Journal of Banking & Finance 30 (2006) 1713–1725

www.elsevier.com/locate/jbf

The effect of heterogeneous risk on the early

adoption of Internet banking technologies
a,* b
Keldon Bauer , Scott E. Hein
a
Department of Finance, Insurance and Law, Illinois State University, 417 College of Business Building,
Campus Box 5480, Normal, IL 61790-5480, USA
b
Department of Finance, Rawls College of Business Administration, Texas Tech University, Lubbock,
TX 79409-2101, USA

Available online 2 November 2005

Abstract

Financial service providers have increasingly offered customers new remote access to such ser-
vices, with Internet banking being the latest example. While Internet banking has been available
for years, the early adoption by customers of this technology was disappointing to most. This paper
examines the demand for remote access to banking accounts by consumers and finds that when the
technology is new, the traditional risk return models including variables allowing for heterogeneous
risk add power in modeling the adoption decision. Perceived risks in Internet banking are seen to be
responsible for some of the hesitation to adopt. Ironically, older consumers are found to be less likely
to adopt Internet banking regardless of their risk tolerances. However, younger consumers are found
to be early adopters only when they have relatively high levels of risk tolerance.
 2005 Elsevier B.V. All rights reserved.

JEL classification: D8; D12; D81; G2; G21

Keywords: Internet banking; Retail banking; Consumer choice; Perceived risk; Heterogeneous risk

1. Introduction

Financial service customers are getting further away from the providers of those ser-
vices. Petersen and Rajan (2002) found that small business lending, which has been the

*
Corresponding author. Tel.: +1 309 438 7991; fax: +1 309 438 5510.
E-mail address: kjbaue2@ilstu.edu (K. Bauer).

0378-4266/$ - see front matter  2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jbankfin.2005.09.004
1714 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

specialty of local relationship lenders, has been moving further from their customers over
time, due in part to remote banking technologies. Remote access technologies in financial
services have long been used to increase the geographic market of the financial service pro-
vider. Hannan and McDowell (1990) showed that banks adopted ATMs in order to
expand their market share or protect their market against those banks that offered ATMs.
Similar arguments have been made about other forms of remote banking technologies by
Bouckaert and Degryse (1995) and Degryse (1996).
New technologies may also reduce costs. Daniel et al. (1973) were one of the first to
show evidence that the production function of banks was improved through technology
adoption. Hunter and Timme (1986, 1991) found that technology did improve industry
wide scale economies, which explains why Humphrey and Pulley (1997) found that larger
banks were more likely to replace people with technology than were smaller ones when
forced to restructure their costs. Wheelock and Wilson (1999) investigated both productiv-
ity and efficiency at banks and found declining technical efficiency. The primary reason for
this inefficiency of banks was the uneven application of technology among all banks; the
most efficient banks were those that adopted new technologies.
However there are disadvantages of new remote access technologies. Pennathur (2001)
found that Internet banking increased operational, legal and reputation risks, and
increased competition. But, financial service consolidation is increasingly dependent on
consumers accepting and adopting remote access technologies.
This paper examines the role perceived risk plays in the early adoption decision of new
remote access technologies. We focus on the adoption decision from the perspective of the
retail banking customer, as opposed to the commercial bank perspective as is traditionally
considered (see Furst et al., 2000). We ask what would cause a retail customer to adopt
Internet banking soon after its introduction. We find that when a technology is new, risk
is likely to be heterogeneous and this heterogeneity leads to divergent optimal adoption
strategies. In Section 2, we present the theoretical model that, shows how consumers max-
imize their utility in selecting new technologies. In Section 3, we develop a methodology of
applying this model to consumer adoption of Internet banking. In Section 4, we present
our empirical results. In Section 5, we summarize and conclude.

