You are on page 1of 14

COMPUTER GENERATED CONJOINT ANALYSIS SURVEYS FOR

INVESTIGATING CITIZEN PREFERENCES

JD Hunt, JE Abraham and DM Patterson


Department of Civil Engineering
The University of Calgary
2500 University Drive NW
Calgary, Alberta, Canada T2N 1N4
jdhunt@acs.ucalgary.ca
jabraham@acs.ucalgary.ca
dmpatter@acs.ucalgary.ca

ABSTRACT

This paper describes the development and application of a purpose-built


computer program that supports conjoint analysis surveys. It also
describes the use of this program in the assessment of citizen attitudes
and preferences in Calgary in Canada regarding elements of
transportation and urban form, including travel times and costs to work
for auto and transit modes, housing densities, air quality, taxes, use of
open space and the impacts of new road construction on open space. The
computer program facilitated the collection of a richer data set by
allowing each interview to be custom-built, which helped the study
provide a more complete understanding of attitudes. In particular, it
helped the survey establish a wide range of specific trade-off rates
involving 11 different elements of urban form and transportation. For
example, the introduction of a new major road in an environmentally
sensitive area was found to have the same impact on household utility as
an increase in driving time to work of 8.6 minutes per trip for the typical
Calgarian household.

1: INTRODUCTION

In 1991 the City of Calgary in Canada began a project to develop a new long-range plan
for its transportation system. This project, called the GoPlan Project, concerned a wide
range of elements of urban form and transportation. This included mobility, air quality,
residential density, use of environmental space and funding sources such as taxes and
user charges. Part of the mandate for the plan was to involve the public effectively and
gauge their attitudes and preferences regarding these different elements.

This paper describes a conjoint analysis survey, called the Core Values Survey, that was
developed and conducted in order to gauge these public attitudes. It also describes a
computer program, called INVIEW, that was developed to support the application of
this survey, presenting some of the capabilities of this program in the context of the
survey. The results of the survey and the insights they provide regarding the
preferences of Calgarians are also considered.

Section 2 describes the conjoint analysis method and compares it with some of the
alternative methods available for assessing public preferences. Section 3 outlines
various aspects of the design of the conjoint analysis survey that was performed in this
study. Section 4 provides a description of the computer program developed to support
the application of the survey. Section 5 describes the execution of the survey and
presents some of the results obtained from the analysis of the survey data. Section 6
offers conclusions regarding the process used, the performance of the supporting
computer program and the meaning of the results.

2: OVERVIEW OF SURVEY METHOD

Description of conjoint analysis

A conjoint analysis survey is a type of stated preference technique where survey


respondents are asked to rank hypothetical alternatives in the order of their preference.
Each alternative is composed of several elements, or 'attributes', with each of these
elements described by a sentence or phrase. A number of different states are possible
for each element. The variation between alternatives is due to a random selection of the
state for each element.

Each survey respondent will have different preferences for the different specified
elements. However, the 'typical' preferences for a group of respondents can be
established in a number of different ways. In this study the preferences for the elements
of urban form and transportation were established by estimating a logit choice model of
the observed ranking behaviour. The logit model assigns probabilities to alternative
rankings using a utility function. This utility function measures the overall level of
satisfaction that individuals expect from alternatives. A coefficient is associated with
each element and the responses from the survey are used to estimate values for these
coefficients using maximum likelihood (McFadden, 1974).

In this work the utility of an alternative i was described with a linear utility function
with the following general form:

where:
ui = utility for alternative i
Xni = numerical indication of state or value regarding element n for
alternative i
φn = utility function coefficient associated with element n.
The estimated values of the coefficients provide direct indication of how the different
values for the elements influence the level of satisfaction with the alternative. This level
of satisfaction is measured in 'utils', which is a synthetic unit, and each coefficient
indicates how many 'utils' are associated with its corresponding element. The ratio of
the coefficients for two elements provides what is called a 'trade-off rate', which
indicates the relative influences of the two elements and suggests the overall rate at
which respondents are willing to accept compromises between more desirable and less
desirable conditions concerning different elements.

