Segmenting Consumers By Part-Worth Utilities:

Evidence For Different Dynamics Of Category

Elements By Segment

Working Paper – Do Not Quote

Howard R. Moskowitz
Moskowitz Jacobs Inc.
White Plains, New York 10604

Abstract

Market researchers use conjoint analysis to discover the utilities and thus the persuasive
power of different words and phrases. When there are many stimuli classified into
categories of related elements, one can measure the importance of each category by an
index. This index is the proportion of the total sum of squares attributable to each
category. The index ranges from 0 to 100%. This study presents empirical analyses of 12
studies, each with more than 10 categories, and more than 100 concept elements.
Segmentation of respondents based upon the pattern of element utilities often shows
dramatic differences in this index. Furthermore, the segments often differ in their
sensitivity to concept elements, as shown by their total sum of squares.

Introduction

Marketing researchers segment consumers on the basis of their preferences in order to

discover new groups of consumers, who can be better targeted. By segmenting consumers
the marketer attempts to identify homogeneous groups of individuals, who presumably
share similar preferences for goods and services, and who therefore can be satisfied by a
more limited set of offerings.

Segmentation is well accepted in the marketing community, but the key issue is how to
segment. What should form the basis of segments? Wells (1975) suggested segmentation
on the basis of psychological attitudes, and coined the term Psychographics. Others took
up the approach and have created batteries of questionnaires that divide consumers into
homogeneous groups on the basis of their attitudes in general. One of the more famous of
these is the VALS segmentation (values, attitudes, lifestyle segmentation (Mitchell,
1983). Still others have eschewed the more global segmentation and have looked into
segmenting consumers on the basis of their responses to concepts (Green & Krieger,
1991).

Sensory Preference Segmentation - The Forerunner Of Concept Segmentation

Moskowitz Jacobs & Lazar (1985) suggested a segmentation of consumers, based upon
the pattern of sensory preferences. This segmentation was subsequently extended to
concepts (Moskowitz, 1996, 1997). What makes this segmentation important and the
topic of this paper is that the segmentation provides both a data analytic approach and at
the same time a clearer insight into the underlying mechanisms driving the segments. In
contrast, most conventional segmentation procedures look for statistically powerful ways
to divide consumers, but do not look for underlying mechanisms.

The organizing principle for concept response segmentation comes from the observation
in food products that as a sensory attribute increases, acceptance changes in a systematic
fashion. In the prototypical relation, liking first increases, peaks, and then drops down
(Moskowitz, 1981). Figure 1 (left panel) shows this organizing principle. However, when
it comes to individuals, the single inverted U shaped curve may not hold, as Figure 1
(right panel) shows. Rather, the sensory-liking curve in its smoothed format may show a
variety of different shaped curves.

Figure 1
The organizing principle behind sensory segmentation: As a sensory attribute
increases, liking changes. The left panel shows fitted curves from the total panel; the
right panel shows fitted curves from different respondents.

Extending The Organizing Principle To Concepts

The sensory segmentation approach may be profitably translated to the world of concept
testing, if the following equivalences are made:

1) The liking assigned to a product is equivalent to either the acceptance of the full
concept, or the part-worth utility of the concept element (from conjoint measurement).
2) The sensory attribute assigned to the product is equivalent either to a communication
attribute assigned to the full concept, or to the location of the concept element on a
semantic differential scale.

If these two equivalences are made, then the same algorithm that was used for products
can be used for concepts. The outcome of the equivalence is that one can plot concept
interest (or element utility) against communication (or location on the semantic
differential scale). The plot will result in a quadratic function, once the scattergram is
smoothed. Figure 2 shows the equivalent set of patterns, this time for element utility
versus location on the semantic differential scale "male versus female."

Figure 2
The organizing principle for concept response segmentation, using concept elements
instead of products, and semantic scale locations instead of sensory attributes. The
organizing principle is exactly like that used for sensory segmentation.

The concept response segmentation for conjoint measurement follows nine steps
described in Table 1.

Table 1
Algorithm For Concept Response Segmentation

Step 1 - Dimensionalize. For each element in the study, profile that element on a limited
set of non-evaluative semantic differential scales. This is called dimensionalization.
Typically there will be 6-12 attributes on which the concept elements are
dimensionalized. Each element will be rated on the full set of attributes, with a 9 point
scale. A group of 8 raters (matched in qualifications to the final sample) generates table
means of elements on the semantic profile scales.
Step 2 - Set Up The Data Matrix For Modeling. For each respondent, array that
respondent's utilities (part worth coefficients) versus the semantic differential scale. Work
on a respondent by respondent basis, a scale by scale basis.

Step 3 - Create The Quadratic Function. On a respondent by respondent basis, and for
each semantic scale, create a quadratic equation relating that respondent's part-worth
coefficients to the semantic scale. If there are 100 people and 12 semantic differential
scales, a total of 1200 quadratic functions will be estimated.

