed, until
the early eighties, the transport contributions were dominated by publications from
Louviere and his colleagues (see Louviere 1979 for a summary) with an almost universal
application to the study of mode choice (Meyer et al 1978)
2
.
Although it is always difficult to pinpoint the major events which heralded in the
beginning of a widespread interest in SP methods, the motivation seems to have evolved
from a number of applications in which the behavioural response involved an alternative
which was either not currently available (e.g. Louviere and Hensher 1983, Hensher
1982) or where there was difficulty in assessing substantially different attribute mixes
associated with existing alternatives to those observed (e.g. Kocur et al 1982, Hensher
and Louviere 1983, Bradley and Bovy 1985, Louviere and Kocur 1983). An important
paper by Lerman and Louviere (1978) demonstrated the theoretical links between
revealed preference and stated preference models.
Prior to the paper by Louviere and Hensher (1983), the emphasis had been on
judgemental tasks in which a respondent was asked to rate or rank a number of attribute
mixes associated with a particular choice context. The modelling of this data using
standard regression-based estimation procedures required simulation of choice
Stated Preference Analysis David A. Hensher
4
environments in order to predict market share. Louviere and Hensher showed how a
preference experiment (i.e. a number of alternative mixes of attributes) could be
extended to incorporate choice experiments in which an individual chooses from among
fixed or varying choice sets, enabling estimation of a discrete-choice model and hence
direct prediction of market share. Stated choice experiments are now the most popular
form of SP method in transportation and are growing in popularity in other areas such as
marketing, geography, regional science and tourism. The papers by Louviere and
Hensher (1982) and Louviere and Woodworth (1983) have become the historical
reference sources for stated choice modelling in transportation.
The introduction of stated choice modelling using the set of established discrete-choice
modelling tools routinely applied with revealed preference data widened the interest in
SP-methods. For the first time travel behaviour researchers could see the benefit of
stated-preference data in enhancing their travel choice methods. This I would argue was
the major watershed which after 10 years has resulted in widespread acceptance of SP
methods in practice in transportation. A number of monographs and special issues of
journals are now available which capture the major contributions up to the late eighties
(Pearmain et al 1991, Louviere 1988, Bates 1988, and Louviere 1992). Louviere,
Hensher and Shocker (1992) run an annual short course, covering all aspects of stated-
preference modelling (i.e. relevance, design, estimation, and application). Batsell and
Louviere (1991) and Louviere (1993) have recently reviewed the state of the art in
experimental analysis of choice experiments. Green and Srinivasan (1978, 1990) are the
recognised review sources in marketing. Louviere and Timmermans (1990) provide an
overview in the context of tourism.
With this brief historical perspective behind us, this paper concentrates on some of the
important developments in recent years which crystallise the state of practice in stated
preference modelling. In particular, we evaluate the pros and cons of alternative
response metrics (namely ranks, rates and choice), the major considerations in the design
of an experiment (i.e. attribute selection, attribute levels, main and interaction effects,
hierarchical designs and making the exercise comprehensive and comprehendible),
approaches to model estimation (especially individual models, and individual choice
models based on a sample of individuals where the data is maintained at a disaggregate
level or aggregated within each observation to choice proportions), and the scaling of
data with different metric dimensions to enable data aggregation and enrichment. We
also refer to the growing software capability for experimental design, model estimation
and market share prediction. Throughout the paper the emphasis is on the practice of
SP analysis.
Stated Preference Analysis David A. Hensher
5
2. Defining the Response Dimension
There are two broad categories of stated response of interest in travel behaviour
research: (i) An individual is asked to indicate his preferences among a set of
combinations of attributes which define services or products. This judgemental task,
usually seeks a response on one of two metric scales - a rank ordering or a rating scale.
(ii) An individual is asked to choose one of the combinations of attributes. Information is
not sought on the ordering or rating of each of the non-chosen combinations. This is
often called a first-preference choice task.
In both stated preference and stated choice experiments, each combination of attributes
can be defined as an alternative in the sense of representing a product or service
specification which may or may not be observed in the market. The attributes can include
not only well-defined sources of (indirect) utility such as travel times and travel costs,
but also aggregators such as name of product (e.g. car, train) which represent the
respondent's perception of the attributes of the alternatives which are not represented by
the explicitly defined attributes. In both preference and choice experiments it is feasible
to vary both the combinations of attributes and levels as well as the subsets of mixes to
be evaluated. This can be achieved by either designing varying numbers of combinations
or asking the respondent to a priori eliminate any combinations which are not applicable
before responding (soliciting criteria for non-applicability - see Louviere and Hensher
1983).
In practice, it is common in preference experiments to hold the number of alternative
attribute mixes constant and only vary the attribute levels. However, in choice
experiments, it is common to vary the number of alternatives, while either holding the
attribute levels associated with each alternative constant, or varying them, producing
varying choice sets (e.g. Hensher et al 1989). Fixed choice set designs are also widely
used (e.g. Louviere and Hensher 1983, Gunn et al 1992).
The decision on which type of response strategy to pursue must be addressed at the
beginning of an SP study, because it will define the available outputs.
3
A major
consideration is the need for predictions of behavioural response, especially market
shares. Rank order and ratings "predictions" must be transformed to accommodate
useful predictive outputs (except where the interest centres on the image of, or attitude
towards, a service or product - see Hensher 1991). Choice responses are directly
translated into predictions, through the application of discrete-choice models such as
multinomial logit (MNL), and are also relatively easier for the respondent. However, the
advantage of the direct translation comes at the expense of information loss. In a first-
preference choice experiment, no information is available on the ordering of all of the
