ArticlePDF Available

User Acceptance of Computer Technology: A Comparison of Two Theoretical Models

Authors:
Article

User Acceptance of Computer Technology: A Comparison of Two Theoretical Models

Abstract

Computer systems cannot improve organizational performance if they aren't used. Unfortunately, resistance to end-user systems by managers and professionals is a widespread problem. To better predict, explain, and increase user acceptance, we need to better understand why people accept or reject computers. This research addresses the ability to predict peoples' computer acceptance from a measure of their intentions, and the ability to explain their intentions in terms of their attitudes, subjective norms, perceived usefulness, perceived ease of use, and related variables. In a longitudinal study of 107 users, intentions to use a specific system, measured after a one-hour introduction to the system, were correlated 0.35 with system use 14 weeks later. The intention-usage correlation was 0.63 at the end of this time period. Perceived usefulness strongly influenced peoples' intentions, explaining more than half of the variance in intentions at the end of 14 weeks. Perceived ease of use had a small but significant effect on intentions as well, although this effect subsided over time. Attitudes only partially mediated the effects of these beliefs on intentions. Subjective norms had no effect on intentions. These results suggest the possibility of simple but powerful models of the determinants of user acceptance, with practical value for evaluating systems and guiding managerial interventions aimed at reducing the problem of underutilized computer technology.
User Acceptance of Computer Technology: A Comparison of Two Theoretical Models
Author(s): Fred D. Davis, Richard P. Bagozzi and Paul R. Warshaw
Source:
Management Science,
Vol. 35, No. 8 (Aug., 1989), pp. 982-1003
Published by: INFORMS
Stable URL: http://www.jstor.org/stable/2632151 .
Accessed: 06/02/2014 14:36
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Management Science.
http://www.jstor.org
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
MANAGEMENT
SCIENCE
Vol. 35, No. 8,
August 1989
Primited
in
U.S.A.
USER ACCEPTANCE
OF COMPUTER
TECHNOLOGY:
A
COMPARISON
OF TWO THEORETICAL
MODELS*
FRED
D.
DAVIS,
RICHARD
P. BAGOZZI
AND PAUL
R. WARSHAW
School
of
Business Administration, University of
Michigan,
Ann Arbor,
Michigan
48109-1234
School
Qf
Business
Administration,
University of
Michigan,
Ann Arbor,
Michigan
48109-1234
School of Business Administration,
California Polytechnic
State University,
San
Lids
Obispa,
California
93407
Computer systems
cannot improve
organizational
performance
if
they
aren't used. Unfortu-
nately,
resistance
to
end-user
systems
by managers
and
professionals
is a
widespread
problem.
To better
predict, explain,
and increase user
acceptance,
we need
to better
understand
why people
accept
or
reject computers.
This
research addresses the
ability
to
predict
peoples'
computer
ac-
ceptance
from
a
measure of their
intentions,
and
the
ability
to
explain
their
intentions
in
terms
of their
attitudes, subjective
norms, perceived
usefulness,
perceived
ease
of
use,
and
related variables.
In a
longitudinal
study
of 107
users,
intentions to
use a
specific system,
measured after
a
one-
hour
introduction
to the system,
were
correlated
0.35 with system use
14
weeks
later. The intention-
usage
correlation
was
0.63
at the
end
of this
time
period.
Perceived usefulness
strongly
influenced
peoples'
intentions,
explaining
more than
half
of the
variance
in
intentions
at the
end
of
14
weeks.
Perceived ease
of use had
a
small
but
significant
effect on intentions
as
well, although
this
effect
subsided
over
time. Attitudes only partially
mediated
the effects of
these
beliefs on
intentions.
Subjective
norms
had
no effect on
intentions. These results
suggest
the
possibility
of
simple
but
powerful
models
of the determinants
of user
acceptance,
with
practical
value
for
evaluating systems
and
guiding managerial
interventions
aimed
at
reducing
the
problem
of underutilized
computer
technology.
(INFORMATION
TECHNOLOGY;
USER
ACCEPTANCE;
INTENTION MODELS)
1. Introduction
Organizational
investments
in
computer-based
tools to
support planning,
decision-
making,
and
communication
processes
are
inherently risky.
Unlike clerical
paperwork-
processing
systems,
these "end-user computing" tools
often require managers
and profes-
sionals
to interact
directly
with
hardware and
software.
However,
end-users
are often
unwilling
to use
available
computer
systems
that,
if
used,
would
generate
significant
performance
gains (e.g.,
Alavi and Henderson
1981;
Nickerson
1981,
Swanson
1988).
The raw power of
computer
technology continues
to improve tenfold
each decade
(Peled
1987),
making sophisticated
applications
economically
feasible. As technical
barriers
disappear,
a
pivotal
factor
in
harnessing
this
expanding power
becomes our
ability to
create
applications
that
people
are
willing
to
use.
Identifying
the appropriate
functional
and interface
characteristics to
be included
in
end-user
systems
has
proven
more chal-
lenging and
subtle than
expected (March
1987;
Mitroff and
Mason 1983).
Recognizing
the
difficulty
of
specifying
the
right
system requirements
based on their own
logic and
intuition,
designers
are seeking
methods for evaluating
the acceptability
of systems as
early
as
possible
in
the
design
and
implementation
process (e.g.,
Alavi
1984; Bewley et
al.
1983;
Branscomb
and Thomas
1984;
Gould
and Lewis
1985). Practitioners
and re-
searchers
require
a
better understanding
of
why
people
resist using computers
in
order
to devise
practical
methods
for
evaluating systems,
predicting
how users
will
respond to
them,
and
improving
user
acceptance
by altering
the nature
of
systems
and the
processes
by
which
they
are
implemented.
Understanding why
people
accept or reject computers
has
proven to be
one of the
most
challenging
issues
in
information
systems
(IS) research
(Swanson 1988).
Investi-
gators
have studied the
impact
of
users' internal
beliefs and
attitudes on
their usage
*
Accepted
by Richard
M. Burton;
received November
10, 1987. This paper
has been with
the authors 4
months for
2
revisions.
982
0025-1
909/89/3508/0982$01.25
Copyright
?
1989, The Institute of Management
Sciences
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE OF
COMPUTER
TECHNOLOGY
983
behavior
(DeSanctis
1983; Fuerst and
Cheney 1982;
Ginzberg 1981; Ives, Olson
and
Baroudi 1983; Lucas
1975; Robey 1979;
Schultz and Slevin
1975;
Srinivasan 1985;
Swanson
1974, 1987), and how
these internal
beliefs and attitudes
are,
in
turn,
influenced
by
various
external
factors, including: the
system's technical
design characteristics
(Ben-
basat
and
Dexter
1986; Benbasat, Dexter
and Todd 1986;
Dickson, DeSanctis
and
McBride
1986; Gould,
Conti
and
Hovanyecz
1983; Malone
1981); user
involvement
in
system
development
(Baroudi, Olson
and
Ives 1986; Franz
and Robey 1986); the
type
of
system development
process used (e.g.,
Alavi 1984; King and
Rodriguez 1981);
the
nature of the
implementation process
(Ginzberg 1978;
Vertinsky, Barth and
Mitchell
1975; Zand and
Sorensen 1975); and
cognitive style (Huber
1983). In general,
however,
these
research findings have
been mixed
and inconclusive.
In
part, this may be
due to
the wide
array
of
different
belief, attitude,
and satisfaction
measures which have
been
employed, often without
adequate
theoretical or
psychometric justification.
Research
progress may be
stimulated by the
establishment of an
integrating paradigm to
guide
theory development and to
provide
a common
frame of reference within which
to
integrate
various research
streams.
Information systems
(IS) investigators
have suggested
intention models from
social
psychology
as a
potential
theoretical
foundation for research on
the determinants of user
behavior
(Swanson
1982;
Christie 1981) .
Fishbein and
Ajzen's
(1975) (Ajzen
and Fish-
bein
1980) theory
of
reasoned action
(TRA) is an especially
well-researched
intention
model that has proven
successful
in
predicting and explaining
behavior across a
wide
variety of domains.
TRA is very general,
"designed to explain
virtually any human
be-
havior" (Ajzen and
Fishbein 1980, p. 4),
and should therefore be
appropriate for
studying
the
determinants of
computer usage
behavior as a
special case.
Davis
(1986) introduced
an adaptation
of TRA, the
technology acceptance
model
(TAM), which is
specifically meant to
explain computer usage
behavior. TAM uses
TRA
as
a
theoretical basis for
specifying
the
causal
linkages between two
key
beliefs:
perceived
usefulness and perceived
ease
of
use,
and
users'
attitudes, intentions and
actual
computer
adoption
behavior.
TAM
is
considerably
less
general
than
TRA, designed
to
apply
only
to
computer usage
behavior,
but because it
incorporates findings accumulated from
over
a
decade of IS
research, it
may
be
especially
well-suited
for
modeling computer acceptance.
