A two-staged regression-neural network approach
Abstract
Purpose – The purpose of this paper is to explore the
main determinants of Internet banking users on the basis of literature of
technology acceptance model (TAM). Understanding and predicting main
determinants of Internet banking is an important issue for banking industry and
users.
Design/methodology/approach – Service quality and trust were incorporated in the TAM together
with demographic variables. The data were collected using Google Docs from 103 Sri Lankan Internet banking users. A two-staged regression-neural network model was applied to understand and predict
Internet banking adoption.
Findings – The results obtained
from multiple linear
regression model were compared with the results from neural network model to
predict Internet banking adoption and the performance of latter model was found
to superior. The neural network model was able to capture relative importance
of all independent variables, service quality, trust, perceived usefulness,
perceived ease of use, attitude and demographic variables.
Practical implications – This study provides
useful insights with regard to development of Internet banking systems to
banking professionals and information systems researchers in Sri Lanka and
similar emerging economies.
Originality/value –The majority of studies
in Internet banking adoption in Sri Lanka and elsewhere usually utilize
modeling methods suited for explanatory purposes.
Keywords
Neural network, TAM,
Sri Lanka, Internet banking, Regression, Service quality
Paper type Research paper
1. Introduction
Internet
banking technologies have been implemented by a large number of banks globally
for providing greater convenience, round the clock availability, reduced
employee and transactional costs, easier customer access to information and
increased accuracy. The majority of the earlier research studies focused on
Internet banking adoption from either a service provider or an employee
perspective (Daniel,
1999; Anandarajan et al., 2000; Aladwani, 2001; Khalfan and Alshawaf, 2004; and Corrocher, 2006). However, the focus of the majority of the studies
subsequently shifted to the adoption of these technologies by the user for two reasons:
(1)
Problems pertaining to provider and employee adoption have
largely been resolved.
(2)
Global banks are relying on Internet technologies as the key to achieve
customer satisfaction and efficiency necessary for surviving in the
highly competitive banking industry.
A
number of research studies have been conducted to understand the user adoption
of online banking
services in the developed
and developing economies (Sathye, 1999; Al-Somali et al.,
2009; Gikandi and
Bloor, 2010; Adesina and
Ayo, 2010; Riffai et al.,
2011; Chong et al., 2012b). In developing economies, particularly the Gulf Cooperation
Council (GCC), the introduction of Internet
banking technologies lagged a few years following their launch in developed
countries. Banking service providers, both
domestic and multinational,
in the GCC have reported large-scale
implementations of Internet
technologies for providing enhanced services for retail
and corporate customers. Despite the
rapid growth in the penetration of Internet technologies, the
use of Internet for banking
purposes is reported to be less than 20 per cent in the GCC countries (Augustine, 2013). The
Internet adoption in developed countries has been well-researched, whereas in the case of developing
countries in the GCC, the resolution
of problems concerning
banking technology acceptance can benefit
from further research contributions (Riffai et al., 2011).
This study intends
to achieve the following objectives. Firstly, this research
focuses on understanding important determinants that are useful in predicting the adoption of Internet banking based
on modified technology acceptance model (TAM).
The original TAM as proposed by Davis (1989) is
modified to incorporate additional variables, namely, trust and service
quality. Along with these variables, demographic variables
such as gender, age, Internet experience, income group and education are also
considered. Secondly, the paper intends
to examine whether
artificial neural networks achieve a superior model fit as
compared to multiple linear regression (MLR) for prediction of Internet banking
adoption. Finally, this research attempts to examine Internet banking
adoption in Sri Lanka, an emerging economy
in the GCC.
Often,
the TAM-based models used for modeling user behavior are explanatory in nature
and are effective in exploring the causal relationships between the variables
concerned. However, Shmueli
(2010) discusses
differences between explanatory and predictive models and points toward the
need to exercise caution when using explanatory models for predictive purposes.
The linear regression methods used in the case of explanatory models for
examining causal relationships often assume linear compensatory decision making
by the user. In reality, users’ decisions with respect to the adoption of
Internet banking technologies can be non-compensatory, and the complex nature
of these adoption decisions may often necessitate the use of nonlinear
statistical methods (Chong,
2013a).
