Benefit Hierarchy Analysis

Benefit Hierarchy Analysis
The steps in the new product development process entail defining the product concept, identifying the
consumer needs and product benefits, and determining the target consumer demographics. Then, an
optimal product formulation (or several alternative formulations) is developed that can satisfy potential
consumer needs, at a manufacturing cost that is low enough to justify a reasonable price.
In every step of the new product development process, researchers are trying to determine what
product benefits, consumer or sensory attributes, ingredients (including their different levels and
combinations) drive product liking, purchase intent or preference. Hierarchy Analysis is a
relatively new data analysis technique that allows researchers to answer these questions by
organizing benefits, attributes or different ingredient levels into hierarchies according to their
relative impact on consumer choice and preference.
The most noticeable difference between Hierarchy Analysis and traditional approaches to product
optimization is the choice of optimization criterion. Let’s consider a typical study where each
respondent tastes several similar products sequentially and uses the following 9-Point Hedonic
Overall Liking Scale to evaluate each product:
9 - Like Extremely
8 - Like Very Much
7 - Like Moderately
6 - Like Slightly
5 - Neither Like nor Dislike
4 - Dislike Slightly
3 - Dislike Moderately
2 - Dislike Very Much
1 - Dislike Extremely
Traditional data analysis methodologies will either calculate the mean Overall Liking score for
each product and use it as a criterion for decision making, thus implying that the best product is
the one with the highest mean Overall Liking score; or calculate for each product a percent of
respondents who rated the product as Like Extremely or Like Very Much, the so called Top 2 Box

score, and use it as a criterion for decision making, thus implying that the best product is the one
with the highest Top 2 Box Overall Liking score.
In contrast, Hierarchy Analysis uses the criterion that the best product is the most preferred
product. Let’s consider the example presented in Figure 1, which are the results from ten
respondents who rated two products using a 9-point Overall Liking scale.
Figure 1 – Product Ratings
Respondent
Product A
Rating
Product B
Rating
Preferred
Product
1 8 9 B
2 8 9 B
3 8 9 B
4 8 9 B
5 8 9 B
6 8 9 B
7 8 9 B
8 8 9 B
9 9 2 A
10 9 1 A
Mean Score 8.2 7.5
Top 2 Box Score 100% 80%
Preference 20% 80%
Using either the mean Overall Liking score or the Top 2 Box Overall Liking score, we would come
to the conclusion that product A is better than product B. However, when analyzing individual
preferences on a respondent by respondent basis, 80% of the respondents preferred product B

over product A. Thus, according to the criterion that the best product is the most preferred
product, we would infer that product B is better than product A.
The main source of discrepancies between the outcomes of different criteria usage comes from
the way that the three different methods use the original 9-Point Hedonic Overall Liking scale:
 Mean Overall Liking score criterion treats the scale as an interval scale, presuming that all
differences between numeric tags assigned to each verbal statement are equidistant.
 Top 2 Box Overall Liking score treats the 9-Point Hedonic Scale as binomial, recognizing
only the difference between a “good rating” (Like Extremely or Like Very Much) and a “bad
rating,” but neglecting all the other differences.
 Preference criterion treats the 9-Point Hedonic Overall Liking scale as ordinal, assuming
that the rating 9 is better than the rating 8, that the rating 8 is better than the rating 7, etc.,
without any assumptions regarding distances between verbal statements and without any
loss of information resulting from aggregating the statements into a “good” and a “bad”
category.
From the measurement theory view point [1], preference criterion is the only correct criterion,
corresponding to the nature of the measurement scale used.
The theoretical behavior background of the technique is based on a model of consumer behavior
known as “bounded rationality.” The term and the concept were originally introduced by Herbert
A. Simon [2], who in 1978 was awarded the Nobel Prize in economics “for his pioneering research
into the decision-making process.” Ideas of bounded rationality were further expanded by Daniel
Kahneman [3], who in 2002 received the Nobel Prize in economics "for having integrated insights
from psychological research into economic science, especially concerning human judgment and
decision-making under uncertainty."