In
the present research we
empirically
examine the ability of
TRA
and
TAM
to
predict
and
explain
user
acceptance
and
rejection
of
computer-based
technology.
We are
par-
ticularly
interested
in
how
well we can
predict
and
explain
future user behavior from
simple
measures taken
after
a
very
brief
period
of interaction with a
system.
This scenario
characterizes the
type
of
evaluations made
in
practice
after
pre-purchase
trial
usage
or
interaction with a
prototype system under
development (e.g.,
Alavi
1984).
After
presenting
the
major
characteristics of
the two
models,
we discuss a
longitudinal
study
of 107
MBA
students which
provides
empirical
data for
assessing
how well the models
predict
and
explain
voluntary usage
of a word
processing system.
We then
address
the
prospects
for
synthesizing
elements
of
the two models
in
order
to arrive at
a more
complete
view
of
the
determinants
of
user
acceptance.
2.
Theory
of
Reasoned Action
(TRA)
TRA
is a widely
studied model from
social psychology which
is concerned with
the
determinants of
consciously
intended behaviors
(Ajzen
and
Fishbein
1980;
Fishbein and
Ajzen
1975).
According to TRA, a person's
performance
of a
specified behavior is
de-
termined
by
his or her behavioral
intention
(BI)
to
perform
the
behavior,
and
BI
is
jointly
determined
by
the
person's
attitude
(A)
and
subjective
norm
(SN)
concerning
the behavior
in
question
(Figure 1), with relative
weights
typically estimated
by regression:
BI =A +SN.
(1)
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
984
FRED D. DAVIS, RICHARD P. BAGOZZI AND PAUL R. WARSHAW
Beliefs and Attitude
Evaluations
o
Toward
(? biei)
Behavior (A)
Behavioral
Actual
Intention
Behavior
(BI)
Normative Beliefs Subjective
and Motivation to Norm
comply (?
nbi
mc
i) (SN)
FIGURE 1.
Theory
of Reasoned
Action
(TRA).
BI
is a measure of
the
strength of
one's intention to perform a specified behavior (e.g.,
Fishbein and
Ajzen
1975, p. 288).
A
is
defined as an
individual's positive
or
negative
feelings (evaluative affect)
about
performing
the
target
behavior
(e.g.,
Fishbein and
Ajzen
1975, p. 216). Subjective norm refers
to "the
person's
perception that most people who
are
important
to
him
think
he should
or should
not
perform
the behavior
in
question"
(Fishbein and Ajzen 1975, p. 302).
According to TRA, a person's attitude toward
a behavior is determined by his or her
salient
beli4fs
(bi)
about
consequences
of
performing
the
behavior
multiplied by
the
evaluation
(ei)
of
those
consequences:
A=
biei.
(2)
Beliefs
(bi)
are defined as the individual's
subjective probability that performing the
target behavior
will
result
in
consequence
i. The evaluation
term
(ei)
refers
to
"an
implicit
evaluative
response"-'
to
the
consequence (Fishbein
and
Ajzen 1975, p. 29). Equation
(2) represents
an
information-processing
view of attitude
formation and
change
which
posits
that external
stimuli
influence attitudes
only indirectly through changes
in
the
person's
belief structure
(Ajzen
and
Fishbein
1980, pp. 82-86),
TRA
theorizes
that
an
individual's
subjective
norm
(SN)
is determined
by
a multi-
plicative function
of
his
or
her
normative beliefs
(nbi),
i.e., perceived expectations
of
specific
referent individuals or
groups,
and
his
or her
motivation to
comply
(mci)
with
these
expectations (Fishbein
and
Ajzen 1975, p. 302):
SN
E
nbimci.
(3)
TRA
is
a
general model, and, as such, it does
not specify the beliefs that are operative
for a
particular
behavior. Researchers
using
TRA
must first
identify the beliefs
that
are
salient for
subjects regarding
the
behavior
under
investigation. Fishbein
and
Ajzen (1975,
p. 218)
and
Ajzen
and
Fishbein
(1980, p. 68)
suggest eliciting
five to nine salient
beliefs
using
free
response
interviews with
representative
members of the
subject population.
They
recommend
using
"modal" salient beliefs for
the
population, obtained by taking
the beliefs most
frequently
elicited
from a
representative
sample
of the
population.
A
particularly helpful aspect
of
TRA
from an IS
perspective
is its assertion that
any
other factors that influence behavior
do
so
only
indirectly by influencing A, SN,
or their
relative
weights. Thus,
variables such
as
system
design characteristics,
user
characteristics
(including cognitive style and other personality
variables), task characteristics,
nature of
the
development
or
implementation process, political
influences, organizational
structure
and so
on would fall into this
category,
which Fishbein
and
Ajzen (Ajzen
and Fishbein
1975)
refer to as "external variables."
This
implies
that TRA
mediates
the
impact
of
uncontrollable environmental
variables and controllable
interventions
on user behavior.
If
so,
then
TRA
captures
the internal
psychological
variables
through
which numerous
external variables studied in IS research achieve their influence on user acceptance, and
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE
OF
COMPUTER
TECHNOLOGY
985
may
provide a common
frame of reference within
which to integrate
various disparate
lines
of inquiry.
A
substantial body of
empirical data in support
of TRA has
accumulated (Ajzen and
Fishbein 1980; Fishbein
and Ajzen 1975; Ryan
and Bonfield 1975;
Sheppard, Hartwick
and
Warshaw
in
press). TRA
has been widely used
in applied research
settings spanning
a
variety of subject areas, while
at the same time
stimulating a great deal of
theoretical
research aimed at
understanding the theory's
limitations, testing key
assumptions
and
analyzing various refinements
and extensions
(Bagozzi 1981, 1982,
1984; Saltzer 1981;
Warshaw 1980a, b;
Warshaw and Davis 1984,
1985, 1986;
Warshaw,
Sheppard and
Hartwick in press).
3.
Technology Acceptance
Model (TAM)
TAM, introduced by
Davis (1986), is an
adaptation of TRA
specifically tailored for
modeling user acceptance
of information
systems. The goal of TAM is
to provide an
explanation
of the
determinants of computer
acceptance that is general,
capable of ex-
plaining user behavior
across a broad range of
end-user computing
technologies and user
populations,
while at the same
time being both
parsimonious and
theoretically justified.
Ideally one would like a
model that is helpful not
only for prediction but
also for expla-
nation, so that researchers
and practitioners can
identify why a
particular system may
be
unacceptable, and
pursue appropriate corrective
steps.
A
key purpose
of TAM, there-
fore, is to
provide
a basis
for tracing the impact
of external factors on
internal beliefs,
attitudes,
and intentions. TAM
was formulated
in
an
attempt to achieve
these goals by
identifying
a
small number of
fundamental
variables suggested by previous
research
dealing
with
the cognitive and
affective
determinants of computer
acceptance,
and
using
TRA
as a
theoretical
backdrop for modeling the
theoretical relationships
among these
variables. Several
adaptations to the basic TRA
approach were made,
supported by avail-
able
theory
and
evidence, based
on
these
goals for
TAM.
TAM
posits that two particular
beliefs, perceived
use/i
ulness
and
perceived
ease
of iuse,
are of
primary
relevance for
computer acceptance
behaviors
(Figure 2).
Perceived
use-
fulness
(U)
is defined as
the
prospective user's subjective
probability that
using
a
specific
application
system
will
increase his or her
job performance
within
an
organizational
context. Perceived ease of use
(EOU)
refers to
the
degree
to which the
prospective
user
expects
the
target system
to be free of
effort. As discussed further
below,
several studies
have
found
variables similar
to these to be linked to attitudes and
usage.
In
addition,
factor
analyses suggest that U and EOU are
statistically distinct
dimensions
(Hauser
and
Shugan
1980;
Larcker
and
Lessig 1980;
Swanson
1987).
Similar
to
TRA,
TAM
postulates
that
computer
usage
is determined
by
BI,
but
differs
in
that BI
is
viewed
as
being jointly determined
by
the
person's
attitude toward
using
the
system (A)
and
perceived
usefulness
(U),
with relative
weights
estimated
by regression:
BI=
A +
U.
(4)
Perceived\
/1
(U)
\
/ j
\ ~~~~~Attitude Behavioral
Actual
External
Toward
Intention
to
System
Variables
\
/
Using
(A)
Use
(BI)
Use
\
Perceived
/
Ease
of
Use
(E)
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
986
FRED
D. DAVIS, RICHARD P. BAGOZZI
AND PAUL R.
WARSHAW
The
A-BI
relationship represented
in TAM
implies
that,
all
else being equal, people
form
intentions to perform
behaviors
toward which
they
have
positive
affect. The
A-BI
relationship is
fundamental to
TRA
and to related
models presented by Triandis (1977)
and
Bagozzi (1981)
.
Although
the direct effect of a
belief
(such
as
U)
on
BI
runs
counter
to
TRA,
alternative intention
models
provide
theoretical
justification
and
empirical
ev-
idence of direct
belief-intention
links
(Bagozzi 1982;
Triandis 1977; Brinberg 1979).