2. Theoretical framework

The main object of this paper is to better understand the adoption decision from the con-
sumerÕs standpoint. This section uses micro-economic theory of consumer utility maximi-
zation to model how consumers decide whether or not to adopt a new remote access
technology. We will assume that consumers derive utility from their traditional bank
accounts according to the utility function f(x). Since traditional accounts have been around
for so long, consumers are relatively certain of the utility they will derive from them.
As remote banking technologies are added to the account delivery mix, the utility
derived can increase or decrease according to the marginal function h(x). Consumers
are not certain of the outcome of their using a remote access account. The source of the
uncertainty is twofold. First, the technology itself poses risks of unauthorized access.
And second, the consumer may be uncertain about their ability to use it. The nature of
the uncertainty itself may differ depending on the remote access service being used. We
assume now k distinct uncertain possibilities. Some of these outcomes only affect one ser-
vice. The first outcome (j = 1) is what happens when all technologies work correctly. Fur-
 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725 1715

ther we assume that each consumer assigns a subjective probability (pj) to each possible
outcome. These probabilities are based on publicly available information, and individual
consumersÕ experience with similar technologies. Over time, these subjective probabilities
would approach a homogeneous set of probabilities as more information is publicly dis-
closed, and as consumers have similar experiences with the technologies. For notational
convenience, it is further assumed that all outcomes other than the first outcome (j > 1)
result in lower utility, and that the disutility grows with the amount of money that is
deposited in the remote accessible account (i.e., h0ij < 0, "j > 1). Since the model focuses
attention on expected utility, either a horrible outcome and/or a high probability of such
adverse outcomes (i.e., personal information theft) could cause the expected utility to differ
greatly from the utility under certainty. The expected utility of a bank account with n dif-
ferent remote banking technologies could be represented as follows:
Xk X
n
U ðxÞ ¼ f ðxÞ þ pj di hij ðxÞ; ð1Þ
j¼1 i

where di is a dummy variable for the ith remote banking method.

The budget constraint will contain information about the costs of the account. In addi-
tion to the base costs, those considering a remote access account may be charged an addi-
tional fee for access to, or use of, these remote facilities. The consumer may have to buy
special equipment or subscribe to a service (for instance Internet access, where a personal
computer would also be required). Depending on the technology, the consumer also may
have to incur a human capital development (training) cost. To some customers, these
added costs are sunk, since they may already have the human capital necessary and any
hardware, etc. To them, therefore, the marginal cost incurred in adopting remote access
over the traditional account would be low.
The base costs of the traditional account can be written as /(x). Again we use a mod-
ular notation to describe the costs associated with the remote cost components as cn(x).
These cost functions can add or subtract costs from the base cost function. Some examples
of cost savings through remote banking include time savings for online banking over tra-
ditional accounts or lower transportation costs. Like the utility function, these modules
are turned on and off by the same zero/one indicator variables, dn. Therefore, the con-
sumer problem would be rewritten as follows:
Xk Xn
Maximize: U ðxÞ ¼ f ðxÞ þ pj di hij ðxÞ
j¼1 i¼1
ð2Þ
X
n
subject to: /ðxÞ þ di ci ðxÞ ¼ m.
i¼1

Our central thesis is that consumers perceive risk in new remote access channels. While
consumers face the risk of loss of purchase price of any online purchase, this financial
exposure may be limited by current credit card laws and/or by FDIC insurance. What
is not covered by any of these devices, however, is the invasion of very private information.
Since financial intermediaries serve as information processors, if security is compromised,
thieves could gain enough information to steal not just deposits, but also customersÕ iden-
tities. There are likely to be a number of key risk dimensions to the adoption of a new tech-
nology such as Internet banking. Perceived risk is further compounded by the consumerÕs
assessment of their own abilities to use the new technology.
1716 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