The main distinction of “bounded rationality” from “full rationality” (which is assumed is such
popular method as conjoint analysis) lies in the recognition that consumers have limited cognitive
abilities and limited time to make decision. Therefore, consumers are not able to evaluate all
product benefits, attributes or ingredients at once, than immediately construct a utility function and
maximize its expected value. There is overwhelming experimental evidence for substantial
deviation of actual consumer behavior from what is predicted by traditional rationality models [3].
Some authors call it “irrationality”, but, in our opinion, the problem is not that people behave
irrationally, but that elegant and beautiful mathematical rationality models do not adequately
explain the consumer’s decisions and choices. According to [4, p.9}, “The greatest weakness of
unbounded rationality is that it does not describe the way real people think.”
According to the bounded rationality concept, consumers employ the use of heuristics or schemas
to make decisions rather than strict rigid rules of decision optimization [5]. A schema is a mental
structure we use to organize and simplify our knowledge of the world around us. We have
schemas just about everything, including ourselves, other people, cars, phones, food, etc.
Schemas affect what we notice, how we interpret things and how we make decisions and act. We
use them to classify things, such as when we ‘pigeon-hole’ people. They also help us to forecast
and predict what will happen in the future. We even remember and recall things via schemas,
using them to ‘encode’ memories. Schemas are often shared within cultures and allow
communication to be shortened. Every word is, in fact, a schema, that we can interpret in our own
way. We tend to have favorite schemas which we use often. They act like filters, accentuating
and downplaying various aspects of the things surrounding us, including different product
attributes and benefits. Schemas are also self-sustaining, and persist even in the face of
disconfirming evidence. If something does not match the schema, such as evidence against it, the
contradictory evidence is often consciously or subconsciously ignored. Some schemas are easier
to change than others, and some people are more open to changing their schemas than others.

Schemas are also referred to in literature as mental models, mental concepts, mental
representations and knowledge structures. The basic proposition of the bounded rationality theory
applied to consumer behavior is that consumers are rational, and when they make choices or
preferences between products, they have some conscious or subconscious reasons for those
choices or preferences that are realized trough their individual schemas.
Hierarchy Analysis presumes that each consumer uses an individual schema for evaluating a
particular category of products and makes choices between products within the category based on
this schema. Hierarchy Analysis represents consumer schema in the form of a hierarchy of
benefits, attributes or ingredient levels arranged in the order of likelihood of their impact on
consumer decisions. By aggregating schemas among random probability sample of consumers,
Hierarchy Analysis allows us to determine the prevalent schema in a population. On other hand,
Hierarchy Analysis methodology allows us to group consumers into clusters based on similarities
or dissimilarities of their individual schemas to discover market segmentation based on consumer
schemas. In addition, Hierarchy Analysis methodology includes procedures for testing statistical
hypotheses related to consumer schemas, for example, if a particular product benefit is more
important than another benefit, or if a particular product benefit is more important for one
consumer group than for another consumer group, or if a particular product benefit is more
important for choice of one product than for choice of another product.
Bounded rationality concept assumes that consumers evaluate product in three steps [5]:
 First they search for some familiar cues.
 When consumers have found enough cues, they stop searching and start evaluating and
organizing these cues in some order of importance to them or the magnitude of the
differences between products.
 Then they make judgments regarding “overall liking,” “purchase intent” and the choice of
product.

The Hierarchy Analysis model relies on the assumption that some of the cues recognized by
consumers are related directly or indirectly, consciously or subconsciously, to the set of product
benefits and attributes that we ask consumers to evaluate (or to the levels and the combinations of
the ingredients and the sensory attributes that are associated with the products, evaluated by
consumers).
Boundedly rational consumers do not necessarily make quantitative choices between alternative
options based on their perceived utilities. Instead, they rely on qualitative expectations regarding
directional changes. For each pair of products, one product could be evaluated by a consumer as
better than or worse than another, or the differences between two products could be negligible.
In this model of consumer behavior, the actual magnitude of the differences between products
does not affect the product choice, only the directional differences matter. On other hand, the
greater the magnitude of the differences between products, the more consumers will recognize the
differences as noticeable and express their preferences. Therefore, the strength of preferences is
measured, not in the magnitude of the differences between products or their utilities, as in the
case of conjoint analysis, but by the proportion of consumers who evaluated the product as
preferred over the alternatives. By considering only the directional differences between products
and benefits, this method essentially treats all scales of measurement used in consumer research
as ordinal, not interval. This corresponds to the actual nature of the scales and makes this
technique conceptually more valid in comparison with traditional statistical methods based on
means and correlations that treat all consumer research scales as if they were interval.
Another important advantage of this approach over traditional statistical techniques is an
acknowledgment of the fact that each respondent has an individual interpretation of the meanings
of different values on psycholinguistic scales. Traditional statistical methods compare ratings
given by an individual respondent to sample averages. This implies that all respondents interpret

