Journal of Electronic Commerce Research, VOL 7, NO.2, 2006
WHAT KEEPS THE E-BANKING CUSTOMER LOYAL?
A MULTIGROUP ANALYSIS OF THE MODERATING ROLE OF CONSUMER
CHARACTERISTICS ON E-LOYALTY IN THE FINANCIAL SERVICE INDUSTRY



Arne Floh
Horst Treiblmaier

Department of Marketing
Department of Information Systems

Vienna University of Economics
Vienna University of Economics
and
Business
Administration
and Business Administration
Arne.Floh@wu-wien.ac.at
Horst.Treiblmaier@wu-wien.ac.at


ABSTRACT

At first sight the Internet is the ideal medium for carrying out banking activities due to its cost savings potential
and speed of information transmission. From a technological and cost-driven standpoint it may seem quite logical
for banks to shift as many banking activities online as possible. At the same time, the question of how to foster
customer loyalty arises when the relationship between the bank and the user becomes a virtual one.
This paper investigates the importance of antecedents of online loyalty such as trust, quality of the Web site,
quality of the service and overall satisfaction. Rather than investigating which factors drive customers to use online
banking instead of offline banking, this paper addresses the problem of how to keep customers online and loyal to a
specific supplier.
A survey among more than 2,000 customers of an Austrian online bank was conducted and a structural equation
modeling approach was used to gain important insights into how customer retention in the online banking business
can be ensured. Satisfaction and trust were identified as important antecedents of loyalty. Additionally, the
moderating role of consumer characteristics (gender, age, involvement, perceived risk and technophobia) was
supported by the data.

Keywords: Loyalty, E-Banking, Structural Equation Modeling, Multigroup Analysis

1. Introduction
In order to reap the benefits of having loyal customers and gaining a competitive advantage online, companies
need to develop a thorough understanding of the antecedents of loyalty on the World Wide Web (e-loyalty), such as
business factors [Bhattacherjee 2001] or personal characteristics [Mägi 2003]. In order to investigate the importance
of e-loyalty, the identification of variables influencing repeat purchasing behavior and word-of-mouth
recommendation is a crucial area of research [Srinivasan et al. 2002]. This holds especially true for those industries
which already depend heavily on their reputations and long-lasting relationships in the offline world, as is the case
with the financial sector. The widespread adoption of online banking services calls for research investigating those
factors which are responsible for keeping customers loyal.
A model explaining the antecedents of loyalty in the online banking industry has to incorporate factors which
take into account the characteristics of the industry as well as those of the medium. Therefore, we consider
antecedents such as trust (being important online and offline) and the perceived quality of the Web site (being
important only online). Besides being topics of scholarly research in the information systems domain, these issues
have been long discussed in marketing. More than two decades have passed since the concept of relationship
marketing was first mentioned in the marketing literature [Berry 1983]. Drivers such as intense competition,
demanding customers and enablers such as the Internet are the reason why relationship marketing has increasingly
attracted the attention of researchers and practitioners alike [Sheth and Parvatiyar 2002]. In relationship marketing
research the concept of customer loyalty plays a central role [Christopher et al. 2004]. The preeminent importance of
retaining customers is supported by several studies [Chen et al. 2002], confirming the relevance of customers'
loyalty to a firm's profitability.
New forms of online communication offer a host of new and promising opportunities for customer retention on
the World Wide Web, while at the same time intensifying competition [Vatanasombut et al. 2004]. In particular, this
applies to company-controlled communication, giving companies the ability to customize information with regard to
the individual needs of a particular customer and to optimize the customer's feedback opportunities [Kierzkowski et
al. 1996]. At the same time, companies also face completely new challenges arising from customer-controlled

Page 97

---

Floh & Treiblmaier: What Keeps the E-Banking Customer Loyal?
Internet communication, such as the growing importance of brand strength, economies of scale and size [Gallaugher
1999].
In view of these changed circumstances in the buyer-seller interaction, researchers and practitioners have to
rethink previous concepts of loyalty [Luo and Seyedian 2003/04]. Empirical studies comparing customer satisfaction
and loyalty in online and offline environments show substantial differences in terms of customer attitudes and
behavior [Shankar et al. 2003]. At the same time, the integration of Internet technology into the customer loyalty
concept is rarely discussed in the relationship marketing literature [Wirtz and Lihotzky 2003].
In the following section we take a look at scholarly literature pertaining to e-banking. In particular, we
concentrate on those papers which have already developed models explaining user behavior in the e-banking
context. Next, we develop a model investigating the antecedents of e-loyalty, including cognitive and affective
constructs (trust, satisfaction) and quality aspects (Web site quality, service quality). Additional variables, such as
gender or involvement, are hypothesized to have a moderating influence on e-loyalty [Seethamraju 2004].

