Content uploaded by Per Møller
Author content
All content in this area was uploaded by Per Møller on Oct 11, 2019
Content may be subject to copyright.
CONSUMER PREFERENCES FOR VISUALLY
PRESENTED MEALS
HANS HENRIK REISFELT1,4, GORM GABRIELSEN2,
MARGIT DALL AASLYNG3, MARIA SCHMIDT BJERRE1and PER MØLLER1
1Department of Food Science
Faculty of Life Sciences
University of Copenhagen
Rolighedsvej 30, 1958 Frederiksberg C, Denmark
2Copenhagen Business School (CBS)
Solbjerg Plads 3, Frederiksberg, Denmark
3Danish Meat Research Institute (DMRI)
Maglegaardsvej 2, Roskilde, Denmark
Accepted for Publication May 21, 2008
ABSTRACT
The aim of the study was to investigate consumers’ preferences for
variations of a visually presented meal. The study was conducted in three
middle-sized Danish towns, including 768 respondents who were presented
with a computerized questionnaire that initially displayed four consecutive
series of photos. The series each consisted of eight unique photos of random-
ized food dishes arranged around the center square in a 3 ¥3 array. Five meal
components, each with two levels, were investigated. One level of each com-
ponent was used for each photo, in total 25=32 combinations.
The respondents were asked to select the meal they preferred the most, the
second most and the least, respectively. Significant interactions were found
between meal components and background variables such as, gender, age,
geographic variables, purchase store and level of education. The current
procedure can be applied to help solve a number of problems involving
consumer choices.
PRACTICAL APPLICATIONS
This study outlines an approach to use visual images for investigations of
food. Our results suggest that rather complex food stimuli of great similarity
4Corresponding author. TEL: +45-35-33-31-73; FAX: +45-35-33-35-09; EMAIL: hhr@life.ku.dk
DOI: 10.1111/j.1745-459X.2008.00202.x
Journal of Sensory Studies 24 (2009) 182–203.
© 2009, Wiley Periodicals, Inc.182
can be used to subdivide consumers based on sociodemographic background
variables. We present an efficient and cheap quick method that provides and
captures more information than an ordinary survey that focuses merely on the
most preferred option. As a prerequisite for success, stimuli should be well
known and appropriately selected. Hence, the present quick method can easily
be applied for several practical purposes, such as pretesting, labeling, product
flop prevention, and for specific optimization and selection tasks, e.g., conve-
nience meals and institutional meal services in various contexts.
The conjoint layout used allows for late-based segmentation. It further
allows for estimation on aggregate as well as individual level. The current
approach is useful for database and/or online implementation.
INTRODUCTION
To satisfy the increasing need for healthy and convenient food products,
scientists and food developers have for a long time been occupied with the
development of appropriate methods addressing customers’ food demands
and food-choice-related topics. However, food-choice behavior is difficult to
predict because it is influenced by several factors such as sociocultural deter-
minants, product differences, situation dependence and availability of food,
sensory properties and individual factors.
It has been established that sensory properties strongly influence the
liking of food (Sørensen et al. 2003). Hence, in order to further understand the
subjects’ food-choice behavior, and develop appropriate and well-liked food
products, and to avoid product failures, it is appropriate to employ and develop
approaches that rapidly and reliably evaluate consumers’ perception of various
sensory properties, e.g., appearance properties, (Jaeger et al. 2001; Köster
2003; Moskowitz and Silcher 2006).
Appearance properties are key factors in food choice, and visual impres-
sions such as color and appearance of food components usually are the first
sensory stimuli that are presented to a consumer, raising expectations about
the food. Appearance properties comprise various visual properties, including
color, physical form or shape of the stimuli, and mode of presentation. The
perceived properties may produce positive sensations leading to acceptance of
a meal, or negative sensations leading to rejection of it. Hence, visual cues and
total appearance are essentials regarding acceptance or rejection of a meal or
a food item (Cardello 1996; Hutchings 2003). Even small changes in food
appearance are vital to our preferences with regard to meals. Similarly, we deal
with many choice situations in our daily day lives in which appearance is
decisive to our choice of food. Usually, you do not taste the meat you purchase
in the supermarket; you are compelled to rely on visual inspection and the
183VISUALLY PRESENTED MEALS
resulting judgment regarding the displayed supplies. In most market research
it is rather the recollection of liking, as expressed in a preference survey, than
an actual taste experience, as expressed in an acceptance test, which is evalu-
ated, because evidently most purchase decisions are not made on the basis of
an immediate taste experience but on the memory of foods.
Most surveys and research concerning food preferences only focus on
consumers’ positive choices, retrieving no information of the remainder, e.g.,
the second most and/or least preferred, which could, however, be important to
food producers for many reasons, such as strategic improvement of food
quality, specification of consumer demands, or avoidance of product flops.
Recently, the advent of the Internet and the conquests in computerization
have opened a new avenue for visual consumer investigations, and some of the
practical implications are that the cost, time and difficulty in creating or
manipulating realistic images keeps diminishing. Thus, the use of visually
founded techniques increases in fields such as consumer preference research,
convenience meal development, packaging, labeling, and advertising. Digital
image manipulation has successfully been applied to a cross-cultural study of
consumer choices of pork meat, assessing consumers’ meat preferences,
reportedly overcoming validity difficulties in consumer surveys, and digital
images have further been used in choice options with respect to ready meal
development (Munkevik et al. 2007; Ngapo et al. 2007).
One general problem that needs to be addressed is that a given product
may be preferred over another, but not actually liked, unless an indication of
overall liking of the stimuli is actually ensured. Hence, in studies of consumer
food-choice behavior, it is important to be aware of the relationship between
appearance and liking or preference of foods. There is a body of short-term
studies that links expected or actual liking with visual exposure, and which
indicates predictive power of visually based methodology (Tuorila et al. 1998;
Cardello et al. 2000; Marcelino et al. 2001; Cardello 2003; de Graaf et al.