scales in the same manner. But, individual interpretations of scales might differ between
respondents based on cultural background, education, age, gender, personal experiences, etc.
Hierarchy Analysis deals with data on a respondent by respondent basis, assuming that each
respondent interprets the scales in an individual manner but consistently across various products,
benefits, attributes, or concepts.
There are multitudes of articles in marketing research literature related to the affect of cross-
cultural differences on scale item interpretations. This issue taints inferences based on the
comparison of mean scores for the same product or benefit across different countries, languages
or cultures. By analyzing data on a respondent by respondent basis, Hierarchy Analysis is free
from this problem and allows the direct comparison of results across countries, languages and
cultures.
Traditional statistical methods usually assume the normal distribution of answers among
respondents for all attributes and criterion ratings. Even if this is not stated explicitly, the mere fact
that traditional statistical methods use only means and standard deviations to describe the
statistical distribution of answers, characterizes the distribution as normal. Moreover, assuming
normality implies that the distributions must be symmetric. In fact, we practically never observe
symmetrical normal distribution in marketing research studies; in many cases answers are skewed
toward high ratings, limited by range, and do not have a symmetrical normal distribution. Also, as
we stated above, a normal distribution could be applied only if we treat all scales as interval, which
actually contradicts the ordinal nature of the scales used. Hierarchy Analysis methodology does
not rely on any assumptions about distributions and accepts all actual distributions “as is”, which
makes it a robust statistical method by definition.
Most of the traditional statistical techniques are based on linear relationships between criterion
and factors (regression and correlation analysis) or linear additive models (conjoint analysis) or

polynomial models (response surface analysis). Hierarchy Analysis presumes only probabilistic
directional relationships between criterion and factors, which makes it independent from the
researcher’s assumptions regarding data.
The integral part of Hierarchy Analysis is the philosophy of Exploratory Data Analysis (EDA),
which was introduced by John W. Tukey [6]. The exploratory approach to data analysis calls for
the exploration of the data with an open mind. According to Tukey, the goal of EDA is to discover
patterns in data. He often likened EDA to detective work; Tukey suggested thinking of exploratory
analysis as the first step in a two-step process similar to that utilized in criminal investigations. In
the first step, the researcher searches for evidence using all of the investigative tools that are
available. In the second step, that of confirmatory data analysis, the researcher evaluates the
strength of the evidence and judges its merits and applicability.
In the classical analysis framework, the data collection is followed by the imposition of a model
(normality, linearity, etc.), and then the analysis that follows is focused on the parameters of that
model. For EDA, the data collection is followed immediately by an analysis that has the goal of
inferring which models are appropriate. Hence, the EDA approach allows the data to suggest
models that best fit the data. Following the spirit of EDA, Benefit Hierarchy Analysis evaluates all
the possible multimodal relationships between product preferences and benefits and estimates the
likelihood that each benefit has an impact on product preference. The result is a hierarchy of
benefits, arranged in the order of likelihood of their impact on product choice and preference.
Another cornerstone of Benefit Hierarchy Analysis is the concept of Probabilistic Causality [7]. A
probabilistic causality approach applied to the analysis of consumer choice and preference data
assumes the following:

 The observed choices and preferences are not spontaneous, but are the results of the
conscious or subconscious use of schemas by consumers in their decision making
process.
 The actual product characteristics, such as various ingredient levels or sensory attributes
could be related to cues discovered by consumers and used in their schemas.
 The perceived product benefits and attributes could be related to cues discovered by
consumers and used in their schemas.
 Consumer schemas represent reasons or causes for their choices.
 Consumers do not use their schemas deterministically and always consistently.
 Consumers do not use their schemas stochastically or completely randomly.
 For each of the possible product benefits, attributes or ingredients, there is an objective
probability that consumers use this particular component in determining their choices and
preferences.
 This causal probability could be estimated from the data.
The process of estimating causal probabilities from observed data starts with the assumption that
all attributes or benefits are mutually independent and a-priori each have an equal chance to be a
cause for the consumer’s choices or preferences. Then, by analyzing evidence of all pairwise
relationships between benefits from the data, and testing, for each pair of benefits, two alternative
hypotheses: (1) that benefit A is more likely to be a cause for the choice and preference between
products than benefit B, and (2) that benefit B is more likely to be a cause for the choice and
preference between products than benefit A, we can estimate for every benefit, the a-posteriori
likelihood that the benefit is a cause of choice and preference between products. The result is a
hierarchy of benefits, arranged in the order of this a-posteriori likelihood of the impact on product
choice and preference.

The following examples illustrate several practical uses of Hierarchy Analysis in consumer
research. The company wanted to develop a new kind of fresh baked bread to sell in stores
nationwide. Their product developers created nine prototypes for the bread using Taguchi
experimental design for four three-level (High-Medium-Low) design factors, as outlined below in
Figure 2.
Figure 2 – Design Factors
Product Factor 1 Factor 2 Factor 3 Factor 4
1 3 1 3 2
2 2 2 3 1
3 2 3 1 2
4 3 2 1 3
5 1 1 1 1
6 1 2 2 2
7 2 1 2 3
8 3 3 2 1
9 1 3 3 3
To identify which of the nine product prototypes is the most preferred by consumers, a nationally
representative sample of 450 consumers were interviewed in 25 locations. Each respondent
tasted 4 of the 9 samples of bread (incomplete block design). To avoid order bias, we
implemented a random balanced rotation algorithm. As a result of the random balanced rotations,
each respondent tasted a unique set of four products. Each product was tasted an equal number
of times in each position balanced by location and each pair of products was tasted an equal
number of times on each sequential position. For each product, respondents were asked Overall
Liking, using a 9-point scale, and 13 diagnostic attributes. The following Figure 3 shows the
results of the Hierarchy Analysis.
Figure 3: Hierarchy Analysis of Products

5.4
7.5
33.9 H
38.0 H
51.9 F
62.6 F
67.1 E
89.0 C
94.6 C
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
PRODUCT 1 (A)
PRODUCT 7 (B)
PRODUCT 8 (C)
PRODUCT 6 (D)
PRODUCT 2 (E)
PRODUCT 4 (F)
PRODUCT 9 (G)
PRODUCT 3 (H)
PRODUCT 5 (I)
In the Hierarchy Analysis, all products are arranged in the order of their preference and labeled
alphabetically, so “A” is a label for the most preferred or best product, while “I” is a label for the
least preferred or worst product. The bars for each product represent the likelihood that the
product is the most preferred by consumers in comparison to the other products being considered.
For PRODUCT 1, which is labeled with the letter “A,” the 94.6% denotes, that based on the
evidence in the data, we have a 94.6% confidence that PRODUCT 1 is the most preferred
product. The letter “C” after the confidence signifies that this product is more preferred than any
product labeled with the letter “C” or below, with at least 94.6% confidence. PRODUCT 7, which is
labeled with the letter “B,” is the second most preferred product. The likelihood that PRODUCT 7
is the most preferred product is equal to 89.0%, which is greater than all the products labeled with
the letter “C” or below. Statistically, PRODUCT 1 and PRODUCT 7 are at parity, despite the fact
that PRODUCT 1 has a numerically greater likelihood of being the most preferred product.