In Section
2, key literature pertaining to Internet banking
adoption, TAM and neural
network modeling are discussed. Section 3 includes the study design and
analyses, followed by the discussion of results in Section 4. Finally, the managerial implications, limitations and future work
are discussed in Section 5.
1.1 Overview
and Internet technology evolution
The Sultanate of Oman is one of the important member
nations within the GCC located in the southeastern region of Arabian Peninsula,
adjacent to Saudi Arabia, Republic of Yemen and the UAE. Subsequent to 2000,
the improvement in oil prices resulted in a tremendous improvement in economic
growth. Currently, the country’s gross domestic product stands at US $80.57
billion (World Bank Report, 2013), and Oman maintains a significant trade
surplus owing to its oil and gas exports. The country had a population of 3.63
million in 2013 (World Bank Report, 2013) of which expatriates contribute a
sizeable percentage of more than 30 per cent.
The launching of Internet by Omantel, then a public
sector organization, marked the beginning of the e-revolution in Oman and is
considered as one of the major milestones in Oman’s modernization drive that
began in the 90s. Omantel and Nawras, both private companies, are the primary
fixed and mobile Internet service providers in Oman. The penetration of
Internet and the growth of network coverage has been quite impressive; the
Internet penetration has increased from 22.8 to 39.44 per cent between the
years 2009 and 2013 (Telecom Regulatory Authority of Oman, 2013). As of 2013,
there were 2,443,296 active mobile broadband subscribers and 154,290 fixed
broadband subscribers (Telecom Regulatory Authority of Oman, 2013). The
development of information and technology infrastructure to lay foundation for
a digital society that would support e-government and e-commerce constitutes
one of the main strategic priorities for the Omani Government. Thus, the
Information Technology Authority, an independent legal body allied to the
Ministry of National Economy, was established in 2006. Omantel has recently
implemented several projects, including the installation of new transmission
stations in remote areas to provide the entire population with mobile services.
2. Literature
review
2.1 TAM and
Internet banking adoption studies
The
impact of user acceptance on the success of technology implementation motivated
research in the area of technology acceptance. Early works that laid the
foundations for understanding the nature of technology acceptance recognized
behavioral intention as the key determinant of technology acceptance and
contributed to the development of the intention-based models, namely, theory of
reasoned action model (Ajzen and Fishbein, 1980), theory of planned behavior
model (Ajzen and Madden, 1986) and TAM (Davis, 1989). These models were
followed by the Unified theory for acceptance and use of technology (Venkatesh
et al., 2003), also referred to as UTAUT, which combined elements of earlier
intention-based models and expectancy theory (Vroom, 1964). Among the models
proposed for understanding user acceptance behaviors, TAM was widely used for
understanding the factors influencing user acceptance with respect to Internet
banking in various countries. Davis (1989) as part of TAM proposed two
constructs perceived ease of use (PEOU) and perceived usefulness (PU). These
are defined as follows:
(1) “PU is the
degree to which a person believes that using a particular system would enhance
his or her job performance”; and
(2) “PEOU
refers to the degree to which a person believes that using a particular system
would be free of effort”.
Klopping
and McKinney (2004) provided evidence in support of the effectiveness of TAM in
predicting behavioral intention when compared to other competing models. A
sample of empirical studies that tested TAM or proposed and tested a modified
TAM include (Sathye, 1999; Thornton and White, 2001; Wang et al., 2003;
Pikkarainen et al., 2004; Sukkar and Hasan, 2005; Guriting and Ndubisi, 2006;
Cheng et al., 2006; Kuisma et al., 2007; Al-Somali et al., 2009; Riffai et al.
2011; Sharma and Chandel, 2013; and Ariff et al., 2014). Recently, Sharma and
Govindaluri (2014) discussed the factors affecting acceptance of Internet
banking technologies in India using a structural equation model based on
extended TAM.