Since all piÕs sum to one, U(x) can be rewritten as

Xn
U ðxÞ ¼ f ðxÞ þ di hi1 ðxÞ þ p; ð3Þ
i¼1

where
! !
X
k X
n X
n
p¼ pj di hij ðxÞ  ð1  p 1 Þ di hi1 ðxÞ .
j¼2 i¼1 i¼1

The formulation in (3) allows us to interpret the p term. The p term (actually p) represents
the risk premium, using the logic in Pratt (1964). The term after the first summation sym-
bol in (3) is the added utility under certainty that is added by remote access under cer-
tainty. As long as the added utility of remote access under certainty is greater than the
risk premium, utility will be higher for a remote access configuration than for a traditional
configuration, and the consumer would demand remote access to their accounts.1
In our model, the risk premium is made up of the utility for adverse outcomes, h(x), and
the subjective probability that those adverse outcomes occurred, pi. Our notation is differ-
ent because our risk premium is not an approximation. Note that in our model, the shape
of the utility function, PrattÕs risk aversion, interacts with the subjective probability to
yield the risk premium.
It is important to note how the risk premium affects the decision to adopt remote access
technologies. If the consumer assesses a remote access configuration Pto provide
 more util-
ity than the traditional account, they will now assess whether Max ni¼1 di hij > p, and if it
is, they will adopt some sort of remote access to their accounts. Because p is in the decision
rule, the risk premium is an important consideration in determining whether consumers
adopt remote banking.
In short, adoption of remote banking, like Internet banking, will depend on the added
utility of the consumerÕs perceived added utility, the cost of that technology and the risk
premium. The risk premium is a composite construct, made up of consumerÕs risk aversion
and the subjective probability for different adverse outcomes. Over time, the subjective
probability set among all consumers will likely tend to converge. But, at the early stage
of introduction of a new technology, these subjective probabilities are likely to diverge
widely.

3. Data and methodology

To examine the influence of perceived risk of a new technology with an old technology,
we examine two different remote access methods, phone-banking and Internet banking.
We examine what influences the selection of one of these new remote access methods,
using data from the 1998 Survey of Consumer Finances (SCF) sponsored by the Board
of Governors of the Federal Reserve System. Internet banking was a new remote access
technology in 1998. In 1998, 43% of all households used phone-banking of some kind,

1
Pratt showed that the distinguishing feature among risk premiums was the individualÕs utility structure. He
approximated the risk premium using a second-order Taylor series expansion. The only difference between the
risk premiums of two people was a measure he called risk aversion, and was a measure of the shape of the
individualsÕ utility functions.
 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725 1717

while only 6% of all households participated in Internet banking. Using the model pre-
sented in this section, the potential factors critical to Internet adoption can be compared
with that of phone-banking to determine what characteristics are most critical in making
the adoption decision.
We categorized the household as using Internet banking if the survey respondent indi-
cated an interaction with a commercial bank, savings and loan or credit union via com-
puter, Internet or online service. Phone-banking utilization was considered in two
different forms. One possible response was phone (voice), and the other was phone (touch-
tone). These responses will be considered separately.
The theory described in the last section implies that Internet banking adoption is a
function of utility added by adoption of remote banking services, a budget constraint
(the added cost of using the new technology), and the risk premium. It was assumed that
those gaining the most utility from Internet banking would be those who had the most
complex banking relationship. To measure complexity, we used number of savings
accounts, and number of checking accounts.2 A second proxy for utility is the amount
of money kept in those accounts.
Since computer literacy is an important portion of the budget constraint for Internet
banking, the next set of variables addresses these costs. To proxy for the budget constraint
we used income, education and age variables. The more one earns, the more likely a com-
puter has been purchased, reducing the marginal cost of Internet banking. The younger
one is the more likely one has computer training. The more educated the customer is
the more likely they had a computer background.3 The survey did not explicitly address
the cost of the online service, the e-banking service, or any phone-banking charges.
Thus, the models estimated below ignore these explicit costs (or assumes them to be zero).
Since in our opinion, these explicit costs were initially set very low by banks to encourage
Internet banking adoption, we believe that this assumption is not too costly.
Familiarization with Internet remote access can be determined by seeing whether the
consumer accesses any other financial services via the Internet. We use the dummy variable
g to indicate if other financial service is accessed via the Internet (labeled Familiarity in the
tabular results presented below). The question of risk aversion was asked in the survey as
respondents were asked ‘‘Which of the statements on this page comes closest to the
amount of financial risk that you and your (spouse/partner) are willing to take when
you save or make investments?’’ Then each subject was shown a card with the following
options:

1. Take substantial financial risks expecting to earn substantial returns.

2. Take above average financial risks expecting to earn above average returns.
3. Take average financial risks expecting to earn average returns.
4. Not willing to take any financial risks.