2005).
We conducted this study to investigate preference of variations of a
visually presented convenience meal in large groups of Danish consumers,
differing in a number of socioeconomic variables. These kinds of products
are less demanded in Denmark compared to neighboring countries such as
Sweden and Norway. Hence, our objectives were to explore effects of back-
ground variables on preference. We further hypothesized that various segments
of consumers could be identified on basis of background variables such as age,
gender, purchase store, sociogeographical variables, educational level and
eating habits. In order to imitate a real-life situation, the investigation was
conducted in shopping centers adjacent to the supermarkets investigated.
Based on the assumptions and principles of conjoint analysis we used
discrete choice experimentation in which different series of photos displayed
184 H.H. REISFELT ET AL.
combinations of variations of a meal (Cattin and Wittink 1982; Carson et al.
1994; Hensher 1994; Jaeger et al. 2001; Train 2003).
Participants were presented with the choice sets one at a time and were
asked to rank the stimuli according to the overall criterion, i.e., relative pref-
erence. In any given context preference indicates a choice between at least two
options, and we here define preference as a (forced) choice between sensory
attributes, expressing relative hedonic responses to the photos. Having respon-
dents rank all options might be tedious; conversely, by investigating, e.g., the
most, second most and least preferred choice from the series of photos, we felt
more likely to elicit valuable information without tiring the respondents.
MATERIALS AND METHODS
The investigation was carried out by means of a computerized question-
naire in shopping centers in three middle-sized Danish towns: Amager, a
suburban part of the Danish capital, Copenhagen; Næstved, a town situated on
the isle of Zealand about 75 km from Copenhagen; and Kolding, situated on
the peninsula of Jutland about 225 km from the Danish capital.
The sample comprised 768 respondents, 470 women and 298 men, who
completed the questionnaire. Respondents were recruited in shopping centers
on location, adjacent to supermarkets of interest for the investigators; “Bilka,”
a relatively large and cheap supermarket (three stores), and “Irma,” a small
supermarket with a reputation for high-quality food (one store).
The program was written in Microsoft Access and functioned interac-
tively, forcing the respondents to pick their choice to any question posed by
pressing the key marker on one of the displayed response options. Conse-
quently, the respondents were allowed progression onto the next question that
appeared on the interface once the former question had been answered and
registered in the computer database.
The program was divided into a pictorial part, which was followed by the
regular questionnaire that included a total of 26 questions that focused mainly
on geo- and sociodemographic and lifestyle variables. The 32 photos were
divided into four series, and each series of eight photos were displayed on the
screen. The series each consisted of eight unique pictures of randomized food
dishes. For each screen display, respondents were instructed to pick their most
preferred choice, second most preferred choice and least preferred choice,
respectively, from eight pictorial variations of a meal presented in randomized
series. Having pressed the key marker, the most preferred image was removed
from the display, and similarly, for the second most preferred choice. When the
respondent had selected the least preferred image of the displayed series, the
sequence was terminated, and the screen display shifted with emergence of a
new series of another eight photos until all of the series had been displayed.
185VISUALLY PRESENTED MEALS
Figure 1 displays the first of four series that respondents were presented
with in the program. The screen display thus consisted of a 3 ¥3 array in
which the questions were posed in the center square.
In order to obtain an indication of respondents’ degree of overall liking of
the visually presented stimuli, they were asked to rate two images that were
diametrically opposed to each other with respect to variations of the com-
ponents. We chose the first photo randomly from a pile of images and
subsequently paired it with its opposite. In this way the majority of meal
components used in the experiment was subjected to a hedonic evaluation. We
used a hedonic 5-point scale, with values ranking from “like very much the
look of the meal” (5) to “don’t like the look of the meal at all” (1).
Meal Components
The five meal components and their corresponding levels (in brackets)
were defined in the following way:
Dish: (modern or traditional level); this component denotes mode of
presentation.
FIG. 1. ONE OF FOUR SERIES CONSISTING OF EIGHT PHOTOS
186 H.H. REISFELT ET AL.
Vegetable mix: (root mix or wok mix).
Meat: (slices or pieces of tenderloin pork); this component represents the
carving of the meat.
Sauce: (dishes with or without sauce).
Herbs: (dishes with or without parsley).
Dish accounted for most of the variation among the components in the
investigation having the highest sum of squares (SSQ), followed by compo-
nents herbs and sauce (see Table 4).
When selecting the stimuli we were cautious to meet popular Danish
choices of meat, vegetable mixes and sauce. Moreover, the energy contents
and constitution of macronutrients of the meals largely were in accordance
with guidelines from Nordic Nutrition Recommendations (NORDEN 2004).
Three base ingredients were used for all photos: tenderloin pork, potatoes
(either white and cooked, or fried, brown boats) and vegetables.
Dish was presented in a modern and traditional way, respectively. The
former was defined as a dish with fried potato boats, and glasses were used if
the component sauce was present. For the component variant meat in pieces,
the meat pieces were arranged on spears and whole leaves of parsley were
used. The traditional dish was arranged with cooked white potatoes, and the
sauce was arranged in a small puddle, and the parsley had been chopped and
poured over the potatoes.
Two different vegetable mixes were used; root mix and wok mix.
Regarding meat, the pork tenderloins were cut in 1-cm-thick pieces and
weighed 50 g on average. The smaller pieces of meat were approximately
1¥1 cm in size, weighing 10 g on average. The slices were fried for 4 min,
and the pieces for 3 min, respectively. Averages of 10 g of virgin olive oil were
used for each pan. The meal components, sauce (whisky sauce) and herbs
(parsley), varied in the series by being either present or absent.
