Journal of Business Research 53 (2001) 75 ± 84
Conceptual and operational aspects of brand loyalty
An empirical investigation
Yorick Odin*, Nathalie Odin, Pierre Valette-Florence
Centre d'Etudes et de Recherches AppliqueÂes aÁ la Gestion (CERAG), UPRES-A/CNRS No. 5046, Ecole SupeÂrieure des Affaires de Grenoble (ESA),
Universite Pierre MendeÁs-France, BP 47-38 040 Grenoble Cedex 9, France
Accepted 1 June 1999
Abstract
The objective of this article is to stress the lack of valid and reliable measures concerning loyalty, and then to conceive, test and validate a
relevant measurement procedure of this concept, by following a rigorous methodology based on the Churchill paradigm [Churchill GA. A
paradigm for developing better measures of marketing constructs. J Mark Res 1979;16(1):64 ± 73]. In the first part, the authors will approach
problems linked to the conceptualization and to the operational aspects of loyalty in the literature. This synthesis will lead to a proposal of
differentiation of the repeat purchasing behavior by the concept of brand sensitivity. The second part will deal with methodological aspects
and will present the main results of this research. Finally, the article concludes on the contributions and limits of this study, as well as on
future research perspectives. D 2001 Elsevier Science Inc. All rights reserved.
Keywords: Churchill paradigm; Brand sensitivity; Brand loyalty
``The success of a brand on the long term is not based on
the number of consumers that buy it once, but on the
number of consumers who become regular buyers of the
brand.'' This sentence (Jacoby and Chestnut, 1978, p. 1)
clearly illustrates the importance for companies to put the
emphasis on their customers' loyalty. Early on, academic
research has also been conscious of the central role played
by the loyalty concept in the consumer-buying process, to
such a point that more than 300 research papers (Jacoby and
Chestnut, 1978), published or not, have dealt with the study
of this concept. The literature brings out two approaches that
apprehend loyalty in two different ways: the stochastic
approach, which is purely behavioral, and the attitudinal
approach that considers loyalty as an attitude. Besides the
numerous divergences about the real nature of this concept,
the literature is characterized by the multitude of available
operational definitions, and by the contradictory results
obtained using the techniques developed to measure loyalty.
These contradictory results can notably be explained by the
* Corresponding author. Tel.: +33-4-72-69-21-62; fax: +33-4-78-5949-12.
E-mail address: [email protected] (Y. Odin).
almost systematic lack of a rigorous study of the reliability
and the validity of the proposed measurement instruments.
The objective of this article is to stress this lack of valid
and reliable measures concerning loyalty, and then to conceive, test and validate a relevant measurement procedure of
this concept, by following a rigorous methodology based on
the Churchill paradigm (1979). In the first part, we will
approach problems linked to the conceptualization and to
the operational aspects of loyalty in the literature. This
synthesis will lead to a proposal of differentiation of the
repeat purchasing behavior by the concept of brand sensitivity. The second part will deal with methodological aspects
and will present the main results of this research. Finally, the
article concludes on the contributions and limits of this
study, as well as on future research perspectives.
1. Brand loyalty: operational and conceptual aspects
The literature on the loyalty concept is characterized by
two divergent streams of research: the stochastic approach
and the deterministic approach. As a consequence of this
divergence, a review of the literature highlights the lack of
clarity about the conceptual nature of loyalty, and also the
0148-2963/01/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved.
PII: S 0 1 4 8 - 2 9 6 3 ( 9 9 ) 0 0 0 7 6 - 4
76
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
large variety of the results obtained using the numerous
existing measurement tools.
1.1. Loyalty: behavior or attitude?
1.1.1. The stochastic approach
For the defenders of the stochastic approach, loyalty is a
behavior: the individual that buys the same brand systematically is said to be loyal to this brand. The problem lies in
the fact that the stochastic approach considers loyalty
behavior as being inherently inexplicable, or too complex
to be comprehended: the number of explanatory variables as
well as their frequency of appearance makes any explanation of this behavior impossible (Bass, 1974; McAlister and
Pessemier, 1982). There is a major disadvantage of such a
point of view: it implies that it is difficult for a company to
influence repeat purchasing behavior, as this company has
no knowledge of the actual cause of loyalty.
1.1.2. The determinist approach
The main postulate of the determinist approach is that
there exists a limited number of explanatory factors generating loyalty: the researcher can isolate these factors and
thus can manipulate them. In the framework of this approach, brand loyalty is treated more as an attitude. The
researcher investigates the psychological commitment of the
consumer in the purchase, without necessarily taking the
effective purchase behavior into account (e.g., Jacoby, 1969,
1971; Jacoby and Olson, 1970; Jarvis and Wilcox, 1976).
Reconciling the two approaches, Jacoby (1971) proposes
to integrate the two notions of behavior and attitude within a
same conceptual definition. He is the first author to propose
a six points definition that integrates behavioral and attitudinal loyalty, and that will influence the definitions proposed later by Engel et al. (1978). According to Jacoby and
Kyner (1973), brand loyalty is the ``(1) biased (i.e., nonrandom) (2) behavioral response (i.e., purchase) (3) expressed over time (4) by some decision-making units (5)
with respect to one or more alternative brands out of a set of
such brands and is (6) a function of psychological (decisionmaking, evaluative) processes.''