The response was coded with the above number of the response given, so the greater the
risk aversion, the larger the coded response.

2
We derive the number of savings accounts as the sum of savings account balances greater than zero.
3
We used dummy variables based on the highest level of education attained by the head of household to proxy
for education with the following categories: less than high school, high school graduate, some college, bachelorÕs
degree, masterÕs degree, and doctorate degree.
1718 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

3.1. The base remote access banking model

The base model used to estimate early adoption of Internet banking is the following:
eA
p¼ ;
1 þ eA
X
10 ð4Þ
A ¼ b0 þ b1 X 1 þ b2 X 2 þ b3 X 3 þ b4 X 4 þ b5 X 5 þ bi d i5 þ b11 g þ b12 X 6 ;
i¼6

where p = Bernoulli probability of using Internet banking, X1 = number of savings ac-

counts held by household members, X2 = number of checking accounts held by household
members, X3 = amount in savings and checking accounts, X4 = reported age of the ‘‘head
of household’’, X5 = natural log of reported income, di = dummy variables for level of
education of the ‘‘head of household.’’ Education levels ‘‘No High School’’ through ‘‘Doc-
torate’’–Bachelors Degree is omitted (acts as base education), g = 1 if subject uses another
form of Internet financial services, 0 otherwise, X6 = response to the risk aversion question
in the survey (1, 2, 3, or 4 – increasing levels of risk aversion).
We also add interaction terms to the base model which allow risk aversion to interact
with the subjective probability variables. We use the risk aversion variable to proxy for
utility function shape. As mentioned before, the variability in subjective probabilities is
expected to be a function of age, education, familiarity with the remote access channel,
and possibly income (since the more one earns, the more likely it is that one has purchased
a computer). We therefore separately interact these variables with the consumerÕs measure
of risk aversion.4

3.2. Joint Internet banking/phone-banking model

To examine how customers decide the remote banking configuration, including phone
banking, we expand the framework to a multinomial model. The following table shows all
possible remote banking configurations:

Category 1 (P11) Category 2 (P12)

Both phone and Internet banking Internet banking, no phone
Category 3 (P21) Category 4 (P22)
Phone, no Internet banking Neither phone nor Internet banking

As in the logistic regressions presented thus far, multinomial logistic regression will esti-
mate the effect of variables on the probability of a consumer falling into each category.
The following model is used5:

4
It should be noted that we tried all models using a set of indicator variables for the risk aversion variable, and
using the numbers as described above. Using a full-reduced framework, there was no loss in prediction power
using just the number as input (treating the risk aversion variable as a continuous variable). Therefore, for
parsimony, we use the reported measure for all models in the analysis section.
5
This notation is from Eq. (11.129) in Kmenta (1986). Elements of Econometrics, Second Edition, Macmillan
Publishing, New York, p. 558.
 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725 1719

eA
pc ¼ P4 ;
A
c¼1 e
ð5Þ
X
10
A ¼ b0 þ b1 X 1 þ b2 X 2 þ b3 X 3 þ b4 X 4 þ b5 X 5 þ bi d i5 þ b11 g þ b12 X 6 ;
i¼6

where pc = Bernoulli probability of customers falling in category c. All other variables are
the same as in Eq. (4). At this point, interactive variables can be added as described above.
To find the individual pc, only c  1 equations (three in this case) are estimated. Just as
in Eq. (4), the coefficients measure how the variable proxies for propensity to fall into the
category studied (phone or Internet banking) over the alternative (of not using those).

