Proceedings of the ASME 2012 International Design Engineering Technical Conferences &
Computers and Information in Engineering Conference
IDETC/CIE 2012
August 12-15, 2012, Chicago, Illinois, United States
DETC2012/DTM-70434
MODELING PRODUCT FORM PREFERENCE USING GESTALT PRINCIPLES,
SEMANTIC SPACE, AND KANSEI
José E. Lugo ⇤
Stephen M. Batill
Design Automation Laboratory
Department of Aerospace and Mechanical Engineering
University of Notre Dame
Notre Dame, Indiana 46556
Email: [email protected]
[email protected]
ABSTRACT
Engineers describe design concepts using design variables.
Users develop their visual judgment of products by mentally
grouping design variables according to Gestalt principles, extracting meaning using semantic dimensions and attaching attributes to the products, as reflected in Kansei methodology. The
goal of this study was to assess how these different sources of information and representations of product form (design variables,
Gestalt variables, Kansei attributes, and semantic dimensions)
could combine to best predict product preference for both designers and users. Sixteen wheel rim designs were created using
four design variables that were also combined into higher-order
Gestalt variables. Sixty-four participants viewed each rim, and
rated it according to semantic dimensions and Kansei attributes,
and provided an overall “like” rating. The most reliable prediction of product preference were developed using Gestalt variables in combination with the meaning and emotion the users
attached to the product. Finally, implications for designers are
discussed.
knowledge, experience and project-specific information derived
from modeling, analysis and experimentation. The decisions are
driven by the roles and responsibilities of groups or individuals
who participate in the process. Some of the decisions are objective and supported by quantitative information and some are
subjective and result from emotional assessments [1].
Traditionally, industrial or product designers tend to focus
on decisions related to form or visual appearance with the intent to exploit the opportunities provided by a new design [2–4],
while engineering designers are faced with the pragmatic decisions associated with assuring that the design meets the functional constraints set forth in design requirements and specifications, as in Zuo et al. [5]. In today’s design challenges these roles
overlap more and more with the realization of the importance of
user-centered design and the subsequent recognition that users
also influence the design process.
For users certain properties can influence how a product is
judged. A property is an attribute, quality, or characteristic of
a product. Furthermore, properties range from materials and dimensions to geometric features. In the design of a product, a designer can specify some of the properties. The set of properties
that the designer can control will be referred to as design variables. Design variables are physical features and dimensions of
a product. Examining how design variables influence the subjective judgment and preference of a product is one of the purposes
1
INTRODUCTION
The engineering design process involves a sequence of interrelated decisions. These decisions are based upon acquired
⇤ Address
all correspondence to this author.
Laura Carlson
Spatial Cognition Laboratory
Department of Psychology
University of Notre Dame
Notre Dame, Indiana 46556
Email: [email protected]
1
Copyright c 2012 by ASME
of this study.
Both the users and the designers have a role in developing
the requirements on both form and function of a new design.
Early in the design process concepts can be expressed in ways
that represent potential form (concept sketches or renderings) or
ways that represent potential function (target design specifications) but in both cases these are only expressions of what the
product may become. Prior to the actual product being built,
most of the representations of a product are visual [6]. They take
the form of sketches, drawings, CAD and Virtual Reality (VR)
models. Decisions made early in the design process based upon
these conceptual expressions of a product are influenced by how
both designers and users react to visual representations of a product [7, 8].
The human visual system has a bias to group certain properties (i.e a visual representation of a product) into higher order
variables following Gestalt principles; this study will be referring
to these variables as Gestalt variables. Therefore, in addition
to the information provided by design variables, designers and
users also perceive the product in terms of these Gestalt variables.
The higher order properties included in this study are proximity,
closure and continuation. However, there is not a unique set of
Gestalt variables for a given visual representation of a product.
The use of Gestalt principles to discretize the visual representation of a product, and to model the predicted product preference,
are the novel applications of these principles presented in this
study.
Although, one would hope that most of the design decisions
are rational and justifiable based upon quantifiable information,
that is not always the case, particularly early in the process.
There are situations where the decisions are intuitive, often justified by experience, but sometimes they are emotional and are
influenced by factors that might not normally be associated with
engineering design. Nevertheless, they are important as they affect the outcome of the process. This study uses Kansei words,
from Kansei engineering methodology, to probe the emotions inferred from a product, and then predict product preference. The
Kansei words used in this study have product-specific meaning
because they are the terms used by users to explain attributes of
the product.
Another attribute that is represented by a product’s form is
its meaning [9]. In this study, the semantic space is introduced
to measure the meaning designers and users assign to a product’s
form. The semantic space is a 3-dimensional space that measures meaning with respect to 3 factors: evaluation, activity and
potency. Each of these dimensions are then used to predict product preference.
