Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships

Abstract

In this paper, we examine aesthetic color combinations in a realistic product self-design task using the NIKEiD online configurator. We develop a similarity-based model of color relationships and empirically model the choice likelihoods of color pairs as a function of the distances between colors in the CIELAB color space. Our empirical analysis reveals three key findings. First, people de-emphasize lightness and focus on hue and saturation. Second, given this shift in emphasis, people generally like to combine colors that are relatively close or exactly match, with the exception that some people highlight one signature product component by using contrastive color. This result is more consistent with the visual coherence perspective than the optimal arousal perspective on aesthetic preference. Third, a small palette principle is supported such that the total number of colors used in the average design was smaller than would be expected under statistical independence.

Full text

Consumer preferences for color combinations: An empirical analysis of
similarity-based color relationships
Xiaoyan Deng
a,
⁎, Sam K. Hui
b
, J. Wesley Hutchinson
c
a
Fisher College of Business, The Ohio State University, Columbus, OH, United States
b
Stern School of Business, New York University, New York, NY, United States
c
Wharton School, University of Pennsylvania, Philadelphia, PA, United States
Received 3 November 2009; received in revised form 14 July 2010; accepted 14 July 2010
Abstract
In this paper, we examine aesthetic color combinations in a realistic product self-design task using the NIKEiD online configurator. We
develop a similarity-based model of color relationships and empirically model the choice likelihoods of color pairs as a function of the distances
between colors in the CIELAB color space. Our empirical analysis reveals three key findings. First, people de-emphasize lightness and focus on
hue and saturation. Second, given this shift in emphasis, people generally like to combine colors that are relatively close or exactly match, with the
exception that some people highlight one signature product component by using contrastive color. This result is more consistent with the visual
coherence perspective than the optimal arousal perspective on aesthetic preference. Third, a small palette principle is supported such that the total
number of colors used in the average design was smaller than would be expected under statistical independence.
© 2010 Society for Consumer Psychology. Published by Elsevier Inc. All rights reserved.
Keywords: Aesthetic self-design; Mass customization; Color combinations; Color relationship
The importance of aesthetics in product design has been
frequently noted (Bloch, 1995; Veryzer & Hutchinson, 1998).
Less frequently noted are the aesthetic problems faced by
consumers who must not only choose products that are pleasing
when considered in isolation, but must combine products to
form a pleasing ensemble. Nowhere is this aesthetic problem
more evident than in the choice of color combinations. Getting
dressed each day, we must combine several articles of clothing,
each of which has one or more colors. We frequently ask each
other questions like, “Does this tie go with this shirt and suit?”
or “What color sofa will go with both the rug and the chairs?”
These questions seem simple because, when we see specific
color combinations, we often have clear impressions of colors
that go well together and colors that “clash.”However, this
apparent simplicity is deceptive. Choosing colors for even a
small ensemble, say four components, using even a small
palette, say seven colors, requires searching over 7
4
= 2,401
possible combinations. Clearly, trial and error is not really
feasible. What principles guide consumers' choices of color
combinations?
In this paper, we propose a similarity-based model of color
relationships and empirically examine which relationships are
chosen most often in a realistic consumer task: the self-design of
athletic shoes. To our knowledge, this is the first such
investigation in the literature of consumer psychology, color
perception, or aesthetics (despite recent calls to study color in its
natural occurring context; e.g., Shevell & Kingdom, 2008).
Testing preferences for color combinations in aesthetic
self-design
Self-design is a form of mass customization in which
consumers partly design a product by specifying certain product
attributes in a product configurator provided by manufacturers.
For mass customization to create value for consumers, those
attributes should be ones on which consumers' preferences
differ sharply and that consumers can easily manipulate and
Journal of Consumer Psychology xx (2010) xxx –xxx
JCPS-00178; No. of pages: 9; 4C: 3, 4
⁎Corresponding author.
E-mail address: deng_84@fisher.osu.edu (X. Deng).