The dishes were arranged placing the vegetables on the plate approxi-
mately between 6:00 and 10:00 a.m., the potatoes approximately between
10:00 a.m. and 1:00 p.m., the sauce approximately between 1:00 and 3:00
p.m., and the meat covering the rest of the plate.
Materials
The meals were placed on white plates with a diameter of 28 cm. A Nikon
F-70 camera (Nikon Inc., Melville, NY) was positioned 0.5 m above the plates
to obtain photos containing all the components on the plate with as good a
resolution as possible. The photos were photographed at 1/30 s, ISO 200 and
exposure 8. The wall behind was covered with a light-gray board to obtain a
uniform background, and a reproduction stand was installed to ensure constant
lighting conditions in the room where the photographs were taken. Four
187VISUALLY PRESENTED MEALS
laptops, Dell C600 (Dell Inc., Round Rock, TX), each with 14-in. interfaces,
were used to run the program. The digital photos were of high-resolution
(1024 ¥768) pixels, stored as bitmap images.
Experimental
The applied design was a 25conjoint layout combining a balanced block
design and an incomplete ranking test. In this factorial design, 25=32 photos
were allocated into four blocks (fractionals, series) of eight, so that each level
of each of the five meal components appeared equally often on the screen, and
at various positions. The same layout was repeated for all respondents. Thus,
the main and two-way effects were balanced with respect to the block, thereby
enabling estimation of the main effects and two-way interaction effects of the
meal components, irrespectively of the block on the individual level as well as
on the aggregate level.
The variable Position described the position of the items in the 3 ¥3 array
used for all blocks (Table 1).
Data Analyses
Statistical analyses were performed using the SPSS statistical software
package version 14.0 (SPSS Inc., Chicago, IL).
The study presented a conjoint layout using linear regressions in which
preference data were analyzed by using analysis of variance. The significance
level was Pⱕ0.05.
The position of the photos on the screen influenced respondents’ choices
and the effect has been corrected for as “design” variables. Likewise, respon-
dents were shown only eight photos at a time, which resulted in a series effect
that has also been corrected for.Age and gender were regarded most important
variables, and effects related to these variables have been corrected for in the
analyses of data as categorical variables (factors) in the analysis.
The five treatment factors (meal components) may be represented as
dummies Xi
C~the factor dish, 0 =traditional and 1 =modern; Xj
V~vegetable
mix, 0 =wok mix of vegetables and 1 =root mix of vegetables; Xk
M~meat,
TABLE 1.
POSITION OF THE PHOTOS
12 3
4 Question posed 5
67 8
188 H.H. REISFELT ET AL.
0=tenderloin meat in pieces and 1 =tenderloin slices; Xl
S~sauce, 0 =without
sauce and 1 =with sauce; Xm
H~herb, 0 =without herbs and 1 =with herbs.
The 32 photos consisted of the 32 combinations of the five factors and
may be indexed by i,j,k,l,m.
We defined a factor Block (B:b=1, 2, 3, 4) allocating the 32 photos to the
four blocks b=b(i,j,k,l,m) indicating to which of the four blocks a photo
belonged.
Dijklm was a dummy indicating the position of the photo (on the screen):
0=lower right corner, 1 =upper left corner, including the diagonal. For con-
venience, Position has been reduced to a two-level “design” factor (level 1 ~
position 1, 2, 3, 4 and 6; and level 2 ~position 5, 7 and 8) because more than
95% of the variation between the eight positions was explained by these two
levels (not shown).
As such, the core of the analysis of variance model – with only main
actions – looked like this:
EY D X X X X X
ijklms b
B
ijklm C i V j M k S l H m
()
=+ + + + + + +μγ δ β β β β β
CV MS H
in which E(Yijklms) denoted the expected response of subject sto photo i,j,k,l,
mcorrected for the effect of photos displayed and the effect of the block.
We have included the 10 two-factorial interactions between the meal
components factors (Table 5), and gender and age were included to study the
effect of these factors on the treatment factors (Table 6). We included
sociodemographic background variables investigating age, gender, family,
economy, geography, and educational level and variables describing eating
habits (Table 7).
Because of limitation considerations, we have included here only the
most significant questions with corresponding response options from the full
wording questionnaire (Tables 6 and 7).
Likewise, we ran the model with background variables for main effects
only, except for the interaction: gender ¥educational level ¥sauce (Table 8)
that was investigated with particular reference to previous Danish findings.
Within each block, subjects ranked the photos they liked the most, the
second most and the least, leaving the last five photos of a block with no
mutual ranking.
The 32 photos were the explanatory variables/factors. To fit a model to all
the data, each of the 32 photos must have a response. Thus, the method used
had to take into account the photos not chosen, representative of “missing
values” in this context.
The values for the hedonic scores within a block were: 2, 1, 0 and -1,
where 2 represented the score of the most preferred choice, 1 represented the
189VISUALLY PRESENTED MEALS
score of the second most preferred choice and -1 represented the score of the
least preferred choice. Finally, 0 represented the score of each of the five
images of a block with no mutual ranking.
To calculate the main effect of a factor, the difference between the means
of the 16 scores on each of the two factor levels was used as the sufficient
statistic. As the blocks are orthogonal to the (treatment) factors, this calcula-
tion could be performed across blocks.
Because of the central limit theorem, therefore, the sufficient statistics for
each subject are approximately normal. Calculating means across subjects
would make the approximation to normality even better.
Furthermore, the sum of scores for a specific factor level was the fraction
of times this level was present in the most preferred photo with weight 2 plus
the fraction of times this level was present in the second most preferred photo
with weight 1 plus the fraction of times this level was present in the least
preferred photo with weight 1.