The U-BI
relationship
in
equation (4)
is based on the idea
that,
within
organizational
settings, people
form intentions toward
behaviors
they
believe
will
increase their
job
performance, over and above
whatever
positive
or
negative
feelings may
be evoked toward
the
behavior per
se.
This
is
because
enhanced
performance
is
instrumental
to
achieving
various rewards that
are extrinsic to the content of the
work itself, such as pay increases
and
promotions
(e.g.,
Vroom
1964).
Intentions toward such means-end behaviors
are
theorized to
be based
largely
on
cognitive
decision rules to
improve performance,
without
each time
requiring
a
reappraisal
of
how
improved performance
contributes to
purposes
and
goals higher
in
one's
goal hierarchy,
and therefore without
necessarily activating
the
positive affect associated
with
performance-contingent
rewards
(Bagozzi 1982;
Vallacher
and
Wegner 1985).
If
affect is not
fully
activated when
deciding
whether to
use
a
particular
system, one's attitude would
not
be
expected
to
completely capture
the
impact
of
per-
formance considerations on one's intention.
Hence,
the
U-BI
relationship
in TAM
rep-
resents the resulting
direct effect, hypothesizing that
people
form intentions toward
using
computer systems
based
largely
on
a
cognitive appraisal
of how it
will
improve
their
performance.
TAM
does not
include TRA's subjective norm (SN) as a determinant of
BI.
As
Fishbein
and
Ajzen
acknowledge (1975, p. 304),
this is one of least understood
aspects
of TRA.
It is difficult to
disentangle direct effects of SN on
BI
from
indirect effects via
A.
SN may
influence
BI
indirectly via A, due to internalization and
identification processes, or
in-
fluence
BI
directly
via
compliance (Kelman 1958;
Warshaw
1980b). Although
it is
gen-
erally thought
that
computer
use
by managers
and
professionals
is
mostly voluntary
(DeSanctis 1983;
Robey 1979; Swanson 1987),
in
some cases
people may use a system
in
order to comply
with
mandates from their
superiors,
rather than
due
to
their own
feelings
and beliefs about
using
it.
However,
as Warshaw
(1980b) points out,
standard
measures of SN
do not
appear
to
differentiate
compliance
from internalization and iden-
tification.
Complicating
matters
further,
A
may
influence
SN,
for
example
due to the
"false consensus" effect
in
which
people project
their own
attitudes to others
(e.g.,
Oliver
and
Bearden
1985).
Because of
its
uncertain theoretical and
psychometric status,
SN
was
not included
in TAM.
However,
since we
measured
SN
in
our study
in
order to
examine
TRA,
we can test whether SN
explains any
of BI's variance
beyond
that accounted
for
by
A
and
U.
Previous IS research
contains empirical evidence
in
favor of the
A-BI
and U-BI rela-
tionships represented
in
equation (4). Although
BI
per
se has seldom
been measured
in
IS
research,
several studies
have measured A, using a
variety
of
measurement method-
ologies,
and
have
observed a
significant
link
between
A
and
usage (for review,
see Swanson
1982). Usefulness,
and variables similar to it such as
perceptions of performance impacts,
relevance and
importance, have also been linked to
usage (DeSanctis 1983; Robey 1979;
Schultz
and Slevin
1975;
Swanson
1987). Although
the
measures
employed
in
these
studies
were
quite
varied,
and often
unvalidated, the
similarity
of the
findings obtained
from
differing
contexts suggests the possibility of fairly
robust underlying relationships.
According to
TAM,
A
is jointly determined by U
and EOU, with relative weights
statistically
estimated
by
linear
regression:
A
=
U
+
EOU.
(5)
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE OF
COMPUTER
TECHNOLOGY
987
This
equation is inspired
by TRA's view
that attitudes
toward a behavior
are determined
by
relevant beliefs. As discussed
above,
TAM
posits
that U has a direct effect
on BI
over
and above
A.
Equation (5)
indicates that U
influences
A
as
well.
Although
we
contend
that one's affect
toward
a
behavior need
not
fully incorporate
affect toward
any
rewards
due to
performance
outcomes
contingent
on that
behavior,
we
acknowledge
that, through
learning and
affective-cognitive consistency
mechanisms
(Bagozzi 1982),
positively valued
outcomes often
increase one's
affect toward the means to
achieving
those
outcomes
(Peak 1955;
Rosenberg
1956;
Vroom
1964).
Hence,
U
is
hypothesized
to have
a
positive
influence on
A
(as
shown
in
equation (5),
above). Previous
IS research contains
empirical
evidence consistent
with
a
U-A link
(Barrett,
Thornton
and Cabe
1968; Schultz and
Slevin
1975).
EOU is
also
hypothesized to have a
significant effect on
A. TAM
distinguishes two
basic mechanisms
by
which EOU influences
attitudes and behavior:
self-efficacy
and
instrumentality. The easier
a system is to interact
with,
the
greater should
be
the
user's
sense
of
efficacy (Bandura
1982)
and
personal
control
(Lepper 1985)
regarding
his or
her ability to
carry out the
sequences of
behavior needed to
operate the
system. Efficacy
is thought to
operate
autonomously from
instrumental
determinants of behavior
(Bandura
1982), and
influences
affect, effort
persistence, and
motivation due to
inborn drives for
competence
and self-determination
(Bandura
1982;
Deci
1975). Efficacy
is one of the
major
factors theorized to
underly intrinsic motivation
(Bandura 1982;
Lepper 1985).
The direct
EOU-A
relationship
is
meant to
capture this
intrinsically
motivating aspect
of
EOU
(Carroll
and
Thomas
1988;
Davis
1986;
Malone
1981).
Improvements
in
EOU
may
also
be
instrumental,
contributing
to increased
perfor-
mance.
Effort saved due to
improved
EOU
may be
redeployed,
enabling
a
person to
accomplish
more work for the
same effort.
To
the extent that increased EOU
contributes
to
improved
performance,
as would be
expected,
EOU would have
a direct effect on
U:
U
=
EOU
+
External Variables.
(6)
Hence,
we view U and EOU as distinct but
related constructs. As
indicated earlier,
empirical
evidence
from
factor
analyses
suggests
these are distinct
dimensions. At the
same
time,
empirical
associations between variables similar to U and EOU
have
been
observed
in
prior
research
(Barrett,
Thornton
and Cabe
1968;
Swanson
1987).
As
equation
(6) implies,
perceived
usefulness (U)
can
be
affected by various external
variables over and above EOU. For
example,
consider two
forecasting systems which are
equally easy
to
operate.
If
one
of
them
produces
an
objectively
more
accurate
forecast,
it
would
likely
be seen as the
more useful
(U) system, despite
the EOU
parity. Likewise,
if
one graphics
program
produces higher
quality graphs
than its
equally
easy-to-use
coun-
terparts,
it
should
be
considered more useful.
Hence,
the
objective design
characteristics
of a
system
can have
a
direct effect on U
in
addition to
indirect effects via EOU.
Several
investigators have found a
significant
relationship
between
system characteristics and
measures similar to
perceived
usefulness
(e.g., Benbasat and Dexter
1986; Benbasat,
Dexter
and
Todd 1986; Miller
1977).
Similarly, educational
programs
designed
to
pur-
suade
potential
users
of the
power
offered
by
a
given system
and the
degree
to which
it
may improve
users'
productivity
could
well
influence U.
Learning
based
on
feedback is
another
type
of external
variable
apt
to
influence usefulness
beliefs.
Perceived ease of
use
(E)
is
also theorized to be
determined
by
external
variables:
EOU
=
External Variables.
(7)
Many system features
such
as
menus,
icons, mice,
and touch screens
are
specifically
intended to enhance
usability (Bewley
et al.
1983).
The
impact
of
system features
on
EOU has been documented
(e.g., Benbasat,
Dexter and
Todd 1986;
Bewley et al.
1983;
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
988
FRED D. DAVIS, RICHARD P. BAGOZZI AND PAUL R.
WARSHAW
Dickson, DeSanctis
and McBride
1986;
Miller
1977). Training, documentation,
and
user
support
consultants are other
external
factors which
may
also
influence EOU.
Despite their similarity,
TAM
and
TRA
differ
in
several theoretical aspects, some
of
which
warrant explanation.
Both
TAM
and
TRA
posit
that
A is
determined by one's
relevant beliefs.
Two
key
differences between how
TAM
and
TRA
model
the
determinants
of A should be
pointed
out.
First,
using TRA,
salient beliefs
are elicited anew for
each
new context. The resulting beliefs
are considered
idiosyncratic to the specific
context,
not
to
be generalized,
for
example,
to
other
systems
and
users
(Ajzen
and
Fishbein
1980).
In
contrast,
TAM's U
and
EOU
are
postulated
a
priori,
and
are meant to
be
fairly general
determinants
of user
acceptance.