Previous work has shown that expertise influences product
form assessment [7]. Thus this study used two subject groups
that varied in expertise. Expertise was defined with respect to
the potential role of the subject in the design process. Experts
were students with mechanical or industrial design majors who
had some level of experience in making design decisions. Nonexperts were students with other majors and were potential users
of the product. This distinction was used to investigate whether
this small variation in expertise altered the influence of these different sources of information: design variables, Gestalt variables,
Kansei words and semantic space.
The first goal of this study was to demonstrate which Gestalt
variables can be used to predict product preference. The hypothesis is that the Gestalt variables (e.g. proximity, etc.) are
representative of the mental process when the product form is
visually explored. When a subject looks at a product their preference mental model is formed by discretizing the product shape
following Gestalt principles. This implies that representing the
design space with traditional engineering design variables may
not be as effective when product preference is the desired outcome. The second goal was to explore whether different sources
of information and representations of product form (design variables, Gestalt variables, Kansei words and semantic space) can
be combined to predict product preference.
This paper employed a study of automobile wheel rim designs to test the hypotheses and accomplish the goals. Different methods were used to gather subjects visual judgment
of the wheel rims. Kansei words were used to capture subjects’emotions towards the wheel rims. Semantic space was used
to gather meanings associated with the wheel rim designs. The
methodology used in this paper is not limited to developing a relationship between the wheel rim form (discretize by design variables) and the subjects judgments as in previous research studies
with other products [4, 10–12]. Rather, explain and justify the
form-judgment relationship, a novel procedure to discretize the
wheel rim form in terms of Gestalt variables is presented. This
new representation of the wheel rim form is more consistent with
the manner that humans process visual images.
2
BACKGROUND
This section introduces concepts and serves as a foundation
to build the study that is presented in this paper.
2.1
Gestalt Principles
When our visual system gathers information through the
eyes, the brain processes the information and interprets it. This
interpretation, visual perception, is a psychological manifestation of visual information. Visual perception is formally defined
as “the process of acquiring knowledge about environmental objects and events by extracting [information] from their emitted or
reflected light” [13]. The information gathered through our eyes
could be interpreted in many ways but we only perceive it one
way at any given time. A simple example is given by Edgar Rubin with the optical illusion known as the Rubin vase in which
you either see a vase or two faces looking at each other. Gestalt
2
Copyright c 2012 by ASME
principles help to dictate which interpretation out of many is perceived visually at a given time.
Gestalt is a German word that means form. Gestalt principles are the factors that govern how we perceive the whole form.
As Kurt Koffka explains, “It has been said: The whole is more
than the sum of its parts. It is more correct to say that the whole is
something else than the sum of its parts, because summing up is
a meaningless procedure, whereas the whole-part relationship is
meaningful” [14]. Gestalt principles consider the basic elements
that compose an image. Then, they consider the relationships
between those basic elements to group them. The relationships
used in this study are: proximity, closure, and continuation (see
Fig. 1). The proximity principle states that elements of an image
(e.g. lines or dots) that are closer to each other will be perceived
as being together, as a group. The closure principle states that
a set of elements can be perceived as a close figure despite the
presence of gaps such as the triangle in Fig. 1. The continuation
principle states that when there is a smooth change from one element to another these elements will be grouped together. Specifically, these three principles will be applied in section 3.2.3 to the
case study product to define Gestalt variables that discretized the
wheel rim form.
tem seems to be much more sensitive to certain kinds of differences than others [13]. For example, the same elements can show
grouping regarding orientation at one angle, but might show less
grouping at another angle. When more than one principle is presented, a process framework could extract the principles that organize the visual perception of the image [15]. The wheel rim
designs presented in this study can show more than one Gestalt
principle in a single image.
2.2
Semantic Space
Charles E. Osgood, a social political psychologist, developed the semantic differential method in the 1950s. This method
measures the connotative meaning of concepts [16]. Some of
the concepts studied by Osgood were general terms like Russians, patriots, and America. In a study to sample the semantic
space, Osgood presented 20 different concepts to 100 subjects;
each subject rated each concept on 50 different scales [17]. The
rating scales used the Likert scale with pair of adjectives that are
antonyms on each end point. A factor analysis summarized the
meaning of concepts into three semantic dimensions: evaluation,
potency and activity.
The three semantic dimensions are orthogonal and define the
semantic space. The evaluation dimension contains word pairs
like beautiful-ugly, good-bad, clean-dirty; the potency dimension
contains word pairs like strong-weak, large-small, heavy-light,
and the activity dimension contains word pairs like fast-slow,
active-passive and hot-cold. The semantic space is a three dimensional space where the meaning of a concept can be located
in space. This study replaces concepts with different product designs to capture the connotative meaning of designs.