This means that although the scores were (incomplete) ranks, the suffi-
cient statistics were means of weighted ranks – or smoothed ranks. By using
the ranks as scores one theoretical difficulty arose: for each subject and each
block the rating or scores of the eight photos comprising the block were given
by the numbers 2, 1, 0, 0, 0, 0, 0, -1. Thus, the mean of the scores in each block
was 0.25 and thereby the mean of the 32 scores given by a subject was also
0.25 – and the grand mean of the 32 scores given by the 768 subjects was
also 0.25.
Furthermore, for each subject and each block the SSQ was (2 -0.25)2+
(1 -0.25)2+5¥(0 -0.25)2+(-1-0.25)2=5.5. For each subject, therefore,
we had the total SSQ =4¥5.5 =22.
This meant that the grand mean was fixed – and no degrees of freedom
were used to estimate it. Furthermore, the total SSQ was fixed being number
of subjects ¥22 =768 ¥22 =16,896. Thus, data could be considered as stan-
dardized, having for each respondent the same mean (=0.25) and the same
SSQ (=22).
This implied that from a theoretical point of view the usual F-test in the
ANOVA was not valid. However, the numerator was still Chi-square distrib-
uted and the usual F-test should have been replaced by a Chi-square test. As
the number of degrees of freedom in the denominator in the present case was
large, the Chi-square test and the F-test were approximately equal – therefore,
we use the F-test as usual.
RESULTS
The data presented have been calculated for all four stores investigated.
190 H.H. REISFELT ET AL.
Regarding respondents’ choice of photos, the most preferred choices were
photos in Fig. 2A–C. The least preferred choices were photos in Fig. 2D–F.
Tables 2 and 3 present a survey that combines data on gender and age
groupings of the sample (n=768) with data about stores and towns where the
investigation was conducted. The most frequently chosen most preferred
photos were characterized by the meal components modern dish; meat in
pieces; and absence of herbs.
The most frequently chosen least preferred photos were characterized by
the traditional dish without sauce and herbs.
TABLE 2.
DISTRIBUTION OF GENDER, STORE AND TOWN
Gender Store and town (S/T) Total
Bilka
Amager
Bilka
Kolding
Bilka
Næstved
Irma
Amager
Female Count 134 132 137 67 470
% within S/T 63.8 61.1 60.9 57.3 61.2
Male Count 76 84 88 50 298
% within S/T 36.2 38.9 39.1 42.7 38.8
Total Count 210 216 225 117 768
% within S/T 100.0 100.0 100.0 100.0 100.0
TABLE 3.
DISTRIBUTION OF AGE OVER STORE AND OVER TOWN
Store and town (S/T) Total
Bilka
Amager
Bilka
Kolding
Bilka
Næstved
Irma
Amager
Age
15–24 Count 37 43 72 19 171
% within S/T 17.6 19.9 32.0 16.2 22.3
25–34 Count 66 50 35 40 191
% within S/T 31.4 23.1 15.6 34.2 24.9
35–44 Count 34 49 38 21 142
% within S/T 16.2 22.7 16.9 17.9 18.5
45–54 Count 37 41 39 15 132
% within S/T 17.6 19.0 17.3 12.8 17.2
55–64 Count 25 25 24 15 89
% within S/T 11.9 11.6 10.7 12.8 11.6
65+Count 11 8 17 7 43
% within S/T 5.2 3.7 7.6 6.0 5.6
Total Count 210 216 225 117 768
% within S/T 100.0 100.0 100.0 100.0 100.0
191VISUALLY PRESENTED MEALS
Notably, all photos were in each choice option, although the frequency of
preference choices varied greatly.
The scores of the most preferred, second most preferred and least pre-
ferred choices (see Fig. 2A–C) are used in Tables 4–8 corresponding to the
description of the model in the data analysis.
FIG. 2. A–C. PHOTOS OF THE MOST PREFERRED CHOICES; D–F. PHOTOS OF THE
LEAST PREFERRED CHOICES
192 H.H. REISFELT ET AL.
TABLE 4.
ANALYSIS OF VARIANCE FOR PREFERENCES
Factor (1 ~success) SSQ DF Pb
Modern dish (1 ~modern) 732,201 1 0.000 0.345
Root mix (1 ~root mix) 31,510 1 0.000 -0.072
Meat in slices (1 ~pork in slices) 8,719 1 0.000 -0.039
Sauce (1 ~with sauce) 36,107 1 0.000 0.077
Herbs (1 ~with herbs) 111,856 1 0.000 0.135
R2=0.065 (Adjusted R2=0.065).
Below: the five meal components (factors) and corresponding levels.
Factors: dish: 0 =traditional and 1 =modern; vegetable mix:
0=wok mix of vegetables and 1 =root mix of vegetables; meat:
0=tenderloin pork meat in pieces and 1 =tenderloin pork in slices;
sauce: 0 =without sauce and 1 =with sauce; herbs: 0 =without
herbs and 1 =with herbs. These factors are referred to in
Tables 4–8.
A positive effect denotes that a factor =1 is preferred more than the
corresponding factor =0, whereas a negative effect denotes the
opposite. The bvalues denote the relationship between estimated
parameters. Data have been corrected for effects of position and
series.
TABLE 5.
TWO-FACTORIAL INTERACTIONS BETWEEN
MEAL COMPONENTS
bSSQ Pvalue Veg. mix Meat Sauce Herb
Dish -0.155 0.106 -0.061 -0.208
37.033 17.404 5.631 46.218
0.000 0.000 0.003 0.000
Veg. mix -0.021 0.093 0.085
0.667 13.406 10.038
0.306 0.000 0.000
Meat -0.007 0.073
0.072 8.240
0.737 0.000
Sauce -0.031
1.500
0.125
Adjusted R2=0.65–0.69 for the models investigated.