Despite
the increasing global importance of the banks operating in the GCC, and the
large-scale implementation of Internet technologies in GCC banking, the number
of research studies that focus on user acceptance of Internet banking are quite
limited. Al-Somali et al. (2009) conducted a study among Saudi Arabian online
banking users that suggests Internet connection quality, awareness, social
influence and computer self-efficacy as significant predictors of online
banking acceptance. In the case of Oman, research has focused primarily on the
provider adoption of Internet banking (Khalfan and AlShawaf, 2004; Al-Hajri and
Tatnall, 2007), except for a study by Riffai et al. (2011) that mentions trust,
usability and perceived quality among the attributes influencing the adoption
of Internet banking in Oman.
2.2 Modified TAM
This
research is an attempt to modify TAM by adding trust, service quality and
demographic variables. The remainder of this section discusses these variables
and supports their inclusion in the model.
Trust
can be considered as one of the most significant factors impacting user
decision to adopt Internet banking, particularly as the user’s financial assets
can be exposed to multiple risks in an online banking relationship. Pavlou
(2003) had defined trust as “a belief that customers entrust upon online
retailers after careful consideration of the characteristics of retailers.” The
degree to which the user is willing to trust in spite of the security and
privacy concerns can depend on a number of variables. Al-Somali et al., (2009)
found trust as one of the key drivers of Internet banking adoption in a study
conducted in Saudi Arabia. An interesting research study by Yeh and Li (2009)
investigates the impact of customization, brand image and satisfaction in terms
of building customer trust toward m-commerce. Riffai et al. (2011) have studied
trust and observed that the levels of trust are low in Oman. In both the
aforementioned studies, trust is part of an explanatory model primarily used
for understanding causal relationships between trust and adoption of Internet
banking. In this research, trust is incorporated into the model with for
predictive purposes.
Service
quality can constitute one of the key determinants of adoption of Internet
banking. The relationship between service quality and service adoption is
well-researched (Dabholkar et al., 2000; Shih and Fang, 2006; and Siu and Mou,
2005). Shostack (1984), Parasuraman et al. (1985) and Grönroos (1984) laid the
conceptual foundations for modeling the quality of services. Two of the models,
namely, SERVQUAL by Parasuraman et al. (1985) and two-dimensional perceived
quality model were widely debated (Carman, 1990; Cronin and Taylor, 1994;
Parasuraman et al., 1994; and Brady and Cronin, 2001).
The
phenomenal growth of e-services in the late 90s led to the development of
specific models dedicated to the measurement of quality in information services
(Kettinger and Lee, 1999; Liu and Kirk, 2000; and Parasuraman et al., 2005). In
addition to the quality of e-services, some studies have addressed the quality
of online banking services (Siu and Mou, 2005 and Sangeetha and Mahalingam,(2011).
The quality issues in the context of Internet banking services have been
discussed by few researchers in the case of GCC countries. For example, Sohail
and Shaikh (2008) investigate the attributes that have a bearing on the quality
judgments of online banking users in Saudi Arabia and list three factors in the
order of importance as efficiency and security, fulfillment and responsiveness.
Furthermore, Sharma et al. (2013) investigated the attributes that have a
bearing on the quality judgments of e-government services in Oman.
Demographic
variables, including gender, age, and education, level have been discussed in
the literature in connection with adoption of Internet technologies (Chong et
al., 2012a, 2012b; Chong, 2013a). These studies show that demographic variables
can influence the Internet banking technology adoption both as an independent
variable or a moderating variable. For example, Riffai et al. (2011) found
demographic variables age, gender and education play the role of moderating
variables in the trust – adoption relationship. In addition to the three
aforementioned demographic variables, two variables, customer’s income group
and their overall Internet experience, were added to explore the impact of
demographic variables on the adoption of Internet banking.
2.3 MLR model
MLR
analysis is used to perform the modeling of the linear relationships between a
dependent variable, adoption with respect to
independent variables considered in this study (Hair et al., 2010). The
variable adoption measures the degree to which users adopt Internet banking
which is a combination of acceptance and usage of the Internet banking service.
One reason for using MLR in this study was to achieve the research objective of
comparison of performance of a linear modeling method MLR with a nonlinear
neural network model.