This
approach
was chosen
in
an
attempt
to arrive at a
belief set that
more
readily generalizes
to different
computer systems
and
user
populations.
Second,
whereas
TRA sums
together
all
beliefs
(bi)
multiplied by corresponding
evaluation
weights
(ei)
into
a
single
construct
(equation (2) above),
TAM
treats
U
and
EOU
as
two
fundamental
and
distinct
constructs.
Modeling
beliefs in this
disaggregated
manner enables
one
to
compare
the relative
influence
of each belief in
determining A, providing important
diagnostic
information.
Further,
representing
beliefs
separately
allows
the researcher to
better
trace the influence
of external
variables,
such
as
system features,
user characteristics
and the
like,
on ultimate behavior.
From a
practical standpoint,
this enables
an
investigator
to
better formulate
strategies
for
influencing
user
acceptance
via
controllable
external
interventions that
have
measurable
influences on
particular
beliefs. For
example,
some
strategies may
focus
on
increasing
EOU,
such as
providing
an
improved
user
interface
or
better
training.
Other
strategies
may target U, by increasing
the
accuracy
or
amount
of
information accessible
through
a
system.
Following the view that
U
and
EOU are distinct constructs, their relative
influences
on
A
are
statistically
estimated
using
linear
regression (or
related methods
such
as
conjoint
measurement or structural
equations).
Within
TAM,
U
and EOU are
not
multiplied
by
self-stated evaluation
weights.
Given
that
neither
beliefs
nor
evaluations are
ratio-scaled,
the
estimated
relationship (correlation
or
regression weight)
between
A
and
the
product
of
a
belief and evaluation
is
ambiguous,
since it
would be sensitive
to
allowable
but
theoretically
irrelevant
linear scale transformations of either
the belief or
evaluation
(for
further
explanation,
cf.
Bagozzi
1984; Ryan
and
Bonfield
1975;
Schmidt
1973).
On the
other
hand,
as
Fishbein
and
Ajzen
(1975, p. 238) point out, omitting
the
evaluation
terms
may
be
misleading
in
cases where some
people
in
a
sample
hold
positive
evaluations
while others hold
negative
evaluations of the same outcome.
However,
we
expect
U
and
EOU to be
positively
valued outcomes for
most
people.
When
the
evaluative
polarity
of
an outcome is
fairly homogeneous
across
subjects,
the
corresponding
belief tends to be
monotonically
related to
attitudes,
and
statistically
estimated
weights
tend to
accurately
capture
the actual
usage
of
information
cues
(Einhorn,
Kleinmuntz and
Kleinmuntz
1979; Hogarth 1974),
and
generally
predict dependent
variables at
least
as
well
as
sub-
jective weights (Bass
and Wilkie
1973;
Stahl and
Grigsby 1987;
Shoemaker and Waid
1982).
A
similar rationale underlies
equation (1)
of
TRA,
where
the
relative
influences
of
A
and
SN
on
BI
are
statistically
estimated as
opposed
to self-stated.
One caveat
is
that,
to
the extent that
individuals
within
a
sample
differ
substantially
with
respect
to
the
motivating impact
of
U and
EOU,
our
statistically
estimated
weights may become
dis-
torted.
In
view of the tradeoffs
involved,
we
chose to use
statistically-estimated weights
within
TAM
to
gauge
the
comparative
influence of
U
and EOU on A.
External
variables, represented
in
equations (6)
and
(7), provide the bridge
between
the internal
beliefs,
attitudes
and
intentions
represented
in TAM
and the
various
individual
differences,
situational constraints and
managerially
controllable
interventions
impinging
on behavior.
TRA
similarly hypothesizes
that external
variables
influence
behavior
only
indirectly
via
A,
SN
or
their relative
weights. Although
our
primary interest in the
par-
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE OF
COMPUTER
TECHNOLOGY
989
ticular
study
described
below is to examine our
ability to predict
and
explain
user behavior
with TAM,
working from U and
EOU forward to
user acceptance,
we explicitly
include
external
variables
in
our
description
of
the model to underscore the fact that one of
its
purposes is to provide a
foundation
for studying the
impact of external
variables on user
behavior. Our
goal
in the
study
reported
below
is to
examine
the
relationships
among
EOU, U, A, BI
and system usage
in order to see
how well we can
predict and explain
user
acceptance with TAM.
In
so
doing, we hope to
gain insight
about
TAM's
strengths
and
weaknesses
by comparing
it
to the
well-established TRA.
4.
Research
Questions
Our analysis of
TRA
and
TAM
raises several research
questions which the
study,
described
below, was
designed to
address:
( 1) How well
do intentions
predict usage? Both
models predict
behavior from
behav-
ioral intention
(BI). Of particular
interest is the
ability to predict
future usage based
on
a
brief
(e.g.,
one-hour) hands-on introduction
to
a
system.
This
would mirror the
applied
situations
in
which
these
models
may
have
particular
value. If, after
briefly
exposing
potential
users
to
a
candidate
system that
is
being
considered for
purchase
and
organi-
zational
implementation, management is
able to
take measurements
that
predict
the
future level
of
adoption,
a
go/
no-go
decision on the
specific system
could
be
made from
a
more informed
standpoint.
Similarly, as
new
systems are being
developed,
early pro-
totypes
can
be
tested,
and
intention
ratings
used to assess the
prospects
of the
design
before
a
final
system
is
built.
(2) How well
do TRA and TAM
explain
intentions to use a
system? We
hypothesize
that
TRA
and
TAM will
both
explain
a
significant
proportion
of
the
variance
in
people's
behavioral
intention
to use a
specific
system.
Although prediction,
in
and
of
itself,
is of
value to system
designers and
implementors,
explaining why people
choose to
use or
not
use a
system
is
also of
great
value.
Therefore,
we are also interested
in
the relative
impact
on
BI of
TRA's
A,
SN
and
Z
b1e1 constructs and TAM's U and EOU.
(3)
Do
attitudes
mediate the effect
of beliefs
on
intentions?
A
key principle
of
TRA
is that attitudes
fully mediate the
effects of beliefs on
intentions. Yet,
as discussed
above,
direct
belief-intention
relationships
have
been observed
before.
One of the theoretical
virtues of the
attitude construct
is that it
purports to capture the
influence of beliefs.
Much of
its
value
is
foregone
if
it
only partially
mediates the
impact
of
beliefs.
(4) Is there
some alternative
theoretical
formulation that better accounts for observed
data? We
recognize
that
any
model
is an
abstraction
of
reality
and is
likely
to have its
own
particular
strengths
and weaknesses. Our
goal
is less that
of
proving
or
disproving
TRA
or
TAM,
than
in
using them to
investigate
user behavior. We are therefore interested
in
exploring
alternative
specifications, perhaps
bringing together
the
best
of both
models,
in
our
pursuit of
a
theoretical account of user
acceptance.
5.
Empirical
Study
In
order to
assess TRA and
TAM, we gathered
data from 107
full-time MBA
students
during
their first of four
semesters
in
the MBA
program at the
University
of
Michigan.
A
word
processing program,
WriteOne, was
available for use by
these students
in
two
public computer
laboratories located
at the
Michigan
Business School.
Word
processing
was
selected as a
test application
because: (
1
) it
is a voluntarily
used package,
unlike
spreadsheets
and
statistical
programs
that
students are
required
to use for one or
more
courses, (2)
students
would face
opportunities
to
use
a
word
processor
throughout
the
MBA
program for
memos, letters,
reports, resumes,
and the like, and
(3) word processors
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
990
FRED D.
DAVIS,
RICHARD P. BAGOZZI AND PAUL R. WARSHAW
are among the most frequently used categories
of software
among practicing managers
(Benson 1983; Honan 1986; Lee 1986).
At the beginning of the semester, MBA students are given a one-hour introduction to
the WriteOne software as part
of a
computer orientation.
At
the end of this introduction,
we administered the first wave of a questionnaire containing measures of the TRA and
TAM variables. A second questionnaire, administered at the end of the semester 14
weeks later, contained measures of the
TAM
and TRA variables as well as a 2-item
measure of self-reported usage.
Salient Belief Elicitation
To determine the modal
salient
beliefs
for
usage
of the WriteOne
software, telephone
interviews were conducted with 40
MBA
students
who were about to enter
their
second
year of the
MBA
program.
We chose to elicit beliefs
from
second-year
students since
they
are
very
similar to the
entering first-year
students
in
terms
of
background
and
abilities,
and had
just completed
a
year
of
study during
which
their
introduction and
access to
the WriteOne
system
was identical to that
which
entering first-year
students would face.