Now, when we know the hierarchy of product preference, we can define the optimal levels of four
Taguchi design factors using a procedure called Non-Parametric Response Surface Analysis. The
principal difference of this analysis from the traditional Response Surface Analysis is the fact that
we do not restrict a set of possible functions describing the relationships between the design
factors and the overall criterion to being the subset of polynomial regression functions, but we
build the response surface as a multitude of points of interest. The following Figure 4 shows the
results of the Non-Parametric Response Surface Analysis.
Figure 4– Non-Parametric Response Surface Analysis
PRODUCT Factor 1 Factor 2 Factor 3 Factor 4
1 3 1 3 2
2 2 2 3 1
3 2 3 1 2
4 3 2 1 3
5 1 1 1 1
6 1 2 2 2
7 2 1 2 3
8 3 3 2 1
9 1 3 3 3
Optimal
Level 3 1 2 3
Confidence 98.2 91.1 88.4 76.5
Cells with the optimal levels of the corresponding factors are highlighted
PRODUCT 1 has the optimal levels for factors 1 and 2, while PRODUCT 7 has the optimal levels
for factors 2, 3, and 4. A product with a high level of factors 1 and 4, a low level of factor 2, and a
medium level of factor 3, which was not part of the original design, could potentially be the best
product.
Figures 5, 6, 7 and 8 represent the non-parametric response surfaces for the factors. The
numbers in the tables represent the likelihood that the corresponding level of the factor is
preferred by consumers over the other levels of the same factor.
Figure 5 - Non-Parametric Response Surface Analysis of Factor 1

5.1
46.7
98.2
0.0
25.0
50.0
75.0
100.0
Low Medium High
91.1
56.0
2.9
0.0
25.0
50.0
75.0
100.0
Low Medium High

0.0
88.4
61.6
0.0
25.0
50.0
75.0
100.0
Low Medium High
6.7
66.8
76.5
0.0
25.0
50.0
75.0
100.0
Low Medium High

As we can see from the results of the four main effect analyses for the four design factors above,
the gradient of differences between the optimal factor level and the second best factor level is
51.6% for factor 1; for factor 2 it is 35.2%, for factor 3 it is 26.8% and, for factor 4 it is 9.6%.
Therefore, by deviation from the optimal factor level, we would be exposed to the highest risk for
factor 1, followed by factor 2, and then factor 3, with factor 4 representing the lowest risk.
Another useful application of Hierarchy Analysis involves linking the consumer preferences to the
sensory attributes of the products. This methodology evaluates, for each sensory attribute, the
likelihood that consumers can recognize different levels of the attribute for different products and
make choices or express preferences between products based on this information. If consumers
do not express preferences between two products with different levels of a sensory attribute, then
we might conclude that the difference between these two levels of a sensory attribute is not
noticeable to the average consumer, but can be discriminated by a trained sensory panel.
In the bread optimization project described above, a sensory panel evaluated 55 various sensory
attributes for each bread sample;
 32 attributes are related to the taste of the bread,
 10 attributes are related to the texture of the bread,
 13 attributes are related to the aroma of the bread.
As a result of applying the Hierarchy Analysis methodology to all 55 attributes, we discovered 11
sensory attributes that affect consumer choices with at least an 80% likelihood. Each of these 11
attributes has a larger impact on consumer preferences with at least a 95% confidence level than
any of remaining 44 attributes. The following Figure 9 illustrates the results of the application of
the Hierarchy Analysis methodology to the sensory attributes.
Figure 9 – Hierarchy Analysis of Sensory Attributes

77.7 oQ
80.9 L
81.6 L
85.5 J
86.0 J
89.0 H
91.0 G
93.5 F
95.5 F
95.5 F
95.5 F
99.2 B
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
FLAVOR27 (A)
FLAVOR6 (B)
TEXTURE5 (C)
TEXTURE9 (D)
FLAVOR26 (E)
FLAVOR9 (F)
FLAVOR23 (G)
AROMA8 (H)
AROMA2 (I)
AROMA7 (J)
TEXTURE7 (K)
FLAVOR4 (L)
The sensory flavor attribute FLAVOR27 has the singular highest impact on consumer choice, with
a 99.2% confidence level. The four attributes, FLAVOR6, TEXTURE5, TEXTURE9, and
FLAVOR26, are statistically at parity on their likelihood to impact consumer preferences, with
confidence levels ranging from 95.5% to 93.5%. Overall, flavor and texture sensory attributes
have a greater impact on consumer choices and preferences between the nine samples of bread
than the aroma related attributes, because the most impactful of the aroma attributes is ranked
only eighth in the hierarchy.
The Hierarchy Analysis for sensory attributes not only identifies which sensory attributes have an
impact on consumer choice, but defines the optimal range for each sensory attribute. The
following Figure 10 illustrates the optimal sensory attribute range for the most impactful sensory
attribute FLAVOR27.