The semantic differential and the semantic space were developed as tools to measure semantic meaning. When first developed, the tool was criticized as a measurement of meaning
from a linguistic perspective but psychologists consider it to be
a useful tool [18]. The semantic differential and semantic space
are used now in studies from other disciplines [7, 19]. In the current study, the semantic space is used to measure the extracted
meaning from the form of the wheel rim and to represent it in the
three semantic dimensions.
2.3
Kansei Engineering
The following example illustrates Kansei Engineering: You
are driving a convertible sports car down a curvy road: the wind
on your face, the short shift leather knob in your hand shifting as
you are going into the turn; the grip of the tires, as the car hits
the apex, the acceleration that gives you tunnel vision, the sound
of the tuned exhausts and the vibrations as the engine revs up
accelerating out of the turn all take the driver and passenger into a
state of excitement. These emotions are the Kansei of most sports
cars. When designers and users see, hear, touch, taste and smell
an artifact, they go through an assessment process to build their
FIGURE 1: Gestalt Principles examples: a.) and b.) Proximity,
c.) Closure, d.) Continuation
When the concept was introduced, Gestalt principles were
only demonstrated one at a time. Also, the principles work better
when everything else is equal, that is there is only one relationship between the basic elements of an image. The visual sys3
Copyright c 2012 by ASME
3
METHODOLOGY
The concepts presented in the background section are applied to a case study. This section describes the case study participants, the design of the study, including the procedures and
tasks carried out by the participants.
judgment of that artifact. Parts of the judgment of the artifact
are feelings and feelings build up to emotions. These subjective
emotional responses to an artifact are what Kansei Engineering
connects with engineering features.
The word Kansei is a Japanese term that is translated by
Nagamachi as“the subjective impressions of an artifact that are
gathered through the senses” [20]. There are other definitions in
literature, all trying to explain this same concept but there is no
direct translation from Japanese to English [21, 22].
Kansei Engineering is commonly translated to English as
emotional or affective engineering. Kansei engineering is a
method that translates the users’feelings into design specifications [19]. K. Yamamoto first used the term Kansei Engineering
in 1986 [23]. Mitsuo Nagamachi developed the Kansei Engineering method in the 1970s at the Hiroshima International University. This method is preceded by other methods that share the
idea of gathering the user needs or emotional impact of an artifact. One of these is the Semantic Differential method introduced
in the previous section. Another method that followed was the
Quality Function Development method created by Misuno and
Akao in the 1960s. After Kansei Engineering was introduced,
the Kano model was developed in the 1980s. One of the key features of Kansei Engineering that differs from these other methods
is its goal of translating the emotional feedback that subjects report into changes in the design variables of the artifacts. This was
used in this study as a starting point to identify the relationship
between design variables and product judgment.
The Kansei Engineering methodology can be summarized in
six steps. The method starts with a collection of what is called
Kansei words, which are words that customers and sellers use to
describe the emotions perceived by the product. This collection
of words should represent the description of the product, but in
order to innovate new words can be added. To reduce the list, the
words can be filtered by different methods like pre-surveys and
affinity diagrams. Once the list is reduced to a quantity subjects
can evaluate, the semantic differential method is used to build
the scales. Schütte has noted that the number of Kansei words
subjects can evaluate changes with culture [21]. Once the words
are selected, a sample of products from the market is collected.
These products are categorized according to the features that the
designer deems important, and the products are chosen such that
there is a variance in the features of interest. In the current study
instead of using a market sample CAD models were generated
to systematically represent different design variations. Then, an
evaluation experiment is conducted where each product is rated
against each scale. The experiment is administered to a desired
number of subjects. The results are interpreted performing a multivariate statistical analysis and using other statistical tools. The
process ends with models that explain the Kansei of a product to
designers. The importance of this method in the current research
is that it determines how design variables influence the subjective
judgment and preference for a product.
3.1
Participants
A total of 64 subjects from the University of Notre Dame
participated in the study (35 males; 29 females). All subjects had
normal or corrected-to-normal vision. They completed a written
consent agreement and were briefed on the experimental protocol before starting. They were compensated with either $10.00
or course credit. The subjects were classified in two groups (according to their design expertise) as either students with design
experiences or students with other majors. Senior students from
the mechanical engineering and industrial design programs were
considered students with design experience. Also, graduate students holding a degree in those disciplines were so classified.
The remaining subjects were students of other majors with diverse academic backgrounds. The design students were recruited
from a senior level design methods course through email. The
students of other majors were recruited from the Department of
Psychology subject pool . The subject groups were composed of
32 design students (25 males; 6 females) and 32 students from
other majors (10 males; 22 females). The distinction between
subjects was made because of previous research showing that the
judgment towards the same product varies with expertise [7].
