See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/314410967

The Technology Acceptance Model

Chapter · January 2005

DOI: 10.4018/9781591404743.ch006.ch000

CITATIONS READS

0 3,822

2 authors:

Qingxiong Ma Liping Liu

University of Central Missouri University of Copenhagen
24 PUBLICATIONS   832 CITATIONS    32 PUBLICATIONS   1,061 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

service-oriented applications View project

Teaching case for System Analysis View project

All content following this page was uploaded by Qingxiong Ma on 26 October 2017.

The user has requested enhancement of the downloaded file.

 IDEA GROUP PUBLISHING
Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 59
701 E. Chocolate Avenue, Suite 200, Hershey PA 17033-1240, USA ITJ2520
Tel: 717/533-8845; Fax 717/533-8661; URL-http://www.idea-group.com

The Technology Acceptance Model:

A Meta-Analysis of Empirical
Findings
Qingxiong Ma, Central Missouri State University, USA
Liping Liu, University of Akron, USA

ABSTRACT

The technology acceptance model proposes that perceived ease of use and per-
ceived usefulness predict the acceptance of information technology. Since its in-
ception, the model has been tested with various applications in tens of studies and
has become the most widely applied model of user acceptance and usage. Never-
theless, the reported findings on the model are mixed in terms of statistical signif-
icance, direction, and magnitude. In this study, we conducted a meta-analysis
based on 26 selected empirical studies in order to synthesize the empirical evi-
dence. The results suggest that both the correlation between usefulness and ac-
ceptance, and that between usefulness and ease of use are somewhat strong. How-
ever, the relationship between ease of use and acceptance is weak, and its signif-
icance does not pass the fail-safe test.

Keywords: meta analysis; technology acceptance model; perceived ease of use;

perceived usefulness; behavioral intention

INTRODUCTION is believed to be more parsimonious, pre-

Information technology (IT) accep-
tance or adoption has received consider-
able attention in the last decade. Several
theoretical models have been proposed to
explain end-users’ acceptance behavior.
Among them, the technology acceptance
model (TAM) proposed by Davis (1989)
is widely applied and empirically tested.
There have been tens of empirical studies
conducted on TAM since its inception.
Compared with its competing models, TAM
dictive, and robust (Venkatesh & Davis,
2000).
Despite the plethora of literature on
TAM, the empirical tests have so far pro-
duced mixed and inconclusive results,
which vary considerably in terms of statis-
tical significance, direction, or magnitude.
Although they are not uncommon in social
sciences where human behavior is diffi-
cult and complex to explain, the mixed find-
ings not only undermine the precision of
TAM, but also complicate efforts for IT

Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
60 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004
practitioners and academicians to identify
the antecedents to user acceptance behav-
ior.
The goal of this study is to understand
to what extent the existing body of litera-
ture reflects substantial and cumulative
validity of TAM. In particular, we synthe-
size the existing findings on TAM by con-
ducting a meta-analysis. We hope that, by
integrating existing empirical findings, we
can better understand how TAM is appli-
cable to different technologies as a whole.
We will be able to examine the relation-
ships between the constructs of TAM with
a larger sample of subjects than any indi-
vidual studies. The results of this study can,
we hope, be used as a benchmark for fu-
ture tests of TAM.
Beside its potential theoretical contri-
butions, a meta-analysis on TAM is also
significant to IT management practice. By
understanding the substantive antecedents
to user acceptance, IT managers can take
more effective interventions to achieve
greater technology acceptance or usage.
As Robey and Markus (1998) and
Benbasat and Zmud (1999) noted, IT
management needs prescriptions. IT re-
searchers should not only apply rigorous
methodology best suited to their research
objectives, but also produce relevant and
consumable research for practitioners.
There can be many possible ways for ac-
ademic research to contribute to practice.
Benbasat and Zmud (1999, p. 9) noted as
a successful example, “IT research based
on Theory of Reasoned Action and its ex-
tensions, such as the Theory of Planned
Behavior, to the study of IT adoption, im-
plementation, and use.” They suggested
that once a sizable body of literature exists
regarding a phenomenon, “it does become
possible to synthesize this literature”
(Benbasat and Zmud, 1999, p. 9). Thus,
they recommended that the “IS research
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
community produce cumulative, theory-
based, context rich bodies of research.”
In a sense, the current study answers this
“rigor and relevance” research call.
The outline of this paper is as follows.
We first review the literature on TAM and
indicate major inconsistencies and discrep-
ancies in the existing findings. Then, we
describe how we collected and recorded
the sample of empirical findings and re-
port the results of our meta-analysis based
on 26 selected empirical studies. Finally,
we conclude this study by including a dis-
cussion on its limitations and some sugges-
tions for future research.
LITERATURE REVIEW
The Technology Acceptance Model
(TAM), introduced by Davis (1986), is one
of the most widely used models to explain
user acceptance behavior. This model is
grounded in social psychology theory in
general and the Theory of Reasoned Ac-
tion (TRA) in particular (Fishbein, & Azjen,
1975). TRA asserts that beliefs influence
attitudes, which lead to intentions and there-
fore generate behavior. Correspondingly,
Davis (1986, 1989) introduced the con-
structs in the original TAM (see Figure 1)
as follows: perceived usefulness (PU),
perceived ease of use (PEOU), attitude,
and behavioral intention to use. Among
the constructs, PU and PEOU form an
end-user’s beliefs on a technology and
therefore predict his or her attitude toward
the technology, which in turn predicts its
acceptance.
Davis (1989) conducted numerous
experiments to validate TAM by using
PEOU and PU as two independent vari-
ables and system usage as the dependent
variable. He found that PU was signifi-
cantly correlated with both self-reported
current usage and self-predicted future
 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 61
Figure 1: The Original Technology Acceptance Model
Perceived
Usefulness
External
Variables
Perceived
Ease of Use
usage. PEOU was also significantly cor-
related with current usage and future us-
age. Overall, he found that PU had a sig-
nificantly greater correlation with system
usage than did PEOU. Further regression
analysis suggested that PEOU might be an
antecedent of PU rather than a direct de-
terminant of system usage. That is, PEOU
affects technology acceptance (TA) indi-
rectly through PU. Figure 2 shows the vali-
dated TAM.
In the last decade, TAM has received
considerable attention and empirical sup-
port (e.g., Davis, 1989; Mathieson 1991;
Taylor & Todd, 1995a). We estimate that
there were about 100 studies, published in
journals, proceedings, or technical reports,
related to TAM between 1989 and 2001.
In these studies, TAM was extensively
tested using different sample sizes and user
groups within or across organizations, an-
alyzed with different statistical tools, and
compared with competing models (Gefen,
2000). It was applied to many different
end-user technologies such as email
(Adams, Nelson & Todd, 1992; Davis,
1989), word processors (Adams, Nelson
& Todd, 1992; Davis, Bagozzi & Warshaw,
Figure 2: A Validated Technology Acceptance Model
Perceived
Ease of Use
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
Attitude Behavioral System
Toward Use Intention to Use Usage
1989), groupware (Taylor & Todd, 1995b),
spreadsheets (Agarwal, Sambamurthy &
Stair, 2000; Mathieson, 1991), and World
Wide Web (Lederer, Maupin, Sena &
Zhuang, 2000). Some studies also extend-
ed TAM by including additional predictors
such as gender, culture, experience, and
self-efficacy. Overall, researchers tend to
suggest that TAM is valid, parsimonious,
and robust (Venkatesh & Davis, 2000).
Davis (1989) developed and validat-
ed the scales for PEOU and PU and found
six highly reliable items for each construct
with a Cronbach’s alpha of .98 for PU and
.94 for PEOU respectively. In succeeding
studies, the measurement items for these
constructs varied from researcher to re-
searcher (Adams, Nelson & Todd, 1992).
As a result, the cumulative number of
items for measuring PU has increased
from the original six to currently about 50,
and that for PEOU has increased from six
to 38. Appendix shows nine different in-
struments for PU and PEOU employed in
the existing studies1. Upon closer scrutiny
of the list, we found that the differences in
measurement items between studies tend
to be the result of adapting TAM to differ-
Perceived Technology
Usefulness Acceptance
62 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004
ent technologies. The essential definitions
of the constructs to be measured are still
the same. Therefore, we conclude that the
empirical findings on the relationships be-
tween the constructs in TAM are not af-
fected much by how the constructs are
measured.
Existing empirical findings on TAM
are not consistent and conclusive (Moore
& Benbasat, 1991). For instance, some
studies indicated that PEOU has no signif-
icant impact on TA, while others found that
such an impact is significant (Hendrickson
& Collins, 1996; Subramanian, 1994;
Venkatesh & Davis, 1996). Many studies
found that the impact of PEOU on PU is
stronger than that of PEOU on TA, whereas
others found a much larger effect of PEOU
on TA than PU (Lim, 2001). More per-
plexing is the fact that even in the same
study, when the subjects were tested with
different applications, PEOU was nega-
tively related to TA in some cases, but pos-
itively in others (Adams, Nelson & Todd,
1992).
With regard to the divergent findings,
many possible explanations have been pro-
vided. However, they tend to be qualita-
tive and subjective. What is needed, we
believe, is to integrate these findings and
generate a quantitative and objective syn-
thesis.
RESEARCH METHODOLOGY
Meta-analysis is defined as the “sta-
tistical analysis of a collection of analysis
results from individual studies for the pur-
pose of integrating the findings”
(DerSimonian & Laird, 1986). An individ-
ual test typically provides summary statis-
tics that indicate the significance of the test
results. In meta-analysis, we need to con-
vert the statistics into a common metric
called effect size, which is usually in the
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
form of the Pearson Product Moment Cor-
relation. Essentially, an effect size repre-
sents the degree to which the phenome-
non is present in the population (Cohen,
1977). In this section, we explain how we
select individual studies for our meta-anal-
ysis and how we estimate the effect size
for each sample study.
Selection of Individual Studies
To be included in our meta-analysis,
a study has to meet four requirements: 1)
it involved empirical testing of TAM direct-
ly or indirectly; 2) it reported a sample size;
3) it reported correlation coefficients be-
tween the constructs of TAM or other val-
ues that can be converted to correlations;
4) it was published or dated after 1989,
which is the year TAM was first published.
One widely documented concern on
selecting sample studies for meta-analysis
is the publication bias. It is well known that
journals are more likely to publish research
results that are statistically significant.
Therefore, effect sizes in published jour-
nals are larger than those that have not
been published. Because one of the pro-
cedures in meta-analysis is to average the
effect sizes of individual studies, the result
may be inflated if only published studies
are reviewed (Schafer, 1999). To avoid this
so-called file drawer problem, we
searched for related studies at all levels of
publications including refereed journals,
unpublished dissertations, and conference
proceedings. In this study, our source pa-
per collection was inclusive. We included
all the related studies in major journals such
as MIS Quarterly, Information Systems
Research, Information and Manage-
ment, etc. We searched digital libraries of
the Association for Computing Machinery
and the Association for Information Sys-
tems as well as major international con-
 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 63
ference proceedings. We also searched
academic databases such as ProQuest,
EBSCO, and ResearchIndex at Google.
One of the important assumptions in
meta-analysis is the independence of indi-
vidual findings; effect sizes such as corre-
lations in different studies are statistically
independent. This assumption is frequent-
ly violated because some studies often re-
port more than one correlation or effect
size based on the same sample
(Martinussen & Bjjornstad, 1999). To ob-
serve this assumption, when we selected
studies and calculated the effect sizes, we
carefully checked the sample to make sure
they were not based on the same data. If
multiple tests based on the same sample
were conducted, we selected only one of
them and recorded its statistics.
In this project, initially we found a
total of 91 empirical studies. Among them,
65 studies did not report correlation coef-
ficients or other statistics that we could
convert into correlation coefficients. Thus,
we dropped those studies and selected the
remaining ones. Among the 26 selected
studies, seven of them are working papers
or published in conference proceedings.
Since some studies reported test results
based on multiple samples (e.g., Davis,
1989; Subramanian, 1994), we obtained 102
correlation coefficients in total from the 26
selected studies.
Estimation of Effect Sizes
It is natural that different studies may
report different statistics such as correla-
tion coefficients, F, t, or chi-square values.
Consequently, the results may not be com-
parable enough to provide an insight into
the strength of a relationship or the effect
of interest (Wolf, 1986). Therefore, it is
necessary to convert different statistics into
a common metric before conducting a
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
meta-analysis. In this study, the Pearson
Product Moment Correlation is used as the
index of effect size to represent the em-
pirical strength of a relationship between
each pair of the constructs in TAM. We
selected the statistic because of its ease
of interpretation and the availability of for-
mulae for converting other test statistics
into correlation coefficients (Lipsey &
Wilson, 2000). In addition to effect size,
we also encoded the sample size for each
study and whether an effect size is posi-
tive and statistically significant.
For each pair of the constructs in
TAM: PU, PEOU, and TA, we calculated
the effect sizes as follows: an effect size
is simply a correlation coefficient, if it was
reported; otherwise, it is obtained through
a conversion by using a formula. For ex-
ample, if a t-value is reported, we convert-
ed it into a correlation using the formula
t2
r = , where df is the degree of
t 2 + df
freedom. Wolf (1986) provides guidelines
for converting the most common test sta-
tistics to r. Cohen (1965, 1977), Friedman
(1968), Glass, McGaw ans Smith (1981),
and Rosenthal (1984) discuss the conver-
sion process and provide guidelines for
transforming some less common statistics.
This procedure is widely used in many oth-
er studies (e.g., Szymanski & Henard,
2001).
DATA ANALYSIS
In this section, we analyze the data
and report findings in two steps. First, we
describe the correlations in terms of range,
direction, statistical significance, and sam-
ple size. These data will reflect the nature
and diversity of the existing findings on
TAM. Second, we present the findings
64 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004