Figure 10 – Hierarchy Analysis for the Most Impactful Sensory Attribute
PRODUCT 9
PRODUCT 8
PRODUCT 7
PRODUCT 6
PRODUCT 5
PRODUCT 4
PRODUCT 3
PRODUCT 2
PRODUCT 1
R2
= 0.4551
0.0
25.0
50.0
75.0
100.0
8 9 10 11 12 13 14 15 16
The optimal range for this attribute is below 9.5. Only two the most preferred products, PRODUCT
1 and PRODUCT 7, have this sensory attribute in the optimal range. As we can see from Figure
10, in this case, the application of the standard polynomial regression to the data would give a
similar conclusion: products with smaller levels of the sensory attribute are more preferred;
however, the Hierarchy Analysis reveals the two ranges of the attribute that are recognisable by
consumers. Products in the optimal range have relatively high average likelihood (91.8%) of being
the most preferred product, while products with sensory attribute levels higher than 9.5 have a low
average likelihood of being the most preferred product (only 38.1%).
During the product evaluation, respondents were asked the overall liking rating for each product
and the ratings of 13 diagnostic attributes, using the same 9-point scale mentioned above. The
following Figure 11 demonstrates the application of the Hierarchy Analysis to the ratings of the 13
bread diagnostic attributes.

Figure 11 – Hierarchy Analysis of Diagnostic Attributes
99.8 B
85.6 D
82.8 D
61.2 hI
60.2 hI
56.4 hJ
55.5 hJ
48.1 J
44.8 J
30.4 K
13.5 L
6.3
5.4
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Taste of bread (A)
Texture of bread (B)
Crust of bread (C)
Appearance of bread (D)
Moistness of bread (E)
Aroma of bread (F)
Thickness/denseness of bread (G)
Crispiness/crunchiness of crust (H)
Color of bread crust (I)
Color of bread interior (J)
Liking of particulates (K)
Amount of crumbs from bread (L)
Amount of particulates within bread (M)
The liking of the taste of the bread is the singular best predictor of overall product preferences,
with 99.8% likelihood. The liking of the texture of the bread and the liking of the crust of the bread
are statistically at parity with an 85.6% and 82.8% likelihood, respectively. These results closely
match the sensory attribute Hierarchy Analysis, where the two sensory attributes with the highest
likelihood of impact were the flavor attributes and five out of the top seven attributes were the
flavor related sensory attributes, while the remaining two were the texture related attributes.
Figure 11 represents the prevalent schema in a population for choosing between samples of
bread. As mentioned above, while assessing the prevalent schema in a population, we calculated
the individual schema for each respondent. Now we can use these results to evaluate the
homogeneity of the consumer schemas. Applying traditional Ward’s algorithm of cluster analysis
to individual schemas, we discovered two different consumer segments with different schemas.
The following Figure 12 illustrates the statistical comparative analysis of two schemas.

