Food Quality and Preference 18 (2007) 133138

www.elsevier.com/locate/foodqual

How important intrinsic and extrinsic product attributes

aect purchase decision
Ulrich Enneking
b

a,*

, Claudia Neumann

b,c

, Sven Henneberg

a
Fachhochschule Osnabruck, University of Applied Sciences, Oldenburger Landstrasse 24, D-49009 Osnabruck, Germany
Technical University Munich, Agricultural and Food Marketing, Alte Akademie 14, D-85350 Freising-Weihenstephan, Germany
c
SAM ASAP GmbH, Drachenseestrasse 1, D-81373 Munich, Germany

Available online 3 November 2005

Abstract
We simultaneously evaluated intrinsic and extrinsic product attributes by means of a choice-based conjoint experiment. A sample of
621 consumers tasted three soft drinks and chose the product most preferred. Test products were systematically varied across choice sets
with respect to sweetening system, calorie reduction label, price and brand. Choice data and data on consumption patterns, attitudes and
socio-demographics were analysed by a conditional logit model. This approach enabled us to model product choice as a function of one
intrinsic and three extrinsic attributes and of consumer characteristics. The latter permits market segmentation of preference data. The
main results showed that consumers preferences of sweetening systems are heavily dependent on brand information. Simulations of market shares based on the total sample reveal the preference of sugar over sweetening systems. However, in some specic consumer segments, sugar is not signicantly superior to sweetening systemsindicating how useful market segmentation can be in sensory analysis.
2005 Elsevier Ltd. All rights reserved.
Keywords: Purchase decision; Market segmentation; Choice-based conjoint; Soft drinks

1. Introduction and literature review

Traditional sensory analysis, which focuses on intrinsic
product attributes alone, is not sucient to meet the
requirements of todays fast moving markets. An optimised
product formulation is necessary for a successful innovation, however, consumers are also inuenced by extrinsic
product information such as brand, price or labelling.
Especially psychologists have long been interested in the
eects of the combination of sensory stimuli, both intrinsic
and extrinsic, in product evaluation. Studies from Pronko
and Bowles (1948a, 1948b, 1949), Brown (1958) and
Makens (1965) show the role of brand recognition and
preference in both the ability to identify products and the
rating of food products on quality and freshness (Sheen

Corresponding author. Tel.: +49 541 9695126; fax: +49 541 969220.
E-mail address: [email protected] (U. Enneking).

0950-3293/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.foodqual.2005.09.008

& Drayton, 1988). Based on these ndings, Allison and

Uhl (1964) observed, in a study on the inuence of beer
brand identication on taste perception, that brand loyal
users assigned signicantly higher ratings to their preferred
brands in an identied test compared to a blind test.
Cheng, Clarke, and Heymann (1989) conducted a conjoint
analysis with restructured beef steaks and were able to
show how three attributesproduct preference, packaging
and brand identicationinuenced the consumers
hedonic responses. Also, Moskowitz (1994) shows in a
product-concept test with fat-reduced cheese, how other
information, besides the physical characteristics of the
products themselves, can modify consumer acceptance.
Consumers stated purchase intent could be increased only
by incorporating a health message. Those ndings indicate
that sensory analysis has to be combined with modern market research methods in order to develop integrated
approaches that are able to evaluate extrinsic as well as
intrinsic product attributes and possible interactions

