Quantitative methods in marketing 203
gated (for example, products, brands, character-
istics, etc.). When the number of variables
investigated is n, the number of all possible
relationships among these variables (along k
dimensions) is n(n– 1)/2. In order to visualize
and quantify the overall attitudinal data of these
respondents with regard to the n variables
investigated along (k) dimensions, the data
should be inputted onto one of the available
software packages.
The solution (output) of the MS computer
program is of a metric nature, consisting of a
geometric configuration, usually in two or
three dimensions. The distances between the
variables (objects) and/or respondents (sub-
jects) investigated, which are presented as
points in the configuration, represent the (dis)
similarity, substitutability, relationship, etc.
Multidimensional scaling is used particularly
in its non-metric version, non-metric multi-
dimensional scaling (NMS). The advantage of
NMS in relation to, say, factor or cluster
analyses is the ability to see the entire structure
of variables together and to obtain metric
output from attitudinal (non-metric) input
data. In addition, NMS enables easy compre-
hension of the results, since the decision maker
can visualize and assess the relationships
among the variables.
Multidimensional scaling and non-metric
multidimensional scaling in particular have
been successfully applied in investigating vari-
ous marketing problems (for example, market
research, sales and market share, market seg-
mentation, determination of marketing mix,
consumer buyer behaviour, brand positioning,
branch preference, export marketing, etc.). An
introduction to multidimensional scaling is
presented by Diamantopoulos and Schlegel-
milch (1997). Discussion on when to use NMS
techniques in marketing research is offered by
Coateset al. (1994).
Conjoint analysis
This technique is concerned with the joint
effects of two or more independent variables
on the ordering of a dependent variable. Con-
joint analysis, like multidimensional scaling, is
concerned with the measurement of psycho-
logical judgements, such as consumer prefer-
ences. Products are essentially bundles of
attributes, such as price and colour. For exam-
ple, conjoint analysis software generates a
deck of cards, each of which combines levels
of these product attributes. Respondents are
asked to sort the cards generated into an
order of preference. Conjoint analysis then
assigns a value to each level and produces a
‘ready-reckoner’ to calculate the preference
for each chosen combination. The preference
logic of conjoint analysis is as follows. The
respondent had to base his or her overall
ranking of the versions on an evaluation of
the attributes presented. The values that the
individual implicitly assigns each attribute
associated with the most preferred brand
must, in total, sum to a greater value than
those associated with the second most-pre-
ferred brand. The same relationship must
hold for the second and third most-preferred
brands, the third and fourth most-preferred
brands and so forth. The computation task
then is to find a set of values that will meet
these requirements.
Potential areas of application for conjoint
analysis include product design, new product
concept descriptions and testing, price–value
relationships, attitude measurement, promo-
tional congruence testing, the study of func-
tional versus symbolic product characteristics,
and to rank a hypothetical product against
existing competitors already in the market and
suggest modifications to existing products
which would help to strengthen a product’s
performance.
The limitations of conjoint analysis are
quite clear when, for example, we are using this
technique to predict trial rate. These include:1 Utility measurement rather than actual
purchase behaviour is used as the predictor.
2 The configuration of elements used in the
concepts may not be complete.