Step 4 - Optimize The Quadratic Function. For each quadratic function in Step 3, find
the value of the semantic scale (X axis of the function) where that equation peaks. Stay
within the upper and lower range of the scale. Each respondent generates a vector of
numbers corresponding to the optimum level on the semantic scales where that
respondent's utilities (part worth coefficients) maximize. Although each individual's
utilities may not be easily compared, one can compare the locations on the semantic
differential scales.

Step 5 - Create A Full Matrix Of Optimal Levels For Semantic Scale x Respondent. The
data matrix comprises a rectangular matrix of semantic scales (columns) by respondents
(rows). Each number in the body of the table corresponds to an optimum level for the
respondent on the specific semantic differential scale.

Step 6 - Reduce Redundancy In The Matrix By Factor Analysis (Principal

Components). The semantic scales may exhibit a great deal of redundancy. In order to
reduce the redundancy, perform a principal components analysis on the columns. Save
the factor scores of the rows (respondents). Typically 2-5 factors emerge from the
principal components analysis.

Step 7 - Cluster The Respondents By K-Means Clustering. Clustering will locate

individuals who have the same set of factor scores, and therefore the same pattern of
optimal semantic scale values.

Step 8 - Average The Data Within A Cluster. Average the part-worth coefficients of
individuals in the same cluster (segment) to yield a set of part-worth utilities, one mean
utility per concept element per cluster.

Step 9 - Compare The Clusters (Segments) On Elements To Interpret The Meaning Of

The Clusters. Assign a label to each cluster, based upon the pattern of elements most
differentiating one cluster (segment) from the other(s), or based upon which elements are
most positive versus most negative for the cluster.

The Importance Of An Element Versus A Category Of Elements

When there are multiple categories of elements (e.g., name, price, heritage), one of the
first questions which arises is the relative importance of the category. Each element in the
category generates its own utility, so it is technically more correct to ask about the
importance of single elements. An element is important when its utility is either highly
positive or highly negative. The utility value for the IdeaMap method (the conjoint
method discussed here) is defined as the incremental probability that a respondent will go
from indifferent to interest in the concept if the concept element is present. A positive
utility value means that the concept element adds to the probability. A negative utility
value means that the concept element subtracts from the probability.

A category comprises many related elements. A better way to analyze the importance of
elements is to combine the elements into a single number. For this paper the key measure
is the sum of squares of the element utilities within a category versus the total sum of
squares of all element utilities across all categories. The squaring of the utility value
removes the importance of sign. Thus, two elements which have utilities of +10 and -10
would become +100 each, rather than canceling each other out.

The importance of a category is determined both by the size of the utility values and the
number of elements in the category. The IdeaMap technique allows the researcher to
explore categories comprising different numbers of elements. Thus, relative importance
of categories can only be compared across different subgroups of respondents who have
seen the same sets of elements and same number of categories. In this paper we will deal
with relative importance in this comparative fashion.

Category Importance And Concept Response Segments

In most cases, when there are many concept elements, the majority of elements hover
around a utility value of 0, meaning that these elements contribute little to category
importance because they generate squares. [A category can become important if it has
many elements, each of which is unimportant, but because of the sheer number of
concept elements the total sum of squares for the category becomes large]. Things
become more interesting, however, when the researcher segments the respondents by the
method of concept response segmentation above. Quite often the small utilities values for
the total panel can be shown to be the consequence of combining two or more opposing
groups of consumers, with opposite utility values. One segment may love elements in a
category, whereas the other segment may hate those very same elements.

The segments each generate a full set of utilities, one utility for each element in each
category. There are two key indices which show the differences between categories, and
the importance of categories:

1) The total sum of squares of elements within a single category. This index shows the
degree to which the elements in a category drive interest in the concept. If one segment
shows a total sum of squares (of utilities in a category) far greater than does the other
segment, we conclude that the first segment ascribes greater importance to that category.

2) The proportion of the total sum of squares for a category versus the total sum of
squares across all elements. Each segment generates a total sum of squares of all
elements, independent of category. This proportion shows the percentage of total sum of
squares accounted for by the particular category. The greater the proportion, the more
important the category as a driver for that segment.

Method

This paper presents results from 12 studies done using the IdeaMap method, with each
study comprising many dozens or even hundreds of elements. All elements in each study
were used in the concept response segmentation (Table 1). The studies were run over a
five year period, for different manufacturers. Key phrasing for the concept elements are
disguised to maintain confidentiality.

Developing The Concepts By Experimental Design & Test Execution

Table 2 shows how the concepts were developed, and then tested. All testing was done
using a computer to present the concepts and to acquire the data.

Table 2
Concept Design, Field Execution, Initial Modeling

1) Select The Basic Concept Design. Each study uses variations of a single design (5
Level, Plackett Burman Screening Design; Plackett & Burman, 1946). The experimental
design is fairly unsaturated, and allows the researcher to estimate the utility of every
concept element. When multiple designs are used for the same respondent, the researcher
can estimate the utilities of 20N, where N is the number of designs.