Advantages of conjoint analysis

Conjoint analysis has several advantages over other available methods for measuring
public attitudes and preferences. Since it is a survey it is possible to solicit opinions
from a representative random sample of the population, unlike in a public meeting,
where only those parties who are sufficiently motivated decide to attend and make
themselves heard. Since each alternative is presented within the framework of a
hypothetical situation, things are under the control of the survey designer, and other
elements that might confuse the issue are explicitly removed from consideration. The
familiar context of choice among alternatives makes it easy for respondents to
participate; whereas the requirement to trade off among a number of attributes within a
specific context ensures that all the attributes are considered from the same perspective.
The trade-off rates established by the estimation process provide numerical indications
of preferences.

The use of conjoint analysis in the investigation of attitudes regarding urban form in this
way, using what is referred to as 'full-profile' construction of alternatives (Green and
Srinivasan, 1978), appears to be fairly new. A somewhat similar approach was used
recently in Seattle (Decision Data, 1993), the use of similar kinds of approaches in the
assignment of monetary values to environmental impacts is reviewed by Turner et al
(1992) and the potential for this approach seems to be acknowledged in the description
of a special course to be offered concerning the use of stated preference techniques
(Portland State University, 1994).

3: SURVEY DESIGN

Survey elements

The attitudes and preference of Calgarians regarding four areas concerning


transportation and urban form were of interest in the GoPlan Project:
v mobility
v built form
v environment
v costs and taxes.
In order to obtain observations of the relevant attitudes and preference, the hypothetical
alternatives considered by respondents were defined by combining states for specific
elements related to these areas. The nature of each of these elements and the related
possible states were determined as part of the design of the interview. In order to get
accurate indications of attitudes, it is important for these elements and their different
possible states to be understandable and relevant to respondents to the fullest extent
possible. It also helps in the interpretation and further use of the results if the
individual states are quantified to the extent possible.

Mobility was represented by the travel times to work for both auto and public transport
modes. A trip purpose was specified in order to provide context. The work purpose
was selected because it represents the largest demand for transportation facilities during
daily peak periods. Travel times for the automobile ranged from 5 to 60 minutes in
increments of 5 minutes, and for public transit times ranged from 25 to 90 minutes in
increments of 5 minutes. Public transit travel times were said to include walking times
and average waiting time.

Housing type was selected to represent the density of the built form, with categories
ranging from single family dwelling to highrise. These were felt to be understandable
and relevant proxies for more abstract measures such as dwelling units per acre or
employees per 1000 square feet. Photographs of examples of each dwelling type were
used in the interviews.

Three elements were selected to represent the environment:

1. frequency of noticeably bad air quality as defined by Alberta Environment;

2. proximity to a river valley or an environmentally significant area (ESA) - with ESAs


as defined by Calgary Parks and Recreation Department; and

3. loss of open space because of roadway construction through river valleys or ESAs.

The states for frequency of bad air quality ranged from never bad to bad once per week.
The states for proximity of the home location to a river valley or an ESA were near or
far, with near defined as being within reasonable walking distance. The states for loss
of open space because of roadway construction included: a new major road is
constructed over a river valley or through an ESA near to the respondent's home, a new
major road is constructed over a river valley or through an ESA far from the
respondent's home and no new major roads are constructed over a river valley or
through an ESA anywhere in the City of Calgary. Photographs of what was meant by
river valley, ESA and major road were used in the interviews, as were written
definitions of air quality and ESA.

Money costs were included in order to get relevant indications and also in order to
obtain trade-off rates expressed in money amounts. Money costs for travel to work
using either auto or public transport and the level of municipal residential taxes were
included. The money costs for the use of auto were defined to exclude parking charges
and the possible states ranged from $1.00 to $5.00 CDN in increments of $0.25. The
money costs for the use of public transport were defined to be fares, with possible states
ranging from $0.50 to $2.50 in increments of $0.25. Taxes were expressed as either an
increase or decrease in municipal taxes or rent with respect to the respondent's current
situation and the possible states ranged from down $600 per year to up $1800 per year
in increments of $300 per year.