3.2
Design of Case Study
This section describes the development of the case study
and includes a description of the different wheel rim designs that
were generated and how the judgment dimensions were chosen.
Some of the judgment dimensions are taken from Osgood’s semantic space and the remainder were Kansei words. The general
dimension, “like”, was included as a measure of product preference.
3.2.1 Product Selection An automobile wheel rim
was chosen as the case study product because it is geometrically
simple, easy to represent visually with few design variables and
has emergent Gestalt variables; at the same time it plays an important role in the appearance and function of an automobile.
The wheel rims are associated with an emotional reaction, as reflected in comments on blogs and web pages. In addition, they
can be rated with respect to the three principal meaning dimensions for semantic space. Another reason to use wheel rims in
the case study is because it is a very common way to change the
appearance of a vehicle.
4
Copyright c 2012 by ASME
3.2.2 Design Variables The product under study was
characterized using engineering design variables. These variables correspond to features manipulated in the product design.
A parametric CAD model was used to render the images of
wheels rims with a normal tire and a white background. The images were a front view of the wheel rim, thus the design variables
selected were the ones that changed the visual appearance of the
rim from this view. The design variables selected to parameterize the CAD model were: number of spokes (Spokes), number of
bolts (Bolts), spoke width angle (Width), and spoke blend radius
(Radius), that is the radius where the spoke meets the outer and
inner rim (see Fig. 2). The blend radius to the outer rim is double the blend radius of the inner rim. Each design variable was
selected at two levels: low and high. See Tab. 1 for each variable level values. In practice a designer would select values for
each of the design variables in order to define a specific wheel
rim design.
factorial DOE generates a total of 16 different wheel rims designs, shown in Tab. 2. (The reader is encouraged to look at the
16 designs and identify the one that he or she likes most to later
compare with the results of the study.) In order to show each
subject a complete set while limiting subject fatigue, the study
only used these four design variables at two levels. The CAD
model was used to generate virtual renderings of the 16 wheel
rims designs with a normal tire and a white background.
3.2.3 Definition of Gestalt Variables Gestalt variables were defined to describe the wheel rims designs in terms of
Gestalt Principles. The first Gestalt Principle to be applied was
proximity. A general definition of the proximity principle that
can be applied to most objects is that elements of the objects that
are close to each other may be perceived as being grouped together. This general definition can be applied to the wheel rim to
develop various quantitative measurements of proximity, accordingly three Gestalt variables are defined based on this principle.
Two of them refer to the angular proximity between spokes and
bolts, the other to the angular proximity between a spoke and the
adjacent spokes.
The coding defined for these variables is as follows:
Proximity Bolts to Spokes = (number of bolts closely aligned
with spokes) / (number of bolts)
Proximity Spokes to Bolts = (number of spokes closely
aligned with bolts) / (number of spokes)
Proximity Spokes to Spokes =1 / (number of spokes)
The Gestalt Principle of closure was used to develop an additional Gestalt variable. There are two wheel rim designs where
the spoke radii blends are tangent with the adjacent spoke radii
blends forming a closed figure. The contrast between a wheel
design with and without closure can be seen in Fig. 3 where the
design on the right shows closure and the one on the left does not
show the effect. Without closure the radius blends are not tangent
and it is hypothesized that the subjects fixate more on the spokes
than the rim. The designs for this variable, closure, were coded
1 for the designs 8 and 16, and coded 0 for all other designs.
The last Gestalt variable was developed from the continuation principle. An initial evaluation of the survey data helped
identify this variable. It was noted that designs with the Width
set at the low level, and the Radius and Spokes set at the high
level, behaved effectively as the designs with the Width set at
the high level. One Gestalt principle that can explain this phenomena is the continuation principle. Specifically the larger radius blends and the number of spokes increase the continuation
(smoother changes between sections) in the front surface of the
wheel. Considering these 2 factors and adding the actual width
of the spokes, the continuation Gestalt variable was defined as:
FIGURE 2: Design Variables: a.) spoke width angle, b.) spoke
blend radius, c.) number of Bolts, d.) number of Spokes)
TABLE 1.
WHEEL RIM DESIGN VARIABLES
Design Variable
Low Value
High Value
Bolts
4
8
Radius
0.7 in. (17.8 mm)
1.5 in. (38.1 mm)
Spokes
4
10
Width
10
16
Perceived Width of Spokes = Width ⇥ Spokes ⇥ Radius
With these four design variables, each at two levels, a full
5
Copyright c 2012 by ASME
3.3
Closure = 0
Case Study Procedure
The study began after the subjects completed a consent
agreement, and were briefed on the experimental protocol. The
subjects started the experiment by entering their major field of
study and a subject number. Then, they were presented with onscreen instructions followed by one example of the task to be
completed. When they completed the task (responded to all the
questions) for the same wheel rim design they moved on to complete the same task for the remaining wheel rim designs. The order in which the wheel rim designs were presented to the subject
was randomly selected within subjects. Each subject evaluated
all 16 wheel rim designs.