2) Create Multiple Concept Stimuli From A Set Of Experimental Designs. In a given

test session, the respondent evaluated concepts created from at least two, or more
generally, from four different experimental designs. Two designs comprise 40 elements
embedded in 50 concepts. Four designs comprise 80 elements embedded in 100 concepts.
(viz., 80 elements in 100 combinations).

3) Run The Study With Consumers. The experimental designs were interdigitated with
each other in order to reduce any boredom or fatigue. The respondents were pre-recruited
to participate, and paid for their time. Typical interviews lasted 20-40 minutes. The
respondents evaluated the concepts, which were shown on a computer screen. The
computer also acquired the data and then processed the data to create the model.

4) Create The Additive Model. By means of a dummy variable regression a model was
created relating the presence / absence of the elements tested to that respondent’s ratings.

5) Estimate The Utility For Untested Elements On An Individual Basis By

Interpolating In The Semantic Space (Smoothing). A smoothing algorithm was used to
estimate consumer's responses to elements not tried. The smoothing algorithm uses non-
evaluative semantic differential scales to locate the concept elements in a geometrical
space. The utilities of untested elements are replaced by the average of the utilities of
their eight closest neighbors in the space. The smoothing algorithm repeats on the data set
until the maximum change in utility value of an untested element is less than 0.01%. At
this point the algorithm is considered to have reach stability (Moskowitz & Martin,
1993).

6 Use Each Respondent’s Data. The data from each respondent in the study was then
used as input for the segmentation algorithm (see Table 1).

Method

Specifics For The 12 Studies

Table 3 shows specifics for the 12 studies. The table can only summarize the studies, due
to the large number of concept elements, categories, and resulting segments.

Table 3

Summary Of Key Specifics For The 12 Studies

Product Or Service Total Number Of

Elements Categories Base Size Segments

Mid-Sized Car 300 14 125 3

Luxury Car 216 15 369 6

Car Oil Additive 180 11 100 2

Laundry Product 278 9 200 4

Baby Sanitary Product 267 7 103 2

OTC Regimen 246 18 150 2

Credit Card 349 19 388 5

Condiment 236 11 150 2

Mexican Condiment 168 14 105 2

Frozen Pastry Snack 226 15 120 3

Carbonated Soft Drink 195 18 150 2

Meat 160 20 120 3

Results

Example Of Additive Model For Elements - Total Panel

Table 4 shows a typical example of the part worth contributions of the elements to
interest. The product category is a cooking oil, whose results are not discussed further.
What is important is the nature of the winning elements. Sometimes these elements differ
across segments quite dramatically, other times they are identical. The winning elements
for each segment are shaded in the appropriate column.

Table 4
Example Of Winning Elements For A Cooking Oil - By Total And Concept Response
Segments

Element Total Seg1 Seg2

Total Panel - Best Three And Worst Three Elements

After 30 years of testing--proven to significantly reduce cholesterol 7 1 12

Contains 90% unsaturated fat 7 5 9

Lets you eat healthy and keep your cholesterol under control 6 3 9

Segment 1 - Best Three And Worst Three Elements

The best oil for salads 3 8 1

The right oil for sautéing, salads and baking 4 7 2

Lets you prepare a healthy, delicious meal your whole family will 3 7 1
enjoy

Segment 2 - Best Three And Worst Three Elements

Has more of a blood pressure-reducing ingredient than canola & olive 4 -6 12

oils

Per American Heart Association, reduces cholesterol 17% when part 5 -3 12

of a healthy diet

After 30 years of testing--proven to significantly reduce cholesterol 7 1 12

Category Importance - Relative Sum Of Squares

We can go one step further in the analysis by looking at the relative importance of the
different categories. To measure relative importance we look at the following formula
(Pioche, 1993):

Importance Of A Category = (Category SS)/(Total SS)

SS = sum of squares = sum of squared coefficients

This formula corrects for the fact that the utilities can be both positive and negative. It is
sensitive to the number of elements in a category. The greater the number of elements, the
greater the sum of squares, all other factors held equal. Thus, this organizing principle is
appropriate for comparing two segments from a single study, where both segments
generate utilities on the same elements.

Table 5A - 5L show the sums of squares for the total panel and segments, for each of the
12 studies. For total panel and segments, the key categories which are most important are
bolded and shaded. There are a variety of different patterns across segments. The simplest
pattern is the one in which the corresponding segments for a single product (or service)
simply like different categories. That is, some categories are important for one segment,
whereas other categories are important for the other segment. An example of this is the
mid-size car study (Table 5A). Segment 1 shows no major likes. Segment 2 likes safety,
Segment 3 likes luxury.
A second type of pattern emerges, wherein the segments appear to like the same
categories. One segment is very sensitive to elements whereas another segment is far less
sensitive to these same elements, even though both segments find the same categories to
be most important. An example of this is credit card (Table 5G). All segments are
interested in fees, credit lines, and interest statements, but the relative importance of those
three categories differs by segment.