This resulted in 11 elements to be used to describe alternative scenarios. Each of these


elements had numerous possible states. The state for each element in an alternative was
selected randomly according to a uniform distribution among the possible states.
Certain logical dependencies were maintained as appropriate: for example, if the home
location was not near an ESA, it could not be allowed to be near new road construction
within an ESA.

Survey context

The hypothetical scenarios created by combining randomly-selected states for elements


were presented to respondents as alternative possible home locations. Respondents
were asked to rank these hypothetical scenarios in order of preference as future home
locations according to the states for the elements. In order to provide context,
respondents were told to imagine that the indicated differences arise because of
differences in city form and residential location as appropriate, and that the situations
were for a point in time two years in the future. This was done in anticipation of
difficulties where a respondent finds the exercise unrealistic because the selected state
for an element is markedly different from the current situation for the respondent
individually or for the City of Calgary generally.

Survey format

Face-to-face interviews were used because of the complexities of the element and state
definitions and the ranking task: interviewers would be able to explain the task, and
would be available to answer questions and clarify issues as required. Interviews were
conducted in homes or offices according to the preferences of the respondents. Home
locations were found to facilitate interviews better: time pressures were reduced and the
respondents were more relaxed generally.

Socio-economic information was collected in the interview, including present dwelling


type, present dwelling tenure, taxes or rent paid, total number of private vehicles owned,
and total household income for the household together with the age, gender, drivers
licence status, employment status, workplace location, and mode typically used to travel
to work for each person in the household.
Each respondent completed three separate rounds of ranking. This improved the
efficiency of the data collection and the second and third rounds were considered likely
to provide more accurate indications in that they were performed by 'experienced'
respondents. The maximum number of rankings was set at three in an attempt to keep
the duration of the interview under thirty minutes - a duration greater than thirty minutes
was considered excessive. Four different hypothetical situations were ranked in each
round, thereby providing six pairwise comparisons per round. More than four situations
would have provided more comparisons, but the ranking task would be more difficult
for respondents.

The descriptions for each alternative were printed on separate sheets of paper, one for
each alternative. This allowed respondents to physically arrange the alternatives in
order of preference.

Of the 11 elements available, only 9 were presented in a given interview. An interview


considered either a river valley or an ESA, not both. The element for proximity to the
open space not considered and the element for road construction through or over the
open space not considered were also not presented in an interview.

Previous research has indicated that the task of considering and comparing conditions
regarding nine elements across four alternatives is too complex. Many respondents
facing such a task simply give up or ignore some attributes, thereby adversely affecting
the quality of the results (Bates, 1988). Accordingly, the number of elements varying in
a given ranking was reduced to only five by holding the other four elements constant
across the four alternatives. The four elements to be held constant in a given interview
were selected randomly by the computer program developed for generating the
interviews at the time the survey forms were being generated. Each of these four was
held constant at a state selected randomly while the other five were allowed to vary
among the alternatives. The states for nine were displayed on each sheet, with the five
varying elements displayed in bold typeface to allow them to be identified more easily.
This simplified the ranking task while still allowing nine elements to be considered in
each interview and also providing a more complete description and associated context
for each respondent.

It was anticipated that elements at the top of a sheet would receive more attention than
those further down. In order to avoid biasing the results, the order of presentation of the
elements was selected randomly for each interview by the computer program. This
order was maintained for all rankings in a given interview, thereby providing
consistency for each respondent together with variation across the whole sample of
respondents.

Figure 1 shows an example of an alternative displayed on a sheet of paper.


4: COMPUTER PROGRAM FUNCTIONALITY

The INVIEW computer software was developed to generate the alternatives needed in
the survey. This section provides details concerning the structure and capabilities of the
program - and can be skipped without loss of continuity.