The bvalues denote the relationship between estimated parameters,
i.e., positive values denote an additional or synergetic effect,
whereas negative values denote a substitution effect. Significant P
values are bolded, bvalues in italics. DF =1 for all SSQ. Data have
been corrected for effects of position and series.
193VISUALLY PRESENTED MEALS
Dish accounted for most of the variation among the components in the
investigation having the highest SSQ, followed by components herbs and
sauce. The modern dish being the more popular one, components herbs
and sauce were in general preferred by respondents, whereas the wok mix and
meat pieces were the more popular choices with respect to these variants.
Interactions involving the component dish explained more of the varia-
tion in the models, however, the component vegetable mix covaried signifi-
cantly with sauce and herb.
The modern dish was selected significantly more often than the tradi-
tional one, and variables gender and age significantly influenced choice of
dish; women preferred the modern one more than men, whereas preference for
the traditional dish was positively correlated with age, and elderly subjects
preferred this variation more than young subjects.
Customers from Irma Amager preferred the modern dish more often than
customers from Bilka, and comparing urban respondents (Amager ~Copen-
hagen) to rural respondents (Næstved and Kolding) the former group preferred
the modern dish the more, whereas Jutlanders preferred the traditional dish
significantly more often than Zealanders.
TABLE 6.
EFFECTS OF INTERACTIONS BETWEEN MEAL COMPONENTS AND AGE, AND
BETWEEN MEAL AND GENDER
Question
posed
Response
options
Dish
success =
modern
Veg. mix
success =
root mix
Meat
success =
meat slices
Sauce
success =
with sauce
Herbs
success =
with herbs
Gender Man (b)* 000 0 0
Woman (b)0.158 0.005 0.021 -0.178 -0.002
SSQ 36.347 0.043 0.635 46.214 0.006
F(1,24558) 56.254 0.064 0.936 68.631 0.009
Pvalue 0.000 0.921 0.333 0.000 0.925
Age 15–24 (b)0.261 -0.002 -0.245 0.125 0.055
25–34 (b)0.414 0.041 -0.249 0.087 0.064
35–44 (b)0.322 0.067 -0.152 -0.058 0.054
45–54 (b)0.148 0.052 -0.135 -0.211 0.038
55–64 (b)0.249 0.088 -0.101 -0.146 0.003
65+(b)* 000 0 0
SSQ 75,791 5.593 31.624 97.422 2.796
F(5,24558) 23.46 1.65 9.32 28.94 0.827
Pvalue 0.000 0.144 0.000 0.000 0.530
The success criterion in Tables 6, 7 and 8 is defined as the most preferred component representing a
positive bvalue, whereas the less preferred component is denoted by a negative bvalue. Significant P
values are bolded. bvalues are italicized. SSQ- and Fvalues are included in the table. Data in Table 6
have been corrected for effects of position and series.
*brepresents the reference group and is thereby fixed to zero.
194 H.H. REISFELT ET AL.
TABLE 7.
INTERACTIONS BETWEEN MEAL COMPONENTS AND OTHER VARIABLES
Question posed/response options Dish Veg. mix Meat Sauce Herbs
When did you last eat a meal? b(slope) 0.007 -0.003 0.005 0.032 -0.008
(breakfast/brunch/lunch/supper) SSQ 0.357 0.064 0.234 8.416 0.482
0–1/1–2/2–3/3+hours F(1,24556) 0.553 0.094 0.344 12.503 0.713
P0.457 0.759 0.558 0.000 0.399
How many times did you prepare the b(slope) -0.006 -0.003 -0.017 -0.002 -0.003
main meal at home last week? SSQ 1.209 0.345 8.746 0.127 0.316
0/1/2/3/4/5/6/7+F(1,24556) 1.871 0.509 12.888 0.189 0.468
P0.171 0.476 0.000 0.664 0.494
How many times did you eat dinner b(slope) 0.016 -0.005 -0.002 -0.011 0.000
alone last week? SSQ 4.348 0.381 0.040 2.084 0.001
0/1/2/3/4/5/6/7+F(1,24556) 6.731 0.562 0.058 3.095 0.001
P0.009 0.454 0.809 0.079 0.972
How much time do you usually spend b(slope) 0.051 0.026 0.016 0.018 -0.003
preparing the main meal? SSQ 10.852 2.906 1.006 1.333 0.031
0–15/15–30/30–45/45+minutes F(1,24556) 16.805 4.281 1.481 1.980 0.046
P0.000 0.039 0.224 0.159 0.830
Do you have any children? Yes (b)0.110 -0.028 -0.099 -0.033 -0.018
Yes-No No (b)* 00 000
SSQ 12.187 0.818 9.896 1.130 0.321
F(1,20364) 18.917 1.205 14.581 1.676 0.474
P0.000 0.272 0.000 0.195 0.491
Level of education: ⱕ10 years 1 (b)-0.025 -0.001 -0.026 0.077 -0.033
Youth education 2 (b)0.192 0.057 -0.003 0.068 0.014
Advanced (2–4 years) 3 (b)0.229 0.039 0.012 0.064 0.047
Academic or similar (5 years or more) 4 (b)0.386 0.041 0.057 -0.060 0.030
Other educational options 5 (b)* 00 000
SSQ 98.208 2.807 2.667 9.892 5.270
F(4,24550) 38.223 1.033 0.982 3.674 1.949
P0.000 0.9388 0.416 0.005 0.099
Purchase supermarket Irma (b)0.164 0.109 -0.022 -0.058 -0.008
Irma–Bilka Bilka (b)* 00 000
SSQ 20.945 9.224 0.370 2.595 0.047
F(1,24576) 32.457 13.590 0.545 3.854 0.069
P0.000 0.000 0.460 0.050 0.793
Respondents place of residence Urban (b)0.109 0.049 -0.032 -0.070 0.017
Urban–Rural Rural (b)* 00 000
SSQ 17.299 3.469 1.486 7.099 0.442
F(1,24556) 26.800 5.110 2.189 10.546 0.655
P0.000 0.024 0.139 0.001 0.419
Region of residence Zealand (b)0.147 0.069 -0.005 -0.074 0.000
Zealand–Jutland Jutland (b)* 00 000
SSQ 26.666 5.949 0.027 6.819 0.000
F41.338 8.763 0.039 10.129 0.000
P0.000 0.003 0.843 0.001 0.996
Significant Pvalues are in bold. bvalues are italicized. Sum of squares (SSQ)- and Fvalues are
included in the table. Data in Table 7 have been corrected for effects of position and series.