1.2. Operational definition
Measures of brand loyalty are so numerous and varied in
the literature that it would be too long to give an ordered and
exhaustive list of them.1 However, three observations can be
raised with regards to existing measures. In general, it is
worth noting the high heterogeneity in the results obtained
using the different instruments. Moreover, the development
of brand loyalty measures fails by the lack of investigation
1
The reader interested in a detailed presentation of existing
measurement tools might consult Jacoby and Chestnut (1978) or Filser
(1994).
of their reliability and their validity, despite the attempts of
Olson and Jacoby (1971) to verify the reliability of some
instruments using the test ± retest technique. Finally, the
operational definition is often developed without any preliminary reflection on the conceptual nature of brand loyalty: this could also explain the diversity of existing
measures, which are often ``irrational and arbitrary'' (Jacoby
and Chestnut, 1978, p. 41).
Following a behavioral or an attitudinal approach, each
of them possesses its advantages and drawbacks, that will be
highlighted in the next paragraphs.
1.3. Advantages and limits of existing measures
The major interest of behavioral measures resides in the
fact that they measure effective behaviors. However, they do
not enable the researcher to tell whether repeat buying has
been done out of habit, for situational reasons, or for more
complex psychological reasons. Furthermore, the processing
of loyalty is made in a dichotomous way Ð loyalty vs.
disloyalty Ð which is singularly short of nuance, and
requires a very arbitrary judgement as for the allocation of
a consumer to one or the other of the two categories.
As an example, Fig. 1 presents the study of loyalty for
four consumers, according to two measurement methods.
In the example reported in Fig. 1, the use of the
percentage of purchase corresponds to the way Cunningham (1956a,b, 1967) measures brand loyalty, that is to say
by the purchase proportion of a same brand on a same
sequence of purchase (incidentally undetermined). The
problem of this measure is that it fixes an arbitrary loyalty
threshold: above 50% of purchase proportion devoted to
the same brand, the author estimates that there is brand
loyalty. Following a slightly different approach, Tucker
(1964) and McConnell (1968) propose the ``3 in the
sequence'' criterion: the consumer is said to be brand-loyal
when the sequence of purchase includes consecutively three
identical brands. As shown in Fig. 1, the measurement
methods used in this example do not converge to a same
result: as an example, consumer 1 is loyal in the framework
of the percentage of purchase, but unloyal using the ``3 in
the sequence'' procedure.
Determinist measures allow to circumvent a certain
amount of criticism addressed to behavioral measures. In
the first place, most of them are constructed around interval
type scales, which facilitate data collection. Moreover,
attitudinal scales are no longer based on a loyalty/disloyalty
opposition, but on a degree of loyalty: thus, the goal is not
to know whether an individual is absolutely loyal or not,
but to know the intensity of his loyalty to a branded
product; the nuance of this type of scales is therefore far
more important.
Despite these advantages, this type of scales suffers
from some major drawbacks. The first criticism to be
addressed to this group of measures is that it only relies
on consumer declarations, and not on the observed beha-
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
77
Fig. 1. Heterogeneity of results using two different measurements of brand loyalty.
vior. The second criticism rather concerns the operational
aspect of the loyalty concept: in most cases, researchers use
either antecedents, or consequences of loyalty to measure
the former, and not loyalty in itself. It is notably the case of
the scales used very recently by Bloemer and Kasper
(1995) or Massad and Reardon (1996) that mix notions
of variety-seeking behavior, behavior of consumers when
an article is out of stock, and loyalty. Similarly, when
Cunningham (1967) evaluates loyalty, he measures it
through one of its antecedents, that is to say through the
behavior of consumers when an article they are used to
purchasing is out of stock: a consumer that tends to
repurchase the same brand is said to be loyal if he waits
or goes to another shop to get the brand he is used to
buying. This example also puts the emphasis on another
problem: the confusion frequently made between shop
loyalty and brand loyalty (for example, Burford et al.,
1971; Bellenger et al., 1976). Yet, Carman (1970) shows
that these are two different notions, and concludes his
research by giving evidence that shop loyalty is an important antecedent of brand loyalty.
This synthesis of the literature emphasizes the great
diversity of measurement approaches and procedures of
the brand loyalty concept. The objective of the following
part is not to develop a new definition of loyalty, but rather
to find a satisfying conceptual positioning for this research,
in the maze of existing definitions.
1.4. Positioning of the research
Before developing a measurement instrument, the construct that has to be measured must be clearly defined. The
only definition that exists, and that makes the bond between
stochastic and cognitivist approaches is Jacoby's (1971),
which postulates that brand loyalty is a behavioral answer,
function of an evaluative psychological process.