The study equipment used was a PC computer, a Qwerty
keyboard, mouse and a monitor (18 inches (45.7 cm) measured
diagonally and set to a resolution of 1280 x 960 pixels). The
study procedure was administered using E-Prime 2.0 software
(Psychology Software Tools, Pittsburgh, PA).
In the study task subjects were shown an image of one of the
wheel rim designs with a tire. Below the image was a sentence
describing a specific judgment dimension (e.g. On the scale
below rate your perception of how STRONG is the wheel). At
the bottom of the screen was the 7-point Likert scale, as shown
in Fig. 4. The judgment dimension were composed of Kansei
words, semantic space terms, and the general judgment dimension that asked how much the subject “like” the design. The judgment dimensions were displayed in random order within subjects, with the exception of the “like” dimension that was always
asked last. The subjects provided their responses using the keyboard, and these responses were recorded in the computer. The
software package SPSS 19 was used to build the regression models presented in the Results section.
Closure = 1
FIGURE 3: CLOSURE PRINCIPLE EXAMPLE
3.2.4 Definition of the Semantic Space To be able
to locate each design in semantic space, one word was selected
from each semantic space dimension; see section 2.2 for examples. From the evaluation component “Beautiful” was selected,
for the activity component “Fast” was selected, and for the potency component “Strong” was selected. It was a subjective decision that these three terms were the ones that applied most directly to wheel rims. In an industry setting the designer would
have to choose the most applicable word for each dimension.
3.2.5 Selection of Kansei Words In order to capture
the attributes that subjects attach to the wheel rims designs, Kansei methodology is used, specifically the use of Kansei words. To
find the Kansei words for a rim, descriptions of the product were
collected from web pages and blogs where customers and users
discussed this product. Also, point of sale verbatim from web
pages that indicated how the seller approached the consumer and
sold the product were collected. In absence of this information
other sources of information could be used such as customer and
expert interviews, store intercepts and observation. All the words
(2,201) were analyzed for frequency of occurrence. The words
that had higher frequency and were used to describe the wheel
rims were selected. Eight words were selected to be included in
the study to avoid subject fatigue. The Kansei words included in
the judgment dimensions were: Aggressive, Performance, Quality, Racing, New, Clean, American, and European.
3.2.6 Product Preference Scale Product preference
was established by asking: “Do you like this wheel?” This question was answered on a 7-point Likert scale, where 1 was “not at
all” and 7 was “very”. The other judgment dimensions discussed
above (Kansei words and semantic space) where also rated on
the 7-point Likert scale. To maintain consistency among Kansei
words and semantic space only the positive word of the semantic
differential pair was used.
FIGURE 4: EXAMPLE OF STUDY TASK.
6
Copyright c 2012 by ASME
TABLE 3. JUDGEMENT DIMENSIONS PREDICTED BY DESIGN VARIABLES
4
RESULTS AND DISCUSSIONS
The key results of the study are presented and discussed in
this section. First, the results indicate that changes in the form
or shape of the wheel rims change the subjects design preference
and judgment. Various forms for the parametric models used to
characterize the subjects judgment for the wheel rims are presented. Initially, these models were based upon design variables
alone but when the model form was based upon Gestalt variables, the models produced better predictions of product preference. Additional improvements to the prediction model were realized when Kansei words and semantic space were added to the
Gestalt variables. Lastly, information from subject expertise was
added to the model; however, this knowledge did not improve the
preference prediction model.
Judgment
4.1
Wheel Rim Form Variation and Judgment
A linear regression using the design variables to predict the
“like” scale resulted in a good fit and statistically significant
model (R2 = 0.931, P < 0.05). A linear regression was performed for the rest of the judgment dimensions using the design
variables as predictors. An example of the functional form of the
equation used in the regression is shown below, b are the coefficients that the regression specifies. This information is summarized in Tab. 3. This table illustrates and confirms that variations
in the form of the wheel rim caused by design variables influenced the product preference and all other judgment dimensions.