1057-7408/$ - see front matter © 2010 Society for Consumer Psychology. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.jcps.2010.07.005
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

evaluate with the configurator. Aesthetic self-design, in which
consumers choose only the product's aesthetic specifications
such as shape, texture, and color, meets these two conditions. A
brief examination of over 500 Web-based configurators (www.
configurator-database.com) reveals that about 50% are from
fashion industries (e.g., apparel, accessories, footwear). That
consumers use fashion to signal self-identity (Venkatesh, Joy,
Sherry, & Deschenes, 2010) also explains the popularity of
self-design in this domain. For manufactures, varying the
product shape greatly increases the cost of customization.
Changing product color, in contrast, is an economical solution
to satisfy consumers' individual aesthetic preferences. The
configurators offered by manufacturers in the apparel and
footwear industries (e.g., Adidas, Lands' End, Nike) are
characterized by providing different color palettes for different
product components and a variety of color options in each
palette. In this paper, we use the NIKEiD configurator to
examine revealed preferences for color combinations (see Fig. 1).
A similarity-based model of color relationships
Color perception and the color space
Psychologists have known for a long time that human color
perception can be represented in a three-dimensional space,
such that (1) physically different colors that appear identical
are located at the same point and (2) colors that appear to be
similar are near each other in the space and dissimilar colors
are far apart (Krantz, 1975a,b). The coordinate system for this
space is sometimes given as hue, saturation, and lightness
(e.g., Munsell, 1969), and this accords well with common
language. However, this system does not correspond well with
the neurophysiology of color vision or with mathematical models
of perceived similarity. Therefore, modern researchers have
adopted a Cartesian system in which colors are represented by
three orthogonal dimensions: one representing lightness and two
representing hue and saturation (i.e., red–green and blue–yellow
axes; see Abramov & Gordon, 1994; Mollon, 1982). The two
hue/saturation dimensions are based on Hurvich and Jameson's
(1957) opponent theory of color processing, in which neural
information from three types of cone receptors in the eye (red,
green, and blue) are transformed by subsequent processing in the
brain into a red–green system, a yellow–blue system, and a
lightness system. Opponent processes enhance color contrasts at
the boundaries of objects, such that red on one side of a boundary
intensifies responses to green on the other side of a boundary
(and vice versa, and similarly for yellow and blue). In the field of
colorimetry, the opponent processing theory underlies the
CIELAB coordinate system (Lrepresents lightness, arepresents
red–green, and brepresents yellow–blue). We use the CIELAB
system in all of our analyses.
Consumer responses to colors
Most research on color in the consumer psychology literature
has focused on the effects of specific dimensions of colors on
consumer responses (Gorn, Chattopadhyay, Yi, & Dahl, 1997).
On the hue dimension, red was found to elicit a higher level of
arousal than blue; however, products presented against blue-
colored backgrounds were liked more than products presented
against red-colored backgrounds (e.g., Bellizzi & Hite, 1992).
On the other two dimensions, Gorn et al. (1997) found that a
higher level of lightness led to relaxation, whereas a higher level
of saturation led to excitement. All of this research, however,
focused on single colors, not color combinations, and thus did
not consider the relationships between colors.
Color relationships
The most basic relationship among colors is similarity, and
the most common way of modeling color similarities is as
distance in the color space (for more general discussions of
theories of similarity as a determinant of psychological
relationships, see Medin, Golstone, & Gentner, 1993;Tversky,
1977). The exact metric for psychological distances among
colors is still a matter of active research; however, most recent
research favors the “city-block”metric (Abramov, Gorden, &
Chan, 2009; Logvinenko, & Maloney, 2006). Therefore, we use
a city-block metric to compute distance, and it is defined as
follows.
dðcolor1;color2Þ=w
j
L1−L2
j
+
j
a1−a2
j
+
j
b1−b2
j
;ð1Þ
where (L
1
,a
1
,b
1
) denotes the CIELAB coordinates of color 1,
(L
2
,a
2
,b
2
) denotes the CIELAB coordinates of color 2, and wis
an attentional weight on lightness relative to the opponent-color
plane (i.e., red–green and blue–yellow, which also represent
hue and saturation).
The city-block metric gets its name by analogy with walking
through a city, in which travel is assumed to be restricted to a
two-dimensional grid of streets, say East/West and North/South.