*brepresents the reference group and is thereby fixed to zero.
195VISUALLY PRESENTED MEALS
With regard to preference for vegetable mixes, no substantial differences
determined by age or gender were observed. On aggregate, respondents pre-
ferred the wok mix significantly more than the root mix, however, respondents
who preferred the root mix spent more time preparing meals.
In general, respondents preferred meat in pieces more often than meat in
slices, and dishes with herbs were selected significantly more often than dishes
without herbs. Women clearly preferred dishes without sauce compared with
men and, for respondents with children, a similar preference was observed.
Preference for sauce decreased with increasing age, however, for respondents
aged 65+years (the reference group with respect to age) the preference was on
a level similar to that of the group aged 35–44 years. Notably, respondents’
preference for sauce was positively correlated with amount of time passed
since consumption of their last meal.
Educational level leads to significant differences in preferences. Respon-
dents’ educational level significantly covaried with preference for sauce such
that length of education was positively correlated with preference for absence
of sauce. Jutlanders preferred sauce significantly more than Zealanders; this
was also the case for rural versus urban respondents.
Overall Liking
The ratings of the questions on overall liking of the stimuli are based on
a modern dish and a traditional one, using opposite levels of the other four
meal components.
Figure 3A,B shows the two randomly chosen meal photos that were used
for assessment of respondents’ overall-liking of the meal attributes.
TABLE 8.
INTERACTIONS BETWEEN GENDER AND EDUCATIONAL LEVEL AND SAUCE
Question posed/response option Sauce
Level of education/response options: Man Woman
ⱕ10 years 1 (b)0.000 -0.064
Youth education 2 (b)0.061 -0.035
Advanced (2–4 years) 3 (b)0.036 -0.095
Academic or similar (5 years or more) 4 (b)-0.127 -0.084
Other educational options 5 (b)* 0*-0.120
Level of education ¥Gender ¥Sauce SSQ 14.192
Level of education ¥Gender ¥Sauce F(4,24547) 5.248
Level of education ¥Gender ¥Sauce P0.000
Significant Pvalues are in bold. bvalues are italicized. Sum of squares (SSQ)- and Fvalues are
included in the table. Data in Table 8 have been corrected for effects of position and series.
*brepresents the reference group and is thereby fixed to zero.
196 H.H. REISFELT ET AL.
Table 9 shows the total distribution of respondents’ hedonic scores for the
two randomly chosen meal photos that were assessed with respect to overall-
liking of appearance. Respondents’ mean hedonic scores of the two photos
were 4.0 (⫾SD 0.829) for the modern dish variant and 3.1 (⫾SD 1.029) for the
traditional variant, indicating general acceptance of overall liking of the
stimuli (1 =do not like at all, 5 =like very much).
DISCUSSION
Meal appearance is mostly influenced by cultural factors and is as such
not defined by global, uniform consensus. In fact, some of the most valued
restaurant cuisines in the world (Indian, Chinese) have rather common self-
FIG. 3. A AND B – THE TWO PHOTOS USED FOR THE HEDONIC RATING
From left to right. Figure 3A – “modern dish” and Figure 3B – “traditional dish.”
TABLE 9.
DISTRIBUTION OF HEDONIC SCORES FOR THE TWO
LEVELS OF THE COMPONENT DISH
Traditional dish Total
12345
Modern dish 1 2 1 0 3 0 6
219624545
35202244596
4 23 111 102 160 28 424
51576305323197
Total 46 217 160 284 61 768
197VISUALLY PRESENTED MEALS
service for composition of meals, whereas in the French and Italian cuisines
the mode of presentation is rather the chef’s privilege (Hutchings 2003).
To provide a resonant background for the discussion of our results, we
therefore focus on a national report based on data from Danish consumers
(n=1837). The aims of this report, a.o., were to describe dietary habits of the
Danish population and to analyze the influence of social background on
dietary habits and to analyze the association between dietary habits and other
lifestyle determinants, hereby reviewing relevant Danish consumption fre-
quencies and preference studies with food and meals (Groth and Fagt 2003).
With regard to gender, women are more focused on vegetables in a meal
context, while men are more focused on meat, and educational level is an
established determinant for Danish men’s food choices with regard to health,
whereas no isolated factor can predict Danish women’s food-choice behavior,
according to Groth and Fagt (2003).
Concordantly, the present group of most educated men had a higher
preference for meals without sauce than other groupings of men (Table 8).
Moreover, a significant positive correlation was found between the amount of
time passed since last meal and choices of meals with sauce, suggesting that
sauce might be associated with a state of hunger. This observation holds in
particular for men. The sauce component was the only one that deviated
appreciably in energy density in the presented meals. The variable educational
level certainly influenced respondents’ choice of dish and sauce, and this
might be caused by the association between absence of sauce in a meal and
low-fat and health benefices. It is likely that respondents with a higher edu-
cational level are more interested in food, or food-health-related issues and
trends. They might be more susceptible to be influenced by health campaigns
than less-educated respondents (Groth and Fagt 2003).