Standarized Coefficients
R2
Dimension
Bolts
Radius
Spokes
Width
Like
0.058
0.147*
0.944*
0.123*
0.931
Fast
0.088
0.079
0.780*
-0.191
0.660
Strong
0.074*
0.124*
0.937*
0.288*
0.983
Beautiful
0.066
0.184*
0.929*
0.029
0.903
Aggressive
0.201*
0.126
0.820*
0.189
0.765
Performance
0.098
0.125
0.914*
0.163*
0.888
Quality
0.114*
0.146*
0.933*
0.194*
0.943
Racing
0.087
0.089
0.807*
0.052
0.669
New
0.004
0.229
0.717*
0.021
0.567
Clean
-0.537*
-0.097
-0.076
0.220
0.352
American
0.055
0.043
0.885*
0.283*
0.869
European
0.062
0.317
0.156
-0.301
0.219
Significant Coefficients (P < 0.05)*
4.2
Wheel Rim Design Preference
Now that it is known that the form of the wheel rim design
influences how it is judged, how these forms are preferred will
be discussed. The wheel rim design that subjects liked most
was design 15, and the least liked was design 1. In Fig. 5 the
designs are ordered by increasing average “like” ratings. This
Figure also includes the results from three models that were developed to predict “like” and are discussed in the next sections.
Inspecting how the wheel rim designs increase in the “like rating,
a discontinuity is apparent in the plot that separates the wheel rim
designs in two groups. One group is composed of designs 1, 2,
5, 6, 9, 10, 13, and 14; they have a lower average “like” rating
and all designs share the lower bound in the Spokes variable. The
other group is composed of designs 3, 4, 7, 8, 11, 12, 15, and 16;
they have a higher average “like” rating and all designs share the
higher bound in the Spokes variable. This shows that the Spokes
variables has a strong effect in the “like” rating.
JudgementDimension = b1 + b2 (Spokes) + b3 (Bolts)
+ b4 (Width) + b5 (Radius)
Table 3 shows that for the judgment dimension Like all design variables play a significant role except Bolts. However for
other judgment dimension like Clean then Bolts is the only significant design variable. Also the Table shows that more complex
judgment dimensions such as European can not be effectively
model by design variables. For most of the judgment dimensions
Spokes is significant. In ten out of twelve judgment dimensions
it is the design variable that had the most effect in all judgment
dimensions. In each of these dimensions more Spokes improved
judgment of the subjects, the only exception was the Clean dimension where less Spokes improved judgment of the subjects.
Overall the results indicate that design variables do have an
influence on the product judgment and preference. This information is important for all disciplines involved in the design of the
product, but in particular for engineering designers because they
might select values for some of the product design variables to
meet other objectives (e.g. function or weight) and be unaware of
their consequences regarding product judgment and preference.
4.3
Linear Regression Models to Predict Product
Preference
The first model presented in Fig. 5 to predict “like” employs
design variables alone and the second Gestalt variables alone.
7
Copyright c 2012 by ASME
5
Reported
Design Variables Model (Predicted)
Gestalt Variables Model (Predicted)
Gestalt + Semantic + Kansei Model (Predicted)
Like Rating
4.5
4
3.5
3
2.5
1
9
5
13
2
10 14 6 3 4 11 12 16
Wheel Rim Designs
8
7
15
FIGURE 5: WHEEL RIM DESIGNS VS LIKE RATING.
Both models provided statistically significant results and good
fits to the reported data, with the Gestalt variables model more
closely approximating the average subjects “like” data in nine
out of sixteen wheel rim designs. Table. 4 presents the contrast
between models; design variables alone do not perform as well
as Gestalt variables, therefore Gestalt variables were chosen to
develop an alternative prediction model. All models in Tab. 4
are linear models where the variables specified are multiplied by
regressed b coefficients.
The next step was to explore the effect of adding meaning and emotional information to the model. This was done by
including the semantic space and Kansei words to the Gestalt
variables model. The semantic space adds meaning information
to the models and Kansei words add emotional information to
the models. First, a regression model was built in two steps:
the first step consisted in performing the linear regression model
with Gestalt variables to predict product preference, the second
step was a stepwise regression to add the semantic space. This
stepwise regression chose the predictive variables to be included
in the model using a F-test with criterion of P  0.05 to enter
and P 0.10 to exit. This model included all Gestalt variables.
The dimensions chosen by the stepwise regression from semantic space were Beautiful and Strong. The same procedure was
implemented to build a model with Gestalt variables and Kansei words. This next model included all Gestalt variables. The
Kansei words selected by the stepwise regression were Quality,
European, Aggressive, and Performance. Table 4 also shows that
when semantic space was added to Gestalt variables the model
was improved as indicated by increased correlation coefficients,
the same trend is shown when adding Kansei words.
8
Copyright c 2012 by ASME
TABLE 4.
PRODUCT PREFERENCE MODELS SUMMARY
Model (Like =)
R2
F
Sig.
Std.
Rˆ2
Sig.
Error
D
D
design of today’s products with the ease of customization and
personalization of products.
