European experts have developed the

European experts have developed the ECSI methodology, based on a set of requirements
(ECSI Technical Committee, 1998), e.g. comparability, reliability, robustness and structural
modelling approach.
The basic ECSI model (see Fig. 1) is a structural equation model with latent variables.
The model links customer satisfaction to its determinants and, in turn, to its consequence,
namely customer loyalty. The determinants of customer satisfaction are perceived company
image, customer expectations, perceived quality and perceived value (`value for money’ ).
Perceived quality is conceptually divided into two elements: `hard ware’ , which consists of
the quality of the product /service attributes, and `human ware’ , which represents the
associated customer interactive elements in service, i.e. the personal behaviour and atmosphere
of the service environment. Main causal relationships are indicated; actually there can
exist many more points of dependence between the variables.
These seven variables are seen as latent, i.e. non-observable. Each of the latent variables
is operationalized by two to six measurement variables (indicators), observed by survey
questions to customers.
The latent variable customer satisfaction is measured through three indicators, empirically
observed by the three questions, that have dominated theory and practice within customer
satisfaction measurement (Ryan et al., 1995, pp. 11± 12).
First, overall satisfaction is measured as the answer to a question such as ``Considering
all your experience of company X, how satis® ed are you, in general?’’ on a scale from
`completely dissatis® ed’ to `completely satis® ed’ . This approach is perhaps the most common
in customer satisfaction measurement practice (Ryan et al., 1995, p. 12).
Second, satisfaction is measured by a question of this type: ``To what degree did
Downloaded by [Monash University Library] at 07:59 11 September 2013
RELATIONSHIP BETWEEN CUSTOMER SATISFACTION AND LOYALTY S511
company X ful® l your expectations?’’ on a scale from `much less than expected’ to `much
more than expected’ .
Third, there is used an ideal-point scale asking. ``Imagine a company which is perfect in
all aspects. How close to this ideal do you consider the company X to be?’’ on a scale from
`very far away’ to `very close’ .
Each question captures diVerent facets of an underlying satisfaction perception. In
combination, the answers to these three questions give a reasonably accurate measure of the
latent variable for an individual. A customer satisfaction index is calculated by a weighted
average of scores from the three questions, and this approach will be more useful that a single
measure from any of the three questions. The use of multiple questions for each latent
variable increases the precision of the estimate, compared to use of a single question, and
there is empirical support for using such an approach within satisfaction measurements
(Fornell & Cha, 1994; Fornell et al., 1996; Ryan et al., 1995).
The latent variable customer loyalty is operationalized by four indicators: the customer’ s
intention to repurchase; intention of cross-buying (buy another product from the same
company); intention to switch to a competitor (price tolerance); and intention to recommend
the brand/company to other consumers.
Data for the model estimation come from data collected through telephone interviews
from a national, representative sample of customers who are recent buyers and/or users of
speci® c products and services. For most companies interviews are conducted with about 250
of their customers. The sample size is determined by a precision requirement: a 95%
con® dence interval for customer satisfaction (on a 0± 100-point scale) should not be wider
that + 2 points. Another accuracy requirement is that R2 of customer satisfaction should be
at least 0.65, i.e. the model must be able to explain at least 65% of what drives customer
satisfaction (ECSI Technical Committee, 1998, pp. 20± 21).
The entire model is estimated using partial least squares (PLS) (Fornell & Cha, 1994).
The latent variables are operationalized as weighted indices of their measurement variables,
and the PLS estimation method weights the survey measures such that the resulting
model has maximum explanatory power, i.e. maximum prediction accuracy of the ultimate
dependent variable customer loyalty. The loyalty measure is a survey-based proxy for
economic results, and therefore the estimated measure for customer satisfaction should be a
forward-looking indicator of economic performance. Many empirical studies have demonstrated
that customer satisfaction, based on the ACSI/SCSB approach, has an impact on
economic results (EkloÈ f et al., 1999; Fornell, 1999), and therefore the ECSI measure of
customer satisfaction also has a linkage to economic performance.