2. Conceptual Framework and Hypotheses
While there is a rich body of literature on online financial services and their adoption, little is known about how
to keep customers loyal to an online bank. This section provides an overview of previous e-banking research and
introduces the concept of loyalty. A framework relating loyalty to important antecedents and a number of
moderating variables is introduced.
2.1 The E-Banking Sector
Online banking, which can be defined as the provision of information or services by a bank to its customers
over the Internet [Daniel 1999], has been one of the major developments in the financial service sector in recent
years. According to a survey conducted by Pew Internet and America Life, online banking has been the fastest
growing Internet activity in the U.S. over the last five years. As of November 2004, a total of 53 million Americans
(44% of all U.S. Internet users) use some form of online banking service [McGann 2005]. In Germany the number
of online accounts has increased almost tenfold between 1999 and 2004, with 40% of all accounts now being online
[Association of German Banks 2004].
A short review of the literature on electronic banking briefly illustrates the major issues that researchers and
practitioners have dealt with in recent years. Security turned out to be a major obstacle for many customers who
were otherwise willing to switch to the online world [Martin 1998]. Besides assuring customers that their privacy is
protected, Hamlet [2000] suggests that banks should not over-animate or clutter their Web sites with too much
advertising. In addition, care should be taken not to over-personalize the online-experience in order to avoid the
impression that personal financial information is freely available. Bhattacherjee [2001] uses online banking
customers to test his expectation-confirmation model of IS continuance. His results suggest that users' continuance
intention is determined by their satisfaction and perceived usefulness of the application. Table 1 shows a number of
research papers that empirically validate models in the context of e-banking. Tan and Teo [2000] found that
attitudinal factors such as Internet experience, the relative advantage of online banking and perceived risk, and
perceived behavioral control factors predict the intention to adopt Internet banking services. The survey by
Karjaluoto et al. [2002] showed that prior experience with computers and technology as well as people’s attitudes
toward computers influences both their attitude toward online banking and their actual behavior. Mukherjee and
Nath [2003] found that communication had a moderate influence on trust, while opportunistic behavior had a
significant negative effect and trust in general led to a higher level of commitment in online banking. Information
sharing and distrust in the Internet were identified as the two major drawbacks for Thai Internet banking adoption by
Rotchanakitumnuai and Speece [2004]. Based on a survey amongst Finnish banking customers, Pikkarainen et al.
[2004] found perceived usefulness and information on online banking on the Web site to be the main drivers for the
acceptance of online banking. A recent study by Lassar et al. [2005] showed that Internet related innovativeness is
positively related to the adoption of online banking.
With the number of online banking consumers steadily increasing, the focus of attention shifts from enticing
customers to the online world to retaining them. While the focus of the aforementioned papers lies on offline versus
online, this paper deals with the problem of how to keep a customer online and loyal to a specific supplier.
Therefore, we analyze which antecedents might induce customers to stay with a particular online bank instead of
switching suppliers. We will start discussing our model by referring to the concept of loyalty, which has been
investigated extensively in the offline world.
2.2 Loyalty
The concept of loyalty is defined from three different angles. Besides discussing previous research to
conceptualize the construct, we present a structural definition of loyalty and introduce our framework for measuring
important antecedents of loyalty in an online context.

Page 98

---

Journal of Electronic Commerce Research, VOL 7, NO.2, 2006
2.2.1 Behavioral and Attributional Definitions
The modeling of loyalty has a long tradition in academic literature research [Jacoby and Knyer 1973]. The
majority of early studies define loyalty as the repeat purchasing of a particular service or product [Homburg and
Giering 2001]. This approach has been long criticized by numerous scholars for the missing differentiation between
true and spurious loyalty: "The key point is that these spurious loyalty buyers lack any attachment to brand
attributes, and they can be immediately captured by another brand that offers a better deal…" [Day 1969]. In order
to avoid the pitfall of equating repeat purchasing with loyalty, the combination of attitudinal and behavioral
attributes is recommended [Grisaffe 2001]. This paper therefore applies a two-dimensional conceptualization of
loyalty consisting of both attitudinal and behavioral elements, with recommendation and repeat purchasing acting as
sub-dimensions of the construct.