134

U. Enneking et al. / Food Quality and Preference 18 (2007) 133138

between them. Understanding the relative importance of

product attributes inuencing food choice at the point of
sale is important to the success of new product development. For many years conjoint analysis has been used to
estimate the importance of various product attributes for
consumers purchasing decisions (Green & Srinivasan,
1978, 1990). Although the taste of a food product is widely
assumed to inuence buying decisions, the product attribute taste has rarely been incorporated in conjoint analysis. Vickers (1993) describes a conjoint analysis of
strawberry yogurt that examined the importance of brand,
taste, health claim, and price to buying intent with a multilevel approach.
It was evident at the 6th Pangborn Sensory Science Symposium in Harrogate, that some authors see the future in
combining sensory experiences with consumer perception,
expectation and attitudes and, therefore, postulate the
expansion of the model beyond acceptance.1 One possibility to meet the demand to combine intrinsic factors, like
taste, with extrinsic factors is to use the conjoint method,
which was also seen as a future tool for the sensory
science.2
The objective of this research was to incorporate food
tasting into the test design and to evaluate extrinsic and
intrinsic product attributes simultaneously. We conducted
a choice-based conjoint analysis because this method
allows us to examine possible interactions between taste
and brand and also allows a segment specic analysis.
2. Materials and methods
2.1. Methodological background
Market research on food choice is predominantly based
on conventional surveys where respondents are directly
interviewed with respect to brand awareness, price sensitivity, etc. Interviewees attention is thus strongly directed to
specic product features, which increases the danger of
overestimating the eect of certain characteristics compared to real market behaviour. In line with this argument,
Teichert (2000) has experimentally shown that unimportant product attributes are overestimated by respondents
when separately evaluated. This can probably be attributed
to a tendency of yes saying observed when respondents
want to please interviewers. Overrepresentation of single
product features can be reduced when interviewees compare several product attributes and brands within a
choice-based conjoint experiment. In this case, respondents
are not aware of which attributes researchers are interested
in. Furthermore, if respondents are not willing to pay a
high product price, they may vote for a cheaper alternative
1

Sensory and consumer scienceWhat have we achieved? Where are we

going? Halliday MacFie; Keynote presentation at the 6th Pangborn
Sensory Science Symposium.
2
Sensometrics overviewchallenges and opportunities. Pascal Schlich;
Keynote presentation at the 6th Pangborn Sensory Science Symposium.

without disappointing the interviewer by choosing no

product at all.
Because of these (and other) advantages, choicebased conjoint experiments are increasingly being used in
research. In a recent study, Burton, Rigby, Young, and
James (2001) calculated willingness-to-pay for several food
attributes, namely reductions in chemical use, food risk,
and in transportation distance but also a label indicating
a voluntary ban on genetically modied ingredients. However, this and other choice-based conjoint experiments have
not modelled taste as an experimental variable. The study
presented in this paper is the rst to evaluate intrinsic
(taste) and extrinsic (brand, labelling, etc.) attributes in a
single choice experiment.
Choice-based conjoint experiments are usually analysed
within a random utility framework which assumes that an
individual, n, maximises his or her utility when choosing
between alternatives, J. The researcher is not completely
informed about all elements considered important by
respondents, so utility observed from a researchers perspective can be broken down into two components, V
and e
U in V in ein ;

where
Uin is the overall utility of choice i for individual n,
ein is the random utility component which comprises
unobserved individual observations, measurement
errors and unobserved attributes,
Vin is the systematic or measurable utility, which is a
function of Xin and bi and an unknown parameter vector
to be estimated. Xin denes: (i) a matrix of attributes
that pertain to choice options, (ii) a matrix of characteristics that pertain to individuals, (iii) a matrix of interactions of attributes with individual characteristics or (iv)
a vector of interactions of individual characteristics with
choice option intercepts (Louviere, 2001). In most practical applications, Vin takes a linear-in-parameters additive form.
If A is dened as the universal choice set of discrete
alternatives, and J the number of elements in A, then individual n will choose alternative i over some other option j
if, and only if,
Ui > Uj

where all j 6 i 2 A

and the probability that individual n chooses i from set A is

given by
P in P fejn  ein g < fV in  V jn g for all j 6 i.

In order to specify the choice probabilities in (3), assumptions must be made with regard to the distribution of the
random components. From the outset of choice-based conjoint experiments (McFadden, 1974), the independent and
identically distributed type I extreme-value distribution
proved convenient for computational ease. This distribu-

U. Enneking et al. / Food Quality and Preference 18 (2007) 133138

135

Table 1
Systematically varied product attributes
National brand

Regional brand

Private label

Price

0,76/0,86/0,96 EURO

0,46/0,56/0,66 EURO

0,26/0,36/0,46 EURO

Labelling

Big label
Small label
Without label

Big label
Small label
Without label

Big label
Small label
Without label

Taste

100% sugar
67% sugar
Conventional sweetener
Diet sweet Up

100% sugar
67% sugar
Conventional sweetener
Diet sweet Up

100% sugar
67% sugar
Conventional sweetener
Diet sweet Up

TM

TM

tion leads to the popular multinomial (conditional) logit

model (MNL)3:
eV in
P in PJ
;
V jn
j1 e

j 1; . . . ; J ; j 6 i.