3.3. Marginal propensity to adopt Internet banking

To find the maximum likelihood estimator of the probability of Internet adoption if the
customer has already adopted phone banking, we turn to the table on the previous page
and note that the conditional probability of Internet banking given phone-bank usage is
P 11
P ðI j F Þ ¼ ; ð6Þ
P 11 þ P 21
where I = uses Internet banking, F = uses phone-banking, P11 = probability of using
Internet banking and phone, P21 = probability of using phone but not Internet banking.
P11 and P21 are estimated in the multinomial logit described in the last section using
maximum likelihood. Therefore, this estimate is a maximum likelihood estimator of the
conditional probability. Note that P11 and P21 can be expressed as equations from the
multinomial logistic regression as follows:
eL11
P 11 ¼ ;
1 þ eL11 þ eL12 þ eL21
eL21
P 21 ¼ ;
1 þ eL11 þ eL12 þ eL21
where Lij = linear combination described in the last section to estimate the joint
probabilities.
Combining those definitions into Eq. (6), yields the following relationship:
eL11
P ðI j AÞ ¼ . ð7Þ
eL11 þ eL21
Further simplification of the terms yields the following:
0 1
    eL11
P 11 P 11 =ðP 11 þ P 21 Þ B1 þ eL11 þ eL12 þ eL21 C
logit ¼ ln ln B
@
C ¼ L11  L21 .
A
P 11 þ P 21 P 21 =ðP 11 þ P 21 Þ eL21
1 þ eL11 þ eL12 þ eL21
ð8Þ
Therefore, the maximum likelihood estimate of coefficients for the conditional probability
of Internet banking given phone-bank usage is just the difference between the coefficients
1720 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

of the P11 and P21. Since these new coefficients are estimated by merely subtracting one
estimate from another, the variance or each estimate can be estimated from the covariance
matrix:
s2ci ¼ s21i þ s22i  2s12i ;
where s2ci = variance estimate of conditional coefficient i, s21i = variance estimate of P11
coefficient i, s22i = variance estimate of P21 coefficient i, s12i = covariance estimate of coef-
ficient i.
These new coefficients no longer estimate the influence of those variables on the mar-
ginal probability, but on the conditional probability of adopting Internet banking, given
the customer is a phone-bank user. The same procedure can be used to estimate the con-
ditional probability of phone-banking given Internet banking. The estimate is simply
L11  L12, with a similar derivation. Using this procedure, we should be able to determine
the sensitivity of consumers to perceived risks in moving back and forth, or comparing the
relative effect of perceived risk to the two channels of banking services.

4. Empirical results

The first column of Table 1 shows the coefficient estimates for the base Internet banking
adoption model. The figures in parentheses are the Wald statistic. All significant coefficient
estimates have the hypothesized sign, except the dummy variable of masterÕs degree. The
base model shows that the larger the number of checking accounts a consumer has the more
likely they are to adopt Internet banking. The number of savings accounts does not appear
to have any significant effect on the adoption rate. The age variable coefficient is significantly
negative, indicating that, ceteris paribus, older consumers are less likely to adopt Internet
banking. Income also appears to have a significant impact on the adoption decision. As sug-
gested by the theory, the higher a consumerÕs income, everything else equal, the more likely
the consumer is to adopt Internet banking. Also, our proxy of Internet familiarity has a sig-
nificant influence on the adoption decision. Finally, the more risk averse a consumer consid-
ers his or herself to be the less likely that individual is to adopt Internet banking.
Table 1, columns two through five, separately adds a different interactive subjective
probability variable, by interacting risk aversion and four different variables. Panel B
shows the overall statistics of fit. Note that all of these models are statistically significant.
Panel C shows whether any model improves upon the base model (whether subjective
probability estimates are in fact heterogeneous). It appears that the age variable is the best
predictor of subjective heterogeneous probability. The other models do not appear to pro-
vide an improvement over the base model.
To better interpret the interaction between risk aversion and age based on the model
estimates, we graph the expected probability of adoption for an average-age consumer,
allowing risk aversion to vary. The average of all variables was used (with the median
amount of education). Two functional relationships were then graphed: (1) the upper
age quartile (61 years old) and, (2) the lower age quartile (36 years old). All other variables
were plugged at their mean/median values. Fig. 1 clearly shows that risk aversion is an
important consideration in adoption of Internet banking for younger bank customers.
The theory suggested that the more expensive is human capital, or the higher the fixed
expense of the adoption (proxied in this instance by age), the less likely that risk aversion
is an important consideration. This seems to hold true as shown by the almost perfectly
Table 1
Logistic regression of Internet banking (1998 SCF)
Base Age Ln(income) Familiarity Education
Panel A—Coefficient estimates
Constant 6.9267 (5.6282)*** 5.3565 (3.8668)*** 5.6348 (1.6129) 7.0839 (5.6824)*** 7.4052 (5.6766)***