INVIEW is written in C++ using object-oriented design in order to provide maximum


flexibility and potential for re-use. The basic design consists of a hierarchy of object
classes where an interview contains one or more games, a game contains two or more
alternatives and an alternative contains one or more attributes.

Attribute class

The attribute class is used to determine the elements of an alternative. Attributes have
methods to sample values from specified distributions, to write text descriptions of
these values and to put numerical indications of these values to datafiles. At present,
there are four different types of attributes: realatt for real numbers within certain ranges,
roundatt for real numbers within certain ranges rounded to some specified level of
precision, intatt for integers, and enumatt for integer codes.

In practice, the roundatt class is usually used instead of the realatt class, which avoids
the confusion that may occur when respondents are presented with accuracy beyond
what can be perceived. Enumatt is used as a subclass of intatt where each integer value
for intatt corresponds to a certain textual description in enumatt. This could be used, for
instance, to represent the different dwelling types or different levels of air quality.

Alternative class

An alternative is a vector of attributes. It has a method called random that causes it to


choose random values for its attributes by invoking the random method for each
attributes. Each alternative has a unique 'tag' that separates it from the other
alternatives in an interview.

Alternatives also have a method called change_some_attributes that randomises only a


subset of their attributes - for use when certain attributes are held constant between the
various alternatives in one game - as well as a method for printing the text descriptions
of their attributes.

Game class

The game class stores the alternatives to be presented in a given round of ranking. It
has a data item that identifies the attributes to be held constant among alternatives and a
method called buildchoices that creates the alternatives for the ranking. Buildchoices
first invokes the random method to generate the first of the alternatives and then
invokes the change_some_attributes method to generate the rest of the alternatives that
share some attribute values with each other as designed.

The alternatives generated by the game can be inspected to ensure that there are trade
offs involved. For most attributes there is an a priori expectation that particular states
regarding an element are perceived as better or worse than others. For instance, it is
expected that lower auto costs are perceived as better. This information can be used to
make sure that all of the alternatives in a game have at least one attribute that is better
than in the other alternatives. This makes it possible to avoid asking respondents to
make completely obvious choices. However, in the Core Values Survey it was not
known with certainty in advance what the order of preference would be for roadway
construction in different types of open space, so this capability was not used.

The game class also includes the specification of the layout for the sheets displaying the
alternatives.

Interview class

The interview class contains all the information associated with one respondent,
including the vector of games for the interview and the socio-economic information
provided by the respondent. A unique interview identification number is added, with an
extra digit provided for error detection. The interview class owns the order vector that
is used to control the order of presentation of attributes, which means that the order of
presentation is maintained throughout the interview.

Specialisations of classes

One aspect of the flexibility of the program is that specialised versions of the classes can
be created for individual surveys. Two such special classes were created for the Core
Values Survey. A special order class was created where the random method for this
class was rewritten so that the order of attribute presentation would always keep the two
transit attributes (travel time and fare) next to each other and the two driving attributes
(drive time and cost) next to each other. A special alternative class was created in order
to control the dependency where the home location could not be near new road
construction over a river valley or through an ESA if the home location was not itself
near a river valley or an ESA.

Other Applications

INVIEW has been used for a number of other stated preference surveys; see Hunt et al
(1994) and Hunt and Patterson (1994) for examples. In most of these surveys there
have been special design issues that could not have been implemented reasonably
without modifying the program code. The ease with which specialised versions of the
classes are created to address these special design issues has proven to be a great asset.
5: SURVEY EXECUTION AND RESULTS

Interview Logistics

A total of 4450 randomly-selected households in Calgary were contacted to participate


in the Core Values Survey. Selection was based on a random sample generated from
both telephone listings and the Calgary civic census database (City of Calgary, 1993).
Twenty-four percent agreed to complete the survey, twenty-six percent declined and
fifty percent could not be reached. In the end, 961 interviews were usable for analysis.