2.2. Survey and data

From April to June 2004, 621 consumers of carbonated
soft drinks participated in the choice-based conjoint experiment, which took place in the region of Munich, Germany.
The data were collected in shopping centres and in a sensory laboratory using personal computer assisted interviews.4 The interview was divided into an experimental
part and a questionnaire part, beginning with the experimental choice task. In an identied test, participants were
asked to taste three carbonated soft drinks (CSD) with
orange avor and to choose the most preferred from a specic choice set. Each choice set is a dierent combination
of several product attributes. In order to design those
choice sets, the mix and match approach, described in
Chrzan and Orme (2000), was employed, which permits
estimation of the main eects as well as rst order interaction eects (i.e. brand * taste). The choice modelling literature recommends at least 500 choices to allow for valid
maximum likelihood estimations (e.g. Long, 1997). Therefore, choice tasks were repeated twice by respondents in the
shopping centres and ve times by consumers in the sensory lab. Having deleted a few outliers from the sample,
a total of 1529 choices was ready for analysis.
The products oered to respondents varied systematically with respect to taste, brand, labelling and price (Table
1).
The rst product is an international premium brand
which is distributed nation-wide and marketed with the
aid of TV advertising. The second brand represents a med-

3
For an exhaustive derivation of the MNL model see Louviere,
Hensher, and Swait (2000, 45).
4
Potential dierences between the sub-samples interviewed in shopping
centres (N = 520) and those consumers who were invited to the sensory lab
(N = 101) are not analysed in this paper. They are going to be discussed in
a separate publication.

TM

ium-priced product distributed on a regional level. Our

third brand represents a low-priced private label and acts
as a reference category in the statistical model. Prices were
varied across all three brands using three brand-specic
price levels. In addition to brand and price, a calorie reduction label was varied across the product by using either a
big label, a small label or the product was shown completely without a label. To include the intrinsic product
attribute taste into the choice set, we used four dierent
sweetening systems to sweeten the CSD: 100% sugar, 67%
sugar, a conventional sweetener (consisting of a sweetener
mixture) and Diet Sweet Up , a new improved sweetener
mixture. These taste variants did not correspond to the original tastes of the three brands displayed to respondents.
All extrinsic product information on brands, prices and
calorie reduction labels was presented via photographs
(Fig. 1).
In addition to the experimental design, a decision had to
be made on whether or not a no-choice option should be
included into the choice design. On one hand, such a base
alternative leads to better predictions of market penetration and is considered more realistic since, in many
purchase situations, consumers can defer purchase or purchase elsewhere. On the other hand, the base alternative
gives no information about the impact of label and price
on choice.5 Furthermore, Dhar (1997) shows that when
alternatives are similar in preference, respondents will
choose the no-choice option more often as compared to
when alternatives are very distinct from each other. In this
paper, respondents were not oered an opt-out alternative
because the tendency to choose none of the products seems
to be rather weak: First, only users of CSD with orange
avor were interviewed, implying a general acceptance
of this product category. Moreover, the selected products covered most types of preferences in this product
category, leading to very low incentives to opt for a nochoice category. Second, most consumers are not highly
involved when buying carbonated soft drinks, thus
deferring purchase or buying elsewhere will probably not
TM

For a more detailed discussion of the no-choice alternative in choicebased conjoint/discrete choice analysis, see Haaijer, Kamakura, and
Wedel (2001).

136

U. Enneking et al. / Food Quality and Preference 18 (2007) 133138

increases the probability to choose the premium brand.