K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

#Checks 0.2454 (2.7609)*** 0.2555 (2.8344)*** 0.2431 (2.7296)*** 0.2490 (2.7909)*** 0.2554 (2.8563)***
#Savings 0.0480 (0.6421) 0.0441 (0.5891) 0.0482 (0.6460) 0.0509 (0.6783) 0.0508 (0.6781)
Ln(liquid) 0.0052 (0.0731) 0.0155 (0.2158) 0.0064 (0.0900) 0.0016 (0.0220) 0.0005 (0.0064)
Age 0.0365 (4.7504)*** 0.0866 (3.6757)*** 0.0363 (4.7095)*** 0.0364 (4.7373)*** 0.0372 (4.8308)***
Ln(income) 0.5936 (4.8346)*** 0.6385 (5.1045)*** 0.4761 (1.4791)* 0.5941 (4.8333)*** 0.5958 (4.8368)***
NHS 0.9942 (1.8835)** 1.0297 (1.9418)** 0.9857 (1.8686)** 1.0152 (1.9212)** 1.3554 (0.6705)
HS 0.9247 (3.3819)*** 0.9228 (3.3625)*** 0.9251 (3.3856)*** 0.9461 (3.4417)*** 0.8138 (0.9607)
SC 0.4427 (1.9811)** 0.4328 (1.9297)** 0.4461 (1.9961)** 0.4446 (1.9867)** 0.7196 (1.0316)
MA 0.7333 (2.4381) 0.6732 (2.2372) 0.7277 (2.4220) 0.7564 (2.4909) 0.5438 (0.6078)
DR 0.5747 (1.5537) 0.5520 (1.4789) 0.5660 (1.5310) 0.5822 (1.5682) 0.0348 (0.0285)
Familiarity 1.7670 (7.5956)*** 1.7407 (7.4211)*** 1.7760 (7.6070)*** 2.3214 (3.5791)*** 1.7972 (7.6520)***
Risk aversion 0.3653 (3.3459)*** 1.0980 (3.2677)*** 0.8613 (0.6807) 0.3194 (2.6509)*** 0.1909 (0.9990)
Age * risk 0.0180 (2.3042)**
Ln(income) * risk 0.0455 (0.3936)
Familiarity * risk 0.2654 (0.9104)
NHS * risk 0.0751 (0.1255)
HS * risk 0.0572 (0.1908)
SC * risk 0.4664 (1.7300)*
MA * risk 0.0717 (0.1944)
DR * risk 0.2608 (0.5090)
Panel B—Overall statistics of fit
R2 0.2339 0.2381 0.2340 0.2346 0.2369
v2 276.4901 281.7004 276.6470 277.3386 280.2397
p-Value >.0001 >.0001 >.0001 >.0001 >.0001
Panel C—Comparing interactive variables to base model
v2 5.2102 0.1568 0.8485 3.7495
p-Value 0.0225 0.6921 0.3570 0.5860
Note: ***, **, * represent significance at the 1%, 5% and 10% levels, respectively.

1721
1722 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

0.08
0.07
Age 36
0.06
F(x) 0.05
0.04
0.03
Age 61
0.02
0.01
0
0 1 2 3 4 5
Risk Aversion
Fig. 1. Effect of age and risk.