The walker can only move in one direction at a time and never
on the diagonal. Thus, the distance traveled is the sum of the
East/West distance and the North/South distance. This
contrasts with Euclidean distances, for which travel “as the
crow flies”is possible. The analogy also has an important
psychological meaning. It is a natural representation of
processes that focus on one dimension at a time and has
been empirically associated with analytic information proces-
sing and with the ability to attend differentially to specific
psychological dimensions (e.g., Garner, 1976; Medin et al.,
1993). In the context of aesthetic color combinations, it is of
particular value to be able to separate the contributions of the
lightness dimension (L)fromthe“opponent-color plane”(a,b)
because changes in lightness result in differing shades of the
same color (and, therefore, reductions in attention to lightness
make shades of the same color more similar). This type of
differential attention is representedbytheweight,w,inEq.(1).
Fig. 2 shows the color palette of the NIKEiD configurator
plotted in the opponent-color plane.
Theoretical perspectives on color aesthetics
Two theoretical perspectives have emerged in the literature
on empirical aesthetics that are relevant for color combination
2X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

Fig. 1. The seven shoe components and corresponding color palettes used in the NIKEiD configurator for the “shox”model of shoe.
3X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

in consumer aesthetic self-design. They share a focus on the
perceptual component of aesthetic experience (for a useful
integrative model of the many factors that have been shown to
affect aesthetic experience, see Leder, Belke, Oeberst, &
Augustin, 2004). They differ in the central principle proposed
as the primary determinant of aesthetic preference. The first,
which we will call the visual coherence principle, has its roots
in the Gestalt theories of figural goodness (Koffka, 1935). A
“good”figure is one that is easily perceived as a coherent
object. One of the main Gestalt principles is similarity (i.e., the
parts of a good figure should be similar to each other). A
related, but more general, principle is unity (Veryzer &
Hutchinson, 1998). Unity refers to a congruity among the
elements of a design such that they look as though they belong
together or as though there is some visual connection beyond
mere chance that has caused them to come together (Lauer,
1979). Visual matching of design components is a common
way of achieving unity. Both figural goodness and unity are
associated with the prototypicality of an object as member of
some natural category (for a recent inquiry of the relationship
between unity and prototypicality, see Kumar & Garg, 2010),
and prototypicality also should increase aesthetic preference
(Hekkert & Wieringen, 1990; Martindale, Moore, & Borkum,
1990).
From the visual coherence perspective two color relation-
ships should increase aesthetic preference. First, two colors
could be identical matches (i.e., the same point in the color
space). Exact matches among component parts of a product help
unify the design. Second, two colors could be distinct, but
closely related (i.e., very close in the color space), and this
should increase figural goodness. For example, navy blue,
aquamarine, and sky blue are different shades of the same color
and are close to each other in the color space (especially in the
opponent-color plane).
The second perspective is Berlyne's theory in which
aesthetic preference is determined by the arousal potential of
a stimulus, and the optimal arousal potential occurs at moderate
levels of arousal (Berlyne, 1974; Hekkert & Wieringen, 1990).
Most scholars (including critics) would agree that Berlyne's
theory was the most influential perspective of the twentieth
century and that it continues to have its followers today despite
a number of problematic empirical findings (e.g., Hekkert &
Wieringen, 1990; Martindale et al., 1990; Veryzer & Hutch-
inson, 1998). The key assumption of this perspective that puts it
in opposition to the visual coherence perspective is that arousal
potential is assumed to be strongly related to visual complexity
(Berlyne, 1974; Hekkert & Wieringen, 1990), and in most cases
increasing complexity implies decreasing similarity, unity, and
prototypicality. In terms of color combinations, this means that
distinct colors (i.e., moderately distant from each other in the
color space, such as red and yellow) should be preferred to
identical matches and closely related colors, which are favored
by the visual coherence perspective. Neither perspective
predicts that contrastive colors (i.e., maximally separated in
the color space, such as red and green) should be most preferred.