For that reason we have distinguished between consumers
who prefer shopping for CSD in supermarkets, beverage
stores and at lling stations. All variables we have modelled in the present study are displayed in Table 2.
2.3. Statistical analysis

Fig. 1. Choice set.

be a real option for them. Thirdly, it is not the purpose of

this study to predict market penetration, which is often
quoted as an important argument to include a no-choice
option.
The experimental part of the interview was followed by
questions on the respondents personal characteristics and
purchasing motives/patterns. Age and size of the household were personal factors weve collected to explain their
inuence on decision making. Since we expected product
choice to be inuenced by the occasion on which consumers drink CSD, respondents who drink CSD mainly on the
way, or during a workout were identied by the variables
way and sports. To explain product choice at the point of
purchase, factors such as brand loyalty, price orientation
andin the specic case of CSDthe sugar content may
also be important. Therefore, we asked consumers which
factors are most important for them when buying CSD.
Furthermore, consumers are not all acquainted with
exactly the same products because they use dierent distribution channels. It can thus be hypothesised that, ceteris
paribus, a preference for shopping in the supermarket

Based on the described variables, a conditional logit

model was estimated by the software package STATA
7.0. Having compared several linear utility functions of
the conditional logit model, the specications displayed
in Tables 3 and 4 respectively were chosen based on LR
tests. By means of a Hausmann test, the assumption of
independence of irrelevant alternatives could be armed.
Tables 3 and 4 show the brand-specic estimators (logit
coecient) of the experimental and of the exogenous variables. In the remaining sections, the results will be discussed focusing on brand-specic eects.

Table 3
Estimation results (t-ratios in parentheses)

Label_small
Label_big
Taste_67%
Taste_dietsweetup
Taste_conventional
Constant
Household
Supermarket
Filling station
Price
Brand
Sugar
Age
Way
Sports

National brand

Regional brand

Private label

0.33
0.24
0.74
0.39
0.82
0.29
0.03
0.00
0.04
0.69
0.73
0.54
0.14
0.34
0.54

0.04
0.31
0.70
0.55
0.90
0.28
0.08
0.38
0.62
0.29
0.28
0.22
0.04
0.28
0.13

0.23
0.08
0.55
0.26
0.58

(2.04)
(1.47)
(3.94)
(2.33)
(4.4)
(0.78)
(0.77)
(0.01)
(0.15)
(3.7)
(3.4)
(2.7)
(2.1)
(2.43)
(2.41)

(0.3)
(2.17)
(4.12)
(3.56)
(5.56)
(0.83)
(2.01)
(3.03)
(1.93)
(1.76)
(1.39)
(1.31)
(0.72)
(2.32)
(0.66)

(1.6)
(0.6)
(3.45)
(1.61)
(3.57)

Number of obs = 4587; LRv2 = 294.29 (p = 0.0000); log likelihood =

1532.6318.

Table 2
Summary of exogenous variables in the estimated models
Question wording

Variable name

Scale

Mean

Where do you prefer shopping for CSD?

Supermarket
Beverage store
Filling station

Supermarket
Beverage store
Filling station

0/1
0/1
0/1

0.39
0.40
0.41

When buying CSD I pay attention to . . .

. . . price
. . . brand
. . . sugar content

Price
Brand
Sugar

0/1
0/1
0/1

0.51
0.11
0.44

When do you drink CSD?

. . . on the way
. . . during work out
How old are you?
How many people live in your household?

Way
Sports
Age
Household

0/1
0/1
04
19

0.47
0.13
2.80
2.61

U. Enneking et al. / Food Quality and Preference 18 (2007) 133138

Table 4
Estimation results for sub-segment of calorie sensitive consumers (t-ratios
in parentheses)

Label_small
Label_big
Taste_67%
Taste_dietsweetup
Taste_conventional
Constant
Household
Beverage store
Brand
Age

National brand

Regional brand

Private label

0.32
0.19
0.86
0.46
0.85
0.41
0.08
0.29
1.22
0.25

0.21
0.69
0.66
0.35
0.82
0.57
0.17
0.78
0.15
0.06

0.15
0.39
0.00
0.04
0.59

(1.1)
(0.7)
(2.57)
(1.58)
(2.69)
(0.75)
(0.96)
(0.94)
(3.47)
(2.16)

(0.88)
(2.72)
(2.31)
(1.3)
(2.9)
(1.14)
(2.23)
(2.95)
(0.4)
(0.62)

(0.62)
(1.61)
(0.01)
(0.18)
(2.06)

Number of obs = 1581; LRv2 = 106.11 (p = 0.0000); log likelihood =

525.9114.