1.4.1. Brand loyalty: behavior or attitude?
The definition of Jacoby (1971) clearly identifies
loyalty as a behavior. However, the author tends to
consider loyalty on a two-dimensional basis, by adding
to the loyalty concept an attitudinal component (evaluative psychological process). If the behavioral aspect of
loyalty seems clear Ð loyalty is the repeated purchasing
of the same brand Ð the attitudinal component remains
relatively vague. In his definition, Jacoby does not give
any indication about the nature of the psychological
factors that would influence repeat purchasing. For some,
this component can be resumed as the attitude towards
brands (e.g., Day, 1969); for others, it rather concerns the
preference towards brands (e.g., Guest, 1955; Jacoby and
Kyner, 1973). Whatever the identified attitudinal component is, it concerns concepts as such, that are antecedent
to the repeat purchasing behavior. Actually, these antecedents represent a motive of repurchase. Thus, the ``out
of stock situation'' measure mentioned previously apprehends loyalty on the basis of the consumer attitude in the
case where the researched brand is out of stock (e.g.,
Cunningham, 1967; Jacoby and Kyner, 1973): the intention of buying again the same brand or the actual rebuying of the same brand are not measured directly, but
through an antecedent.
In the framework of this research, loyalty is measured
directly, i.e., as a repeat purchasing behavior. However, the
78
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
repurchase of the same brand can be considered in two
different ways; it concerns either:
a reflective loyalty, as a result of brand commitment or
a favorable attitude towards the brand,
or an inertia of purchase, that is to say a repeat
purchasing of the same brand without a real motive
for the choice made.
Thus, when a repeat purchasing measure is available,
measuring loyalty consists in isolating real, reflective loyalty, from inertia. In this purpose, the concept of brand
sensitivity developed in France by Kapferer and Laurent
(1983) will be used to distinguish loyalty and inertia. Brand
sensitivity is a psychological individual variable, which is
defined as the degree to which the brand name plays a key
role in the choice process of an alternative in a given
product category (Kapferer and Laurent, 1983, p. 17).
1.4.2. Brand loyalty, inertia, and brand sensitivity
Studies on the difference between the loyalty and inertia
concepts are, to our knowledge, almost non-existent in the
literature. This can notably explain the lack of conceptualization of inertia. Nevertheless, the inertia concept seems
to be quite different from the loyalty concept. Indeed, for
Filser (1994), the repurchase of the same brand under
conditions of strong perceived differences between brands
and strong involvement characterizes brand loyalty. This
approach allows to distinguish loyalty from inertia, the
latter appearing in a situation of weak involvement and
weak perceived differences between brands. Kapferer and
Laurent (1983) have shown that the belief in differences
between brands is a major determinant of brand sensitivity.
Similarly, they give evidence that the level of involvement
influences the level of brand sensitivity positively. Thus,
the present research plans to use the level of sensitivity to
differentiate loyalty from inertia (see Fig. 2). In other
terms, a repeat purchasing behavior under conditions of
strong sensitivity will be considered as brand loyalty: a
consumer who tends to repurchase the same brand and who
attaches great importance to brands in his choice is said to
be brand loyal. To the opposite, a repeat purchasing
behavior under conditions of weak brand sensitivity is
considered as purchase inertia. In this case, the consumer
does not give any importance to the brand of the product
he is to buy, as he is not able to make any difference
between existing brands, and also as he is not involved in
the product category.
After having examined what had to be measured, i.e.,
loyalty as an observable repeat purchasing behavior, under
condition of strong brand sensitivity, the following step of
this research is to investigate the way this concept has to
be measured.
2. Methodology
Preceding paragraphs have allowed to determined what
the scale has to measure, namely the repeat purchasing
behavior of the same brand, as declared by consumers.
The hypothesis formulated here is that the nature of
repeat purchasing behavior can change relatively to the
level of brand sensitivity: a strong brand sensitivity
accompanied by a high declared repeat purchasing behavior corresponds to what will be defined as loyalty.
Conversely, a weak brand sensitivity accompanied by a
high declared repeat purchasing behavior will in turns
correspond to purchase inertia.
2.1. Development of the measurement scale
With regards to the problems previously stressed concerning existing behavioral measures, the use of interval
scales appears to be far more interesting: easier in their
practice for data collection, they also allow a more shaded
measure, using degrees of loyalty, and they do not evaluate
the latter on the basis of the loyalty/disloyalty opposition.
Therefore, the scale proposed in this research does not
measure a real behavior, but rather the consumer's perception of his own repeat purchasing behavior.
2.1.1. Development of measurement items
Churchill (1979) suggests a rigorous methodological
framework to create a reliable and valid measurement
instrument. The first stage is the specification of the constructs that are studied, which was the objective of the
preceding part. The next step is then to generate a set of
items, stemming from the literature, from group interviews,
or from the researcher's intuition. In the present case, an
important review of the literature, completed with inter-
Fig. 2. Repeat purchasing behavior under conditions of brand sensitivity.
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
views with experts and consumers, have lead to propose a
set of 18 items to the test: these items represent most of the
measurement methods observed in the framework of the
determinist approach, and are presented in Appendix A.