Mid-Size Car (Table 5A)

1) Total panel shows the greatest interest in safety (26% of the total SS)

2) Segment 1 is interested primarily in elements which convey a "personal statement"

(viz., about the driver of the car).

3) Segment 2 is primarily interested in safety. Furthermore, the utilities for Segment 2 are
far greater than the utilities for Segment 1 (as suggested by the total SS for Segment 2)

4) Segment 3 is primarily interested in value, equipment and prestige.

5) In order to total sum of squares, Segment 2 (safety) shows the greatest range of
utilities, followed by Segment 3 (prestige), followed by Segment 1.

Table 5A

Distribution Of Relative Sums Of Squares Across Categories - Mid-Size Car

Mid-Size Car Total Seg1 Seg2 Seg3

Total Sum Squares 2121 2265 5150 3109

Safety 26% 4% 24% 9%

Interior 12% 5% 11% 10%

Treatment 12% 9% 12% 8%

Ride/Handling 10% 11% 10% 6%

Versatility/Range 8% 6% 7% 4%
Value/Luxury Balance 7% 3% 7% 14%

Life Stage/User 5% 5% 5% 3%

Quality 5% 6% 4% 6%

Fully Equipped 5% 2% 2% 12%

Personal Statement 4% 16% 5% 4%

Prestige 2% 6% 7% 15%

Environment 2% 9% 3% 1%

Exterior Style 2% 12% 2% 5%

Product Range 1% 6% 1% 1%

Luxury Car (Table 5B)

1) Total panel shows greatest interest in safety and service.

2) Segment 1 shows greatest interest in safety and ongoing product development /

improvement

3) Segment 2 shows greatest interest in safety and service

4) Segment 3 shows greatest interest in ride

5) Segment 4 shows greatest interest in safety, and design of interior and exterior

6) Segment 5 shows greatest interest in safety

7) Segment 6 shows greatest interest in interior and value for money

8) The sums of squares of the utilities are dramatically different across segments.
Segment 2 (safety and service) and Segment 4 (safety, design) are by far the most
conscious of the elements because their sums of squares are very high. Segments 3
(safety and service) and 6 (interior, value for money) are conscious of the elements, but
far less. Segment 1 is hardly conscious of the elements at all, because it has a very low
total SS.

Table 5B

Distribution Of Relative Sums Of Squares Across Categories - Luxury Car

Luxury Car Total Seg1 Seg2 Seg3 Seg4 Seg5 Seg6

Total Sum Squares 4522 504 15801 7764 15506 13869 7448

Safety 17% 14% 21% 11% 11% 19% 4%

Service 13% 4% 20% 11% 3% 7% 3%

Ride 12% 16% 16% 19% 9% 7% 8%

Performance 10% 9% 15% 10% 4% 5% 3%

Interior 8% 7% 2% 8% 11% 11% 20%

Value For Money 7% 3% 3% 12% 7% 3% 20%

Family 6% 2% 1% 4% 5% 5% 12%

Environment 6% 7% 10% 1% 3% 11% 2%

Product 6% 10% 2% 10% 10% 6% 6%

Development

Exterior 4% 7% 1% 5% 15% 7% 8%

Up To Date 4% 5% 5% 1% 4% 9% 2%

Comfort 3% 1% 0% 3% 4% 6% 9%

Consumer Image 1% 4% 0% 2% 4% 1% 1%
Quality 1% 10% 2% 2% 2% 2% 1%

Successes 1% 1% 1% 1% 7% 2% 1%

Car Oil Additive (Table 5C)

1) The total panel is interested primarily in statements about reducing engine wear and
friction, as well as statements about longevity of the engine, and cleanliness of the
engine.

2) Segment 1 is interested in the same things as the total panel.

3) Segment 2 is interested in the same things as the total panel, along with user imagery.

4) Both Segment 1 and Segment 2 show substantially total greater sum of squares
compared to the full panel. The total panel shows less sensitivity than do the segments.

Table 5C

Distribution Of Relative Sums Of Squares Across Categories - Car Oil Additive

Car Oil Additive Total Seg1 Seg2

Total Sum Squares 8084 13003 11219

Engine Wear/Friction 44% 45% 43%

Longevity 12% 13% 10%

Clean Engine 12% 10% 10%

User Imagery 7% 8% 14%

Quality/Value 5% 4% 3%

Visuals 5% 5% 3%
Influential Support 5% 5% 5%

Smooth Running 4% 3% 5%

Taglines 3% 2% 4%

Descriptor 2% 2% 1%

Heritage 1% 1% 2%

Laundry Detergent Product (Table 5D)

1)The total panel is interested in sensory statements about the fragrance, and in the
visuals.