3. Results and discussion

Analysing all choice decisions made by consumers in
the two separate samples, we found signicantly high negative eects for all alternative sweetener systems as compared to the reference category 100% sugar (Table 3).
This is true for all three dierent brand segments. This
indicates that sugar remains the queen of sweeteners.
Additionally, the fact that the estimated t-values are much
higher compared to other model variables reveals a considerable impact of taste on product choice. As for the calorie
reduction label, the impact is not that clear. Interestingly, only the choice probability of the national brand
increases signicantly (t-value = 2.04) when attached
with a small label, whereas the regional brand is the only
product in the choice set which signicantly prots from
a big calorie reduction label. We can not explain this phenomenon by the information collected within this study.
The results may be inuenced by design aspects. Nevertheless, the estimated coecients generally suggest that there
is a (brand specic) positive labelling eect on product
choice.
As was shown in Table 1, a brand-specic price was also
included as an experimental variable. However, no signicant inuence of price on the product choice decision could
be estimated, so this variable was dropped from the estimated models. The simultaneous evaluation of intrinsic
and extrinsic product attributes may have led to this nding. Since consumers usually do not evaluate both taste
and marketing attributes at the point of purchase, interviewees may have overvalued taste compared to price. This
interpretation is conrmed by an unpublished study where
respondents had to make a choice between the same products. However, the three products were only varied with
respect to extrinsic attributes. This test partly showed a signicant inuence of price on brand choice.
Apart from experimental and product specic variables
which can be systematically varied by researchers, some
individual specic variables were used to model preferences. This approach permits a deeper understanding of
brand choice and gives marketers important information
for market segmentation.

137

Respondents who judged themselves as price sensitive

have a signicantly lower probability to choose the premium national brand compared to the private label. The
opposite is true for those who pay attention to brands.
Interestingly, respondents who look out for sugar content
prefer the private label over the national brand. Probably,
the premium brand, which rather attracts younger consumers, is conceived as a sweet CSD. The model also reveals
that the place of purchase has a signicant eect on product choice. Respondents who primarily choose the supermarket as place of purchase prefer the regional brand
whereas consumers who buy CSDs at lling station behave
in the opposite way. Apart from age, sociodemographics
do not have a big impact on product choice.
With the chosen conditional logit model, it is also possible to focus on subsets of the total group of consumers.
Through our questionnaire, we analysed consumer attitude
toward calorie reduced products and were able to identify
34% of the total group as calorie sensitive consumers. This
group can be seen as the target group for a new calorie
reduced orange avored CSD. For this subgroup of consumers we found that 67% sugar and Diet Sweet Up
can compete with sugar for the private label segment
(Table 4). Compared to the conventional sweetener, all
other considered sweeteners are better in the eyes of real
consumers.
By looking to the results for the calorie reduction labels
(Table 4), we nd a clear indication that labelling improves
the probability to choose each of the three considered
brands. The bigger the label the more gain in market shares
can be expected.
We conclude that the considered subgroup of calorie
sensitive users of orange avored CSD prefer even new
improved sweetening systems over pure sugar products if
they are clearly indicated as calorie reduced products.
Hence, the development of new sweetening systems allows
calorie reduced CSDs to invade even the premium segment.
Moreover, calorie reduction is a feature that is relevant in
all market segments (premium, regional brand, private
label). As calorie sensitive consumers react favorably to
new sweetening systems if they are clearly labelled, the
results indicate that a new sweetening system should not
substitute a 100% sugar product. A diversication addressing one third of the potential buyers of CSD should be
preferred.
TM

4. Conclusion
Choice-based conjoint methods applied to sensory market research allows a focus on markets and consumer segmentation. With the chosen model, we could analyse a
subgroup that is interesting in terms of a diversication
of the orange avored carbonated softdrinks market. This
calorie sensitive consumer group has no signicant preference for sugar compared to new sweetening systems. New
sweeting systems combined with labelling the product as
calorie reduced will enhance the probability of choice of