Women chose a meal without sauce more often than men did and they
preferred the modern dish. This might be because the modern dish was more
associated with a healthy choice, as some of the female respondents indicated
when completing the questionnaire. A difference in energy intake can not be
claimed as such, however, it is an interesting observation that some of the
present stimuli might associate differently in terms of health perception.
It has been suggested that Danish women in general are more conscious
of health than Danish men; this could explain why they more often chose a
healthier-looking and less energy-dense meal option (Groth and Fagt 2003).
We can not exclude the possibility that respondents’ preference choices have
been influenced by the meal construction, using a tripartite entrée, starch and
vegetables structure that might be associated with a “proper meal” (Marshall
and Bell 2003).
Age is the most important variable with respect to choice of dish in the
present experiment. The components, traditional dish, meat in slices and sauce
198 H.H. REISFELT ET AL.
are positively correlated with age. This is most probably because of cultural or
habitual determinants. Traditionally, tenderloin pork is more expensive and
well estimated than meat in pieces which is often provided by inferior meat
cuts. It has been suggested that elderly people in general tend to be more
reluctant toward new foods, relying on habit and thus choosing the more
old-fashion looking traditional dish consisting of a proper slice of meat and
sauce (Groth and Fagt 2003).
Another explanation might be that the mode of presentation of the
modern dish by some respondents might have been perceived as inappropriate,
e.g., the meat pieces on spears and the use of glasses for sauce, in line with
findings from other studies (Wheately 1973; Cardello et al. 2000; Kennedy
et al. 2005).
The significant differences between the results found with respect to
place of residence and purchase store, e.g., Jutland versus Zealand and the
purchase stores Bilka versus Irma, reflect a pattern in which urban respondents
tend to prefer the modern dish without sauce, whereas the opposite choice
tends to be more pertinent to rural respondents, especially those living in
Jutland.
Respondents from Copenhagen prefer the root mix more than the
Zealanders, who in turn prefer it more than the Jutlanders. The number of
restaurants and stores is larger in towns and urban areas than in rural areas
which are more dominant in Jutland. This, and cultural traditions of food
choice, might explain these results. With respect to the purchase supermarkets,
customers from Irma are associated with better overall education than Bilka
customers. Irma customers spend more time cooking and are better informed
about food and more interested in cooking. In line with this, Irma customers
prefer the root mix associated with vegetables that are relatively more time
consuming to prepare than the wok mix, generally favored by Bilka customers.
CONCLUSION
The consumer investigation revealed significant results and subtle differ-
ences between genders, age groups, regions and purchase stores, and levels of
education.
The mode of presentation, as expressed by the component dish,
accounted for most of the variation among the meal components; the modern
dish was significantly more preferred by women compared with men, urban
with rural respondents, Zealanders with Jutlanders, “Irma” with “Bilka” cus-
tomers, respondents with children with childless, and positively correlated
with respondents’ educational level. However, preference for the traditional
dish and the component meat in slices was positively correlated with age.
199VISUALLY PRESENTED MEALS
Women preferred absence of sauce significantly more than men. Notably,
in terms of educational level, the sole group of men who likewise preferred
absence of sauce was the highest one.
Urban respondents and “Irma” purchase store customers preferred root
mix more than rural respondents and “Bilka” purchase store customers,
respectively.
This study does not establish underlying causes for the current cultural
differences in meal preferences.
PERSPECTIVES
Would the results differ if subjects actually tasted the meals in the
experiment?
Because this investigation deals with well-known stimuli, it is reasonable
to assume that results would not differ substantially, if it was possible to have
people actually taste all the meals used in the experiment. Hence, it is a
prerequisite for success that measurements should apply well-known and
appropriately selected stimuli in order to ensure realism.
To warrant that the consumers’ (quality) expectations are met, the pro-
cedure described in this paper should be followed by an acceptance test before
launching a product on the market. Turner and Collison (1988) found that
anticipated preference scores were lower than acceptance scores regarding
complete meals and conclude that preference surveys may underestimate
acceptance of good quality food.
If, however, the aim of an investigation is to rank order a set of stimuli, with
due consideration to the prerequisite mentioned earlier, it is of no importance if
absolute acceptance can not be deduced from the experiment. If there is a
one-to-one relationship between rank-ordered preferences, based on photos of
stimuli, and the eventual rank order of acceptabilities, a procedure like the one
used here produces all the necessary information needed to make decisions
about which of a set of variants of a product should be launched on the market.
Other prospective steps to be taken into consideration in the elaboration
of the procedure are the assigning of monetary value to the stimuli used in the
procedure, a clarification of the adequate number(s) of stimuli, and the posi-
tioning and size of stimuli on the screen. When putting this procedure to the
test, we pondered upon the number of rankings requested, e.g., the two most
and least liked, or can one obtain valuable information from ranking, say, all
eight or even more choice options?
Notably, the present experimental design allows for a posteriori or latent-
based segmentation, based on the individual scores in the investigation, pos-
sibly expanding the potential when examining more complex data from an
200 H.H. REISFELT ET AL.
investigation (Ben-Akiva and Lerman 1985; Green and Krieger 1991; Train
2003). The potential depends on the investigated variables and the accuracy of
the response options formulated in the questionnaire, because the combination
of these parameters is decisive with regard to the degree of detailed informa-
tion that can be gained from the investigation. In contrast, the more traditional
approach, in which a priori segments are selected by theory-driven assump-
tions about the preferences of the respondents (Hahn 1997), is likely to be
outperformed by the latent-based segmentation in most cases because it pre-
sumably fails to elicit or gather the same amount and/or quality of information
as provided by the former approach, which is more viable to catch uncalcu-
lated segmental surprises and unravel perspicuous (h)in(d)sights.