The
general model for MLR can be written as follows:
Yi = ß0 + ß1Xi1
+ ß2Xi2+
... +ßp-1Xi,p-1
+ εi, (1)
Where þ0, þ1, þp-1 are parameters; Xi1
…, Xi, p-1 are known constants; and si are statistically independent error
terms with N (0, o2). The MLR equation for the adoption of Internet banking is
as follows:
Adoptioni
= ß0 + ß1 * PUi
+ ß2 * PEOUi
+ ß3 *
Integrityi + ß4 *
Servicequalityi
+ ß5 *
Attitudei + ß6 * Gender
+ ß7 * Age
+ ß8 *
Income, where, i = 1,... n
2.4 Neural network
An
artificial neuron is a processing unit, which is similar to the biological
neuron of the human brain:
Artificial
neural network or neural network is a massively parallel distributed processor
made up of simple processing units, which have a natural propensity for storing
experimental
knowledge
and making it available for use (Haykin, 2007).
An
artificial neural network is a computational model resembles a human brain
where acquisition of new knowledge from the environment is achieved through
learning processes. The mathematical model of neural network consists of a
hidden layer feed forward network with x1,· · ·, xn,
as inputs and yk as the output. Synaptic weights are assigned to
each input and are transferred to the hidden layer made of a number of hidden
neurons. The weighted summation of the inputs is performed by each neuron and
transferred to a nonlinear activation function. The output of neural network is
given as follows:
Where
wkn is the synaptic weight between output of neuron k and input of
neuron n and Ý(v) is the
nonlinear activation function. The activation function used in the neural network
model was hyperbolic tangent. Neural network model offers various advantages
over traditional statistical models. A linear or nonlinear nature of neural
networks is a major advantage over traditional statistical models. Neural
network models have outperformed the traditionally used MLR method in
information systems research (Chong, 2013a). Applications of neural networks
are increasing in business research due to its computational power, flexibility
and ease of use (SPSS, 2010) (Figure 1).
3. Study
design and analysis
The
methodology in this paper is similar to the approach adopted by Chong (2013a).
This research uses neural network model as the primary tool for analysis and
compares the results obtained with respect to prediction of adoption of Internet
banking by neural network model with those obtained by the MLR model.
3.1 Sample and procedure
A
survey questionnaire was developed to explore the various determinants
affecting the adoption of Internet banking with the help of Google Docs, a free
online survey design
Service
provided by Google. A pilot study was conducted with 25 Internet banking users
and 3 experts of information systems. The survey was modified based on the
pilot study and opinions of experts. Convenient sampling procedure was adopted
to distribute survey. A Web page link of survey was sent to respondents using
mainly three channels, namely, friends, relatives and colleagues. The data were
collected using convenient sampling approach in the month of October 2019. In
total, 103 usable responses were obtained from Internet banking users. The
minimum sample size of ten samples for each independent variable was achieved,
as recommended by Hair et al. (2010).
3.2 Variables and measurements
This
paper attempts to predict the important determinants affecting the adoption of
Internet banking based on users’ perceptions. These determinants adapted from
the literature include service quality (Sohail and Shaikh, 2008), PEOU (Davis,
1989; Cheng et al., 2006), PU (Davis, 1989), attitude (Venkatesh et al., 2003;
Cheng et al., 2006), trust (Al-Somali et al., 2009; Riffai et al., 2011) and
demographic variables (Chong, 2013a) of Internet banking users. The demographic
variables include age, gender and income .All determinants other than
demographic variables were measured on a five-point Likert scale ranging from 1
(strongly disagree) to 5 (strongly agree). For measurement purposes, a total of
17 indicators were used for independent variables and 3 indicators in case of
the dependent variable adoption (Sharma and Govindaluri, 2014). All measurement
scale items and its descriptive statistics can be found in the Appendix
3.3 MLR analysis
The
MLR model is used to determine the statistically significant independent
variables that predict the adoption of Internet banking. MLR results yielded a
low value of coefficient determination (R2 = 0.63). As the R2 value is low
indicating the weakness of linear model in explaining the variance, a nonlinear
model may be considered. Hence, a neural network model, which is nonlinear in
nature, was selected to model the relationship between adoption and the
independent variables. The results obtained from regression model are compared
with the results of the neural network model in the next section. The summary
of MLR results is given in Table I. The results obtained from MLR shows that
Attitude (X Variable 4), Service Quality (X Variable 5) and PEOU (X Variable 8)
are the significant factors in determining the adoption of Internet Banking as
the p-values are less than 0.05 . Other variables are not statistically
significant as the p-values are greater than 0.05.