138

U. Enneking et al. / Food Quality and Preference 18 (2007) 133138

the respective product. This result is valid also for the

national brand segments, which indicates a possibility for
new sweetening systems to invade even the premium market segments. The choice-based conjoint method is a new
approach to analyse interesting segments of consumers.
However, from a methodological point of view, it has to
be mentioned that a simultaneous evaluation of extrinsic
and intrinsic product attributes seems to be quite demanding on respondents and, therefore, unsuitable for pricing
research. However, respondents showed considerable
brand-specic taste evaluations. By combining intrinsic
(taste) and extrinsic factors (brand and labels), we can
describe not only main eects, but also interactions
between taste and marketing mix elements, allowing us to
determine key drivers for product preference. Future
research might be conducted as a two step evaluation of
consumer preferences where pricing research is a separated
task. This is expected be less demanding on respondents
and will probably enhance data quality.
References
Allison, R. I., & Uhl, K. (1964). Inuence of beer brand identication on
taste perception. Journal of Marketing Research, 1, 3639.
Brown, R. L. (1958). Wrapper inuence on the perception of freshness in
bread. Journal of Applied Psychology, 42, 257260.
Burton, M., Rigby, D., Young, T., & James, S. (2001). Consumer attitudes
to genetically modied organisms in the food in UK. European Review
of Agricultural Economics, 28(4), 479498.
Cheng, H. W., Clarke, A. D., & Heymann, H. (1989). Inuence of selected
marketing factors on consumer response to restructured beef steaks: a
conjoint analysis. Journal of Sensory Studies, 4, 165178.
Chrzan, K., & Orme, B. (2000). An overview and comparison of design
strategies for choice-based conjoint analysis. Sawtooth Software
Research Paper Series.
Dhar, R. (1997). Consumer preference for a no-choice option. Journal of
Consumer Research, 24, 215231.

Green, P. E., & Srinivasan, V. (1978). Conjoint analysis in consumer

research: issues and outlook. Journal of Consumer Research, 5,
103123.
Green, P. E., & Srinivasan, V. (1990). Conjoint analysis in marketing: new
developments with implications for research and practise. Journal of
Marketing, 54, 319.
Haaijer, R., Kamakura, W., & Wedel, M. (2001). The no-choice
alternative in conjoint choice experiments. International Journal of
Market Research, 43(1), 93106.
Long, J. S. (1997). Regression models for categorical and limited dependent
variables. Thousand Oaks: Sage Publications.
Louviere, J. J., Hensher, D., & Swait, J. (2000). Stated choice methods:
analysis and application. Cambridge.
Louviere, J. J. (2001). Choice Experiments: an overview of concepts
and issues. In J. Bennett & R. Blamey (Eds.), The choice modelling
approach to environmental valuation (pp. 1336). Cheltenham: Edward
Elgar.
Makens, J. C. (1965). Eect of brand preference upon consumers
perceived taste of turkey meat. Journal of Applied Psychology, 49,
261263.
McFadden, D. (1974). Conditional logit analysis of qualitative choice
behaviour. In R. Zerembka (Ed.), Frontiers in Econometrics
(pp. 105142). New York: Academic Press.
Moskowitz, H. R. (1994). Food concepts and products. Just-in-time
development. Trumbull: Food and Nutrition Press.
Pronko, N. H., & Bowles, J. W. (1948a). Identication of cola beverages I:
a rst study. Journal of Applied Psychology, 32, 304312.
Pronko, N. H., & Bowles, J. W. (1948b). Identication of cola beverages
II: a further study. Journal of Applied Psychology, 32, 559564.
Pronko, N. H., & Bowles, J. W. (1949). Identication of cola
beverages III: a nal study. Journal of Applied Psychology, 33,
605608.
Sheen, M. R., & Drayton, J. L. (1988). Inuence of brand label on sensory
perception. In D. M. H. Thomson (Ed.), Food acceptability
(pp. 8999). London: Elsevier Applied Sciences.
Teichert, T. (2000). Auswirkungen von Verfahrensalternativen bei der
Erhebung von wahlbasierten Conjoint-Daten, Befunde einer empirischen Anwendung. Marketing Zeitschrift fur Forschung und Praxis, 2,
145159.
Vickers, Z. M. (1993). Incorporating tasting into a conjoint analysis of
taste, health claim, price and brand for purchasing strawberry yogurt.
Journal of Sensory Studies, 8, 341352.