RESEARCH WITH CONJOINT ANALYSIS
||The goal of Segmentation is to
obtain groups of Greek customers of a specific product category or
service, that react similarly to various product or service categories.
The goal of segmentation is not an easy thing and most methods fall
short of that.
It is a long time, since MARKETECK has abandoned the type of
segmentation, which involves grouping customers on the basis of
demographic variables, because those variables are generally not closely
related to the buying behavior. Such variables account only for 10% or
less of the variance observed in the buying process for products or
We use the so called approach of BENEFIT SEGMENTATION, which
incorporates the critical features, which drive consumers' buying
behavior. Segmenting customers on the basis of their reactions to these
critical features brings MARKETECK's analysis closer to the above goal
Our analysis by the use of appropriate software follows three stages:
|The first stage calculates distances. Two types of calculations
are performed: First, ratings' data is standardized to remove bias
involving respondent's use of the rating scale. Secondly, it
calculates how similar each respondent's utilities are to every
|In the second stage, a hierarchical cluster
analysis on the respondent similarity measures obtained in the
preceding stage, using the Ward's method of clustering, is performed.|
|In the third stage, we have results for solution. The distance
reflects the amount of variability within each segment and there
might point out the elbow of the curve, where distances start to
level out. The elbow indicates the appropriate number of segments to
select the problem. On the other hand, the knowledge of the product
or service is also needed to make the appropriate determination
regarding segments. |
For each segment, it shows the number of respondents, its average
utility function and the relative importance of each feature, then by
using the relative statistic, it makes easier to identify the type of
the important features. For quantitative features like price, we also show
the quantitative model, the vector and ideal-point coefficient.
The refinement procedure assumes that each segment has its own
normal distribution, including its own covariance structure.
Respondents are assigned to the group to which they are most similar.
Initially, the mean and covariance matrix are derived from the cluster
solution. Subsequently, they are recalculated based on the new group
formed by reassignment of observations. At each stage of the iterative
refinement procedure, the likelihood that each respondent belongs to
each group is calculated. This probability is modified to reflect the
size of each group. The procedure performs a quadratic discriminant
analysis at each stage and then uses the discriminant function and Bayes
rule to assign observations to each group.