SUMMARY
OUTPUT
|
|
Regression Statistics
|
|
Multiple R
|
0.79188371
|
R Square
|
0.62707982
|
Adjusted R Square
|
0.59534193
|
Standard Error
|
0.30668903
|
Observations
|
103
|
ANOVA
|
|||||
df
|
SS
|
MS
|
F
|
Significance F
|
|
Regression
|
8
|
14.86727
|
1.858409
|
19.75808
|
3.6847E-17
|
Residual
|
94
|
8.841467
|
0.094058
|
||
Total
|
102
|
23.70874
|
Coefficients
|
Standard Error
|
t Stat
|
P-value
|
Lower 95%
|
Upper 95%
|
Lower 95.0%
|
Upper 95.0%
|
|
Intercept
|
2.76740134
|
0.220308
|
12.5615
|
7.96E-22
|
2.329974266
|
3.204828
|
2.329974
|
3.20482841
|
X Variable 1
|
0.0682367
|
0.062938
|
1.084185
|
0.281056
|
-0.056728644
|
0.193202
|
-0.05673
|
0.19320205
|
X Variable 2
|
0.0148882
|
0.023756
|
0.626707
|
0.532371
|
-0.032280376
|
0.062057
|
-0.03228
|
0.06205678
|
X Variable 3
|
0.02452263
|
0.023177
|
1.058076
|
0.292732
|
-0.021495114
|
0.07054
|
-0.0215
|
0.07054037
|
X Variable 4
|
-0.1358185
|
0.036007
|
-3.77196
|
0.000283
|
-0.207312051
|
-0.06432
|
-0.20731
|
-0.0643249
|
X Variable 5
|
-0.5395387
|
0.190352
|
-2.83443
|
0.005619
|
-0.917486484
|
-0.16159
|
-0.91749
|
-0.16159087
|
X Variable 6
|
-0.030071
|
0.028301
|
-1.06254
|
0.290715
|
-0.086263532
|
0.026122
|
-0.08626
|
0.02612159
|
X Variable 7
|
0.32942539
|
0.186506
|
1.766295
|
0.080593
|
-0.04088756
|
0.699738
|
-0.04089
|
0.69973833
|
X Variable 8
|
-0.0754254
|
0.027168
|
-2.7763
|
0.006636
|
-0.129367316
|
-0.02148
|
-0.12937
|
-0.02148355
|
3.4
Neural network results
The
neural network model was developed using Data Engine software in this research.
Three variables age, gender and income were included in input layer of the
network model along with the five covariates service quality, PU, PEOU, trust
and attitude. The dependent variable adoption which refers to adoption of
Internet banking is included in the output layer of the network model. The
results shows that neural network model is a better choice than regression
model. Furthermore, a nonlinear neural network model shows a better fit of the
model in comparison with MLR and could capture nonlinear relationship of
independent variables with dependent variable in the context of Internet
banking adoption (Table II).
The
training data set included 65 responses and the testing data set included 30
responses obtained from participants of the survey.
When
checking the output neuron value for the above criteria, the network achieved
the following results.
Neural
Network Results for set of 8 responses.
Respondent
|
Desired Output
|
Output Neuron
|
Correct or Wrong
|
1
|
0
|
0
|
Correct
|
2
|
0
|
0
|
Correct
|
3
|
1
|
1
|
Correct
|
4
|
0
|
0
|
Correct
|
5
|
1
|
0
|
Wrong
|
6
|
0
|
0
|
Correct
|
7
|
1
|
1
|
Correct
|
8
|
1
|
1
|
Correct
|
According
to the above results obtained, the neural network provides 7 correct
classifications, while only one classification is wrong. Therefore neural
network provides 88% accuracy level when predicting the Internet Banking
Adoption with a data set obtained from 103 respondents. Whereas the MLR model was able to provide
only 63% accuracy level.