Since we wanted to have
the questionnaire prepared
in
advance
of the first
1 -hour
exposure
the
first-year
students would
have with
WriteOne,
so
we
could track
changes
in
their
beliefs over
time,
it would not have been
practical
to ask
first-year
students
their
beliefs
prior to this initial indoctrination. Although they are likely to have had similar basic
concerns as the second-year students, first-year
students were not
expected
to be
in
a
position
to
articulate
those concerns
as well with
regard
to
the WriteOne
system specif-
ically, since they would be unlikely
to even know
that
such a
system existed. We would
have faced greater risk of omitting beliefs which would have become salient by the time
first-year students completed their initial usage and learning and usage of WriteOne. On
the
other
hand, using
second
year
students increased the risk of
including
some beliefs
that are nonsalient for first
year
students after their
initial one-hour
introduction.
However,
the
consequences
of
omitting
a
salient
belief
are
considered more severe than those of
including a nonsalient one.
To omit a
salient belief, i.e., one that does significantly
in-
fluence
attitude, degrades
the
validity
of the
TRA
belief summation
term
(by omitting
a source of
systematic variance),
whereas
including
a
nonsalient
belief, i.e.,
one that does
not
influence attitude, degrades
the
reliability
of
the belief
summation
term
(by adding
a source of random
variance). Moreover,
beliefs
lower in
the salience
hierarchy
contribute
less
to one's total attitude
than do more
salient ones (Fishbein and Ajzen 1975, p. 223).
In
view of the
tradeoffs involved,
we elected to
pursue
a
more inclusive belief set by
eliciting
it from
second-year
students.
Interviewees were asked to list separately the advantages, disadvantages, and anything
else
they
associate
with
becoming
a user of WriteOne.
(This procedure
is
recommended
by Ajzen
and Fishbein
1980, p. 68.)
Beliefs
referring
to
nearly
identical
outcomes
using
alternative
wording
were
classified as the same
item,
and the most common
wording
was
utilized. The seven most
frequently
mentioned
outcomes
were chosen.
This
belief
set
complied
with
the
criteria for modal
beliefs,
since each belief was
mentioned
by
more
than 20%
of the
sample
and
the set contained more
than 75% of
the beliefs emitted.
The
seven
resulting
belief
items,
in
order of
frequency
of
mention,
are:
1. I'd save time
in
creating and editing documents.
2. I'd find it easier to create and edit documents.
3.
My
documents would be
of
a better
quality.
4.
I
would not use alternative word
processing packages.
5.
I'd
experience problems gaining
access to the
computing
center due to crowdedness.
6. I'd become
dependent
on WriteOne.
7.
I
would not use WriteOne after
I
leave the
MBA
program.
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE
OF
COMPUTER
TECHNOLOGY
991
Questionnaire
Both
TRA
and
TAM are
being
used to
explain
a
specific
behavior
(usage)
toward
a
specific target (WriteOne)
within
a
specific
context
(the
MBA
program).
The
time
period
of
usage, although
not
explicitly indicated,
is
implicitly
bounded
by
the context
of
the
MBA
program.
The definition and measurement of model constructs
correspond
in
specificity
to these
characteristics of the behavioral
criterion,
so that the measures of
intentions, attitudes,
and
beliefs are worded
in
reference to the
specific target, action and
context
elements,
but
are
relatively nonspecific
with
respect
to time frame
(for
further
discussion of
the
correspondence issue,
see
Ajzen
and Fishbein
1980). BI, A, SN,
bi
and
ei
were all operationalized according
to
Ajzen
and
Fishbein's ( 1980, Appendix A)
rec-
ommended
guidelines.
TAM's U and EOU are each operationalized with 4-item instruments resulting from
an extensive measure
development
and
validation
procedure.
As
described
in
Davis
( 1986), the measure development process consisted of: generating
14
candidate items
for
each
construct based on
their
definitions; pre-testing
the items to
refine their
wording
and to
pare
the item sets down to 10 items
per construct,
and
assessing
the
reliability
(using
Cronbach
alpha)
and
validity (using
the multitrait-multimethod
approach)
of the
10-item scales. High levels of convergent and discriminant validity of the 10-item scales
were observed, and Cronbach alpha reliabilities were 0.97 for U and 0.91 for EOU. Item
analyses
were used
to
streamline the
scales
to 6 items
per construct,
and new
data
again
revealed high validity and reliability (alpha of 0.97 for U and 0.93 for EOU). Further
item analyses
were
performed to arrive at the 4-item scales used
in
the present research.
The
four ease
of
use items were: "Learning to operate WriteOne would be easy for me,"
"I
would
find
it easy to get WriteOne to do what
I
want
it
to do," "It would be easy for
me to become skillful at
using WriteOne,"
and
"I
would find WriteOne
easy
to use."
The four usefulness items were: "Using WriteOne would improve my performance
in
the
MBA
program," "Using
WriteOne
in
the MBA
program
would increase
my pro-
ductivity," "Using
WriteOne
would
enhance
my
effectiveness
in
the
MBA
program,"
and
"I
would find WriteOne useful
in
the
MBA
program."
The usefulness and ease of
use
items were
measured
with
7-point
scales
having likely-unlikely endpoints
and the
anchor
points extremely, quite, slightly,
and neither
(identical
to
the
format
used
for
operationalizing
TRA
beliefs
and
recommended
by Ajzen
and
Fishbein 1980, Ap-
pendix A).
System usage
is
measured
using
2
questions regarding
the
frequency
with which
the
respondent currently uses WriteOne. The first
was
a 7-point scale with the adjectives
frequent and infrequent at the endpoints. The second was a "check the box" format,
with categories
for
current use of: not at all; less than once a week; about once a week;
2
or
3
times a week;
4
to 6 times a week; about once a day, more than once a day. These
are
typical
of
the
kinds of
self-reported
measures often used to
operationalize system
usage, particularly
in
cases where
objective usage
metrics
are
not available.
Objective
usage logs
were
not
practical
in
the
present
context since the word
processing
software
was
located
on
personal computers and subjects use different computers, as well as different
applications,
from
one session to the
next.
Self-reported frequency measures should
not
be
regarded
as
precise
measures of
actual
usage frequency, although previous research
suggests they
are
appropriate
as relative
measures (Blair and Burton 1987; Hartley,
et
al.
1977).
Results
Scale Reliabilities.
The two-item
BI
scale obtained a Cronbach alpha reliability of
0.84 at time 1 (beginning of the semester) and
0.90 at time 2 (end of the semester). The
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
992
FRED
D.
DAVIS,
RICHARD
P. BAGOZZI AND PAUL
R. WARSHAW
four-item
A
scale
obtained reliabilities of 0.85
and 0.82 at times
1
and 2 respectively.
The four-item
U scale achieved a reliability
of 0.95 and 0.92 for
the two
points
in
time,
and the four-item
EOU scale obtained reliability
coefficients
of 0.91 and 0.90 for
time 1
and time
2.
SN,
the
bis
and
the
eis,
were each
operationalized
with
single-item
scales,
per TRA,
and hence
no internal
consistency
assessments
of
reliability
are
possible.
The
two-item
usage
scale
administered
in the second
questionnaire
achieved an
alpha
of
0.79.
These scale reliabilities
are
all at levels considered
adequate
for
behavioral
research.
Explaining
Usage.
As
expected,
BI was
significantly
correlated
with
uisage.
Intentions
measured right
after the
WriteOne
introduction
were correlated 0.35 with
usage
frequency
14
weeks
later
(Table
1).
Intentions
and
usage
measured
contemporaneously
at
the end
of the
semester
correlated
0.63. Also consistent with the
theories,
none of the other
TRA
or TAM variables
(A, SN, L b1e1, U,
or
E)
had a
significant
effect on
usage
over
and
TABLE
1
Predicting an?d
Explaining Usage,
Intentions and Attitludes wvith thle Theory of Reasoned Actioni
(TRA)
and the Technology
Acceptance
Model (TAM)
Time
1
Immediately After
Time
2
1
Hr
Intro
14
Weeks
Later
Equation
R
2
Beta R
2
Beta
(1) Explaining Usage
at
Time
2
From BI Measured
at Times
I
and
2
(Common
to both
Models)
Usage (Time 2)
BI 0.12***
0.40***
BI
0.35***
0.63***
(2)
TRA
BI
=
A
+ SN
0.32*** 0.26***
A
0.55***
0.48***
SN 0.07
0.10
A
=
be
0.07** 0.30***
E
bje,
0.27**
0.55***
(3)
TAM
BI
=
A
+ U
0.47*** 0.51
A
0.27**
0.16
U
0.48***
0.6
1
A
=
U
+
EOU
0.37***
0.36***
U
0.6
1***
0.50***
EOU
0.02
0.24**
U
=
EOU 0.01
0.05**
EOU
0.10
0.23**
Note.
*
p <0.05.
** p
<
0.01.
***
p
<
0.001.
BI
=
Behavioral
Intention
A
=
Attitude
SN
=
Subjective
Norm
U
=
Perceived Usefulness
E
bje=
Sum of
Beliefs
Times Evaluations
EOU
Perceived
Ease
of Use
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER ACCEPTANCE OF COMPUTER TECHNOLOGY
993
above intentions at
either time
1
or time 2, which
suggests that intentions fully mediated
the effects of these
other variables on usage.