Households initially were contacted by letter inviting them to participate. Very few
households called on their own initiative to set up an appointment. Temporary staff
were hired to take calls as well as to contact as many households as possible in order to
book appointments. At least five attempts were made to contact each household.

Principal Findings

Estimation results based on all 961 households are shown in figure 2. Figure 2 indicates
how changes in conditions regarding each of the elements impact the typical household.
It does this using utility values measured in utils to indicate levels of satisfaction. This
allows the typical household's sensitivities to different elements to be compared directly
and allows trade-off rates among the elements to be established.

The trade-off rate between auto drive time and money costs for these results, which
implies the value of auto drive time for a work trip, is $11.06 per hour. This can be
compared to another study done in Calgary based on revealed preference data that
calculated a value for auto drive time to work of $10.76 per hour (Hunt, 1992). A
similar comparison for the value of transit time to work implies a value of $6.95 per
hour. The value of transit time is approximately half that for auto time which is
consistent with other findings (Hunt, 1992). The close match between the
corresponding values found in these different studies suggests that they are accurate
estimates of the value of drive time to work for the population. It also suggests that
people are responding to hypothetical situations in this study in much the same way
people responded to real situations, which adds more general credence to the full range
of results obtained in this study.

The levels of utility for different housing types are shown with respect to a deviation
from a single family dwelling. As expected, all other categories of housing are
considered less desirable. The infill, duplex, and townhouse categories show values
approximately 4 times as large as that for a 10 minute auto drive time to work. This
indicates that the average Calgarian would rather drive an additional 40 minutes to work
than live in one of the more dense housing types. Clearly, housing type and density are
very important to Calgarians.
An increase in taxes by $100 per month is perceived in roughly the same manner as
switching from a single family dwelling to a townhouse. The ratio of auto drive time to
municipal taxes is approximately 2.5 times. This indicates that a one minute further trip
to work is valued approximately the same as $2.50 in municipal taxes per month. This
trade-off rate can be used when considering whether municipal tax dollars should be
spent on transportation improvements to provide travel time savings.

An increase in the frequency of poor air quality has a negative effect on utility. The
frequency of bad air quality going from never bad to bad once per week has
approximately the same impact on utility as $100 per month of municipal tax dollars.
The statistically significant perception between air quality that is never bad and air
quality that is only bad one day per year demonstrates a fine sensitivity to air quality.

The survey results also measure Calgarians' perceptions of environmental areas and
river valleys. As shown in figure 2, being close to a river valley or an ESA is seen as a
good thing. Disturbing these areas with major road development is seen as positive if
the road crosses a river valley and negative if it passes through an environmental area.
This may be because rivers are viewed as barriers to transport and bridges are seen to
improve accessibility. Road development in open spaces near the home location is
perceived to be worse than such development far away. The disutility of road
development through ESAs far from the home location is the same as the disutility of an
8.6 minute increase in driving time to work.

Different sub-samples of households within the full sample were considered separately
in the same manner as the full sample. These sub-samples were defined according to
socio-economic information such as income, family size, dwelling type, geographic
location, employment status. These further results are presented by Hunt (1994).

6: CONCLUSIONS

This study has shown that conjoint analysis can be an effective way of gauging attitudes
and preferences regarding various elements of transportation and urban form in Calgary.
More generally, it has shown that the method itself provides a means for measuring
what is sometimes dismissed as 'too difficult' to measure. The use of a hypothetical
context involving new housing locations with the mix of elements considered seems to
have been effective. All this suggests that there is considerable potential for conjoint
analysis to be used to gain further indication of attitudes and preferences regarding a
range of other factors sometimes dismissed as 'too difficult' to measure.

The various trade-off rates between the elements of urban form and transportation
considered in the study are themselves the principal findings of the study. They are
discussed above and need not be reviewed here.
The consideration of 11 elements was managed and displayed to respondents effectively
using the purpose-built INVIEW program. The treatments that it facilitated, where only
9 of the 11 elements were considered in a given interview with 4 of these 9 held at a
constant state in a given ranking, appears to have simplified the potentially complex
task of ranking according to 11 variables into a form that many respondents could
handle.