Some items relate directly to repeat purchasing behavior (``I
often switch from one brand of ___ to another.''), while
other are more like antecedents of it (e.g., ``I like switching
from one brand of ___ to another.''). Some items that were
originally of nominal type have also been used. It is notably
the case of the items used by Kapferer and Laurent (1983,
p. 64), which measures consumer's loyalty on a three-point
nominal scale (loyal/multiloyal/disloyal); the multiloyal
modality, which concerns loyalty to a limited number of
brands, has been excluded of the 18 items inventory.
Indeed, the study of multiloyalty would probably require
the development of a specific measure, as some authors
such as Dufer and Moulins (1989) have done it. A study of
Olson and Jacoby (1971) confirms this point of view, after a
principal components analysis on 12 brand loyalty/multiloyalty items: items of loyalty and multiloyalty loaded on
two distinct dimensions.
2.1.2. Data collection and measurement scale purification
In accordance with the procedure suggested by Churchill,
a first data collection has been carried out on a sample of
109 undergraduate students. The category of jeans has been
retained, a pretest showing that these products benefited
from a good variability in terms of repeat purchasing
behavior and brand sensitivity. A questionnaire was administered, which included the 18 loyalty items, mixed with
several other items such as those of the involvement scale of
Kapferer and Laurent (1983). Instead of purifying the whole
set of the 18 items by successive factor analyses and through
the examination of Cronbach a as Churchill recommends it,
a series of confirmatory analyses has been realized, more
robust and frequently recommended in the literature (e.g.,
Gerbing and Anderson, 1988). A four-item solution was
retained at the end of this phase, and is presented in
Appendix B. To evaluate the reliability and the validity of
the scale, a second data collection was organized on a
sample of 334 persons.
2.2. Scale evaluation
Commonly, a scale is reputed to be of a good quality
if it is reliable and valid. Before studying the validity of
the scale, its dimensionality and its consistency need to
be investigated.
2.2.1. Reliability
The unidimensionality of the scale was checked through
a series of factor analyses using maximum likelihood
estimation: the results bring out strong loadings and communalities, and around 90% of the four-item information is
extracted by the factor. The most frequently used method to
assess scale consistency is the computation of the Cron-
79
Table 1
Results of the confirmatory analyses
lF1
lF2
lF3
lF4
r
rvc
Maximum likelihood estimation
Bootstrap ADF estimation
Parameters*
Standard error
Parameters*
Standard error
0.949
0.920
0.948
0.942
0.968
0.884
0.007
0.010
0.007
0.008
0.948
0.912
0.937
0.937
0.964
0.872
0.003
0.004
0.002
0.001
* d parameters are not reported: (d = 1 l2).
bach2 a, even if this coefficient remains criticized, especially because of its dependence on the number of items of
the scale (e.g., Roehrich, 1993, p. 352). The a is very high
(0.96), and tends to confirm the consistency of the scale.
The average inter-item correlation is to a maximum of 0.88,
and the strongest correlation between two items is around
0.9, which means around 80% of shared variance between
the two items. It is worth noting that the exclusion of any of
the four items would not bring down the value of the
Cronbach a.
An interesting and recommended alternative is to use
structural equations modeling to carry out confirmatory
analyses on the items of the scale, and to compute the
JoÈreskog r (1971).3 In a first time, a confirmatory analysis
using maximum likelihood estimation was used to compute the JoÈreskog r. The r, equal to 0.96, points out the
very good validity of the scale (see Table 1). The Student
t-test on each parameter shows that all of them are
significant at p < 0.01.4
The results obtained using maximum likelihood estimation are based on the hypothesis that the measurement
variables follow a multinormal distribution. As a conse2
aˆ
0
k
B
@1
k 1
1
s2i
P 2i P C
A
si ‡ 2 sij
P
i
i;j
with k as the number of items, si2 as the variance of i item and sij as the
covariance between items i and j.
3
rˆ
2
k
P
iˆ1
k
P
iˆ1
2
li
‡
li
k
P
iˆ1
1
l2i
with k as the number of items and l as the standardized factor loadings.
4
Steiger (1995, p. 3591) also suggests to assess whether there is any
redundancy in the items used to measure the construct, i.e., whether these
items are too correlated to be distinguished. The estimation procedure did
not detect any redundancy, which points out that each variable holds a
sufficient part of its specific information.
80
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
quence, the estimation method employed is likely to poorly
estimate the parameters in the case of non-multinormal
distributions.5 To circumvent this problem, a second series
of confirmatory analyses were carried out following an
asymptotically distribution free estimation procedure, which
leads to releasing us from the problem of the violation of
multinormality conditions (Browne, 1984).
Valette-Florence (1993) also recommends to check that
the results obtained using different estimation methods
converge to the same conclusion. The multinormality
condition is also bypassed thanks to the use of a bootstrap
procedure carried out on the sample (100 replications with
N = 325); this procedure also leads to confirming the
reliability and the stability of the obtained results.
Table 1 presents the results computed with the confirmatory analyses. Whatever the estimation procedure used, the
parameters just slightly varied. Loadings are high (significant at 0.01) and ensure a good reliability of the scale, with r
of more than 0.9.