Explaining
Behavioral Intention
(BI).
As
theorized,
TRA
and
TAM
both
explained
a
significant proportion
of the variance
in
BI
(Table
1).
TRA
accounted
for
32%
of the
variance at time
1
and 26% of the variance at time 2. TAM
explained
47%
and 51% of
BI's variance at
times
1
and
2
respectively. Looking at the individual
determinants
of
BI,
within
TRA,
A
had a
strong significant influence
on
BI
(f =
0.55,
time
1;
f =
0.48,
time
2),
whereas SN
had no
significant
effect
in
either time
period (b
=
0.07
and
0.10,
respectively). Within
TAM,
U has a
very strong
effect
in
both time
periods (f
=
0.48
and
0.61,
respectively),
while A
had a smaller effect
in
time
1
(/
=
0.27)
and a
nonsig-
nificant effect in
time
2
(/ =
0. 16). The increased
influence
of U from
time
1
to time
2
is noteworthy.
Equation ( lb), Table 2, shows that U adds
significant explanatory power
beyond
A and
SN,
at both time
1
and time
2,
underscoring
the
influential
role of
U.
In both models,
unexpected direct belief-intention
relationships were observed. Counter
to
TRA,
the
belief summation
term,
Z
b,ej,
had
a
significant
direct
effect
on
BI
over and
above
A
and SN
in
time
period
2
(3
=
0.21
)
but
not
in
time
period
1
(3
=
0.08) (Table
2). Counter to
TAM, EOU had a significant direct
effect on
BI
over and above
A
and
U
in
time period
1
(3
=
0.20)
but not time
period
2
(3
=
0.11) (Table 2). Hence,
attitude
appears
to
mediate the effects of beliefs
on intentions
even less than
postulated
by
TRA
and
TAM.
TABLE
2
Hierarchical
Regression Tests for Relationships Expected
to be
Nonsignfficant
Time I
Time
2
Equation
R
2
Beta R
2
Beta
(1)
Behavioral Intention
(BI)
(a)
BI
=
A
+
SN
+
E
bhej
0.33***
0.30***
A
0.53*** 0.37***
SN
0.06
0.08
E
bjej
Q008a
0.21*
(b)
BI =
A
+
U
+
SN 0.47***
0.51
A
0.27**
0.16
U
0.48***
0.63***
SN
0.02a
-0.04a
(c)
BI
=
A
+
U
+ E
0.5
1***
0.52***
A
0.26**
0.
19*
U 0.47***
0.62***
E
0.20**
-0.1
la
(2)
Attitude
(A)
A
=
U
+
E
+
-
bie,
0.38*** 0.44***
U
0.58*** 0.35***
E
0.01
0.18*
E
bje-
0.0Oa
0.32***
*
p
<
0.05.
**
p
<
0.01.
***
p
<
0.00
1.
a:
Expected
and
found
nonsignificant.
b:
Expected nonsignificant
but
found significant.
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
994
FRED D.
DAVIS,
RICHARD P.
BAGOZZI AND PAUL
R. WARSHAW
ExplainingAttitude.
As
expected,
both
TAM and TRA
explain
a
significant
percentage
of variance
in
attitude
(Table
1).
TRA
explained
7%
of A's
variance
at time
1
and 30%
at time
2. TAM
explained
37% and 36% at times
1
and
2,
respectively.
U
has a
strong
significant effect
on
A
in
both
time
periods
(
= 0.61
and
0.50, respectively),
although
EOU is
significant
at
time
2
only
(f
=
0.24).
In
both
models,
there were some
interesting developmental
changes over time
in
the
relationship
among beliefs,
A
and
BI.
Within
TAM,
at
time
1
EOU
appears
to
have a
direct
effect on
BI
(1 =
0.20),
with
no indirect effect
through
A or
U,
at
time
2
EOU's
effect is
entirely
indirect via
U,
and the
A-BI link
becomes
nonsignificant.
TRA's
belief
summation
term, E b1e1,
has a
significant
effect on A above and
beyond
U and
EOU
in
time
period
2
( =
0.32)
but
not in
time
period
1 (I =
0.10) (Table 2).
Our
analysis
below
investigates
the nature of these
patterns
further
by
analyzing
the internal
structure
of TRA's
beliefs
and
analyzing
their
relationship
to U and
EOU,
A
and
BI.
Further
Analysis of Belief
Structure.
In
order to
gain greater
insight
into the
nature
of TRA's
beliefs,
as
well
as their
relationship
to
U and
EOU,
a
factor
analysis
was con-
ducted.
Table
3 shows
a
varimax rotated
principal components
factor
analysis
of
TRA's
7
belief
items and
TAM's
4
U items
and 4
EOU
items,
using
a
1.0
eigenvalue
cutoff
criterion.
For time
period 1,
a five-factor
solution
was
obtained,
with
the
7
TRA
beliefs
factoring
into three
distinct
dimensions,
the
other two factors
corresponding
to
TAM's
U and EOU.
TRA
beliefs
1,
2
and 3 load on a common
factor which
taps specific
aspects
of
"expected performance
gains."
Whereas TAM's U
is
a
comparatively general
assessment
of
expected performance
gains
(e.g.,
"increase
my
productivity"),
TRA 's
first
three items
are more
specific
aspects (i.e.,
"saving
time
in
creating
and
editing documents",
"finding
it
easier
to
create
and edit
documents",
and
"making higher
quality
documents"). We
will
refer
to this
specific usefulness construct
comprised
of TRA's
first three belief
items
as
Us.
Consistent
with
this
interpretation,
Us
correlates
significantly
with
U
(r
=
0.46, p
<
0.001
for time
1
and
r
=
0.65, p
<
0.001
for time
2).
At time
period 2,
a
four-factor
solution
was
obtained,
with
Us
converging
to
TAM's
U to form a
single
factor. We
will
denote this combined
7-item
usefulness index
Ut,
for
total
usefulness. Cronbach
alpha
reliabilities for
Ut
were 0.85 and 0.93
for
time
1
and
2,
respectively.
In
both time
periods,
TRA
beliefs
4
and
6 loaded on
a common factor
which
has
to
do
with
becoming dependent
on WriteOne
("would
become
dependent
.
.
.",
"would
not
use alternatives
.
.
."),
which we
will
denote D.
TRA
items
5
and
7
loaded on a
common factor at time
1, and
are concerned
with access
to
WriteOne,
both
while
in
the
MBA
program
(item 5),
and
after
leaving
the
program (item
7).
We
will
denote
this
factor
ACC. At time
2, only
item 5 loaded
on this
factor,
with
item
7
showing a
tendency
to
load
on
Ut
instead
(loading
=
-0.45).
Hence,
the factor
analysis
of TRA
and
TAM
beliefs
suggests the existence
of belief
dimensions
concerning
usefulness,
ease of
use, dependency,
and
accessibility. Overall
perceived
usefulness
(U,) appeared
to have
separate specific
(Us)
and
general
(U) di-
mensions at time
1
which
converged to
form
a common
dimension at time 2.
Perceived
accessibility
(ACC)
was
comprised
of
2
items
(TRA
beliefs
5
and
7) at time
1
and only
1
item
(belief
5) at
time 2.
Hybrid
Intention
Models. The factor
analysis
above
provided some
interesting insights
into the
dimensional structure
of
the
beliefs
underlying user
acceptance.
Combining the
beliefs
of TRA
and
TAM
into a
single analysis
may yield
a
better
perspective on the
determinants
of BI
than
that
provided
by
either model
by
itself. Given
that
A
was
generally
not
found to
intervene between beliefs
and
intentions, our
approach in this
section is to
first
assess
the
impact
on
intentions
of
the beliefs
identified
in
the factor
analysis above,
and then
test whether
A
mediates these
belief-intention
relationships.
We
estimated the
effect
on
BI
of the five beliefs
identified
by
the
factor
analysis:
U,
US,
EOU, D
and ACC
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER ACCEPTANCE OF COMPUTER
TECHNOLOGY
995
TABLE
3
Factor Analysis of TAM and TRA Belief Items
Time
I
Factors
Time 2 Factors
Belief
Item 1
2
3 4
5 1
2
3 4
(a)
TRA
Items
TRAI
0.28 0.05
0.89 0.10 -0.01 0.82
0.08 0.08 0.34
TRA2
0.27 0.13
0.88 0.11 -0.02 0.84
0.13 0.05 0.34
TRA3 0.18 0.03
0.80 0.13 -0.01 0.74
0.12 0.14 0.42
TRA4
0.17 -0.11
0.09 0.81 -0.04 0.20
0.01 0.88 -0.02
TRA5 0.02 0.09
0.06
0.24
0.83 -0.02
0.05 0.10 0.82
TRA6 0.08 -0.09
0.17 0.79 0.07 0.32
-0.06 0.69 0.30
TRA7 -0.26 0.00
-0.12 -0.34 0.66 -0.45
0.32 -0.07 -0.06
(b)
TAM
Usefulness (U) Items
Ul
0.90 -0.03
0.18 0.06 -0.02 0.79
0.11 0.31 -0.15
U2
0.90 -0.03
0.26 0.14 -0.04 0.82
0.07 0.22 -0.16
U3 0.91 0.01
0.16 0.06 -0.05
0.84
0.08 0.12 -0.24
U4
0.85 0.03
0.24
0.17 -0.13 0.85
0.17
0.18 -0.07
(c)
TAM
Ease
of
Use (EOU)
Items
EOU1
-0.08
0.84
0.08 -0.12 0.08 -0.09 0.88 -0.06
0.07
EOU2 0.01
0.90
0.03 -0.05 0.01 0.15 0.85 0.12 -0.05
EOU3
-0.05 0.91
0.03 -0.09 0.03 0.05
0.89
-0.04 0.07
EOU4 0.10 0.91 0.07
0.00 -0.01 0.30 0.84 -0.02 0.03
Eigen.