2) Segment 1 is interested in the visuals, and in clothes protection.

3) Segment 2 is interested in the fragrance

4) Segment 3 is interested in the fragrance, the visuals, and the non fragrance experience

5) Segment 4 is interested primarily in protection.

6) Again, the segments show greater sums of squares of elements than does the total
panel. Segment 2 (fragrance) shows the greatest sensitivity to elements, followed by
Segment 3 (experience, visuals), followed by both Segments 1 and 4.

Table 5D

Distribution Of Relative Sums Of Squares Across Categories - Laundry Product

Laundry Product Total Seg1 Seg2 Seg3 Seg4

Total Sum Of Squares 1115 2822 8622 4995 2887

Fragrance Experience 28% 12% 32% 20% 8%

Visuals 23% 22% 18% 17% 14%

Non Fragrance 17% 12% 15% 19% 5%
Experience

Generation X 9% 5% 3% 7% 15%

Protection 8% 18% 5% 9% 26%

Personal Care 7% 11% 8% 12% 9%

Control 6% 9% 9% 10% 13%

Purity/Rinsing 2% 5% 6% 5% 8%

Baby/Family 1% 6% 4% 1% 2%

Baby Sanitary Protection Product (Table 5E)

1) The total panel is interested in the physical construction of the product.

2) Segment 1 is interested in the product form.

3) Segment 2 is interested in the fit of the product to the baby’s body.

4) Segments 1 and 2 radically differ from each other in what elements drive interest.

5) Segment 2 (fit to body) shows the greater sensitivity to elements.

Table 5E
Distribution Of Relative Sums Of Squares Across Categories - Baby Sanitary Protection

Sanitary Protection Total Seg1 Seg2

Total Sum Squares 1495 1910 4839

Ingredients/Construction 22% 15% 13%

Fit 16% 7% 34%

Material 17% 10% 12%

Form 16% 44% 13%

Benefits to Mom 7% 6% 7%

Baby Benefit 7% 11% 11%

Environmental 16% 7% 9%

OTC Regimen For Smoking Cessation (Table 5F)

1) The total panel is interested in efficacy and brand image.

2) Segment 1 is interested in efficacy and user image.

3) Segment 2 is interested in brand image and efficacy.

4) Segment 3 is interested in user image.

4) Segments 2 (brand image) and 3 (user image) are the most sensitive to elements.
Segment 1 (brand image and user image together) is the least sensitive.

Table 5F

Distribution Of Relative Sums Of Squares Across Categories - OTC Regimen

OTC Regimen Total Seg1 Seg2 Seg3

Total Sum Squares 6048 4636 20295 16093

Efficacy 19% 7% 22% 7%

Brand Image 17% 17% 27% 2%

Dosage 13% 1% 12% 7%

User Image 8% 19% 2% 28%

Lifestyle 7% 2% 4% 6%

Mode Of Action 7% 1% 9% 4%

Support System 6% 6% 4% 5%

Tagline 5% 5% 4% 8%

Program Aids 4% 7% 2% 3%

Visual 3% 3% 3% 3%

Timing 3% 2% 4% 4%

Endorsement 2% 7% 1% 11%

Cost 2% 3% 1% 3%

Safety 2% 5% 2% 3%

Aesthetics 1% 6% 2% 1%

Package 1% 5% 1% 2%

Ease 1% 1% 0% 1%

Manufacturer 0% 2% 1% 2%

Credit Card (Table 5G)

1) The total panel is interested in fees, credit line, and interest.

2) Segment 1 is interested in fees and interest.

3) Segment 2 is interested in credit.

4) Segment 3 is interested in the credit line, but is not particularly involved.

4) Segment 4 is interested in fees, credit line and interest.

5) Segment 5 is interested in feeds and interest.

6) The key differences among segments appears to be the interest in credit lines versus
fees, and the sensitivity to elements. Segments 2 and 4 are most sensitive to elements,
Segments 1 and 5 are moderately sensitive to elements, whereas Segment 3 is fairly
indifferent to elements.

Table 5G

Distribution Of Relative Sums Of Squares Across Categories - Credit Card

Credit Card Total Seg1 Seg2 Seg3 Seg4 Seg5

Total Sum Squares 3437 8573 11446 4014 10696 6236

Fees 17% 19% 11% 8% 14% 15%

Credit Lines 15% 3% 16% 14% 10% 4%

Interest Claims 13% 16% 9% 8% 10% 11%

Cash Advance 9% 5% 9% 9% 7% 3%

Visuals 9% 13% 11% 6% 8% 10%

Service 4% 5% 4% 6% 6% 8%

Incentives - Specific 4% 6% 7% 8% 4% 3%

Cash Back 4% 6% 4% 7% 3% 6%

Incentives - General 3% 3% 6% 6% 5% 4%

Travel Benefits 3% 3% 2% 4% 5% 6%
Rebates 3% 4% 3% 3% 3% 5%

General Claims 2% 2% 2% 2% 4% 1%

Transfers 2% 2% 5% 3% 2% 5%

Availability 2% 2% 2% 2% 5% 3%

Gifts 2% 3% 2% 6% 3% 6%

Taglines 2% 2% 1% 2% 4% 3%

Combination 2% 1% 2% 1% 3% 1%
Services

Value 2% 1% 2% 2% 3% 3%

Security 1% 1% 0% 1% 1% 2%

User Image 1% 2% 1% 1% 1% 2%

Italian Condiment (Table 5H)