For each company in the sample, the PLS method estimates indices for all the seven
latent variables, i.e. customer satisfaction, loyalty and their drivers. Furthermore, the PLS
method estimates the relationships within the entire model, i.e. the impacts between the
latent variables (inner coeYcients in the structural model) and the weights for each
measurement variable associated with a latent variable (outer coeYcients in the measurement
model).
All seven latent variables are transformed from the original 1± 10-point survey measures
to 0± 100-point scales, where zero means lowest possible (for example completely dissatis® ed)
and 100 means highest possible (for example completely satis® ed).
Measuring and estimation are made at the company level. Latent variable indices and
impact scores at industry level are calculated by aggregating company-level estimates,
weighted by market share.
A major advantage of the ECSI is the use of generic questions, which are suYciently
¯ exible to be used across a wide variety of products, services and public sector services, such
as education, healthcare, etc.
Downloaded by [Monash University Library] at 07:59 11 September 2013
S512 L. GRé NHOLDT ET AL.
Data from the Danish part of the ECSI pilot study
Twelve European countries participated in the ECSI pilot study, and across Europe nearly
55 000 interviews were carried out during the spring of 1999. Telecommunication was
covered by all participating countries, and retail banks and supermarkets were covered by
nearly all the countries.
In Denmark, 30 companies within eight speci® c industries were included in the study,
namely four telecommunication industries (® xed net, mobile phones, the Internet and cable
television), retail banks, supermarkets, the soft drink industry and fast food restaurants.
During the spring of 1999, data were collected for the Danish ECSI pilot study. Nearly 9000
customers were interviewed regarding their perceptions of quality and satisfaction with
products and services in these industries.
Comparative results of these co-ordinated studies in 11 European countries have been
reported (ECSI, 1999), and the authors have presented Danish results and cross-industry
® ndings (Martensen et al., 2000).
Our experiences with the application of the ECSI model have been very good. The
model ® ts well and seems to be suYciently ¯ exible for diVerent industries.
Comparability has been highlighted by the ECSI Technical Committee (1998, pp. 7,
11). The ECSI methodology, i.e. the structural model approach, the questionnaire and the
estimation technique will allow comparisons between companies and organizations not only
at national levels but also at Pan-European and global levels. It is possible to make meaningful
comparisons with companies located outside Europe where similar customer satisfaction
indices are already produced, including the US and East Asia. As national customer
satisfaction indices reliably and consistently measure customer satisfaction and quality
perceptions for many companies within a variety of industries, the ECSI has the potential to
be an excellent platform for benchmarking.
Empirical ® ndings
Based on the results of Danish ECSI pilot study, we have analysed the relationship between
company-level customer satisfaction indices and loyalty indices. Figure 2 illustrates the
relationship, where the eight aggregated indices for the eight industries are also marked. The
® gure shows that customer satisfaction has a positive eVect on customer loyalty. This ® nding
comes as no surprise, since it is well supported in the literature. However, in our study the
relationship is estimated with a high explanatory power across the 30 companies plus `all
other’ companies within six industries.
Regression analysis shows that the relationship between customer satisfaction and loyalty
is strongly signi® cant: the estimated regression coeYcient is 1.14, i.e. for every point change
in satisfaction, loyalty changes by 1.14 point, on average (n 5 36, R2 5 0.691, t 5 8,71,
p-value < 0.0005).
Looking at the individual companies as they are positioned in Fig. 2, a very interesting
pattern appears. Very large positive residuals are found for companies with a low price
strategy. The supermarkets Netto and Fakta and Saltum/Rù rkjñ r Brewery (and the gas
stations Uno-X, also measured according to the ECSI methodology) are all companies that
have made price their main competitive weapon. These companies have a much larger loyalty
than expected from their customer satisfaction. On the other hand, we ® nd that companies
that have used a lot of energy on branding indeed have a high customer satisfaction but they
do not have a correspondingly high loyalty. This holds good for, for example, McDonald’s,
PepsiCo and Coca-Cola Company from the international scene and Faxe Brewery and Tele
Denmark from the Danish scene.