Empirical tests and statistical models of preferences for color
combinations
Early empirical research on “color harmony”examined
relationships similar to the four just described (i.e., identical,
related, distinct, or contrastive; see Moon & Spencer, 1944;
Munsell, 1969). However, the details of their theories differed
considerably and empirical tests were infrequent and
conflicting (Ou & Luo, 2006). More recently, researchers
have presented participants with pairs of color patches and
collected ratings of “harmoniousness”or “clashing.”The
presented pairs were typically based on changing one aspect of
color at a time (e.g., brightness or hue) and the empirical
results were often mixed or conflicting (e.g., Chuang & Ou,
2001; Polzella & Montgomery, 1993; Ou & Luo, 2006).
Our approach differs from that of the color harmony tradition.
First, we examine complex, realistic stimuli (i.e., athletic shoes),
not simple color pairs. Second, we use a revealed preference
approach in which participants simply design shoes that appeal to
them, rather than making judgments of harmoniousness (which
is difficult to define, both to participants and in the literature).
Most importantly, we use a similarity-based model of choice
for color pairs. When empirically estimated, this model reveals
the color relationships that exert the greatest influence on
choice, and this in turn reveals whether visual coherence or
optimal arousal provides the better explanation for aesthetic
self-design.
Fig. 3 illustrates our approach. We estimate choice
likelihood as a function of the distance between colors in the
palette for each pair of shoe components. Identical colors have
zero distance, closely related colors have small distances,
distinct colors are moderately distant, and contrastive colors
have large distances. If, for a given pair of shoe components,
exact matching is most influential, then the estimated function
will have only two levels: high choice probabilities when the
distance is zero and low probabilities when the distance is
Fig. 2. The NIKEiD colors plotted in the opponent-process color plane (i.e.,
CIELAB coordinates aand b).
4X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

greater than zero (see Fig. 3a). If closely related color pairs are
preferred, then a downward sloping function should be
observed (see Fig. 3b). The visual coherence perspective is
consistent with these two functions. The optimal arousal
perspective, however, predicts an inverted-U shape (see
Fig. 3c), which is consistent with a preference for distinct
colors. An upward sloping function (see Fig. 3d) is consistent
with a preference for contrastive colors and, as noted earlier,
inconsistent with either theoretical perspective. A U-shaped
function (see Fig. 3e) reflects heterogeneity in preferences:
some people prefer relatedness, while others prefer contrast.
Finally, it could be that color distance is not predictive of
preference (see Fig. 3f).
Empirical analyses
Method
In our experiment we asked 142 participants to go to the
NIKEiD Web site and create a Nike “shox”model shoe for
themselves. Participants first picked a style from the five
starter styles: four gender-specific “inspirations”(existing
professional designs offered by Nike) and a nearly all-white
design. They then chose a color for each of seven shoe
components (i.e., base, secondary, swoosh, accent, lace, lining,
and shox; see Fig. 1). Each component had a palette of 6 to 12
colors (in total there were 16 unique colors used in the NIKEiD
configurator). The configurator was interactive such that
whenever a color in the palette was clicked, it was automatically
applied to the corresponding area of the shoe. Upon finishing the
self-design task, they recorded the colors of their self-designed
shoe and saved it in the NIKEiD myLOCKER.
Before and after the self-design task, participants rated the
four gender-specific professional designs on a 9-point scale
(−4 = dislike a lot; 4 = like a lot). These ratings provide a
predictive validity test for the similarity-based model of color
combination that we estimate from the self-designed shoes
(see subsequent discussion).
Results
Table 1 summarizes the frequencies with which each color
was selected for each of the seven shoe design components. Our
interest, however, is in color combinations and the extent to
Fig. 3. A similarity-based model of color relationships in aesthetic self-design.
Table 1
Frequencies of each color used in each shoe area.
Shoe area Base Secondary Swoosh Accent Lace Lining Shox
Color
(1) Light steel grey 30 10 23
(2) Canteen 9 9 10 5 5
(3) Light sand 17
(4) Twilight blue 13 36 15 16 12 21
(5) Red raspberry 10 14 5 7 7
(6) Shy pink 13 13 10 20 18 24
(7) Varsity red 8 23 23 21 13 23
(8) Light chocolate 18
(9) Linen 12
(10) Treeline 14 5 1
(10) Lymon 8 6 3 5 12
(12) Ice blue 11 15 15 12 15
(13) Shock orange 11 11 7 6 12
(14) Neutral grey 30
(15) Black 21 20 18 16 19 19 26
(16) White 21 42 20 25 23 12
Note. An empty cell represents color choices that are unavailable for the
corresponding area.