4.83
3.35
1.51
1.14 1.06 5.87 2.99
1.27 1.01
%
Var 32.3 22.3
10.1 7.6 7.0 39.2
19.9 8.5
6.7
Cum
%
32.3
54.6 64.7
72.3 79.3
39.2 59.1 67.6 74.3
(see
Table
4).
Together, these variables explained 51%
of
BI's variance
in
time
1
and
61%
in
time 2.
U,
Us
and
EOU were
significant for time 1, but EOU became
nonsignificant
in
time 2.
In
addition,
Us
increased
in
importance
from time
1
(b
=
0.20)
to time
2
(1
=
0.39). Next,
we
combined the two usefulness
subdimensions to form the
U1
index,
and
ran
another
regression.
U1 was
highly significant
in
both time
periods
(1 =
0.59
and
0.71, respectively),
and EOU was
significant
for time
period
1
only
(1 =
0.20).
In
order to
test whether
A
fully mediated
either the EOU-BI or U-BI
relationships,
we
introduced
A
into the
second equation.
This
had
little effect on the coefficients for either
U1
or
EOU, suggesting
that
although
A
may partially
mediate
these
relationships,
it
did
not
fully
mediate
them. The
relationship
between EOU and
U,,
hypothesized by TAM,
was nonsignificant for time
1, but became significant for time
2
(1 =
0.24). Therefore,
the causal
structure
suggested
is that
U,
had a
direct
impact
on
BI in
both time
periods
and EOU
had
a
direct effect
on BI
at time
1
and an indirect effect via U1 at time
2.
In
order to obtain more
precise estimates
of
these significant
effects, regressions omitting
nonsignificant
variables were run
(see
Final
Models,
Table
4).
At time
1,
U1
and
EOU
accounted for 45% of the
variance
in
intention,
with
coefficients
of 0.62 and 0.20
re-
spectively.
At
time
2,
U,
by
itself
accounted
for 57% of BI's variance
(3 =
0.76),
and
EOU
had a small but
significant effect
on
U,
(3
=
0.24).
As
mentioned
earlier,
to the extent that
people
are
heterogeneous
in their
evaluation
of or
motivation
toward
performance,
our
statistical estimate of the usefulness-intention
link
may
be
distorted.
In
order
to test
for
whether differences
in
motivation moderated
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
996
FRED D.
DAVIS,
RICHARD P. BAGOZZI
AND PAUL R. WARSHAW
TABLE
4
Hybrid
Intention Models
Time
1
Time
2
Equation
R
2
Beta R
2
Beta
BI=
U +
Us+ EOU
+
D
+
ACC 0.51
0.61
U
0.48***
0.35***
Us
0.20*
0.39***
EOU
0.21**
-0.04
D 0.09 0.14
ACC
-0.11 -0.12
BI
=
U,
+
EOU
+
D
+
ACC 0.50
0.61
U,
0.59***
0.7 1***
EOU
0.20** -0.06
D 0.09 0.15
ACC
-0.12 -0.12
U,
=
EOU
0.02 0.15
0.06 0.24*
Final Models:
A.
Time
1
BI
=
U,
+
EOU
0.45
U,
0.62***
EOU
0.20**
B. Time 2
BI
=
U,
0.57 0.76***
U,
=
EOU
0.06 0.24*
*
p
<
0.05.
** p
<
0.01.
***
p
<
0.001.
Note.
U
=
TAM's
general perceived
usefulness
scale
(4 items). Us
=
TRA's
specific
usefulness scale
(items
1-3).
U,
=
Total usefulness
index
(comprised
of U and
Us;
7
items).
the
usefulness-intention relationship,
we
asked
subjects
to
report
the
extent to which they
believed
"performance
in
the
MBA
program
is
important
to
getting
a
good
job." By
hierarchical regression,
this question did not significantly
interact
with
U1
in either time
period.
We also used the
sum of the three
evaluation terms
(ei)
corresponding
to
TRA
belief items 1-3 as an
indicant of
subjects'
evaluation of usefulness
as an
outcome. This
also
did not
significantly
interact
with
usefulness
in
either
time
period.
Thus,
in
our
sample,
it
appears
that individuals did
not
differ
enough
in either
( 1)
their
perceived
impact of performance
in the MBA
program
on their getting
a
good job
or (2) their
evaluation of
performance
to
seriously
distort our estimate of
the
effect of U1 on BI.
The picture that emerges
is that U is a
strong determinant
of
BI in
both time
periods,
and that EOU also
has a
significant
effect on
BI
at time
1
but
not at
time
2.
EOU's
direct
effect
on
BI in
time
period
1
developed
into
a
significant
indirect effect, through
usefulness,
in
time
period
2.
6.
Conclusions
Our results
yield
three
main
insights
concerning
the
determinants
of
managerial
com-
puter
use:
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE OF COMPUTER TECHNOLOGY
997
(1) People's computer use can
be predicted reasonably well from their
intentions.
(2) Perceived usefulness is a major
determinant of people's intentions to
use computers.
(3) Perceived ease of use is a significant
secondary determinant of people's
intentions
to
use
computers.
Although our data provided mixed
support for the two specific theoretical
models that
guided our investigation, TRA and
TAM, their confluence led to the identification
of a
more
parsimonious
causal structure that is powerful for predicting and explaining
user
behavior
based on
only
three theoretical constructs: behavioral intention
(BI),
perceived
usefulness
(U)
and
perceived
ease of use
(EOU). Specifically,
after the
one-hour
intro-
duction to the
system, people's
intentions
were
jointly determined by perceived
usefulness
(1
=
0.62)
and
perceived
ease of use
(3
=
0.20).
At the end of 14
weeks,
intention was
directly
affected
by
usefulness alone
(1
=
0.79),
with
ease of
use
affecting
intention
only
indirectly via usefulness (1
=
0.24).
This simple model accounted for 45%
and 57% of
the variance
in
intentions
at
the beginning and end
of
the
14-week
study
period, respec-
tively.
Both TRA and TAM postulated
that BI is the major determinant of usage
behavior;
that behavior should
be
predictable
from
measures
of
BI, and
that
any other
factors
that
influence user
behavior do so
indirectly
by influencing
BI. These
hypotheses
were all
supported by
our
data.
Intentions
measured
after a one-hour
introduction
to a
word
processing system were correlated
0.35 with behavior
14
weeks later.
This is promising
for those who wish to evaluate
systems very early
in
their
development,
and cannot
obtain extensive user
experience
with
prototypes
in
order to
assess its
potential
accept-
ability.
This
is also
promising
for
those
who
would
like to assess user reactions to
systems
used
on a trial basis
in
advance of
purchase
decisions.
Intentions
and usage measured
contemporaneously
correlated 0.63.
Given
that intentions are
subject
to change between
the time
of
intention measurement and behavioral
performance,
one would expect the
intention-behavior correlation to
diminish
with
increased elapsed time (Ajzen
and
Fish-
bein
1975, p. 370).
In
addition,
at time
1, given
the
limited experience
with the
system,
peoples'
intentions would
not be
expected
to
be extremely well-formed
and stable. Con-
sistent
with
expectations,
hierarchical
regression
tests
indicated
that none of
the other
variables studied influenced behavior
directly, over and above intention.
In
order to place these intention-behavior
correlations
in
perspective,
we can compare
them
to
(a) past experience using
intention measures
outside
the
IS domain and
(b)
correlations
between
usage
and
various
predictors reported
in
the IS literature.
In
a
meta-
analysis
of non-IS
studies, Sheppard,
Hartwick and
Warshaw
(in press)
calculated a
frequency-weighted average intention-behavior
correlation of 0.38, based on
1514 subjects,
for
goal-type
behaviors.
The
intention-usage
correlations of
0.35
and
0.63 obtained
in
the
present study compare
favorably
with
this
meta-analysis. Although
the
intention-
usage relationship per
se
has
been
essentially
overlooked
in
the IS
literature, usage pre-
dictions based
on numerous other
variables
have
been
investigated.