Table 1: Models in E-Banking Literature
Author Endogenous
Variables
Exogenous
Variables
Tan and Teo [2000]
Intention to Use Internet
Attitudes
Banking Service
Subjective Norms
Perceived Behavioral Control
Karjaluoto et al. [2002]
Attitude Toward Internet
Prior Computer Experience
Banking
Prior Technological Experience
Internet Banking Usage
Personal Banking Experience
Reference Group Influence
Mukherjee and Nath [2003] Commitment
Shared Value
Trust
Communication
Opportunistic Behavior
Rotchanakitumnuai and
Internet Banking
Web Benefits (Information Quality, Information
Speece [2004]
Adoption
Accessibility, Information Sharing, Transaction Benefit)
Web Barriers (Organization Barrier, Trust, Legal
Support)
Pikkarainen et al. [2004]
Online Banking Use
Perceived Usefulness
Perceived Ease of Use
Perceived Enjoyment
Information on Online Banking
Security and Privacy
Quality of Internet Connection
Lassar et al. [2005]
Online Banking Adoption Consumer Innovativeness
Personal Characteristics

2.2.2 Dispositional Definition
The positive outcomes of loyalty have been the subject of several theoretical articles and empirical studies.
Reichheld and Sasser [1990] found that reducing defections by 5% yields improvements in profitability of 20% to
85%. When Reichheld and Schefter [2000] analyzed customer life-cycle economics in several e-commerce sectors
(e.g. online selling of books, groceries and consumer electronics) they found that on the Internet the same rules
apply as in the offline world. Early losses, which are caused by expenses for acquiring new customers, are followed
by rising profits, caused by a higher willingness to pay and more tolerance on the part of the customer if problems
occur [Zeithaml et al. 1996]. In fact, the success of several online companies is attributed to their high ratio of repeat
sales [Gefen 2002]. Amazon.com, for example, generated 66% of its sales from purchases made by returning
customers a couple of years ago already [The Economist 2000]. Loyal customers are also more inclined to
recommend an online service provider to other customers [Heskett et al. 1994]. Referrals increase the customer base
by lowering the costs of attracting new ones [Reichheld 1996].
2.2.3 Structural Definition
The structural definition of a construct describes the way in which it is linked to other variables [Bagozzi 1984].
Our model, which describes loyalty as the endogenous variable, includes one exogenous (Web site quality), three
mediating (service quality, overall satisfaction, trust) and five moderating variables (gender, age, involvement,
variety seeking behavior, technophobia) and can be seen in Figure 1. Taking into consideration psychological
attributes of customers such as involvement or technophobia is crucial when measuring loyalty [Mägi 2003]. In the
following sections all influencing variables will be described briefly.

Page 99

---

Floh & Treiblmaier: What Keeps the E-Banking Customer Loyal?
Satisfaction can be defined as a post-choice evaluative judgment concerning a specific purpose decision [Oliver
1979] and is mostly used as part of the confirmation/disconfirmation paradigm [Oliver and Svan 1989]. Although in
previous models customer satisfaction is solely seen as the result of cognitive processes, recent studies and
conceptualizations suggest that affections contribute to the explanation and prediction of customer satisfaction
[Fornell and Wernerfelt 1987, Homburg and Giering 2001]. Furthermore, several authors have argued that
satisfaction is based on the customer's cumulative experience rather than being a transaction-specific phenomenon
[Anderson et al. 1994, Bayus 1992]. Especially in the context of the relationship between loyalty and satisfaction,
conceptualizing satisfaction with a single transaction is too restrictive [Homburg and Giering 2001]. Dissatisfaction
with a single transaction does not lead to customer switching, neither does one satisfying transaction result in long-
term loyalty. For measuring satisfaction we therefore used a single item and asked the respondents to state their
overall level of satisfaction [Kettinger and Lee 1994].
With an increasing number of customers being online, the importance of Web sites for influencing purchasing
decisions is rising steadily. Measuring the quality of Web sites from a user's perspective enables companies to take
corrective actions, develop an appropriate e-business strategy, and improve their operations [Ganapathy et al. 2004,
Seethamraju 2004]. Service quality and Web site quality are incorporated into our structural model to account for
the above-mentioned distinction between cumulative and transaction-specific experiences. Whereas service quality
and Web site quality are supposed to measure the perceived transaction quality on a cognitive basis, overall
satisfaction refers to the cumulative experience based on an affective component. It is important to mention that
service quality and overall satisfaction implicitly include issues such as price perception, which is usually only felt
rather than objectively measurable. This can be argued by the complex and constantly changing pricing system in
the banking industry, which makes it hard for the average customer to determine the overall costs of the banking
product. We therefore hypothesize that service quality and Web site quality positively influence perceived overall
satisfaction (H1, H2). Additionally, we assume a positive effect of perceived Web site quality and service quality on
trust (H3, H4). Finally, Web site quality can be seen as an antecedent of service quality (H5). As the services
marketing literature suggests, customers who are not satisfied with the basic services of the organization are likely to
seek to satisfy their needs elsewhere [Gruen 1995]. Consistent with Bitner [1990] and Rust et al. [1995] we
hypothesize a positive impact of customer satisfaction on customer loyalty (H6).
Trust can be defined as the willingness to rely on an exchange partner in whom one has confidence [Moormann
et al. 1992]. It is an important antecedent in most models dealing with relationships that include loyalty or
satisfaction as dependent variables [Schaupp and Bélanger 2005]. Bryant et al. [2002] state that "trust is an
important consideration in the development and fostering of e-Commerce relationships in the context of the
knowledge-based economy". Lowering perceived risks associated with online transactions as well as maintaining
transaction trust are vital keys to attracting and retaining customers [Verhagen and Tan 2004]. Following Morgan
and Hunt [1994], we hypothesize a positive effect of trust on loyalty (H7).