5X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

which each of the four types of color relationships are used
when consumers self-design shoes. Building on our discussions
in Sections 1 and 2, we seek to explain color combinations by
relating pairwise color choice frequencies to the distance
between the two colors in a pair of shoe components. As
discussed earlier, we characterize the distance between two
colors using the city-block metric (see [1]).
We estimated the similarity-based model of choice for color
pairs by modeling each of the 7 × 6 / 2 = 21 pairwise combina-
tions of shoe components using a Poisson regression model
(with canonical log-link function), a standard methodology used
to analyze two-way contingency tables (Agresti, 2002), which is
the format of our data. That is, for each pair of components, we
specify that the observed count y
ij
(the number of shoes that use
color ifor the first area, and color jfor the second area) follows
a Poisson distribution with rate parameter μ
ij
. We then model
log(μ
ij
)as a linear function of the distance between the colors
in areas iand j.
yij ePoissonðμijÞ
logðμijÞ=β0+β1
˜
dij +β2
˜
d2
ij +β3Ifi=jg
ð2Þ
where
˜
dij denotes the city-block distance between color iand
color j;I
{i=j}
is an indicator variable for identical matches.
Note that to avoid colinearity between the linear and quadratic
terms, the distances were mean centered in each model. Also,
the attentional weight, w, was simultaneously estimated and
was the same for all pairs of shoe components.
1
We estimated
all parameters using maximum likelihood.
Parameter estimates are shown in Table 2. The estimated
value of wwas .084. This indicates that, compared to color
discriminations (i.e., the task upon which the CIELAB
coordinates are based), the aesthetic task of self-designing an
athletic shoe lowers the salience of lightness relative to the
salience of hue and saturation (i.e., the opponent-process plane).
This strong attentional effect receives statistical support from
the fact that the value of the log-likelihood function was −1572
for w= .084 but decreased to −1635 when wwas constrained to
be 1, a significantly worse fit to the data (χ
2
= 126, pb.0001).
Next we turn to color relationships. From Table 2,we
observe the following general results about color relationships.
First, the coefficient β
3
(for identical matching) is positive for 20
of 21 pairs of components and statistically significant for 13 of
21 pairs. Thus, identical matching clearly exerts a strong
influence on aesthetic color combination. Second, the linear
coefficient, β
1
, of the distance function is always negative, and it
is statistically significant in 15 of 21 pairs. This means that as
distance increases, preference decreases; thus, closely related
colors are often favored in aesthetic color combination. Finally,
20 of 21 quadratic coefficients, β
2
, are positive (indicating some
degree of U-shaped), of which 8 are statistically significant, and
the only negative coefficient is small and not significant. Thus,
there is no support for an influence of distinct colors (i.e., no
inverted-U functions), and the visual coherence perspective is
more strongly supported than the optimal arousal perspective.
As discussed earlier, U-shaped functions indicate heteroge-
neity in preferences: some people prefer relatedness, while
others prefer contrast. To fully appreciate the level of
heterogeneity for any specific pair, the linear coefficient
(generally negative) and quadratic coefficient (generally
positive) must be combined to reveal whether the distance
function (i.e., μ
ij
, the likelihood of choice, as a function of
distance, d
ij
) is U-shaped (indicating roughly equal numbers of
each type of person) or more like a reversed J (indicating more
people favor relatedness than contrast). These functions are
plotted for each pair in Fig. 4. These plots also allow us to
appreciate the relative effect sizes of the distance function
across pairs. It is easy to see that the distance functions with the
largest “upturns”for large distances involve the shox compo-
nent. In fact, all 6 pairs involving the shox component exhibit a
U or reverse-J shape, and 5 of these have statistically significant
quadratic coefficients. Thus, there was at least a large minority
of people who wanted the shox to contrast with the other shoes
colors. The shox is a unique component among athletic shoes
and the signature component for this Nike product line.
