preferences will naturally be modified by characteristics of the choice situation
(e.g. choices made for one’s own consumption or for a gift, changes in one’s income
and so on). Finally, aggregation of the choices by all potential consumers will lead
to a prediction of the overall market response (e.g., sales of an item).
1.2 Origins of Conjoint Analysis
While the foundations of conjoint analysis go back to at least the 1920s, it is
generally agreed that the seminal paper by Luce and Tukey (1964) on the theory
of conjoint measurement formed the basis for the applied field of conjoint analysis.
The development of the field was aided considerably by the proliferation of
algorithms for the computations involved.
Conjoint measurement is conce rned with determining the joint effect of levels of
two or more attributes of stimuli on the total evaluative judgmen ts of a set of stimuli
(see Rao 1977 for a review of conjoint measurement in marketing analysis). The
objective is to decompose the total evaluation into component scores, imputable to
each attribute level or combination of attribute levels . The theory is concerned with
the conditions under which there exist measurement scales for both the evaluative
score (dependent variable) and each attribute level (independent variables), and a
pre-specified composition rule. All are based on formal axiomatic system
formulated by Krantz et al. (1971), including the axioms of consistency, transitiv-
ity, and attribut e independence. The evaluative score can be categorical, ordinal or
interval-scaled. For example, consider an individual’s evaluation of a pair of
running sneakers described on two attributes of price and quality (e.g., $70 per
pair and medium quality); these responses can be categorical (e.g. suitable for
serious young runners, for casual young runners, or for retirees), ordinal (e.g.,
very good, good, bad or very bad value for money), or interval-scaled (e.g., a rating
on a 10 point scale on value for money). With such evaluation scores of price and
quality on a number of profiles, an analyst can develop a utility function for the
individual. Calling the functions for price and quality v
p
, and v
q
respectively (called
partworth functions), the composite specification for the evaluation can be additive
as a*v
p
+ b*v
q
or polynomial as a*v
p
+b*v
q
+ c*v
p
*v
q
or some other formulation.
The axioms enable the analyst to choose the appropriate specification.
In the course of implementing conjoint measurement methods to applied busi-
ness problems, such as those encountered in marketing, the emphasis on theoretical
aspects of measurement has given way to the more pragmatic issues of design of
studies and analysis of data. This is due to various intricacies in testing
1
whether the
axioms are satisfied in the data collec ted. The testing procedures require extensive
data and are highly complicated even for a small number of respondents. This
process became frustrating for applied researchers.
1
See Corstjens and Gautschi (1983) for detailed methods for testing these axioms.
1.2 Origins of Conjoint Analysis 3