Web si
eb s t
i e
H [+]
1
Overal
v
l
eral
Quality (W
Quality (
Q)
Satisf
i action
i
(OSAT)
H [+]
H [+]
3
6
H [+]
5
Loyalty (LOY)
H [+]
2
H [+]
7
Service
Serv
Quality (SQ)
Trust (TR)
H [+]
4
MODERATOR VARIABLES
Gender
Age
Involvement
Variety Seeking Behavior
Technophobia

Figure 1: Antecedents of Loyalty


Page 100

---

Journal of Electronic Commerce Research, VOL 7, NO.2, 2006
Based on previous research we included gender, age, involvement, variety seeking behavior, and technophobia
as moderators in our model. The impact of gender on buying behavior in an offline environment has been the focus
of attention in numerous scholarly papers [Jasper and Lan 1992, Slama and Tashlian 1985, Zeithaml 1985]. Slama
and Tashlian [1985] found that women are more involved in purchasing activities and more heavily influenced by
personal interactions than men. Although similar gender studies in an online environment are difficult to find, we
hypothesize that gender plays a moderating role on the Internet, too. Age is another popular demographic
characteristic in previous consumer studies. Most of the studies focus on the differences in people's information-
processing abilities [Gilly and Zeithaml 1985, Roedder and Cole 1986]. Gilly and Zeithaml [1985] found that our
capability to process information declines with age. Similar to Homburg and Giering [2001], we suggest a negative
moderating role of age on Web site quality, service quality and satisfaction.
Involvement can be defined as the degree of personal relevance of an object, product or service to a customer
and affects customer behavior in a number of ways [Beatty et al. 1988, Zaichkovsky 1985, Barki and Hartwick
1989]. Bloemer et al. [1996] state that the level of customer involvement with the product must be considered when
measuring customer retention. We therefore perceive involvement as an important moderating variable in both the
offline and the online world.
McAllister and Pessemier [1982] provided an interdisciplinary review of variety seeking behavior. They divided
the explanations into two groups, namely (a) derived motivation, in which varied behavior is the result of some other
motivation such as multiple uses, and (b) direct motivation, in which varied behavior is the result of a desire for
change per se due to interpersonal or intrapersonal motives. In line with Homburg and Giering [2001], who found
that variety seeking behavior moderates “offline” customer retention, we take into account the importance of variety
seeking in the online world.
Following Dekimpe et al. [2000], technological risk has to be added to the list of variables negatively
influencing the perception of risk. Many users are overwhelmed by the technological complexity of computers
leading to a low level of self-efficacy [Thatcher and Perrewe 2002]. This renders consumers less open to innovative
technology-related products and may lead to an aversion to sophisticated products or technologies. This behavior
can be described as technophobia [Mitchell 1994], which we hypothesize to have a negative moderating effect on
loyalty in the electronic banking sector.

3. Research Methodology and Data
According to Kierzkowski et al. [1996], the online banking and finance industry has a high potential for
building individual relationships on the World Wide Web. We therefore selected the customers of a pure DotCom
bank as our universe. The following sections discuss in detail the methodology of the survey and the development of
the measurement instrument.
3.1 Sample and Data Collection
The survey was conducted in cooperation with the largest Austrian online bank, which mailed a questionnaire to
7,500 randomly selected customers. The scope of the study was explained in a cover letter and an enclosed return
envelope guaranteed the anonymity of the results. After three weeks a total of 2,253 respondents (30.04 %) had
replied, of which 178 questionnaires had more than 10% missing values and were excluded from further analyses.
Missing EM-Algorithm was used for data imputation for the remaining 2,075 data records, leading to a final
response rate of 27.67%. Demographic characteristics of the respondents are listed in the appendix. We compared
the characteristics of our sample to those of the universe of Austrian Internet users published by the Austrian
Internet Monitor [2005] and found no statistically significant differences. Furthermore, T-tests showed no significant
difference between early and late respondents, which otherwise could be seen as an indicator for a non-response bias
in quantitative surveys [Armstrong and Overton 1977]. As non-normality of data occurred in the data file, we used
bootstrapping for testing the effects of non-normal distributed variables on our structural equation model [Efron and
Tibishiran, 1993]. The analyses produced no significant changes in parameter estimation. Finally, the data were split
randomly into two sub-samples (n1 = 1,015, n2 = 1,060). Sample 1 serves as the calibration sample on which the
initially hypothesized model is tested and on which the post-hoc analysis was conducted. Afterwards, the validity of
the structure of the final model was tested based on sample 2.
3.2 Measures
Empirically validated scales were adapted to the context of the study and used to measure the respective
constructs. All items are listed in Appendix B. Additionally, a confirmatory factor analysis was used to assess
construct measurement. Four items were removed after the analysis based on inadequate factor loadings and
theoretical arguments. A 6-point Likert scale was used to measure all items. In the case of Web site quality, item
parceling was used to reduce the total of 15 items to three subscales (design, structure and content).