Ginzberg (1981)
obtained a correlation of
0.22 between
a
measure
of users "realism of
expectations"
and
usage.
DeSanctis
(1983)
obtained correlations around
0.25
between "motivational
force"
and DSS
usage.
Swanson
(1987)
obtained
a
0.20
correlation between
usage
and a
variable
referred to as "value"
which is
similar
to
perceived
usefulness.
Robey
obtained a
striking
0.79 between
usage
and Schultz and Slevin's
(1975) performance
factor,
which is
also
similar to
perceived
usefulness.
Baroudi,
Olson
and Ives
(1986)
found both user
infor-
mation
satisfaction
and user
involvement
to be correlated 0.28
with
system
usage.
Sri-
nivasan
(1985) found relationships
varying
from -0.23 to 0.38
between various measures
of
user
satisfaction
and
usage.
Overall,
the
predictive
correlations obtained
in
IS
research
have
varied widely,
from -0.23
up
to
the 0.79 correlation
obtained
by
Robey (1979),
with
typical
values
falling
in
the 0.20-0.30 range.
The 0.35 and 0.63 correlations
obtained
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
998
FRED D.
DAVIS,
RICHARD P. BAGOZZI AND PAUL R.
WARSHAW
for the
two
time
periods investigated
in the
present
research
compare favorably with
these
previous
IS
findings.
Both
TRA and TAM hypothesized
that expected
performance impacts due to using
the
specified
system, i.e., perceived
usefulness,
would be a
major
determinant
of BI.
Interestingly, the models arrived
at this
hypothesis by very
different
lines of reasoning.
Within
TAM, perceived
usefulness was
specified
a
priori,
based
on
the
observation that
variables
having to do with performance
gains had surfaced as
influential determinants
of user
acceptance
in
previous
IS studies.
In
contrast,
TRA
called
for
eliciting
the
specific
perceived
consequences
held
by specific
subjects concerning
the
specific system under
investigation.
Using
this
method,
the
first
three beliefs
elicited were
specific performance
gains.
These three
TRA
beliefs,
which
were much more
specific
than TAM's
perceived
usefulness measures
(e.g.,
"save time
in
creating
and
editing
documents" versus
"increase
my
productivity")
loaded
together
on a
single
dimension
in
a factor
analysis. Although
TRA's specific usefulness dimension
(Us)
was
factorially
distinct from TAM's
U at time
1
(just after the one-hour
demonstration), they
were
significantly
correlated
(r
=
0.46).
Fourteen weeks later
(time 2),
the
general
and
specific
items
converged
to load on
single
factor.
But why was
it
the case
that
U had more influence on
BI
than
Us right after the one-
hour
introduction,
whereas
Us
increased
in
influence, and
converged to U, over time?
One
possibility
relates to
the
concreteness-abstractness distinction from
psychology (e.g.,
Mervis
and Rosch, 1981).
As Bettman and
Sujan (1987) point
out,
novice
consumers
are more
apt
to
process
choice alternatives
using abstract, general
criteria,
since
they
have
not
undergone
the
learning
needed to
understand
and
make
judgments about more
concrete,
specific
criteria.
This
learning
process
could account for
the increased
importance
of
Us
over
time,
as
well
as its
convergence
to
U,
as the
subjects
in
our
study gained
additional
knowledge
about the
consequences
of
using
of
WriteOne
over the 14-week
period
following
the
initial
introduction. The
implication
is
that,
since
people
form
general
impressions
of usefulness
quickly
after a brief
period
of
using
a
system,
the more
general
usefulness
construct
provides
a somewhat
better
explanation
of
intentions at such
a point
in
time.
Combining
the 3
specific
TRA
usefulness beliefs and the
4
general
TAM
usefulness
beliefs
yielded
a total
index
of
usefulness
U,
that
had
a
major
impact
on BI in
both time
periods.
Indeed, subjects appeared
to
form
their
intentions toward
using
the
word pro-
cessing system
based
principally
on their
expectations
that it
would
improve
their
per-
formance
within
the
MBA
program.
Among
the
other beliefs
studied, only EOU
had
a
significant
effect
on
BI,
and
only
at time
1.
Over
time,
as users
learned to
effectively
operate
the word
processor,
the direct effect of ease
of use on
BI
disappeared, being
supplanted
by
an
indirect
effect
via
U,.
Following
our
theorizing,
early on, people appeared
to
process
EOU from a
self-efficacy
perspective, appraising
how
likely they
would
be
to
succeed
at
learning
to
use
the
system
given they
tried. As
learning progressed
over
time,
this concern
became
less
salient,
and
EOU
evolved into a
more instrumental
issue,
re-
flecting
considerations of how the relative effort of
using
the
system
would
affect the
overall
performance
impact
the
system
offered
(U1).
The lack
of
a
significant
SN-BI
effect was surprising, given previous
IS research stressing
the
importance
of
top management
support
and user
involvement.
There
are two
reasons
to
interpret
this
finding narrowly.
First,
as
pointed
out
in
our
discussion
of
TAM,
com-
pared
to
other
measures recommended for
TRA
(Ajzen
and Fishbein
1980),
the
SN
scale
is
particularly
weak from a
psychometric
standpoint.
More
sophisticated
methods
for
assessing
the
specific types
of social influence
processes
at work
in
a
computer accep-
tance
context
are
clearly
needed.
Second,
the
specific application
studied,
word
processing,
is
fairly personal
and
individual,
and
may
be driven
less
by
social influences
compared
This content downloaded from 130.184.237.6 on Thu, 6 Feb 2014 14:36:30 PM
All use subject to JSTOR Terms and Conditions
USER
ACCEPTANCE OF COMPUTER
TECHNOLOGY
999
to
more
multi-person
applications
such
as
electronic
mail, project
management
or
group
decision
support
systems.
Further research
is needed
to address the
generalizability of
our
SN findings,
to better
understand the nature of social
influences,
and
to
investigate
conditions and
mechanisms
governing the impact
of
social influences on
usage
behavior.
The absence of
a significant
effect
of
accessibility on
intentions
or
behavior was also
surprising
in
light
of the
importance
of
this variable
in
studies
of
information
source
usage (Culnan
1983; O'Reilly 1982). Since our measure
of
accessibility
was
nonvalidated,
having been
developed by
exploratory factor
analysis, psychometric
weaknesses may be
partly
at fault.
In
addition, although
access was a salient
concern
frequently
mentioned
in
the belief
elicitation, the system
under investigation
was fairly
uniformly accessible to
all
respondents.
Accessibility may
well have
played
a
more
predominant
role
if
greater
variations in
system accessibility
were present in the
study. Also
surprising was the finding
that
attitudes intervened between beliefs
and intentions
far
less than
hypothesized by
either TRA or TAM.
Although
some work on the direct
effect
of beliefs
has been done
(e.g., Bagozzi
1982; Brinberg 1979;
Triandis
1977),
more
research
is
needed
to
identify
the conditions under which
attitudes mediate the belief-intention link.
In
either
case,
the
attitude construct did little
to
help
elucidate the causal
linkages
between beliefs
and
intentions
in
the
present study since,
at best, it only
partially mediated
these relationships.
There are
several aspects of the
present study which
circumscribe the extent
to
which
our
findings generalize.
MBA
students are not
completely representative
of the entire
population
of
managers
and
professionals whose
computer usage
behavior we would like
to model. These students
are
younger
and,
as
a
group, probably
more
computer
literate
than their
counterparts
in
industry.
Hence, EOU
may
have
been
less an
issue
for
this
sample
than it would
have been
for
managers
and
professionals
more
generally.
The
WriteOne system,
while typical of
the types of
systems available to end
users, is still only
one
system.
With more
complex or difficult
systems,
ease of
use may
have had a
greater
impact on
intentions. These subjects
were also
probably more highly
motivated
to
perform
well than the
general population,
which may have
caused perceived
usefulness to take
on
greater importance
than it
generally
would. Further research on these variables and
relationships
in
other settings
will
sharpen our
understanding
of
their
generalizability.
Additional
theoretical constructs
such as computer
anxiety and
instrinsic motivation
may profitably be
brought into the
analysis. There is reason for
optimism,
however.
Extensive
experience with intention
measures
in
other
contexts has
consistently supported
their role as
predictors of an individual's behavior
(e.g., Ajzen
and Fishbein
1980).
In
addition, the usefulness-intention
relationship observed
in
the
present
data is so
strong
that
it seems
unlikely
to be
totally
idiosyncratic.
If
models similar to the
final models
presented
in
Table
4
do generalize to
other contexts,
we
will
be moving to
a
situation
in
which
powerful
yet simple models
for predicting
and explaining user
acceptance
are
available.
7.
Practical
Implications
What
do our
results
imply for
managerial
practice? When planning
a
new
system, IS
practitioners
would like to be able
to predict whether
the new system
will be acceptable
to
users, diagnose
the reasons
why a
planned system
may not be fully
acceptable to users,
and to
take corrective