A major advantage of confirmatory analysis is that it
gives a large set of formal indices to assess the quality of the
model to be tested. Among the multitude of adequation
indices proposed, those that are recommended in the literature (Browne and Cudeck, 1992) were chosen. This is
notably the case of indices that are given with a confidence
interval, such as the RMSEA of Steiger and Lind (1980), or
the Gamma of Tanaka and Huba (1989). For information,
GFI, AGFI and Bentler± Bonnett NFI were also computed;
all indices are reported in Table 4. The Steiger and Lind
RMSEA must be higher than 0.08 (Browne and Cudeck,
1992), and the Gamma equal to or above 0.9.6
The results obtained on indices are reported in Table 2,
and are globally encouraging, even if the estimated RMSEA
has just the accepted level for a model of good quality. The
Gamma, GFI/AGFI and NFI are, on the other hand, rather
high, which brings out a good adequation of data to the
tested structure and which therefore limits the importance of
the RMSEA.
In conclusion of these analyses, the reliability of the scale
appears to be good. However, the evaluation of the quality
of the proposed instrument could not be limited to the
simple study of its reliability. Indeed, the literature also
strongly recommends the evaluation the scale validity.
2.2.2. Validity
Peter (1981) suggests to structure the evaluation of
validity in four points: content validity, trait validity, predictive and nomological validity. Thus, content validity will
be studied first.
5
A prior assessment of variables using the approach proposed by
Steiger (1995, p. 3643) brought out that the multinormality hypothesis was
probably violated.
6
See Didellon and Valette-Florence (1996) for a detailed review of
adequation indices in structural equations models.
Table 2
Evaluation of the quality of the structural model
Maximum likelihood estimation
Bootstrap ADF estimation
Lower
bounda
Upper
bound
Lower
bound
Estimateb
Upper
bound
0.17
0.993
0.04
0.097
0.17
RMSEA 0.038
Gamma 0.854
GFI
AGFI
NFI
a
b
Estimate
0.10
0.950
0.987
0.935
0.995
Confidence interval estimation of the indices (90%).
Statistics does not give some of the indices in bootstrap results.
2.2.2.1. Content validity. In the framework of content
validity, the researcher must assess that each item of the
scale deals effectively with the content of the construct that
has to be measured. The development process of the scale
was based on the purification of a list of items, which
stemmed from a large review of the literature and also from
qualitative interviews with experts and consumers about
repeat purchasing behavior. These precautions should ensure a good content validity.
2.2.2.2. Trait validity. Campbell and Fiske (1959) propose
to assess trait validity through the evaluation of both
convergent validity and discriminant validity. In our case,
the study of discriminant validity is useless, as the measure
tested here is unidimensional. Didellon and Valette-Florence
(1996) suggest to evaluate convergent validity through
confirmatory analysis, by using Student t-tests to evaluate
the significance of the loadings (li) that link measures (Xi)
to their latent exogeneous variable (xi): li significantly
different from 0 mean that items converge to a same
construct. In addition, Fornell and Larcker (1981, p. 46)
recommend to check that the construct shares at least 50%
of variance with its measures, to ensure a good convergent
validity (rvc).7 Results reported in Table 1 point out the good
convergent validity of the scale. All l parameters are
statistically significant, whatever the estimation procedure
used. Moreover, in both cases, the shared variance between
the latent construct and its measures (rvc) largely exceeds
the 50% threshold.
2.2.2.3. Nomological validity. Only nomological validity
will be studied here, as the latter does not practically
distinguish from predictive validity. Indeed, Roehrich
(1993, p. 355) underlines that ``practically or theoretically,
it is not easy to distinguish predictive validity from nomo-
7
k
rvc ˆ i ˆ 1
k
l2i
with k as the number of items and l as the standardized factor loadings.
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
81
Table 3
Results of the structural model using maximum likelihood estimation
The d and e parameters are not reported (d/e = 1 l2); n.s.: not significant parameter at p < 0.05.
Maximum likelihood estimation
Strong brand sensitivity group
(loyalty) N = 209
Bootstrap maximum likelihood estimation
Strong brand sensitivity group
(loyalty) N = 209
Weak brand sensitivity group
(inertia) N = 102
Scale of perceived risk: dimensions of risk importance and risk probabilitya
l1
0.537
0.559
l2
0.751
0.556
l3
0.548
0.621
l4
0.738
0.380
0.690
0.856
l5
l6
0.705
0.732
l7
0.636
0.533
0.536
0.742
0.549
0.731
0.696
0.703
0.635
0.551
0.593
0.601
0.388
0.857
0.764
0.515
Scale of declared repeat purchasing behavior
l8
0.938
l9
0.913
l10
0.937
0.935
l11
0.911
0.877
0.933
0.915
0.932
0.913
0.936
0.935
0.918
0.872
0.931
0.911
rvc
f11
g11
g12
0.826
0.260
n.s.
n.s.
0.863
n.s.
0.258
0.266
0.825
0.260
n.s.
n.s.
0.866
n.s.