Page 101

---

Floh & Treiblmaier: What Keeps the E-Banking Customer Loyal?
A comprehensive pretest, including qualitative interviews and focus groups with customers, was carried out to
ensure the understandability of the items. In total, 18 persons of different gender, age and educational backgrounds
were asked to fill out the questionnaire and, at the same time, to comment on the questions. Their comments were
written down and led to a complete revision of the questionnaire in order to increase its understandability.

4. Results
In the following sections we report the results of our survey. First, the measurement model is evaluated and
local fit indices are discussed. Second, the standard regression weights and global fit indices for both samples are
given. Finally, we present the multigroup analyses based on the respective indicator variables and the latent mean
structures.
4.1 Measurement Assessment
Prior to assessing the fit of the global model, it is necessary to check the quality of the construct measurement.
This was done by following the standard procedures for scale development and assessment postulated by Anderson
and Gerbing [1988].
Positive factor loadings are a necessary condition for adequate construct measurement. As can be seen in Table
2, the indicator reliability of all items exceeds .4 [Bagozzi and Baumgartner 1994], and the factor reliability is higher
than .6 [Bagozzi and Yi 1988].

Table 2: Local Fit Indices
Latent Variable
Item Name Indicator Reliability T-Values Factor Reliability AVE FLR
x1 0.708 ---
Web Site Quality (WQ)
x2 0.590
20.102
0.93 0.74
0.52
x3 0.451
20.872
y1 0.636
30.856
y2 0.679
32.465
Service Quality (SQ)
y3 0.816
37.955
0.87 0.69
0.52
y4 0.767
35.953
y5 0.776 ---
y7 0.798
25.551
Trust (TR)
y8 0.849
26.008
0.80 0.58
0.57
y9 0.470 ---
y10 0.520 ---
Loyalty (LOY)
0.70 0.55
0.56
y11 0.821
27.963

Alternative measures of how well a construct is measured by its indicators are the average variance extracted
[Anderson and Gerbing 1988] and the Fornell-Larcker Ratio [Fornell and Larcker 1981]. Values greater than 0.5 for
the average variance extracted and lower than 1 for the Fornell-Larcker Ratio are recommended [Bagozzi and Yi
1988]. In this study, all of the mentioned fit indices meet the recommended levels. Further analyses can be
conducted as values suggest strong construct and discriminant validity.
4.2 Model Testing
Structural Equation Modeling (SEM) appears to be the best available statistical technique for testing the
hypotheses since it includes the indirect effects of one latent variable on another [Nidumolou 1989]. In Figure 2 an
overview of the hypothesized effects is given and all parameters are labeled. As was discussed in previous sections,
we hypothesize that loyalty (LOY) is positively influenced by overall satisfaction (OSAT) (H6) and trust (H7),
which are in turn affected by Web site quality (WQ) and service quality (SQ) (H1 – H4). Additionally, the quality of
a Web site affects the perception of service quality (H5).
AMOS 5.0 was used to estimate the main effects. Table 3 shows the standardized regression weights with their
relevant t-values for both samples, all of which are significant (p<.01), supporting all hypotheses. Loyalty is
therefore significantly affected by satisfaction and trust. Additionally, effects of Web site quality on service quality,
satisfaction and trust were observed, as was a significant effect of service quality on overall satisfaction.
The overall fit measures indicate that the hypothesized model is a good representation of the structures
underlying the observed data.
The goodness-of-fit index and the adjusted goodness-of-fit index, two descriptive overall measures, both meet
the recommended value of .9 [Bagozzi and Yi 1988]. The RMSEA, a measure that is based on the concept of
noncentrality for both samples is .059, which is slightly below the recommended upper limit of .6 [Hu and Bentler
1999]. In both cases the χ2/df ratio is higher than 2.5, which can be attributed to the comparatively large sample

Page 102

---

Journal of Electronic Commerce Research, VOL 7, NO.2, 2006
sizes. However, both incremental fit measures meet the recommended levels, which are .95 for the NFI [Hu et al.
1999] and .9 for the TLI.