4.5
f (DV )
0.931
91.1
0.000*
0.213
-
-
f (GV )
0.949
96.7
0.000*
0.187
-
-
f (GV, SS+ )
0.978
153.0
0.000*
0.127
0.029
0.013
f (GV, KW + )
0.987
182.8
0.000*
0.103
0.038
0.012
f (GV, KW + , SS+ )
0.990
210.4
0.000*
0.091
0.041
0.015
Significant Models (P < 0.0005)*
Variables filtered by Stepwise regression+
D reference model is f (GV )
The last model constructed used Gestalt variables, semantic space and Kansei words. This model was built in three steps:
the first step consisted of determining the linear regression model
with Gestalt variables to predict product preference. The second
step was a stepwise regression that used the same criterion used
previously to choose Kansei words; the third step also consisted
of a stepwise regression with the same criterion previously used
but with the semantic space. In addition to Gestalt variables this
last model included Kansei words Quality, European, Aggressive, and Performance, and from semantic space only Beautiful.
This is the best model achieved in this study with a R2 of 0.990
(P < 0.05, F = 210.401, n = 64). This implies that when trying
to understand how the form of a product influences the product preference, a designer can gather information from Gestalt
variables, and attach meaning and emotions that the shape might
invoke for the users. In practice a designer will be most interested in a model that predicts the most preferred design. This
last model is the one that when compared to the rest of the models presented better predicts the most preferred design (wheel rim
15).
Implications For Designers
This study focused on modeling product form preference using Gestalt principles, semantic space and Kansei words for a
specific product. The results indicate that designers could follow
the methodology presented to model form preference of other
products. However the intent of this work was to establish a base
for future formal methods that incorporate Gestalt principles, semantic space and Kansei words to aid engineering designers to
design the form of a product taking into account these dimensions in conjunction with the functionality of the product. This
paper establish two important points for the application of such
a method, first product form does influence the judgment of the
product, and second, the use of appropriately selected Gestalt
variables could better predict the judgment of the form of the
product.
In order to apply the methodology used in this study a designer would have to follow these steps. For the Gestalt variables
the designer should apply the Gestalt Principles to the product
form. A systematic method should be developed to apply Gestalt
Principles to a product form. In the Semantic Space the designer
can choose the words that apply best to the product from each
semantic space dimension. Words from the seller of the product and from the users that describe the product can be collected
from the sources mentioned, then filtered by frequency and then
by the higher frequency words used as the Kansei words. Finally
a parametric CAD model is built to generate alternative designs.
With these items a study can be built to gather the data to use to
model the product form judgment. Then the model can inform
other designers involved in the design of the product. Note that
the complexity of the models and number of subjects would most
likely have to increase in order to apply the methodology in an
industry setting.
5
CONCLUSIONS
The results of this study showed that when a designer needs
to gauge product preference from the form or shape of a product,
information from design variables, visual perception, meaning
and Kansei are useful to predict product preference. Furthermore, both design variables and Gestalt variables were able to
individually predict product preference. However, when comparing the models developed using either design variables or Gestalt
variables, the later one showed better fit. This indicates a potential advantage of discretizing the product form using a method
similar to the way subjects mentally group the design variables
of the product form. Once these models are established for a
specific product they can be used in the early design stages of the
product at the studio, but more important these models can be
4.4
Expertise Effect on Product Preference
The difference in subject expertise was explored as another
source of information to help predict product preference. An
additional variable of expertise was added to all the previously
presented models. However, this expertise variable was not significant in any of the models. This might be because the students
identified as designers have little real experience in product design particularly with regard to the wheel rim product. Subjects
with more experience and specific knowledge of the product under study could have demonstrated a difference between designers and users. Another factor that might attenuate a distinction
between the two groups is that users are more involved in the
9
Copyright c 2012 by ASME
used in the product development process by engineer designers
that are solving engineering problems that affect the final form
of the product.
[10] Orsborn, S., Cagan, J., and Boatwright, P., 2009. “Quantifying aesthetic form preference in a utility function”. Journal of Mechanical Design, 131(6), June, pp. 061001–10.
[11] Kelly, J. C., Maheut, P., Petiot, J., and Papalambros, P. Y.,
2010. “Incorporating user shape preference in engineering design optimisation”. Journal of Engineering Design,
22(9), pp. 627–650.
[12] Reid, T. N., Gonzalez, R. D., and Papalambros, P. Y., 2010.
“Quantification of perceived environmental friendliness for
vehicle silhouette design”. Journal of Mechanical Design,
132(10), Oct., pp. 101010–12.
[13] Palmer, S. E., 1999. Vision Science: Photons to Phenomenology. MIT Press.
[14] Koffka, K., 1935. Principles of Gestalt Psychology,. Harcourt Brace and Co.