from the univariate analysis of the corre-
lations. The purpose here is to show the
central tendencies of the existing findings
and their statistical significance.
Descriptive Statistics
Based on the 26 selected studies, we
obtained 102 correlations as summarized
in Table 1. Note that not all studies report-
ed all the three correlations or equivalents.
Among the 102 correlations, 37 PU-TA
correlations were obtained from 21 stud-
ies, 32 PEOU-TA correlations from 20
studies, and 33 PEOU-PU correlations
from 21 studies. The number of studies for
each of the three relationships is approxi-
mately the same. According to the range
of correlation coefficients, it is easy to see
that the strength of each relationship var-
ies greatly from insignificant to strongly
significant. For instance, the correlation
between PEOU and PU changes from
0.003 to 0.92. In addition, the correlation
coefficient between PU and TA were in-
significant in some instances, although most
studies found otherwise.
As expected, most studies reported
positive significant findings and few non-
significant or negative ones. According to
Table 1, the percentage of positive signifi-
cant correlations of PEOU-PU is the high-
est among the three relationships, while
PEOU-TA has the highest percentage of
positive non-significant correlation. The
sample size varies from study to study. In
some studies, the sample size is as low as
36; while in others, it is as high as 1370. Of
course, the extreme cases are few in num-
ber. The average sample size indicates that
the number of subjects used in the select-
ed studies is very close across all three
relationships.
The Analysis of Direct Effect
In Table 2, we reported the findings
on the mean effect sizes using three meth-
ods: simple mean, sample size adjusted-
mean (Mosteller & Bush, 1954), and the
Fisher r to Z transformation (Fisher, 1932).
A simple mean is simply the average of all
individual effect sizes. A sample size ad-
justed-mean is the sample size-weighted