ε2
y6
λy
ε
ε
δ
62
ζ
x
10
1
2
λx
1
1
11
1
λx
γ
21
WQ
δ
21
OSAT
x
2
2
ξ
y
y
10
11
η
1
2
β
δ
λx31
42
3
x
λy
λy
3
10 4
11 4
β21
ε
γ
LOY
11
1
y1
η4
γ
ε
y
λy
31
21
2
2
λy11
ζ
β
4
λy
SQ
β
43
ε
31
TR
y
31
3
3
η
η
λy
1
3
41
ε
λy51
λy
ζ3
4
y
λy
4
73
ζ
λ
93
y
1
83
ε5
y
y
y
y
5
7
8
9
ε
ε
ε
7
8
9

Figure 2: Structural Equation Model Measuring the Antecedents of Loyalty

Table 3: Standard Regression Weights and T-values
Sample 1
Sample2
Parameter Parameter Value (standardized) t-value Parameter Parameter Value (standardized) t-value
γ11
0.472 13.328
γ11
0.402 11.183
γ21
0.425 10.617
γ21
0.457 11.460
γ31
0.431 10.902
γ31
0.365 9.312
β21
0.548 14.204
β21
0.525 14.095
β31
0.290 8.318
β31
0.287 8.159
β42
0.690 13.890
β42
0.600 11.817
β43
0.250 6.599
β43
0.270 6.698

Table 4: Global Fit Indices
Goodness-of-Fit Measure
Sample 1
Sample 2
Stand Alone Fit Measures
χ2 (d.f.)
315.363 (70) 349.342 (70)
AGFI 0.936
0.934
GFI 0.957
0.956
RMR 0.038
0.038
RMSEA 0.059
0.059
Incremental Fit Measures
NFI 0.966
0.960
TLI 0.965
0.958

Thus, although the χ2 statistic is significant for both samples (p < .01) we conclude that the model has been
validated successfully and can be seen as appropriate for the explanation and prediction of loyalty in the context of
online banking.

Page 103

---

Floh & Treiblmaier: What Keeps the E-Banking Customer Loyal?
4.3 Multigroup and Latent Mean Analysis
Once support for the main effects had been found, the next step was to include the suggested moderator
variables into the model in order to gain further insights. Median splits were conducted in this study based upon the
values of the moderator variables. Furthermore, multiple group analyses were calculated in a hierarchical approach
comparing two sub-samples which were selected according to gender or based on a median split of the respective
moderating variable.
In a first step, an overall Chi-square difference was calculated for each of the moderator variables. Technically,
a model with equality constraints is compared to a model that allows the parameters to vary. This test imposes the
null hypothesis that the moderator variables do not have any effect on the seven parameters. As can be seen from
Table 5, these hypotheses were rejected for each of the moderator variables (Δχ2 ≥ 14.07, ΔDF =7). Afterwards,
constraints were imposed to test the invariance of the model across various subgroups. According to Steenkamp and
Baumgartner [1998], who propose a hierarchical procedure in multigroup analysis, the equivalence of measurement
weights was analyzed in a second step [Steenkamp and Baumgartner 1998]. Because these models are nested with
the general model having one degree of freedom less than the restricted model, the χ2-value will always be lower
for the general model [Homburg and Giering 2001]. Significant differences (CR = 3.84 at the 5% level) indicate that
the hypotheses of the moderator effect are supported.
In a last step, the invariance of the means of the latent variables was tested. The purpose of testing latent mean
structures is to test the equivalence of means related to each underlying factor. Because these factors cannot be
observed directly, latent means can be calculated for one group only. The results of the multigroup analyses are
shown in Table 5 and Table 6.

Table 5: Results of Multigroup Analyses
Gender
Chi-Square Difference
Male Female
(ΔDF =1)
γ11 = 0.592 (14.962)
γ11 = 0.455 (7.752)
Δχ2 = 4.402*
γ21 = 0.384 (12.254)
γ21 = 0.372 (8.367)
Δχ2 = 0.063
γ31 = 0.458 (11.388)
γ31 = 0.526 (8.296)
Δχ2 = 0.011
β21 = 0.414 (19.915)
β21 = 0.311 (9.507)
Δχ2 = 7.760*
β31 = 0.261 (9.447)
β31 = 0.317 (7.124)
Δχ2 = 0.220
β42 = 0.801 (16.030)
β42 = 0.843 (8.514)
Δχ2 = 0.936
β43 = 0.252 (7.819)
β43 = 0.231 (5.525)
Δχ2 = 0.093
Δχ2 = for all gammas set equal across subgroups (DF = 7): 15,161*

Age
Chi-Square Difference
Low High
(ΔDF =1)
γ11 = 0.514 (12.273)
γ11 = 0.602 (12.446)
Δχ2 = 1.971
γ21 = 0.299 (8.425)
γ21 = 0.468 (13.584)
Δχ2 = 11.645*
γ31 = 0.438 (9.664)
γ31 = 0.517 (10.388)
Δχ2 = 1.303
β21 = 0.464 (15.170)
β21 = 0.315 (13.511)
Δχ2 = 16.503*
β31 = 0.344 (9.699)
β31 = 0.226 (6.982)
Δχ2 = 7.284*
β42 = 0.834 (12.681)
β42 = 0.828 (13.918)
Δχ2 = 0.079
β43 = 0.182 (5.080)
β43 = 0.294 (8.153)
Δχ2 = 5.085*
Δχ2 = for all gammas set equal across subgroups (DF = 7): 35,727*