The development of INVIEW is judged to have been worthwhile when compared


against the alternative: to purchase a commercially-available package. The availability
of the program in code form provides complete flexibility to make modifications in
order to implement design features as desired. In addition, it has proven to be useful to
have the program because it provides a framework for the design of conjoint analysis
surveys. It has been used in a number of other applications, and has acted as a
repository of the experiences gained and approaches developed for these other
applications.

Acknowledgements

The authors wish to acknowledge the assistance provided by the GoPlan Project Team
and David Watson of the City of Calgary in particular. Funding for the work was
provided by the City of Calgary and the Natural Science and Engineering Research
Council of Canada.

REFERENCES

Bates, J.J., (1988) Stated preference technique for the analysis of consumer choice. In
Store Choice, Store Location and Market Analysis, N. Wrigley, ed. Routledge Chapman
and Hall, New York, pp:187-202.
City of Calgary (1993) 1993 City of Calgary Civic Census, Calgary Planning and
Building Department, Calgary, Canada.
Decision Data Inc (1993) Seattle Planning Department Residential Preference Study,
Seattle Planning Department, Seattle, USA.
Green P.E. and Srinivasan V., (1978) Conjoint analysis in consumer research: issues
and outlook, Journal of Consumer Research 5:103-123.
Hunt J.D. (1992) Report on EMME/2 Logit Mode Split Model Development for the City
of Calgary, City of Calgary, Calgary, Canada.
Hunt, J.D. (1994) Evaluating Core Values Regarding Transportation and Urban Form
in Calgary, BGS No. 22-10-94. Calgary GoPlan Project, Calgary, Canada.
Hunt J.D., McMillan J.D.P. and Abraham J.E. (1994) A stated preference investigation
of influences on the attractiveness of residential locations, Transportation Research
Record (in press).
Hunt J.D. and Patterson D.M. (1994) An investigation of factors influencing time of
travel selection behaviour for a recreational trip, Compendium of Technical Papers for
the 1994 Annual Conference of the Institute of Transportation Engineers, Canadian
District, Windsor, Canada, June 1994, pp:169-193.
McFadden D. (1974) Conditional logit analysis of qualitative choice behavior, in
P. Zarembka (ed.) Frontiers in Econometrics, Academic Press, New York, pp105-142.
Pearmain D.B., Swanson J., Kroes E. and Bradley M. (1991) Stated Preference
Techniques: A Guide to Practice - Second Edition, Steer Davies Gleave, London,
Hague Consulting Group, Netherlands.
Portland State University (1994) Announcement for "Stated Preference Applications in
Land Use, Transportation Planning and Environmental Economics Short Course" to be
held at Portland State University December 1994; copy available from the authors.
Turner R.K., Bateman I.J., Hargest K.W., Freeman A.S. and Scanlon D.A. (1992)
Environmental Appraisal: A Review of Monetary Evaluation and Other Techniques,
Contractor Report 290, TRRL, UK Department of Transport, Crowthorne, UK.
* housing taxes/rent: do not change
(relative to your present taxes/rent)

* transit walk, wait and ride time to work: 40 minutes

* fare per transit trip to work: $1.75

* local air quality: noticeably bad 1 day per month

* situation: environmentally significant area / major park not nearby


(not within reasonable walking distance)

* road development: 1 new major road crossing environmentally signifi-


cant area / major park - not nearby

* auto drive time to work: 40 minutes

* cost per auto trip to work: $3.00 (excludes parking costs)

* housing type: single family dwelling


(with your present floorspace and $ value)

Figure 1: Example hypothetical alternative considered in the Core Values Survey


Figure 2: Graphical representation of estimated values for coefficients in utility
function for full sample of respondents

You might also like