Table 1: A Summary of Selected Correlations

# of Range of Positive Positive Cumula-

# of Range of
Cor- Correlation Significant Insignificant tive
Link Stud Sample Sizes
rela- Coefficients Correlations Correlations Sample
ies
tions From To # % # % From To Average Size
PU-TA 21 37 .09 .91 23 62.2 2 5.4 36 1370 179 6058
PEOU-
20 32 .07 .59 17 53.0 12 37.5 36 1370 194 5744
TA
PEOU-
21 33 .003 .92 21 63.6 6 18.2 39 1370 169 5421
PU

Table 2: Means and Variances of Correlations

Sample-Size- Simple Mean Correlation Sample Fail Safe Confidence

Link
Adjusted Correlation Correlation from Zr Variance Nfs.05 Interval

PU-TA 0.4113 0.49 0.54 0.0323 131 (.41, .57)

PEOU-TA 0.2759 0.27 0.28 0.0172 -0.7 (.21, .33)

PEOU-PU 0.4679 0.50 0.54 0.0380 71 (.41, .59)

Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 65
average of individual effect sizes, i.e.,
r= ∑ (N r ) ∑ N
i i i , where Ni and ri are
the sample size and the effect size of test
i, respectively. To use the Fisher r to Z
transformation, three steps are followed
(Wolf 1986). First, each correlation is trans-
formed into Fisher’s Z score using the for-
1+ r
mula Z = 0.5 * ln ( ) , where r is an
1− r
individual correlation coefficient. Next, we
compute the sample size weighted aver-
age of the individual Z scores for each pair
of the constructs in TAM. Finally, we con-
vert the weighted average Z score back
into a correlation coefficient.
There are some discussions regard-
ing these methods. Some suggest the ne-
cessity of using the Fisher r to Z transfor-
mation in meta-analysis, while others feel
there is not much difference between sim-
ple mean and the Fisher r to Z transforma-
tion (Wolf, 1986). Schmidt, Gast-
Rosenberg and Hunter (1980) discussed
the issue and reported a study based the
Fisher transformation. In the current study,
we employed both techniques.
It is also commonly believed that cor-
relations estimated from larger samples and
more reliable data sources can produce a
mean correlation closer to the population
mean, all else being equal (Hunter &
Schmidt, 1990; Szymanki & Henard, 2001).
Thus, it is desirable to calculate reliability-
adjusted mean. However, we found that it
is difficult to do so due to the fact that many
source studies failed to report reliability
data. Therefore, in the current paper, we
chose sample-size-weighted mean instead
of reliability-adjusted mean. Szymanki and
Henard (2001) did the same in their recent
meta-analysis on customer satisfaction.
Table 2 shows some differences that
result from using different methods. First,
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
we see that the Fisher r to Z transforma-
tion consistently results in larger means
than the other two methods. This inflation
phenomenon has been reported previously
(Hunter, Schmidt & Jackson, 1982;
Schmidt, Gast-Rosenberg & Hunter, 1980).
Second, we found that the mean effect siz-
es obtained using the Fisher r to Z trans-
formation and simple average are almost
identical, while the results from the sam-
ple-size-adjusted method are smaller. We
rechecked sample sizes and re-calculated
the means, and found that some extreme
sample sizes have an apparent effect on
the means. For example, the correlations
for a study with a sample size of 1,370 are
all relatively small (between 0.31 and 0.37).
When it is removed from the meta-analy-
sis, the averages become larger and com-
parable with those obtained from the other
two methods. It indicates that the sample
size adjusted method may not be appropri-
ate for the current study. Thus, we will
interpret the results of this study based on
the Fisher r to Z transformation method.
According to Cohen (1977), the mag-
nitude of an effect size is small when it is
close to 0.10, medium when it is close to
0.30, and large when it is close to 0.50.
By this rough guideline, our meta-analysis
suggests a medium-sized effect for the re-
lationship between PEOU and TA, and
large effect sizes for PU-TA and PEOU-
PU. Also, note that the effect sizes for
PU-TA and PEOU-PU are almost identi-
cal to each other. It is different from the
general perception; our study does not sug-
gest that the PU-TA relationship is stron-
ger than the PEOU-PU relationship.
To show the statistical significance of
the mean effects, we computed the 99%
confidence intervals for each mean esti-
mate, based on the assumption that indi-
vidual correlations are normally distribut-
ed. These intervals portray the range of
66 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004
effects that might exist in the true popula-
tion given the presence of errors and vari-
ation in the calculation of sample effects.
According to the results in Table 2, no in-
terval contains zero, which therefore sug-
gests that all three mean effects are sig-
nificantly different from zero.
To further test the significance of
the findings, given the possibility that we
may miss the studies that report null ef-
fects (r = 0), we calculated the fail-safe
N for p = 0.05 using the formula
N fs .05 = (ΣZ
1.645 − , where ΣZ is
)2 N
the sum of individual Z scores and N is the
number of tests. A fail-safe N represents
the number of additional studies confirm-
ing the null hypothesis (r = 0) that would
be needed to reverse a conclusion that a
significant relationship exists (Cooper,
1979). Table 2 shows that the mean corre-
lations for PEOU-PU and PU-TA are sig-
nificantly different from zero, to the extent
that 71-131 of null effects would have to
exist to bring the respective mean estimates
down to a level not considered statistically
significant. However, the mean correlation
for PEOU-TA does not pass the fail-safe
test as indicated by the negative Nfs.05.
As we pointed out before in this study,
we selected the individual correlations re-
ported for the model rather than the aver-
age of the correlations reported within a
study. The former is often referred to as
individual-level analysis, while the latter is
considered study-level analysis. Hunter
and Schmidt (1990) raised the possibility
that an individual-level analysis might un-
derestimate the sampling-error variance
and the generalizability of the estimates.
To address this concern, we computed the
variance due to sampling error and stan-
dard deviation for each relationship. The
results show that the variances of sam-