Involvement
Chi-Square Difference
Low High
(ΔDF =1)
γ11 = 0.586 (13.563)
γ11 = 0.475 (10.258)
Δχ2 = 4.223*
γ21 = 0.387 (11.341)
γ21 = 0.390 (10.288)
Δχ2 = 0.045
γ31 = 0.470 (10.477)
γ31 = 0.494 (9.659)
Δχ2 = 0.008
β21 = 0.367 (14.345)
β21 = 0.416 (14.128)
Δχ2 = 1.396
β31 = 0.265 (8.449)
β31 = 0.309 (8.281)
Δχ2 = 0.753
β42 = 0.835 (13.510)
β42 = 0.724 (11.768)
Δχ2 = 1.641
β43 = 0.276 (7.793)
β43 = 0.203 (5.715)
Δχ2 = 2.063
Δχ2 = for all gammas set equal across subgroups (DF = 7): 14,317*

Page 104

---

Journal of Electronic Commerce Research, VOL 7, NO.2, 2006
Variety Seeking Behavior
Chi-Square Difference
Low High
(ΔDF =1)
γ11 = 0.586 (13.563)
γ11 = 0.475 (10.258)
Δχ2 = 17.577*
γ21 = 0.387 (11.341)
γ21 = 0.390 (10.288)
Δχ2 = 0.289
γ31 = 0.470 (10.477)
γ31 = 0.494 (9.659)
Δχ2 = 2.025
β21 = 0.367 (14.345)
β21 = 0.416 (14.128)
Δχ2 = 58.690*
β31 = 0.265 (8.449)
β31 = 0.309 (8.281)
Δχ2 = 15.078*
β42 = 0.835 (13.510)
β42 = 0.724 (11.768)
Δχ2 = 0.307
β43 = 0.276 (7.793)
β43 = 0.203 (5.715)
Δχ2 = 0.361
Δχ2 = for all gammas set equal across subgroups (DF = 7): 121,778*



Technophobia
Chi-Square Difference
Low High
(ΔDF =1)
γ11 = 0.524 (11.232)
γ11 = 0.569 (13.241)
Δχ2 = 0.351
γ21 = 0.378 (9.249)
γ21 = 0.388 (12.239)
Δχ2 = 0.192
γ31 = 0.513 (9.305)
γ31 = 0.449 (10.717)
Δχ2 = 0.838
β21 = 0.447 (13.304)
β21 = 0.345 (15.156)
Δχ2 = 4.013*
β31 = 0.309 (7.446)
β31 = 0.268 (9.315)
Δχ2 = 0.496
β42 = 0.721 (11.985)
β42 = 0.866 (13.592)
Δχ2 = 1.520
β43 = 0.232 (6.829)
β43 = 0.265 (7.054)
Δχ2 = 0.283
Δχ2 = for all gammas set equal across subgroups (DF = 7): 9,013*
*Chi-square difference is significant at the 5% level.

Table 5 shows a moderating effect of gender for two parameters (γ11 and β21). These results suggest that the
influence of Web site quality on service quality and service quality on overall satisfaction is significantly higher for
men than for women. In other words, service quality for men is more important in explaining satisfaction. This result
is supported by a significantly lower Web site quality estimate for men.
Age is our next moderator variable. The analysis shows a positive moderator effect on the relationship between
Web site quality – satisfaction, service quality – satisfaction, service quality – trust and trust – loyalty. Additionally,
elderly people rate the importance of Web site quality and service quality significantly lower than younger
respondents. These results are similar to the findings of the study of Homburg and Giering [2001], who found that
service quality has a stronger impact on satisfaction for younger people than for the elderly.

Table 6: Latent Mean Structures; standardized estimates (t-values)
Web Site
Service
Overall
Moderator Variable
Group
Trust Loyalty
Quality
Quality
Satisfaction
Male+ 0
0
0
0 0
Gender
-0.267
0.002
0.009
0.028
0.025
Female
(-8.639)*
(0.058)
(0.266)
(0.0753)
(0.530)
Low+ 0
0
0
0 0
Age
0.068
-0.119
-0.019
0.017
0.013
High
(2.381)*
(-3.442)*
(-0.655)
(0.505)
(0.299)
Low+ 0
0
0
0 0
Involvement
0.109
0.171
-0.022
-0.018
0.169
High
(3.767)*
(4.928)*
(-0.736)
(-0.549)
(3.962)*
Low+ 0
0
0
0 0
Variety Seeking
-0.241
-0.174
-0.077
-0.062
0.008
Behavior
High
(-8.372)*
(-4.720)*
(-2.531)*
(-1.768)
(0.186)
Low+ 0
0
0
0 0
-0.112
Technophobia
-0.017
-0.109
-0.009
0.042
High
(-
(-0.585)
(-3.151)*
(-0.312)
(1.251)
2.539)*
+ Reference Group
* Chi-square difference is significant at the 5% level.