The present procedure can easily be implemented on the Internet, and in
conjunction with detailed information about consumers held in databases,
this opens up a vast range of possibilities for investigations of consumer
preferences. Further, the procedure can be developed in various ways, and
further studies and theoretical work are needed for elaboration of the approach
and to further explore the field of visual preference in food choice.
ACKNOWLEDGMENTS
The authors would like to thank Torben Kvamm (DMRI) for developing
and facilitating the computer program. We are thankful to Irma and Bilka for
assistance and to Agrova; Danish Crown; Nestlé, Denmark; Lipton, Denmark;
and LU, Denmark for providing samples and incentives used for the
investigation.
REFERENCES
BEN-AKIVA, M. and LERMAN, S.R. 1985. Discrete Choice Analysis:
Theory and Application to Travel Demand, MIT Press, Cambridge, MA.
CARDELLO, A.V. 1996. The role of the human senses in food acceptance. In
Food Choice, Acceptance and Consumption (H.L. Meiselman and H.J.H.
Macfie, eds.) pp. 1–82, Blackie Academic and Professional, London.
CARDELLO, A.V. 2003. Consumer concerns and expectations about novel
food processing technologies: Effects on product liking. Appetite 40,
217–233.
CARDELLO, A.V., SCHUTZ, H., SNOW, C. and LESHER, L. 2000. Predic-
tors of food acceptance, consumption and satisfaction in specific eating
situations. Food Qual. Prefer. 11, 201–216.
201VISUALLY PRESENTED MEALS
CARSON, R.T., LOUVIERE, J.J., ANDERSON, D.A., ARABIE, P.,
BUNCH, D., HENSHER, D.A., JOHNSON, R.M., KUHFELD, W.F.,
STEINBERG, D., SWAIT, J. ET AL. 1994. Experimental analysis of
choice. Mark. Lett. 5, 351–368.
CATTIN, P. and WITTINK, D.R. 1982. Commercial use of conjoint analysis:
A survey. J. Mark. 46, 44–53.
DE GRAAF, C., KRAMER, F.M., MEISELMAN, H.L., LESHER, L.L.,
BAKER-FULCO, C., HORSCH, E.S. and WARBER, J. 2005. Food
acceptability in field studies with US army men and women: Relationship
with food intake and food choice after repeated exposures. Appetite 44,
23–31.
GREEN, P.E. and KRIEGER, A.M. 1991. Segmenting markets with conjoint
analysis. J. Mark. 55, S.20–S.31.
GROTH, M.V. and FAGT, S. 2003. Danskernes Kostvaner. Ministeriet for
Fødevarer, Landbrug & Fiskeri, Fødevaredirektoratet, København.
HAHN, C. 1997. Conjoint- und Discrete Choice-Analyse als Verfahren zur
Abbildung von Präferenzstrukturen und Produktauswahlentscheidungen.
Ein theoretischer und computergestützter empirischer Vergleich. In
Betriebswirtschaftliche Schriftenreihe, Bd. 80. Lit-Verlag, Münster.
HENSHER, D.A. 1994. Stated preferences analysis of travel choices – The
state of practice. Transportation 21, 107–133.
HUTCHINGS, J.B. 2003. Expectations and the Food Industry: The Impact of
Color and Appearance, Kluwer Academic/Plenum Publishers, NewYork.
JAEGER, S.R., HEDDERLEY, D. and MACFIE, H.J.H. 2001. Methodologi-
cal issues in conjoint analysis: Acase study. Eur. J. Mark. 35, 1217–1237.
KENNEDY, O.B., STEWART-KNOX, B.J., MITCHELL, P.C. and THURN-
HAM, D.I. 2005. Flesh colour dominates consumer preference for
chicken. Appetite 44, 181–186.
KÖSTER, E.P. 2003. The psychology of food choice; some often encountered
fallacies. Food Qual. Prefer. 14, 359–373.
MARCELINO, A.S., ADAM, A.S., COURONNE, T., KÖSTER, E.P. and
SIEFFERMAN, J.M. 2001. Internal and external determinants of eating
initiation in humans. Appetite 36, 9–14.
MARSHALL, D. and BELL, R. 2003. Meal construction: Exploring the rela-
tionship between eating occasion and location. Food Qual. Prefer. 14,
53–64.
MOSKOWITZ, H.R. and SILCHER, M. 2006. The applications of conjoint
analysis and their possible uses in Sensometrics. Food Qual. Prefer. 17,
145–165.
MUNKEVIK, P., HALL, G. and DUCKETT, T. 2007. A computer vision
system for appearance-based descriptive sensory evaluation of meals. J.
Food Eng. 78, 246–256.
202 H.H. REISFELT ET AL.
NGAPO, T.M., MARTIN, J.F. and DRANSFIELD, E. 2007. International
preferences for pork appearance. I. Consumer choices. Food Qual. Prefer.
18, 26–36.
NORDEN. 2004. Nordic Nutrition Recommendations. Nord 2004:13. Nordic
Council of Ministers, Copenhagen.
SØRENSEN, L.B., MØLLER, P., FLINT, A., MARTENS, M. and RABEN,
A. 2003. Effect of sensory perception of foods on appetite and food
intake: A review of studies on humans. Int. J. Obes. 27, 1152–1166.
TRAIN, K. 2003. Discrete Choice Methods with Simulation, Cambridge
University Press, Cambridge.
TUORILA, H., MEISELMAN, H.L., CARDELLO, A.V. and LESHER, L.L.
1998. Effect of expectations and the definition of product category on the
acceptance of unfamiliar foods. Food Qual. Prefer. 6, 421–430.
TURNER, M. and COLLISON, R. 1988. Consumer acceptance of meals and
meal components. Food Qual. Prefer. 1, 21–24.
WHEATELY, J. 1973. Putting colour into marketing. Marketing 67, 23–29.
203VISUALLY PRESENTED MEALS