pling-error are very close to each other,
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
and therefore suggest that an individual
level analysis is appropriate within the con-
text of this meta-analysis.
CONCLUSIONS
This meta-analysis was designed to
synthesize and analyze the existing empir-
ical findings on the Technology Acceptance
Model (TAM). It examined the relation-
ships in TAM with a larger sample size,
which is impossible to achieve in a tradi-
tional empirical study. The results of our
study in general confirm Davis’ original
findings: Among the three constructs in
TAM, both the relationships between
PEOU and PU, and between PU and TA
are strong, while the relationship between
PEOU and TA is weak. Here we mea-
sured the strengths of the relationship from
three perspectives. First, with respect to
the magnitude of a mean effect, we found
the mean effects for PEOU-PU and PU-
TA are large, while that for PEOU-TA is
medium. Second, with respect to the sta-
tistical significance of a mean effect, we
found that all three mean effects are sig-
nificantly positive at the level α = 0.01.
Finally, with respect to the fail-safe test
significance, we found that between 71-
131 null effects would have to be hidden
away in file drawers for the mean correla-
tions between PEOU and PU, and between
PU and TA to be non-significant, which
seems unlikely. However, the mean ef-
fect for PEOU and TA does not pass the
fail-safe test, in the sense that one addi-
tional study reporting a null effect would
lead to the effect being non-significant.
This study contributes to information
technology (IT) management in several
important ways. First, usefulness is shown
to be critical for IT adoption based on the
accumulated evidence. It implies that de-
velopers should focus on system
 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 67
functionalities and features to improve the
acceptance of a system to be developed.
Second, the relationship between ease of
use and usefulness cannot be ignored.
Therefore, when IT professionals devel-
op, test, or adopt a new system, they should
keep in mind that the ease of use of the
system has a strong impact on the end-
users’ perception of its usefulness.
Of course, when interpreting or ap-
plying the results of this research, some
caution is advised. As with any other re-
search methodology, meta-analysis has its
assumptions and limitations. One of the
major difficulties to applying meta-analy-
sis to the studies on TAM is that the find-
ings of many previous researchers are gen-
erated by multivariate analyses such as
multiple regression, factor analysis, and
structural equation modeling. Meta-ana-
lysts have not yet developed effect size
statistics that adequately represent this
form of research findings (Lipsey & Wil-
son, 2000). Consequently, many sample
studies were dropped from our list. Other-
wise, the results of this study would be
more accurate.
A second limitation of this study is
the lack of the analysis of potential moder-
ators. Besides documenting the distribu-
tion, central tendencies, and magnitudes of
correlations, a meta-analysis can identify
whether the variation of correlations is due
to chance or measurement and method
factors. Although we planned to do the
analysis on moderating effects and collect-
ed data on potential moderators including
the technologies applied, the types of par-
ticipants, and whether the researchers use
an experimental or survey approach, the
small size of the selected studies prevent-
ed us from running multiple regressions to
investigate their possible effects.
Our finding on the relationship be-
tween PEOU and TA is uncertain. More
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
future studies are needed to resolve the
uncertainty. One possible direction is to
investigate whether the relationship is mod-
erated by a third variable such as gender,
culture, experience, self-efficacy, complex-
ity of a technology, or the state of knowl-
edge on a technology. Several existing
studies do suggest the possible existence
of moderating effects. Gefen and Straub
(1997) found that although women tend to
perceive the PEOU of email to be higher
than men, their usage of email is actually
less than that of men. In their more recent
study with Internet technology (Gefen &
Straub, 2000), they found that PEOU in-
fluences TA when a web site involves in-
quiries, but does not influence TA when a
web site is used for a purchasing task. Sim-
ilarly, in a study of examining PEOU with
objected-oriented analysis technique, Liu
and Grandon (2002) found that, compared
to the subjects with only partial training in
structured technique, those without train-
ing felt more positive about the PEOU of
object-oriented techniques, but performed
worse in object-oriented analysis tasks;
those with full training also felt more posi-
tive about their PEOU, but performed bet-
ter too. Finally, Chircu, Davis, & Kauffman
(2000) studied the role of expertise in the
adoption of electronic commerce interme-
diaries and found that transaction complex-
ity is an important moderator for the rela-
tionships between PEOU and PU and be-
tween PU and TA. By synthesizing these
existing findings, we noticed a common phe-
nomenon that individual and task charac-
teristics moderate the PEOU-TA relation-
ship. Gefen and Straub (1997) and Liu and
Grandon (2002) emphasized individual dif-
ferences such as gender and experience,
whereas Gefen and Straub (2000) and
Chircu, Davis and Kauffman (2000)
stressed task characteristics. Therefore, we
suggest the future research should look into
68 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004
Appendix: Sample Measurement Scales for PU and PEOU
Study PU
Davis (1989) -Work more quickly
-Job performance
-Increase productivity
-Effectiveness
-Makes Job Easier
-Useful
Gefen & Keil (1998) -Using … enables me to accomplish
configuration tasks more quickly
-Using … improves the quality of the
work I do
-Using … improves my job performance
-Overall, I find … to be advantageous in
my job
-Using … increases my sales productiv-
ity
Heijden (2000) -I find … primarily a useful site
-The information on the site is interesting
to me
-I find this a site that adds value
Agarwal & Prasad -Accomplish tasks more quicktly
(1999) -Improve my job performance