0.260a
0.265
a
Weak brand sensitivity group
(inertia) N = 102
Items of the risk importance dimension have been reversed to make interpretation easier.
logical validity,'' highlighting that both use correlation and
both intend to validate a scale by leaning on existent theory.
According to Roehrich (1993, p. 371), nomological validity
``refers to the ability of the measure to behave as, on a
theoretical basis, the trait it measures should behave.''
Roselius (1971), followed by several authors, gave
evidence that brand loyalty is one of the most important
risk reduction strategies to eliminate risk in the purchase of a
product. Thus, a consumer who perceives an important risk
associated with a product category will be more prone to
remaining brand loyal. To check the nomological validity of
the scale proposed here, the two dimensions of perceived
risk used in the involvement scale of Laurent and Kapferer
(1986) have been retained: the probability to make a mistake
and the risk importance. A series of confirmatory analyses
have been carried out to assess the psychometric qualities of
this scale.
From a conceptual point of view, the research proposal is
as follows: under conditions of strong brand sensitivity, a
consumer who perceives an important risk in the product
category should show more loyalty than those people who
perceive low risk, in accordance with Roselius' results. On
the other hand, under conditions of weak brand sensitivity,
perceived risk would not influence purchase inertia. Therefore, research hypotheses are formulated as follows.
H1. The two dimensions of perceived risk influence
significantly and positively brand loyalty.
H2. The two dimensions of perceived risk do not influence
significantly purchase inertia.
The methodology relied on a multigroup analysis, which
allows to investigate one structural model on two or more
groups of respondents, and to test whether the direction and
the strength of the relationships between latent variables
differ from one group to the other.8 The grouping variable
was computed after a K-means cluster analysis on one item
adapted from Kapferer and Laurent (1983, p. 78) brand
sensitivity scale. A two-group solution was extracted from
the data:9 one group constituted of brand-sensitive consumers and a second group of brand-insensitive consumers.
As previously mentioned, two estimation procedures have
been used to confirm the reliability of results. However, the
sub-sample formed by the low sensitivity group appeared to
be too small: thus, the bootstrap procedure used to bypass
the violation of the multinormality criterion was only carried
out using maximum likelihood estimation, and not using
ADF estimation. Results are reported in Table 3: parameters
are stable whatever the estimation method used. Adequation
indices are good (Gamma and RMSEA), and highlight a
good adequation of data to the specified model (Table 4).
The results obtained and reported in Fig. 3 show that
the two dimensions of risk influence brand loyalty significantly and positively. In the opposite, no significant
8
To get more valid results with multigroup analysis, a constrained
model was also tested. In that framework, all measurement variables (li)
have been constrained to remain equal on the two groups; only the g
parameters have been left free for estimation: the results obtained were not
different between the constrained and the unconstrained model.
9
A one-way analysis of variance lead to confirm that the two groups
were effectively different on the brand sensitivity item.
82
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
Table 4
Adequation indices of the structural model
Maximum likelihood estimation
RMSEA
Gamma
a
Bootstrap max. likelihood estimation
Lower bounda
Estimate
Upper bound
Lower bound
Estimate
Upper bound
0.030
0.961
0.053
0.979
0.074
0.994
0.028
0.043
0.056
Confidence interval estimation of the indices (90%).
relationships were found between these two dimensions
and purchase inertia.
The validation of H1 is consistent with the theory that
postulates that a consumer who perceives a strong risk
within a product category will tend to be brand loyal: this
conclusion leads to confirm the nomological validity of the
proposed scale. The validation of H2 gives evidence that
perceived risk in a product category does not influence
consumer purchase inertia. This result is consistent with the
idea that the only apparent motive of inertia repurchase is
the pure and simple repetition of behavior (Kapferer and
Laurent, 1983, p. 125).
3. Conclusion
The aim of this research was to clarify the concept of
loyalty and to propose a reliable and valid measurement
procedure. The psychometric properties of the proposed
scale have been rigorously tested, and in terms of reliability
and validity, results appear to be encouraging. This objective was important, as it remains difficult to find rigorous
tests of loyalty measurement scales in the literature.
Besides this operational objective, this research has
managed to clarify the loyalty concept, and notably to point
out the importance of the distinction between loyalty and
purchase inertia. It has also led to highlight the interest of
focusing rather on degrees of loyalty, which is far more
shaded and more reliable, and not on a dichotomy based on
the loyalty/disloyalty opposition.
However, measurement through degrees of loyalty lead
to a scale based on consumer declarations, which undoubtedly constitutes a limit to this research. Nevertheless, the
recourse to panel data would have brought out some other
limits, notably the limit dealing with undifferentiation
between the decision maker and the actual buyer: in many
cases, the consumer who decides the purchase of a brand
and the effective buyer of this brand are two different
persons, which probably biased loyalty results.
The sample used, slightly under-dimensioned in the
framework of multigroup analysis, is a convenience sample,
constituted of undergraduate students. However, the product
category used in this research (i.e., jeans) seems relevant
enough considering the nature of the sample, which reduces
the scope of this methodological limit.