1) The total panel is interested in positioning, flavor, usage, and the visual pictures of the
product and usage conditions.

2) Segment 1 is interested in positioning and pricing.

3) Segment 2 (the more sensitive segment) is interested in flavor and usage.

4) For condiment, there does not appear to be the same dramatic polarization of interest
by segment (which leads to vastly higher sums of squares for segments in contrast to total
panel).

Table 5H

Distribution Of Relative Sums Of Squares Across Categories - Italian Condiment

Condiment Total Seg1 Seg2

Total Sum Squares 5181 5720 7195

Positioning 17% 14% 20%

Flavor 15% 9% 19%

Usage 14% 9% 18%

Visual 13% 13% 10%

Pricing 10% 18% 6%

Health Statement 7% 9% 5%

Sub-Brand # 1 6% 5% 7%

Sub-Brand # 2 5% 4% 7%

Product Line 5% 6% 3%

Light Flavor 4% 8% 2%

Brands 4% 5% 3%

Mexican Condiment (Table 5I)

1) The total panel is interested in fresh and natural.

2) Segment 1 (the less sensitive to elements) is interested in fresh and natural.

3) Segment 2(the more sensitive to elements) is interested in the Mexican imagery.

Table 5I
Distribution Of Relative Sums Of Squares Across Categories - Mexican Condiment

Mexican Condiment Total Seg1 Seg2

Total Sum Squares 2544 3999 7121

Fresh/Natural 20% 25% 8%

Image 14% 12% 16%

Claims 9% 8% 7%

Texture 9% 6% 6%

Freshness Benefits 8% 8% 4%

Flavor/Taste 7% 5% 8%

Heritage 6% 4% 3%

Ingredient 6% 4% 11%

Texture Benefits 4% 4% 3%

Usage 4% 14% 13%

Flavor Strengths 4% 1% 7%

Product Types 3% 3% 4%

Visuals 3% 4% 7%

Brand Names 2% 3% 4%

Frozen Pastry Snack (Table 5J)

1) The total panel is interested in taste description and product image.

2) Segment 1 is interested in taste description.

3) Segment 2 is interested in convenience, heritage and brand names. This is the most
sensitive segment to elements.

4) Segment 3 is interested in product heritage, and is almost as sensitive as Segment 2.

Table 5J
Distribution Of Relative Sums Of Squares Across Categories - Frozen Pastry Snack

Frozen Pastry Snack Total Seg1 Seg2 Seg3

Total Sum Square 5094 6700 10024 9510

Taste Description 16% 17% 7% 11%

Product Image 12% 12% 4% 7%

Ingredients 10% 11% 3% 5%

Nutritional Claims 10% 11% 7% 5%

Flavor 8% 8% 9% 9%

Texture 8% 9% 4% 4%

Convenience 7% 4% 15% 4%

Product Quality 6% 6% 4% 3%

Product Heritage 5% 6% 13% 24%

Bread Mentions 5% 5% 3% 5%

User Type 4% 4% 7% 6%
Satisfying 3% 2% 7% 3%

Brand Names 2% 4% 11% 12%

Visuals 2% 1% 4% 2%

Use Occasions 1% 1% 2% 1%

Nutritionally Oriented Carbonated Beverage (Table 5K)

1) The total panel is interested in visual.

2) Segment 1 and Segment 2 are interested in ingredients, but in different ways, and to
different degrees. Segment 2 is more sensitive to elements than is Segment 1.

Table 5K
Distribution Of Relative Sums Of Squares Across Categories - Nutritional Beverage

Nutritional Beverage Tot Seg1 Seg2

Total Sum Squares 1172 2340 3850

Visual 18% 12% 18%

Ingredients 12% 22% 32%

Taste Descriptors 11% 7% 6%

Choice 8% 9% 7%

Product Benefits 8% 10% 9%

User Image 7% 6% 5%

Health Claims 7% 7% 6%
Heritage 5% 4% 2%

Texture/Mouthfeel 4% 3% 2%

Flavor Descriptor 4% 2% 3%

Carbonation 4% 2% 2%

Taste Image 3% 6% 2%

Comparative Taste 2% 1% 2%

Color 2% 1% 1%

Physical Benefits 1% 3% 1%

Product Descriptor 1% 2% 1%

Emotional Benefits 1% 3% 2%

Usage 0% 0% 0%

Smoked Meat - Deli Product (Table 5L)

1) The total panel is interested in smoked flavor, health benefits, and flavor/taste.