Overall, these results suggest that most people combined
colors that either matched or were closely related to each other
(i.e., were visually coherent), and some then chose a
contrastive color to highlight some components, especially
the shox. In a sense, they chose a relatively small region of the
color space for most of the colors (e.g., shades of brown, like
canteen, chocolate, and linen) and possibly one highlight
color. This suggests a small palette hypothesis (i.e., each
individual worked with a relatively small set of colors,
although that set might differ greatly across individuals).
1
We also estimated a version of the model that included fixed effects for
each color in each component. The model in [2] had a better fit than the more
saturated version. Details are available from the authors upon request.
Table 2
Parameter estimates from the distance-only model, w= .084.
Area 1 Area 2 β
1
(dist) β
2
(dist
2
)β
3
(match)
base secondary −0.0132* 0.0000 0.1838
base swoosh −0.0048 0.0001 0.3556
base accent −0.0137* 0.0001 0.3047
base lace −0.0086* −0.0001 0.6128*
base lining −0.0132* 0.0002* 0.7466*
base shox −0.0134* 0.0002* 0.7339*
secondary swoosh −0.0039 0.0003* 0.4579
secondary accent −0.0092 0.0003 0.5261
secondary lace −0.0096* 0.0003* 1.0736*
secondary lining −0.0146* 0.0001 0.3450
secondary shox −0.0149* 0.0008* −0.5059
swoosh accent −0.0067 0.0001 0.7071*
swoosh lace −0.0063 0.0000 1.5348*
swoosh lining −0.0076* 0.0001 1.0082*
swoosh shox −0.0074* 0.0002 0.9758*
accent lace −0.0151* 0.0002 1.1232*
accent lining −0.0162* 0.0001 0.6050*
accent shox −0.0223* 0.0003* 0.1410
lace lining −0.0064 0.0000 1.9463*
lace shox −0.0181* 0.0004* 0.9244*
lining shox −0.0187* 0.0003* 0.9595*
Note. *pb.05; no marker: n.s.
6X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

Using a small palette has the advantage of reducing the
complexity of the task.
In principle, of course, many closely related colors might be
used and the use of a highlight color might offset matching, so a
small palette is not a necessary consequence of the modeling
results. To test the small palette hypothesis, we first computed
the average number of colors used in self-designed athletic
shoes in our observed sample of 142 shoes. That average
number is 4.02. We then compared this number with a null
distribution of the average number of colors that would be
expected if color combinations were statistically independent
using a permutation test (details available from the authors upon
request). We find that the observed average is significantly
smaller than the average number that is expected under the null
hypothesis (4.02 vs. 5.48, pb.0001). Thus, we conclude that
consumers prefer to use a small palette of colors and closely
matching colors within this palette.
The final analysis we conducted was a test of predictive
validity for our similarity-based model. The model was estimated
using color choices for a self-designed shoe. If the model
captures some general principle of aesthetic color combination, it
should be predictive of liking for all shoes, not just self-designed
shoes. Thus, we used the estimated model to compute a “color
coordination index”for a specific shoe design by summing over
μ
ij
for all 21 pairs of components for the specific colors used in
the design. This index was computed for the 8 gender-specific
designs presented on the NIKEiD Web site. Each subject had
provided a liking rating for the 4 inspiration designs for their
gender before and after designing their shoe. For each subject, the
Spearman rank order correlation between preference ratings and
color coordination indices across the four inspirations was
computed. For the pre-design ratings, the average correlation
was .11 (t=2.33, p= .02), and for the post-design ratings, the
average correlation was .11 (t=2.19, p=.03). Thus, there was
support for the predictive validity of our similarity-based
model. We note that this is a very strong test given that the
shoes were quite different from the self-designed shoes for
most people, and there are many determinants of shoe liking.
Fig. 4. Plots of log(μ
ij
)versus d
ij
for each model (for d
ij
≠0). Note. *Linear term β
1
significant at .05 level;
#
quadratic term β
2
significant at .05 level.