A certain number of antecedents of loyalty have been
identified and investigated in the literature, notably by
Kapferer and Laurent (1983). In their research, the authors
study the influence of some determinants on loyalty by using
a Probit regression model. The utilization of this type of
Fig. 3. Results of the structural model under conditions brand sensitivity.
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
model was determined by the nature of the dependent
variable, which is a measure based on the loyalty/disloyalty
dichotomy. The study of these antecedents using the scale
proposed in this research could have given interesting results,
as the measurement approach developed seems more
nuanced. Furthermore, the approach used in this research,
which enables to distinguish inertia from brand loyalty, might
allow to refine the conclusions on the relationships between
the concepts studied by Kapferer and Laurent (1983) and the
repeat purchasing behavior declared by the consumer.
On the other hand, relationships that are likely to exist
between concepts such as variety-seeking tendency or exploration tendency and the repeat purchasing behavior could
be investigated. Indeed, Sirieix (1994) shows that existing
researches are far from being commonly shared on the
nature of the relationships between loyalty and variety
seeking. The incorporation of major concepts of varietyseeking, loyalty and inertia within a same study, the latter
measured according to the procedure proposed in this
article, would allow to provide answers to the relationships
which bring these different concepts together.
Appendix A.
1. I like switching from one brand of ___ to another.
2. I often switch from one brand of ___ to another.
3. If the brand I usually buy is not available in a shop, I
go to another shop.
4. On several purchase occasions, it is likely that I will
buy each time the same brand of ___.
5. I am loyal to only one brand of ___.
6. During my next purchase, I will buy the same brand
of ___ as the last time.
7. Even if the price of the brand of ___ I am used to
buying strongly increases, I'll still buy it.
8. Generally, I am loyal to a small number of brands of
___.
9. Even when I hear negative information about the
brand of ___ I usually buy, I still stick to that brand.
10. I've been buying the same brand of ___ for a long
time.
11. I like trying several brands of ___.
12. I always buy the same brand of ___.
13. During my last purchases, I've always bought the
same brand of ___.
14. If I like a brand of ___, I rarely switch from it.
15. If the shop I regularly visit has not got the brand of
___ I usually buy, I go to another shop.
16. Usually, I buy the same brand of ___.
17. Each respondent was also interviewed on the purchase intention probability for them to buy the same
brand of ___ as the last purchase:
18. Finally, the Cognitive Loyalty Index developed by
Jarvis and Wilcox (1976) on the basis of Jacoby's
(1969) researches was also used in the questionnaire.
83
This index was computed as follow: Cognitive Loyalty Index = (RR/AR)(1 [(BA (RR + AR))/BA]),
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
with RR as the number of brands in the rejected
region, AR as the number of brands in the accepted
region, and BA as the number of brands known by
the consumer.
Appendix B.
The items of the repurchasing scale used in this research
are presented in the following table. Each item is evaluated
on a six-point Likert scale ranging from (1) ``Totally
disagree'' to (6) ``Totally agree.''
Item RPB 1
Item RPB 2
Item RPB 3
Item RPB 4
I am loyal to only one brand of jeans
During my next purchase, I will buy the
same brand of jeans as the last time
I always buy the same brand of jeans
Usually, I buy the same brand of jeans
The item used to measure brand sensitivity was adapted
from Kapferer and Laurent (1983, p. 78): ``When I buy a
___, the brand is the first thing I'm looking at.''
References
Bass FM. The theory of stochastic preference and brand switching. J Mark
Res 1974;11:1 ± 20.
Bellenger DN, Steinberg E, Stanton WW. The congruence of store image
and self-image: as it relates to store loyalty. J Retail 1976;52(1):17 ± 32.
Bloemer JMM, Kasper HDP. The complex relationship between consumer
satisfaction and brand loyalty. J Econ Psychol 1995;16:311 ± 29.
Browne MW. Asymptotically distribution free methods for the analysis of
covariance structures. Br J Math Stat Psychol 1984;37:62 ± 83.
Browne M, Cudeck R. Alternative ways of assessing model fit. In:
Bollen KA, Long JS, editors. Testing Structure Equation Models.
Beverly Hills, CA: Sage, 1992. pp. 136 ± 63.
Burford RL, Ennis BM, Paul GW. An index for the measurement of consumer loyalty. Decis Sci 1971;2:17 ± 24.
Campbell DT, Fiske DW. Convergent and discriminant validation by the
multitrait ± multimethod matrix. Psychol Bull (March) 1959;56:81 ± 105.
Carman JM. Correlates of brand loyalty: some positive results. J Mark Res
1970;7:67 ± 76.
Churchill GA. A paradigm for developing better measures of marketing
constructs. J Mark Res 1979;16(1):64 ± 73.
Cunningham RM. Brand loyalty Ð what, where, how much? Harv Bus Rev
(January/February) 1956a;34:116 ± 28.
Cunningham RM. Brand loyalty Ð what, where, how much? J Mark
1956b;21:206.
Cunningham RM. Perceived risk and brand loyalty. In: Cox D, editor. Risk
Taking and Information Handling in Consumer Behavior. Boston: Harvard Univ. Press, 1967. pp. 507 ± 23.