-Give me greater control over my work
-Improve the quality of the work I do
-Improve my productivity
-Make it easier to do my job
-Is useful in my job
Davis, Bagozzi & -Using … would improve my perform-
Warshaw (1989) ance
-Using … would enhance my effective-
ness
-Using … would increase my productiv-
ity
-I would find … useful
Gefen & Straub (2000) -ABC improves my performance in book
searching
-ABC enables me to search and buy
books faster
-ABC enhances my effectiveness in book
searching and buying
--ABC makes it easier to search for and
purchase books
-ABC increases my productivity in
searching and purchasing books
Igbaria, Iivari & Mara- -Using … improves my job performance
gahh (1995) -Using … increases my productivity
-I find … useful in my job
-Using … enhances my effectiveness in
the job
-Using … provides me with information
that would lead to better decisions
Venkatesh & Davis -Using … would improve my perform-
(1996) ance in my degree program
-Using … in my degree program would
increase my productivity
-Using … would enhance my effective-
ness in my degree program
-I find … would be useful in my degree
program
Straub, Limayem & -Voice mail is very important in per-
Karahanna (1995) forming my job
-My decision making is more effective
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
PEOU Technology
-Easy to learn Lab experiment with email
-Clear and understandable and graphics
-Easy to become skillful
-Easy to use
-Controllable
-Flexible
-Learning to operate … was easy for me Field study with expert sys-
-Using … is clear and understandable tem CONFIG
-I believe that … is easy to use
-It is easy to navigate around the site Field study with a web site
-I can quickly find the information that I
need
-I think it is a user-friendly site
-It is easy for me to remember how to Field study with GUI envi-
perform tasks ronment
-It is easy to get … to do what I want it
to do
-My interaction with … is clear and
untreatable
-Overall it is easy to use
-Learning to operate … would be easy Lab experiment with a word
for me processor
-I would find it easy to get … to do what
I want it to do
-It would be easy for me to become
skillful at using …
-I would find … easy to use
-ABC is easy to learn Lab experiment with an on-
-My interaction with ABC is clear and line bookstore
understandable
-It is easy to become skillful at using
ABC
-Learning to operate ABC is easy
-It is easy to interact with ABC
-ABC is flexible to interact with
-Learning to use … would be easy for Field study with microcom-
me puters
-I would find it easy to get … to do what
I want to do
-It would be easy for me to become
skillful at using …
-I would find … easy to use
-My interaction with … is clear and Lab experiment with PC and
understandable word processor
-Interacting with … does not require a lot
of my mental effort
-I find … easy to use
-I find it easy to get … to do what I want
it to do
-I find it easy to get voice mail to do Field study with voice mail
what I want it to do
-I feel very comfortable using voice mail
 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 69
these characteristics in order to better un-
derstand the weak PEOU-TA contingency.
ENDNOTES
1
Many studies used one of these in-
struments with no modification. For exam-
ple, Adams, Nelson and Todd (1992),
Gefen & Straub (1997), and Chircu, Davis
& Kauffman (2000) used the same instru-
ment as in Davis (1989). Similarly, Jack-
son, Chow and Leitch (1997), Subramanian
(1994), and Szajna (1996) used the same
instrument as in Davis, Bagozzi and
Warshaw (1989). Some other studies did
not report their instruments but mentioned
that they adopted an existing one in the
list.
REFERENCES
(*Studies whose findings are included in the meta-
analyses.)
*Adams, D.A., Nelson, R.R. & Todd,
P.A. (1992). Perceived Usefulness, Ease
of Use, and Usage of Information. MIS
Quarterly, 16(2), 227-250.
Agarwal, R. & Prasad, J. (1999). Are
Individual Differences Germane to the Ac-
ceptance of New Information Technolo-
gies? Decision Sciences, 30(2), 361-391.
Agarwal, R., Sambamurthy, V. &
Stair, R.M. (2000). The Evolving Relation-
ship Between General and Specific Com-
puter Self-Efficacy – An Empirical As-
sessment. Information Systems Research,
1(4), 418-430.
Benbasat, I. & Zmud, R. (1999).
Empirical Research in Information Systems:
The Practice of Relevance. MIS Quar-
terly, 23(1), 3-16.
*Chau, P.Y.K. (1996). An Empirical
Assessment of a Modified Technology
Acceptance Model. Journal of Manage-
ment Information Systems, 13(2), 185-204
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
*Chircu, A.M., Davis, G.B. &
Kauffman, R.J. (2000). The Role of Trust
and Expertise in the Adoption of Elec-
tronic Commerce Intermediaries. Tech-
nical Report, WP 00-07, Carlson School
of Management, University of Minnesota,
Minneapolis, MN.
Cohen, J. (1965). Some statistical is-
sues in psychological research, In B.
Wolman (ed.) Handbook of Clinical Psy-
chology. New York, NY: Academic Press.
Cohen, J. (1977). Statistical Power
Analysis for the Behavioral Sciences.
New York: Academic Press.
Cooper, H.M. (1979) Statistically
Combining Independent Studies: A Meta-
Analysis of Sex Differences in Conformi-
ty Research. Journal of Personality and
Social Psychology 37, 131-146.
Davis, F.D. (1986). A Technology
Acceptance Model for Empirically Test-
ing New End-user information Systems:
Theory and Results. Doctoral Disserta-
tion, MIT Sloan School of Management,
Cambridge, MA.
*Davis, F.D. (1989). Perceived Use-
fulness, Perceived Ease Of Use, And User
Acceptance. MIS Quarterly, 13(3), 319-
340.
*Davis, F.D. (1993). User Accep-
tance of Information Technology: System
Characteristics, User Perceptions and
Behavioral Impacts. International Jour-
nal of Man-Machine Studies, 38(3), 475-
487.
*Davis, F.D. & Venkatesh, V. (1996).
A Critical Assessment of Potential Mea-
surement Biases in the Technology Accep-
tance Model: Three Experiments. Inter-
national Journal of Human-Computer
Studies, 45(1), 19-45.
Davis, F.D., Bagozzi, R.P. &
Warshaw, P.R. (1989). User Acceptance
of Computer Technology: A Comparison
of Two. Management Science, 35(8), 982-
1001.
70 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004
DerSimonian, R. & Laird, N.M.
(1986). Meta-Analysis in Clinical Trials.
Controlled Clinical Trials, 7, 177-188.