Page 105

---

Floh & Treiblmaier: What Keeps the E-Banking Customer Loyal?
The multigroup analysis for involvement shows only one moderating effect. The effect of Web site quality on
service quality is lower for low-involvement people. Otherwise, comparisons of the latent mean structure between
low-involvement and high-involvement people indicate a significant difference for the latter. The moderating role of
involvement leads to a higher perception of the two quality aspects. Furthermore, highly involved people stay more
loyal to an online bank than people with low involvement in banking and finance.
Variety seekers are people who tend to switch between several suppliers regardless of their perceived
satisfaction with previous companies or service clients. The current findings are partially inconsistent with the
current interpretation of variety seeking. While analyses show significant differences in latent mean structures of
service and Web site quality, no difference was found between the two groups of low/high variety seekers in terms
of loyalty. These surprising results may be explained by a very low total mean of the variable (mean = 2.66; 1 =
totally agree, 6 = totally disagree).
As the last moderating variable, technophobia has a negative moderator effect on the relationship between
service quality and satisfaction. In other words, service quality is more important for people with low technophobia.
The latent mean structure analyses confirm these findings. The mean estimate of service quality is significantly
lower for this group. Additionally, people who have less anxiety in using the Internet are more loyal toward an
online service provider.

5. Conclusions and Outlook
Our results confirm that loyalty of e-banking customers is directly affected by satisfaction and trust in an online
bank, which in turn are determined by Web site quality and service quality. Moderating variables such as gender,
age, involvement, variety seeking behavior and technophobia exert a significant influence on some of the proposed
relationships. These results have several implications for those banks which want to increase loyalty on the World
Wide Web.
First, the quality of Web sites has a direct and an indirect impact on both satisfaction and trust. Companies have
to redesign their Web sites with a view to enhancing usability and usefulness. Amongst the many factors which
account for the perceived quality of a Web site, the avoidance of downtimes seems to be extremely important to
online banks. Furthermore, based on related literature, we recommend making the sites easy to navigate and giving
them an uncluttered look. Sufficient information should be given on how to conduct transactions and, most
importantly, on how to get help should unforeseen events happen. Similarly to the quality of the site, the perceived
quality of the service exerts a significant influence on overall satisfaction and trust.
Second, trust and overall satisfaction can be seen as major antecedents of e-loyalty. We would therefore
recommend that trust-building actions are paid more attention in scholarly literature, focusing for example on pay-
back guarantees or quality certificates, which are seen as helpful steps in increasing electronic customer retention.
There is a plethora of literature on trust and some exemplary papers have been cited in previous sections. It seems
obvious that the results of many surveys suggest incorporating trust-building measures into online customer
relationships. As far as our research is concerned, the preeminent importance of trust can be explained by both the
core products of the financial industry, which can be seen as the transmission and processing of highly confidential
information, and trust in the medium as such, which again stands for the bank's capability to securely transfer and
store confidential personal information. Unless customers establish personal contacts a in a bank branch, users of
Internet banking in many cases do not have well-known contact persons and must rely completely upon the
capability and trustworthiness of the bank. Therefore, the bank must build a strong brand in order to signal
competence to its customers. As mentioned above, we used a pure DotCom bank as our focus of research in order to
avoid interchannel conflicts. Nonetheless, it must be mentioned that this bank is the subsidiary of a well-established
Austrian bank, which is clearly displayed on the homepage. The trusted brand of the parent company is thus used to
create trust on the part of the consumers.
Third, and somehow a by-product of this research, our empirical findings demonstrate that surveys might be an
adequate instrument for online banks to learn about their customers' attitudes. The comparatively high response rate
for an online survey can be taken as an indicator that customers of DotComs are actually willing to give feedback
and get in touch with their supplier.
Further research should be performed in this area to validate the model across other industries in order to assess
its general stability. While we used a cross-validation approach by taking the same sample population for model
calibration and validation, the use of different populations will extend its general validity. Furthermore, it might be
interesting to see how the parameters change in countries with different legal regulations and varying Internet user
behavior. A possible sample bias limits the the generalizability of results. Additionally, we recommend replication
studies which focus exclusively on the influence of the moderating variables and might include other influencing
variables such as the price of the product. By understanding how different customer segments can be differentiated,

Page 106

---

Document Outline

• ÿ

---