[15] Marr, D., 2010. Vision: A Computational Investigation into
the Human Representation and Processing of Visual Information. The MIT Press.
[16] Osgood, C. E., Suci, G. J., and Tannenbaum, P. H., 1957.
The Measurement of Meaning. University of Illinois Press.
[17] Osgood, C. E., and Suci, G. J., 1955. “Factor analysis of
meaning”. Journal of Experimental Psychology, 50(5),
pp. 325–338.
[18] Carroll, J. B., 1969. “Review of The Measurement of
Meaning”. In Semantic Differential Technique: A Sourcebook, J. G. Snider and C. E. Osgood, eds. Aldine, Chicago,
Illinois, pp. 96–115.
[19] Nagamachi, M., 2010. Kansei/Affective Engineering. CRC
Press.
[20] Nagamachi, M., 2001. “Workshop 2 on Kansei Engineering”. In Proceedings of International Conference on Affective Human Factors Design.
[21] Schütte, S., 2005. “Engineering Emotional Values in
Product Design”. PhD Thesis, Linköpings Universitet,
Linköping, Sweden.
[22] Lévy, P., 2005. “Interdisciplinary design for the cyberspace
by an approach in kansei information - methodology and
workgroup communication tool design”. PhD thesis, University of Tsukuba, Japan.
[23] Yamamoto, K., 1992. “Japans Automotive Industry: It’s
Strength”. In Special Lectures by Top Management, Faculty
of Business Administration.
ACKNOWLEDGMENT
Authors would like to acknowledge the support of the Graduate School of the University of Notre Dame Fernandez Fellowship, and the Department of Psychology at the University of
Notre Dame for access to their subject pool. Also the authors
would like to thank the anonymous reviewers for their insights
and comments.
REFERENCES
[1] Sylcott, B., Tabibnia, G., and Cagan, J., 2011. “Understanding of emotions and reasoning during consumer tradeoff between function and aesthetics in product design”. In
Proceedings of the ASME 2011 International Design Engineering Technical Conference & Computers and Information in Engineering Conference.
[2] Pugliese, M. J., and Cagan, J., 2002. “Capturing a rebel:
modeling the Harley-Davidson brand through a motorcycle shape grammar”. Research in Engineering Design,
13(April), pp. 139–156.
[3] Orsborn, S., Boatwright, P., and Cagan, J., 2007. “Identifying product shape relationships using principal component
analysis”. Research in Engineering Design, 18(4), Oct.,
pp. 163–180.
[4] Achiche, S., and Ahmed, S., 2008. “Mapping shape geometry and emotions using fuzzy logic”. ASME Conference
Proceedings, 2008(43284), Jan., pp. 387–395.
[5] Zuo, Z. H., Xie, Y. M., and Huang, X., 2011. “Reinventing
the wheel”. Journal of Mechanical Design, 133(2), Feb.,
pp. 024502–4.
[6] Westmoreland, S., Ruocco, A., and Schmidt, L., 2011.
“Analysis of capstone design reports: Visual representations”. Journal of Mechanical Design, 133(5), May,
pp. 051010–7.
[7] Hsu, S. H., Chuang, M. C., and Chang, C. C., 2000. “A
semantic differential study of designers’ and users’ product form perception”. International Journal of Industrial
Ergonomics, 25(4), pp. 375 – 391.
[8] Linsey, J. S., Clauss, E. F., Kurtoglu, T., Murphy, J. T.,
Wood, K. L., and Markman, A. B., 2011. “An experimental study of group idea generation techniques: Understanding the roles of idea representation and viewing methods”.
Journal of Mechanical Design, 133(3), Mar., pp. 031008–
15.
[9] Hsiao, S., and Wang, H., 1998. “Applying the semantic
transformation method to product form design”. Design
Studies, 19(3), July, pp. 309–330.
10
Copyright c 2012 by ASME
TABLE 2.
WHEEL RIM DESIGNS
Width L
Width H
Width L
Width H
Radius L
Design 1, Like avg: 2.72
Design 2, Like avg: 3.11
Design 3, Like avg: 4.08
Design 4, Like avg: 4.38
Radius H
Design 5, Like avg: 2.86
Design 6, Like avg: 3.34
Design 7, Like avg: 4.61
Design 8, Like avg: 4.47
Radius L
Spokes H
Design 9, Like avg: 2.73
Design 10, Like avg: 3.22
Design 11, Like avg: 4.39
Design 12, Like avg: 4.41
Radius H
Bolts H
Bolts L
Spokes L
Design 13, Like avg: 2.98
Design 14, Like avg: 3.31
Design 15, Like avg: 4.80
Design 16, Like avg: 4.41
11
Copyright c 2012 by ASME