2) Segment 1 is interested in quality and ingredients.

3) Segment 2 is interested in smoked flavor.

4) Segment 3 is interested in health benefits.

4) Segments 2 and 3 are more sensitive to elements than is Segment 1.

Table 5L
Distribution Of Relative Sums Of Squares Across Categories - Meat
Meat Total Seg1 Seg2 Seg3

Total Sum Squares 3279 3862 10293 8166

Smoked Flavor 13% 7% 16% 5%

Health Benefits 11% 6% 5% 15%

Flavor/Taste 10% 6% 10% 9%

Texture 7% 3% 6% 7%

Ingredients 7% 13% 7% 15%

Quality 7% 14% 7% 6%

Process--Smoking 7% 4% 6% 6%

Heritage/Experience 6% 4% 5% 7%

Convenience 6% 4% 5% 3%

Product Specific 5% 7% 5% 4%

Claims 4% 5% 5% 4%

Value 3% 3% 4% 2%

Versatility 3% 3% 3% 1%

Processing--General 2% 4% 2% 1%

Casing 2% 2% 3% 3%

Emotional Benefits 2% 5% 3% 1%

Visuals 2% 4% 1% 4%
Comparison claims 1% 5% 3% 3%

Brands 1% 2% 1% 2%

Process--Curing 1% 1% 3% 2%

Discussion

Absolute Versus Differential Importance Of Categories

Since there are many possible elements in a category, and since the utilities of concept
elements in the same category may differ (some being very positive, others being neutral
or negative), it is impossible to truly measure absolute category importance. A category
can have high sum of squares even with a few elements because those elements are
highly effective. Yet, a different set of elements for that category may have a low sum of
squares, because those particular elements (expressing the same idea) are ineffective.
Thus, even equalizing the number of elements in a category, or in this case estimating the
average sum of squares of an element would be incorrect, because the importance of a
category would then still depend upon the particular choice of elements used. The
approach presented here is best used to quantify the relative importance of concept-
element categories for two or more segments who have been exposed to the same
elements.

Categories: Primary Versus Secondary, Idiosyncratic Drivers

Table 6
Key Categories With High Sum Of Squares - By Total Panel And First Two-Three
Segments. Where The Segments Agree The Category Is A Primary Driver; Where The
Segments Disagree The Category May Be A Secondary, Idiosyncratic Driver.

Total Panel Segment 1 Segment 2 Segment 3

Mid Size Personal Statement Prestige

Safety Exterior Style Safety
Car Value/Luxury Balance
Ride/Handling
Fully Equipped

Safety Safety Safety

Luxury Car Ride
Service Product Development
 Service

Engine Wear Engine Wear Engine Wear

Car Oil
Additive Longevity Longevity User Imagery

Clean Engine Clean engine Longevity

Clean Engine

Fragrance Visuals Fragrance

Laundry Fragrance
Product Visuals Protection Non Fragrance Exp.

Visuals

Baby Construction Form Fit

Sanitary

Efficacy User Image Efficacy

OTC User Image
Smoking Brand Image Brand Image Brand Image
Cessation

Fees Fees Fees

Credit Card Credit Lines
Credit Lines Interest Claims Credit Lines

Interest Claims Interest Claims

Positioning Pricing Positioning

Italian
Condiment Flavor Positioning Flavor

Usage Usage

Visual

Image
Mexican Fresh/Natural Fresh Natural
Condiment Usage

Taste Descriptor Convenience

Frozen Taste Descriptor Product Heritage
Pastry Product Image
Snack Product Heritage

Brand Names

Nutritional V visual Ingredients Ingredients

Beverage

Ingredients Health Benefits

Smoked Smoked Flavor Smoked Flavor
Deli Met Health Benefit Quality Ingredients
Flavor/Taste

These data show two interesting patterns across segments. Pattern A shows one or a few
categories being most important (viz., greatest sum of squares) across total panel and all
segments. This means that the segments differ in the relative importance of elements, but
the category is equally important. Products or services showing Pattern A may be those
where consumers’ opinions of what is important are deeply entrenched. The specific
elements within the category may differ across consumers, but the category itself is
important. An example of Pattern A is credit cards, where the key interest is in fees,
interest charges, and credit lines. It may be very difficult to come up with new ideas
because the thinking here is rigid.

Pattern B shows differences in the categories that are important. To one segment, it may
be ingredient, to another segment, it may be heritage. Products or services showing
Pattern B may be more open to new ideas, because consumers themselves pay attention
to radically different aspects of the products. Examples of Pattern B are the different
foods (Mexican condiment, Italian condiment, Frozen pastry snack, Deli meat).