flat relationship between risk aversion and adoption at older ages. This suggests that
efforts to better educate younger bank customers about the risks of Internet banking
may be more beneficial than educating older bank customers.
Table 2 presents coefficient estimates of Internet banking given the customer uses
phone-banking (the columns define the form of phone-banking used). Note that the same
variables are significant for Internet adoption in Table 1 as in Table 2. It would appear
that there are greater risks perceived in moving from phone-banking to Internet banking,
since the risk aversion variable is statistically significant.
Table 3 presents coefficient estimates of phone-banking given the customer uses Inter-
net banking (the columns define the form of phone-banking used). In this case, nearly no
variable is statistically significant. Most important to us in this paper is that customers do
not appear to perceive any additional risk moving from Internet banking to phone-bank-
ing, even though one could argue that the risks are in fact different. These results may be
driven by the small number of observations of Internet banking (only 163).
One interesting result from the conditional logit model is that risk aversion, which is
either downward sloping or flat in the bi-variate and multinomial logit models, may actu-

Table 2
Conditional logit P (Internetjphone banking)
Voice phone Touchtone
Intercept 2.6236 (1.3862) 4.4486 (2.2853)**
#Checks 0.2119 (1.7377)** 0.3068 (2.7340)***
#Savings 0.0839 (0.8091) 0.0893 (0.8997)
Ln(liquid) 0.0462 (0.4475) 0.0547 (0.5317)
Age 0.1338 (3.9126)*** 0.0835 (2.4626)***
Ln(income) 0.5912 (3.3324)*** 0.5987 (3.4214)***
NHS 0.6605 (0.8931) 0.5364 (0.5769)
HS 0.6652 (1.7749)** 1.2772 (2.5485)***
SC 0.7652 (2.2530)** 0.0396 (0.1378)
MA 0.3231 (0.8315) 0.7470 (1.7302)
DR 0.6594 (1.1532) 1.3153 (1.7985)
Familiarity 1.4755 (4.2913)*** 1.6488 (4.6244)***
Risk aversion 1.4985 (3.2038)*** 1.2724 (2.5592)***
Age * risk aversion 0.0312 (2.7922)*** 0.0230 (1.9342)**
Note: ***, **, * represent significance at the 1%, 5% and 10% levels, respectively.
 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725 1723

Table 3
Conditional logit P (phone bankingjInternet)
Voice phone Touchtone
Intercept 2.1626 (0.8801) 0.4908 (0.1959)
#Checks 0.1214 (0.7666) 0.4373 (2.4450)***
#Savings 0.0411 (0.3012) 0.1125 (0.8050)
Ln(liquid) 0.0251 (0.1881) 0.1465 (1.0913)
Age 0.0865 (2.0041)** 0.0177 (0.4043)
Ln(income) 0.0617 (0.2745) 0.1230 (0.5416)
NHS 0.4524 (0.4332) 0.5366 (0.4812)
HS 0.5353 (1.0175) 1.2145 (2.0733)**
SC 0.2718 (0.6452) 0.8620 (1.9304)
MA 0.9188 (1.6665)** 0.2176 (0.4010)
DR 0.2563 (0.3828) 1.5284 (1.9308)
Familiarity 0.3246 (0.8178) 0.5104 (1.2623)
Risk aversion 0.9640 (1.5580)* 0.3417 (0.5373)
Age * risk aversion 0.0236 (1.6195)* 0.0024 (0.1578)
Note: ***, **, * represent significance at the 1%, 5% and 10% levels, respectively.

0.1
0.09
0.08
0.07 Age 35
0.06
F(x)

0.05
0.04
0.03
0.02 Age 58
0.01
0
0 1 2 3 4 5
Risk Aversion
Fig. 2. Effect of age and risk P (Internetjvoice phone banking).

ally be upward sloping in a broader context. In other words, as some consumersÕ risk aver-
sion increases, they may be more likely to adopt Internet banking. Fig. 2 uses the same
methodology as described for Fig. 1, but uses the upper and lower quartiles of ages of
those using voice phone banking. The slightly upward sloping relationship between risk
aversion and adoption for the older-aged consumer suggests that the overall risk may
be perceived to go down upon adoption of the Internet given that one has already adopted
voice phone banking. This suggests that a positive step to get older consumers to adopt a
new technology may be first to encourage the use of an older technology.