4. Discussion
The
objective of this research was to examine whether artificial neural networks
achieve a superior model fit as compared to multiple linear regression (MLR)
for prediction of Internet banking adoption. This research study shows that
neural network model is a better choice than MLR model for predicting Internet
banking adoption. The MLR approach was able to explain only 63% of the variance
in Internet banking adoption while the Neural Network model was able to
provides 88% accuracy level when predicting the Internet Banking Adoption. The
limitation in explaining the variance to a greater degree may be because of its
inability to capture nonlinear relationships. On the other hand, the neural
network model is able to capture nonlinear relationship between the independent
variables and Internet banking adoption.
The
results obtained from neural network model show that all the determinants (e.g.
service quality, trust, ease of use, usefulness, attitude, age, gender, income
group) are significant predictors of Internet banking adoption in Sri Lanka.
5. Managerial
implications, limitations and future work
In
this study, an attempt was made to extend TAM to understand and predict the
independent variables of Internet banking adoption. Service quality and trust
were incorporated into TAM together with demographic variables, gender, age,
Internet experience, income group and education. The results obtained from
neural network model and MLR model were compared, and the former fared better
among the two methods. The neural network model was able to capture relative
importance of all independent variables, including service quality, trust, PU,
PEOU, attitude and the five aforementioned demographic variables; whereas MLR
was not able to capture the significance of three independent variables.
The
study has important managerial implications. First, this study can be helpful
for Internet banking professionals to enhance adoption of Internet banking in
Sri Lanka. Second, the application of neural network models enables prediction
of adoption instead of merely establishing a causal relationship. The use of
two-staged neural network regression approach is relatively new and can be an
important tool for researchers intending to study the adoption of Internet
banking study in future. Third, this study is conducted in a developing country
like Sri Lanka, where a number of service providers have adopted Internet
banking, expecting the user penetration to grow rapidly over the next decade.
The user adoption of Internet banking in Sri Lanka is in a nascent stage, and
this study can prove to be useful to Sri Lankan banks seeking to expand the
Internet banking customer base. The findings of this study can assist Internet
banking service providers in developing appropriate strategies for attracting
more users. Lastly, demographic details of users also play an important role in
the adoption of Internet banking. Important variables include age, gender, and
income group. The results pertaining to demographic variables are useful in
determining personalized promotional strategies based on varying preferences of
different demographic groups.
5.1 Limitations and future work
This study has mainly three
limitations. First, the sample size is not large enough to generalize results
to the entire country. Therefore, further studies may have to be conducted for
generalizing results for the entire country. Finally, this study focused merely
on adoption of Internet banking and can be expanded further to understand
continued usage trends by banking users.
Therefore, further studies
may have to be conducted for generalizing results for the entire country.
Finally, this study focused merely on adoption of Internet banking and can be
expanded further to understand continued usage trends by banking users.
Appendix
Indicators
of dimensions
|
PEOU
|
(P1) I
found it easy to learn commonly used tasks in Internet banking
(P2) I
find the Internet banking Web site clear and easy to interact
(P3) I
expect to become skilled at Internet banking
PU
|
(PU1) I am
able to save considerable time and effort using Internet banking
(PU2) I
am able to manage my banking tasks more effectively using Internet banking
(PU3)
Overall, I find Internet banking very useful
Attitude
|
(AT1)
I am likely to recommend the use of Internet banking to my relatives and
colleagues
(AT2)
Internet banking fits with my lifestyle
(AT3)
Overall my attitude toward Internet banking is positive
(AT4) The
Internet banking transactions are error free
Service
quality
|
(S1)
The Internet banking site is managed well and up to date
(S2)
The Internet banking site is supported by prompt 24 × 7 customer service
(S3)
Overall, the Internet banking site satisfies my expectations
(S4)
The Internet banking site is trustworthy
Trust
|
(T1)
I find it easy to complete my tasks using Internet banking
(T2)
The Internet banking site keeps customer’s best interest in mind
(T3)
The Internet banking site is predictable
Adoption
|
(A1) I
plan to use Internet banking in future
(A2) I
recommend use of Internet banking among peers and relatives
(A3) I
think Internet banking is a great idea