Day GS. A two-dimensional concept of brand loyalty. J Advertising Res
1969;9:29 ± 35.
Didellon L, Valette-Florence P. L'Utilisation des Indices d'Ajustement dans
84
Y. Odin et al. / Journal of Business Research 53 (2001) 75±84
les Modeles d'Equations Structurelles: PreÂsentation et Recommandations d'Usage. Actes des 12eÁmes JourneÂes Nationales des IAE, Toulouse,
France, 16 ± 17 Avril 1996;111 ± 25.
Dufer J, Moulins JL. La relation entre la satisfaction du consommateur et sa
fideÂlite aÁ la marque. Rech Appl Mark 1989;4(2):22 ± 35.
Engel JF, Blackwell RD, Kollat DT. Consumer Behavior. Hindsale, IL:
Holt, Reinehart & Winston, 1978.
Filser M. Le Comportement du Consommateur. Dalloz: Collection PreÂcis
de Gestion, 1994.
Fornell C, Larcker D. Evaluating structural equation models with unobservable variables and measurement error. J Mark Res (February)
1981;18:39 ± 50.
Gerbing DW, Anderson JC. An updated paradigm for scale development
incorporating unidimensionality and its assessment. J Mark Res (May)
1988;25:186 ± 92.
Guest LP. Brand loyaltyÐtwelve years later. J Appl Psychol 1955;
39:405 ± 8.
Jacoby J. Toward a multi-brand model of brand loyalty. Purdue Papers in
Consumer Psychology, Paper No. 105, 1969.
Jacoby J. Brand loyalty: a conceptual definition. Proc, Am Mark Assoc
1971;6:655 ± 6.
Jacoby J, Chestnut RW. In: Webster FE, editor. Brand Loyalty: Measurement and Management. New York: Wiley, 1978.
Jacoby J, Kyner DB. Brand loyalty vs. repeat purchasing behavior. J Mark
Res (February) 1973;10:1 ± 9.
Jacoby J, Olson JC. An attitudinal model of brand loyalty: conceptual
underpinnings and instrumentation research. University of Illinois Conference on Attitude Research and Consumer Behavior, Urbana, IL,
December 1970.
Jarvis LP, Wilcox JB. Repeat-purchasing-behavior and attitudinal brand
loyalty: additional evidence. In: KL Bernhart, editor. Educator's Proceedings. American Marketing Association, 1976. pp. 151 ± 2.
JoÈreskog KG. Statistical analysis of sets of congeneric tests. Psychometrika
(June) 1971;36:109 ± 33.
Kapferer JN, Laurent G. La Sensibilite aux Marques: Un Nouveau Concept
pour GeÂrer les Marques. Fondation Jour de France pour la Recherche en
PubliciteÂ, Paris, 1983.
Laurent G, Kapferer JN. Les Profils d'Implication. Rech Appl Mark
1986;1:41 ± 58.
Massad V, Reardon J. Channel surfing, brand loyalty and risk propensity: a
segmentation approach to modelling consumer switching behaviors in
the USA. Int J Advertising 1996;15:250 ± 61.
McAlister L, Pessemier EA. Variety seeking behavior: an interdisciplinary
review. J Consum Res 1982;9:311 ± 22.
McConnell JD. The development of brand loyalty: an empirical study. J
Mark Res 1968;5:13 ± 9.
Olson JC, Jacoby J. A construct validation study of brand loyalty. Proc, Am
Psychol Assoc 1971;6:657 ± 8.
Peter JP. Construct validity: a review of basic issues and marketing practices. J Mark Res (May) 1981;18:133 ± 45.
Roehrich G. Les Consommateurs-Innovateurs: Un Essai d'Identification.
TheÁse pour le Doctorat d'Etat en Sciences de Gestion. Universite P.
MendeÁs-France, Grenoble 2, 1993.
Roselius T. Consumer rankings of risk reduction method. J Mark
1971;35:56 ± 61.
Sirieix L. La Recherche de VarieÂte dans le Choix du Point de Vente: Conception et Analyse. TheÁse Pour le Doctorat eÁs Sciences de Gestion,
Universite de Montpellier 2, 1994.
Steiger JH. Structural equation modeling. Statistica V5.0 (vol. 3). Tulsa,
OK: Statsoft, 1995, pp. 3539 ± 688.
Steiger JH, Lind JC. Statistically based tests for the number of common
factors. Annual Spring Meeting of the Psychometric Society, Iowa City,
IA, May 30, 1980.
Tanaka JS, Huba GJ. A general coefficient of determination for covariance
structure models under arbitrary GLS estimation. Br J Math Stat Psychol 1989;42:233 ± 9.
Tucker WT. The development of brand loyalty. J Mark Res 1964;1:32 ± 5.
Valette-Florence P. Dix ans d'Equations Structurelles: Un Etat de l'Art.
20eÁme SeÂminaire International de Recherche en Marketing, La Londe
Les Maures, IAE, Aix-en-Provence, 1993.