Do these two patterns represent different types of motivation patterns? That is, does
Pattern A suggest that the category is an obvious primary motivator and cannot be
changed by segmentation? Does Pattern B suggest that the category is a secondary
motivator, effective for some individuals but not for other? This question needs more
investigation.

Two Types Of Segments - Interested Versus Non Interested

A recurring pattern in these data is the existence of two types of segment, sensitive and
insensitive, respectively. Sensitive segments comprise individuals who generate high
sums of squares for their utilities. These high sums of squares appear at the top of each of
the tables (5A-5L). Two segments, one sensitive and the other insensitive, may show
similar patterns of what is important to them, based upon the proportion of variability
attributable to each category. Yet, one segment will generate very high sums of squares,
indicating that the utilities are numerically high (either positive or negative or both). The
other segment(s) will show relatively low sums of squares, indicating that the utilities are
numerically low, hovering around 0. In most cases, the total sum of squares for each of
the segments exceeds the total sum of squares for the entire panel. This result suggests
that even in the case of the "less sensitive" segment, however, there is more sensitivity to
the elements than would have been supposed by looking at the entire panel.

All Categories Are Not Created Equal

The data in Tables 5A-5L suggest that there are some categories which continue to be
critical in driving interest, and other categories which continue to be irrelevant.
Emotional benefits, often touted as critical to acceptance, are generally not important.
Nor, in most cases, are visuals (even though some pictures may be important in and of
themselves). What tends to be important are elements which convey specific information
of relevance to the consumer. This may be mode of action for a drug, safety features (but
not necessary claims) for a car, ultimate benefits for a car oil additive, etc.

The same category may appear in different studies, and in one study the category will be
important whereas in the other study the category will be deemed unimportant. There are
two reasons for the study-to-study differences in the importance of the category.

1) Sheer Number Of Elements: The number of elements in the category may vary across
studies. With everything held equal, those categories having the greater number of
elements stand a greater chance of having more importance simply because they have
more elements to contribute to the total sum of squares.

2) Specific Execution Varies Across Categories: The same category may have different
types of elements, depending upon the product or service to which it refers. Thus
nutritional/health benefits may be quite different for a meat product than for a nutritional
beverage product.

Practical Applications In Understanding The Decision Making Process

One of the goals of conjoint measurement is to understand the decision making process,
and by so doing, create either actual products or product/positioning concepts that are
more effective. To the extent that the relative sum of squares represents the relative
importance, the marketer can identify which of the particular categories should be used to
construct the concepts. More importantly, however, the data for the segments shows that
the decision making criteria are different. Categories and elements appropriate for one
segment may be less relevant or irrelevant for another.

Some segments in this set of studies appear to be focused principally on one category of
elements, and are relatively indifferent to the remaining categories. A good example of
this is Segment 2 for smoked deli meat, which appears to be interested primarily in
statements about "smoked flavor". This type of segment may require specific messages.
In contrast, Segment 3 for meat is interested in health benefits and ingredients. This type
of segment may be harder to reach through conventional product-based messages,
because they respond to a different, and perhaps unusual message for the category.
Further research is called for to identify the potential differences in decision processes
between those individuals who are influenced primarily by one category of elements, and
those who can be influenced by any number of different types of categories of elements.
Relevant questions include the differences in the way these types of segments process
information about products, the degree to which their purchase is brand loyal, and the
degree to which their preferences are stable over time.

References

Green, P.E., & Krieger, A.M. 1991. Segmenting markets with conjoint
analysis. Journal Of Marketing, 55, 20-31.

Mitchell, A. 1983. The Nine American Lifestyles, New York, MacMillan.

Moskowitz, H.R., 1981. "Sensory intensity Vs hedonic functions: Classical

psychophysical approaches". Journal Of Food Quality, 5, 109-138.

Moskowitz, H.R. 1996(A) Segmenting consumers world-wide: An application of

multiple media conjoint methods. 49th E.S.O.M.A.R. Congress, Istanbul, 1996,
535-552.

Moskowitz, H.R., 1996(B) Segmenting consumers on the basis of their response to

concept elements: An approach derived from product research. Canadian Journal
Of Market Research, 1996, 15, 38-54.

Moskowitz, H.R., Jacobs, B.E., & Lazar, N. 1985. Product response

segmentation
and the analysis of individual differences in liking, Journal Of Food
Quality, 8,
168-191.

Moskowitz, H.R. & Martin, J.G. 1993. How computer aided design and presentation
of concepts speeds up the product development process. Proceedings Of The
6th ESOMAR Conference, 1993, Copenhagen, Denmark, 405 - 419.

Pioche, A. 1994. Personal communication.

Plackett , R.L., & Burman, J.D., 1946. The design of optimum multifactorial
experiments, Biometrika, 33, 305-325.

Wells, W, D. 1975. Psychographics, A critical review. Journal Of Marketing Research,

12, 196-213.