Fishbein, M. & Azjen, I. (1975). Be-
lief, Attitude, Intention and Behavior.
Reading, MA: Addison-Wesley.
Fisher, R. A. (1932). Statistical
Methods for Research Workers. London,
UK: Oliver and Boyd.
Friedman, H. (1968). Magnitude of
experimental effect and a table for its rap-
id estimation. Psychological Bulletin, 70,
245-251.
Gefen, D. (2000). Structural Equation
Modeling and Regression: Guidelines for
Research Practice. CAIS, 4(7).
*Gefen, D. & Keil, M. (1998). The
Impact of Developer Responsiveness on
Perceptions of Usefulness and Ease of
Use: An Extension of the Technology Ac-
ceptance Model. The Data Base for Ad-
vances in Information Systems, 29(2), 35-
49.
Gefen, D. & Straub, D.W. (1997).
Gender Differences in the Perception and
Use of E-mail: An Extension to the Tech-
nology Acceptance Model. MIS Quarter-
ly, 21(4), 389-400.
*Gefen, D. & Straub, D. (2000). The
Relative Importance of Perceived Ease-
of-Use in IS Adoption: A Study of e-Com-
merce Adoption. Journal of the Associ-
ation for Information Systems, 1(8).
Glass, G.V., McGaw, B. & Smith,
M.L. (1981). Meta-Analysis in Social
Research, Sage Publications, Beverly Hills,
CA.
Heijden, H. (2000). Using the Tech-
nology Acceptance Model to Predict
Website Usage: Extensions and Empir-
ical Test. Technical Report, No. 0025,
http://ideas.uqam.ca/ideas/data/Papers/
dgrvuarem2000-25.html.
*Hendrickson, A.R. & Collins, M.R.
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
(1996). An Assessment of Structure and
Causation of IS Usage, The DATA BASE
for Advances in Information Systems,
27(2), 61-67.
Hunter, J.E. & Schmidt, F.L. (1990).
Methods of Meta Analysis, Sage Publi-
cations, Newbury Park, CA.
Hunter, J.E., Schmidt, F.L. & Jack-
son, G.B. (1982). Meta-Analysis: Cumu-
lating Research Findings Across Stud-
ies, Sage Publications, Beverly Hills, CA.
*Igbaria, M., Iivari, J. & Maragahh,
H. (1995). Why Do Individuals Use Com-
puter Technology? A Finnish case study.
Information and Management, 29(5),
227-238.
*Igbaria, M., Guimaraes, T. & Davis,
G.B. (1995). Testing the Determinants of
Microcomputer Usage via a Structural
Equation Model,” Journal of Manage-
ment Information Systems, 11(4), 87-114.
*Igbaria, M., Zinatelli, N., Cragg, P.
& Cavaye, A.L.M. (1997). Personal Com-
puting Acceptance Factors in Small Firms:
A Structural Equation Model. MIS Quar-
terly, 21(3), 279-305.
*Igbaria, M., Parasuraman, S. &
Baroudi, J. (1996). A Motivational Model
of Microcomputer Usage. Journal of
Management Information Systems, 13(1),
127-143.
*Jackson, C.M., Chow, S. & Leitch,
R.A. (1997). Towards an Understanding
of the Behavioral Intention to Use an In-
formation System. Decision Sciences,
28(2), 357-389.
*Lederer, A.L., Maupin, D.J., Sena,
M.P. & Zhuang, Y. (2000). The technolo-
gy Acceptance Model and the World Wide
Web. Decision Support Systems, 29(3),
269-292.
Lim, K-S. (2001). An Empirical Test
of The Technology Acceptance Model.
Proceedings of Decision Science Insti-
 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004 71
tute Annual Meeting, San Francisco, CA.
Lipsey, M.W. & Wilson,D.B. (2000).
Practical Meta-Analysis, Applied Social
Research Methods Series, Volume 49,
Thousand Oaks, CA: Sage Publications.
Liu, L. & Grandon, E.E. (2002). An
Empirical Study of How Prior Training
in Structured Analysis Affects the Per-
ceptions on Object Orientation. Tech-
nical Report, Department of Management,
University of Akron, Akron, OH.
*Malhotra, Y. & Galletta, D.F. (1999).
Extending the Technology Acceptance
Model to Account for Social Influence:
Theoretical Bases and Empirical Valida-
tion, Proceedings of the 32nd Hawaii
International Conference on System
Sciences, Honolulu, HI.
Martinussen, M. & Bjornstad, J.F.
(1999). Meta-Analysis Calculations Based
on Independent and Nonindependent Cas-
es. Educational and Psychological Mea-
surement, 59(6), 928-950.
*Mathieson, K. (1991). Predicting
User Intentions: Comparing the Technolo-
gy Acceptance Model with the Theory of
Planned Behavior. Information Systems
Research, 2(3), 173-191.
Moore, G.C. & Benbasat, I. (1991).
Development of an Instrument to Measure
the Perceptions of Adopting an Informa-
tion Technology Innovation. Information
Systems Research, 2, 192-222.
Mosteller, F. M. & Bush, R. R. (1954).
Selected quantitative techniques. In G.
Lindzey (ed.) Handbook of Social Psy-
chology, Vol. 1. Cambridge, MA: Addison-
Wesley.
Robey, D. & Markus, L. M. (1998).
Beyond Rigor and Relevance: Producing
Consumable Research about Information
Systems. Information Resources Man-
agement Journal, 11(1), 7-15.
*Rose, G. & Straub, D.W. (1998).
Predicting General IT Use: Applying TAM
Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.
to the Arabic World. Journal of Global
Information Management, 6(3), 39-47.
Rosenthal, R. (1984). Meta-Analyti-
cal Procedures for Social Research.
Beverly Hills, CA: Sage.
Schafer, W.D. (1999). Methods, Plain-
ly Speaking—An Overview of Meta-Anal-
ysis, Measurement and Evaluation in
Counseling and Development, 32(1), 43-
61.
Schmidt, F.L., Gast-Rosenberg, I. &
Hunter, J.E. (1980). Validity Generaliza-
tion: Results for Computer Programmers.
Journal of Applied Psychology, 65(6),
643-661.
*Segars, A. H. & Grover, V. (1993).
Re-Examining Perceived Ease of Use and
Usefulness: A Confirmatory Factor Anal-
ysis. MIS Quarterly, 17(4), 517-525.
Straub, D.W., Limayem, M. &
Karahanna, E. (1995). Measuring System
Usage: Implications for IS Theory Test-
ing. Management Science, 41(8), 1328-
1342.
*Straub, D.W., Keil, M. & Brennan,
W. (1997). Testing the Technology Accep-
tance Model across Cultures: A Three
Country Study. Information and Man-
agement, 33(1), 1-11.
*Subramanian, G.H. (1994). A Rep-
lication of Perceived Usefulness and Per-
ceived Ease of Use Measurement. Deci-
sion Sciences, 25(5-6), 863-872.
*Szajna, B. (1996) Empirical Evalua-
tion of the Revised Technology Accep-
tance Model. Management Science,
42:(1), 85-92.
Szymanki, D.M. & Henard, D.H.
(2001). Customer Satisfaction: A Meta-
Analysis of the Empirical Evidence. Jour-
nal of the Academy of Marketing Sci-
ence, 29(1), 16-35.
*Taylor, S. & Todd, P. (1995a). Un-
derstanding Information Technology Us-
age: A Test of Competing Models. Infor-
 72 Journal of Organizational and End User Computing, 16(1), 59-72, Jan-Mar 2004