7X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

General discussion
Summary of results
In this paper, we examined aesthetic color combinations in a
realistic product self-design task using the NIKEiD online
configurator. We developed a similarity-based model of color
relationships and empirically modeled the choice likelihoods of
color pairs as a function of the distances between colors in the
CIELAB color space. Our empirical analysis revealed three key
findings. First, we found that people greatly reduce the lightness
dimension, essentially squashing the traditional color sphere
into a pizza shape. Second, in this squashed space people
generally like to combine colors that are relatively close or
exactly match. Thus, shades of the same color that differ greatly
in lightness are often combined, but colors of the same lightness
that differ greatly in hue or saturation (i.e., large distances in the
opponent-colors plane) are seldom combined. These results are
clearly more consistent with the visual coherence perspective on
aesthetic preference than the optimal arousal perspective and are
consistent with other empirical results in visual aesthetics (e.g.,
Hekkert & Wieringen, 1990; Martindale et al., 1990; Veryzer &
Hutchinson, 1998).
The main exception to the visual coherence perspective was
that a large minority of people chose to highlight one
component by choosing a contrastive color far from the
“neighborhood”used for the other components. The signature
shox component of the shoe was especially likely to be
highlighted in this way. This suggests some higher order
principles are at work. Perhaps visual coherence determines the
“looks”of the whole shoe, but it is desirable to create a separate
visual figure within the shoe and use the rest of the shoe as the
background for this figure. Such higher order principles are an
interesting topic for future research.
Finally, the total number of colors used in the average shoe
design (4.0) was smaller than would be expected under
statistical independence (5.5). Using a small palette simplifies
the final design and reduces the cognitive effort required during
the self-design process. Additionally, we computed a color
combination index for the four professionally designed
“inspiration”shoes shown to each participant and found that
this index was positively correlated with preference ratings for
those shoes, providing some support for the predictive validity
of our model.
Limitations and future research
This research represents an important first step in our
understanding of aesthetic color combination for consumer
products. However, it has several limitations that highlight
directions for future research. First, we have examined a single
product, self-designed athletic shoes. Color combination (or
more generally, feature combination) is important in many other
consumer products (e.g., cars, computers, “phone-pillows”
(Gibbert & Mazursky, 2009)) and, probably more important, in
the combination of separate products into coordinated ensem-
bles when they are used (e.g., dressing for different occasions,
furnishing different types of rooms). Expanding the analysis
into these domains is an important next step.
Second, our approach used a standard revealed preference
approach that assumes that what is chosen is what is most
preferred, regardless of whether the revealed preference is more
inherent or constructive in nature (Bettman, Luce, & Payne,
2008; Simonson, 2008). Especially in the area of self-design, this
assumption might not hold. We conducted a validity test showing
that the model estimated from self-designed shoes predicted
preferences for professionally designed shoes, but much more
extensive research is needed to show that self-designed shoes are
the most preferred and that most consumers are sufficiently
skilled to be successful at aesthetic self-design. Moreover, if
many consumers do not have sufficient aesthetic skills, how can
they best be assisted? Finally, the similarity-based model of color
relationships used here was successful, but there are many
possible variations from the perspectives of both psychological
theory and mathematical models.
References
Abramov, I., & Gordon, J. (1994). Color appearance: On seeing red—Or
yellow, or green, or blue. Annual Review of Psychology,45, 451−485.
Abramov, I., Gorden, J., & Chan, H. (2009). Color appearance: Properties of the
uniform appearance diagram derived from hue and saturation scaling.
Attention, Perception, and Psychophysics,71, 632−643.
Agresti, A. (2002). Categorical data analysis, 2nd Edition. Wiley.
Bellizzi, J., & Hite, R. E. (1992). Environmental color, consumer feelings, and
purchase likelihood. Psychology and Marketing,9(5), 347−363.
Berlyne, D. E. (1974). Studies in the new experimental aesthetics. New York:
Wiley.
Bettman, J. R., Luce, M. F., & Payne, J. W. (2008). Preference construction and
preference stability: Putting the pillow to rest. Journal of Consumer
Psychology,18(3), 170−174.
Bloch, P. H. (1995). Seeking the ideal form: Product design and consumer
response.Journal of Marketing,59,16−30 July.