5. Conclusions

As technology becomes more ubiquitous in the financial services industry, it is imper-

ative that institutions adopt new technology to improve their cost structure, their efficiency
and enhance their competitive position. Increasingly, financial services consumers are
1724 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725

being asked to adopt these technologies. But, much of the discussion of these new technol-
ogies presumes a ‘‘field of dreams’’ mentality. That is if the institution builds the technol-
ogy the customer is assumed just to come. This paper takes a very different focus and asks
what will drive a consumer to adopt a new technology, by presenting a framework for ana-
lyzing the adoption decision from the consumersÕ standpoint.
The early adoption decisions for Internet banking allow us to examine the implications
of the utility adoption model. First, we find risk perception is not homogeneous and risk
matters in deciding to adopt. Customers having experience with similar technologies dem-
onstrate much less risk aversion, than those with no experience. We also find that not only
does risk aversion matter in the adoption decision, it matters more for younger as opposed
to older decision makers. Second, if it is relatively different from other technologies, a
familiar technology should be offered as a parallel remote access choice until the consumer
is comfortable with the new channel (e.g. phone banking to supplement Internet banking).
Once the consumer becomes comfortable with the new channel, older channels may be
seen as an added risk to consumers, and they could then be eliminated.
This paper has focused on the decision to adopt Internet banking as a new remote deliv-
ery device early in its evolution by using the 1998 Survey of Consumer Finance. We focus
on the early adoption decision because of our conjecture that when technologies are new
expectations about risks are likely to be more heterogeneous and more important. While
our evidence shows that risk aversion is important in influencing the adoption decision, we
do not have evidence that risk aversion is lessening in relative importance. We leave this
question for future research. In addition, we focus attention on the adoption decision and
leave left unanswered the question about Internet banking utilization, which might be
considered as important a matter today, as the adoption decision was a few years ago.
In summary there remains much to be learned about Internet banking in our financial
system.

Acknowledgements

The authors wish to thank Naomi Boyd, Ken Cyree, Tim Koch, participants at the
2003 Southern Finance Association, participants at the 2005 Global Finance Conference,
the editors, Anthony Saunders and Barry Scholnick, and two anonymous referees for
helpful comments on earlier versions of this paper.

References

Bouckaert, J., Degryse, H., 1995. Phonebanking. European Economic Review 39, 229–244.
Daniel, D., Longbrake, W., Murphy, N., 1973. The effect of technology on bank economies of scale for demand
deposits. Journal of Finance 28, 131–146.
Degryse, H., 1996. On the interaction between vertical and horizontal product differentiation: An application to
banking. Journal of Industrial Economics 44, 169–186.
Furst, K., Lang, W., Nolle, D., 2000. Internet banking: development and prospects. Economic and Policy
Analysis Working Paper 2000-9. Office of the Controller of the Currency.
Hannan, T., McDowell, J., 1990. The impact of technology adoption on market structure. Review of Economics
and Statistics 72, 164–168.
Humphrey, D., Pulley, L., 1997. BanksÕ responses to deregulation: Profits, technology, and efficiency. Journal of
Money, Credit, and Banking 29, 73–93.
Hunter, W., Timme, S., 1986. Technical change, organizational form and the structure of bank production.
Journal of Money, Credit and Banking 18, 152–166.
 K. Bauer, S.E. Hein / Journal of Banking & Finance 30 (2006) 1713–1725 1725

Hunter, W., Timme, S., 1991. Technological change in large US, commercial banks. Journal of Business 64, 339–
362.
Kmenta, J., 1986. Elements of Econometrics, Second ed. Macmillan Publishing, New York.
Pennathur, A., 2001. Clicks and bricks: E-risk management for banks in the age of the Internet. Journal of
Banking and Finance 25, 2103–2123.
Petersen, M., Rajan, R., 2002. Does distance still matter? The information revolution in small business lending.
Journal of Finance 57, 2533–2570.
Pratt, J., 1964. Risk aversion in the small and in the large. Econometrica 32, 122–136.
Wheelock, D., Wilson, W., 1999. Technical progress, inefficiency, and productivity change in US banking, 1984–
1993. Journal of Money, Credit and Banking 31, 212–234.