Figure 12 – Comparative Schema Analysis
COMPARATIVE SCHEMA ANALYSIS
95% Significant Differences
Amount of particulates
within bread
Liking of particulates
Aroma of bread
Amount of crumbs from
bread
Moistness of bread
Thickness/ denseness of
bread
Crispiness/
crunchiness of crust
Crust of bread
Texture of bread
Color of bread interior
Color of bread crust
Appearance of bread
Taste of bread
0.0
25.0
50.0
75.0
100.0
0 25 50 75 100
SEGMENT 2 (48%)
SEGMENT1(52%)
For all consumers, the most impactful product attribute is the taste of the bread. However for 52%
of consumers (SEGMENT 1), the second most impactful attribute is the aroma of the bread. For
the other 48% of consumers (SEGMENT 2), the aroma of the bread is ranked very low on the
schema hierarchy; this is why aroma was not placed high on an average consumer schema
presented on Figure 11. We can clearly see that the taste and the crust of the bread are equally
important for both consumer segments. However, the aroma of the bread is significantly more
impactful for SEGMENT 1, while the texture of the bread and the crispiness/crunchiness of the
crust are significantly more impactful for SEGMENT 2, with at least 95% confidence.
As result of applying two different schemas to the product evaluation, consumers belonging to the
different segments prefer different products. The following Figure 13 illustrates the results of the
statistical comparative product choice analysis.

Figure 13 – Comparative Choice Analysis
COMPARATIVE CHOICE ANALYSIS
95% Significant Differences
PRODUCT 1
PRODUCT 2
PRODUCT 3
PRODUCT 4
PRODUCT 5
PRODUCT 6
PRODUCT 7
PRODUCT 8
PRODUCT 9
0.0
25.0
50.0
75.0
100.0
0.0 25.0 50.0 75.0 100.0
SEGMENT 2 (48%)
SEGMENT1(52%)
Consumers in SEGMENT 1 preferred PRODUCT 8, PRODUCT 4, and PRODUCT 9 with a
significantly greater likelihood than the consumers in SEGMENT 2, with PRODUCT 8 being the
most preferred product in SEGMENT 1, with 93.4% likelihood. Consumers in SEGMENT 2
preferred PRODUCT 1, PRODUCT 6, and PRODUCT 2 with a significantly greater likelihood than
consumers in SEGMENT 1, with PRODUCT 1 being the most proffered product in SEGMENT 2,
with 98.5% likelihood. Interestingly, PRODUCT 7 is the second most preferred choice for both
segments, and should be chosen if the manufacturer decides to introduce just one new product to
the market. Alternatively, the introduction of two new products corresponding to PRODUCT 1 and
PRODUCT 8 will better satisfy both segments. Product optimization, based on experimental
design and sensory attributes, illustrated above, could be performed for every segment for more
insight.

Hierarchy Analysis is a versatile and robust statistical methodology that helps to solve many tasks
of consumer research. It has more than 15 years of history of usage for hundreds of consumer
research projects by leading consumer packaged goods manufacturers. It is based on the
bounded rationality consumer behavior theory and treats all consumer research scales as ordinal.
The analyses are performed on a respondent by respondent basis, without unjustified
assumptions of interval scales, respondent uniformity, linearity and normality. It provides
quantifiable recommendations for choosing the best product prototype and the best levels of
design factors or sensory attributes. It reveals the reasons for consumer choice and preference
between products (known as consumer schemas), provides statistical tests for homogeneity of
schemas in the population and discovers consumer segments in the cases of heterogeneous
consumer schemas.

Bibliography
1. Coombs C. H. (1964) A Theory of Data. New York: John Wiley & Sons, Inc
2. Simon H. A. (1957) Models of Man: Social and Rational (Mathematical Essays on Rational
Human Behavior in a Social Setting). New York: John Wiley & Sons, Inc
3. Kahneman D., Slovic P., Tversky, A. (Eds.). (1982) Judgment Under Uncertainty:
Heuristics and Biases. New York: Cambridge University Press.
4. Gigerenzer, G., Todd, P.M., & the ABC Research Group. (1999). Simple Heuristics that
make us smart. New York: Oxford University Press.
5. Gigerenzer G. Selten, R. (Eds.). (2001) Bounded rationality: The adaptive toolbox.
Cambridge/MA: MIT Press.
6. Tukey J. W. (1977) Exploratory Data Analysis. Addison-Wesley
7. Pearl, J. (2000) Causality: Models, Reasoning, and Inference. New York: Cambridge
University Press.

Benefit Hierarchy Analysis

More Related Content

What's hot (14)

Similar to Benefit Hierarchy Analysis (20)

Recently uploaded (20)

Benefit Hierarchy Analysis