Chuang, M. C., & Ou, L. C. (2001). Influence of a holistic color interval on color
harmony. Color Research & Application,26(1), 29−39.
Garner, W. R. (1976). Interaction of stimulus dimensions in concept and choice
processes. Cognitive Psychology,8(1), 98−123.
Gibbert, M., & Mazursky, D. (2009). How successful would a phone-pillow
be: Using dual process theory to predict the success of hybrids
involving dissimilar products. Journal of Consumer Psychology,19(4),
652−660.
Gorn, G., Chattopadhyay, A., Yi, T., & Dahl, D. (1997). Effects of color as an
executional cue in advertising: They're in the shade. Management Science,
43(10), 1387−1400.
Hekkert, P., & Wieringen, P. C. (1990). Complexity and prototypicality and
determinants of the appraisal of cubist paintings. British Journal of
Psychology,81, 483−495.
Hurvich, L. M., & Jameson, D. (1957). An opponent process theory of color
vision. Psychological Review,64, 384−404.
Koffka, K. (1935). Principles of Gestalt psychology. New York: Harcourt,
Brace.
Krantz, D. H. (1975a). Color measurement and color theory: I. Representation
theorem for Grassmann structures. Journal of Mathematical Psychology,12,
283−303.
Krantz, D. H. (1975b). Color measurement and color theory: II. Opponent-
colors theory. Journal of Mathematical Psychology,12, 304−327.
Kumar, M. and Garg, N. (2010). Aesthetic principles and cognitive emotion
appraisals: How much of the beauty lies in the eye of the beholder? Journal
of Consumer Psychology (this issue).
Lauer, D. A. (1979). Design basics. New York: Holt, Rinehart, & Winston.
8X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005

Leder, H., Belke, B., Oeberst, A., & Augustin, D. (2004). A model of aesthetic
appreciation and aesthetic judgment. British Journal of Psychology,95,
489−508.
Logvinenko, A. D., & Maloney, L. T. (2006). The proximity structure of
achromatic surface colors and the impossibility of asymmetric lightness
matching. Attention, Perception and Psychophysics,68,76−83.
Martindale, C., Moore, K., & Borkum, J. (1990). Aesthetic preference:
Anomalous findings for Berlyne's psychobiological theory. The American
Journal of Psychology,103,53−80.
Medin, D. L., Golstone, R. L., & Gentner, D. (1993). Respects for similarity.
Psychological Review,100, 254−278.
Mollon, J. D. (1982). Color vision. Annual Review of Psychology,33,41−85.
Moon, P., & Spencer, D. (1944). Geometric formulation of classical color
harmony. Journal of the Optical Society of America,34(1), 46−59.
Munsell, A. H. (1969). A Grammar of Color. New York: Van Nostrand
Reinhold (From the original version of 1921).
Ou, L. C., & Luo, M. R. (2006). A colour harmony model for two-colour
combinations. Color Research & Application,31(3), 191−204.
Polzella, D. J., & Montgomery, D. A. (1993). Dimensions of color harmony.
Bulletin of the Psychonomic Society,31(5), 423−425.
Shevell, S. K., & Kingdom, F. A. (2008). Color in complex scenes. Annual
Review of Psychology,59, 143−166.
Simonson, I. (2008). Will i like a “medium”pillow? Another look at constructed
and inherent preferences. Journal of Consumer Psychology,18(3),
155−169.
Tversky, A. (1977). Features of similarity. Psychological Review,84, 327−352.
Venkatesh, A., Joy, A., Sherry Jr., J. F., and Deschenes, J. (2010). The aesthetics
of luxury fashion, body and identify formation. Journal of Consumer
Psychology (this issue).
Veryzer, R. W., & Hutchinson, J. W. (1998). The influence of unity and
prototypicality on aesthetic responses to new product design. Journal of
Consumer Research,24(4), 374−385.
9X. Deng et al. / Journal of Consumer Psychology xx (2010) xxx–xxx
Please cite this article as: Deng, X., et al., Consumer preferences for color combinations: An empirical analysis of similarity-based color relationships,
Journal of Consumer Psychology (2010), doi:10.1016/j.jcps.2010.07.005