mation Systems Research, 6(2), 144-173. cision Sciences, 27(3), 451-480.

*Taylor, S. & Todd, P. (1995b). “As-
sessing IT Usage: the Role of Prior Expe-
rience. MIS Quarterly, 19(4), 561-570.
*Teo, T.S.H., Lim V.K.G. & Lai,
R.Y.C. (1999). Intrinsic and Extrinsic Mo-
tivation in Internet Usage. Omega, 27(1),
25-37.
*Venkatesh, V. & Davis, F.D. (1996).
A Model of the Antecedents of Perceived
Ease of Use: Development and Test. De-
*Venkatesh, V. & Davis, F.D. (2000).
A Theoretical Extension of the Technolo-
gy Acceptance Model: Four Longitudinal
Field Studies. Management Science,
46(2), 186-204.
Wolf, F. M. (1986). Meta-Analysis—
Quantitative Methods for Research Syn-
thesis. Newbury Park, CA: Sage Publica-
tions.

Qingxiong Ma is an assistant professor of Computer Information Systems at Central

Missouri State University, and currently a Ph.D. candidate in Management
Information Systems at Southern Illinois University at Carbondale. He received
his MBA from Eastern Illinois University, Charleston, IL, in 1999. His research
interests include information technology adoption/diffusion, electronic commerce,
and information security management. He has published an article in International
Journal of Healthcare Technology and Management and presented numerous
papers at the America’s Conference on Information Systems and Decision Sciences
Institute Annual Meetings. Information Systems classes he has taught include:
Systems Analysis and Design, Data Communication and Networks, Management
of Information Systems, and E-commerce.
.

Liping Liu is an Associate Professor of Information Systems at the University of

Akron. He received his Master in Systems Engineering from Huazhong University
of Science and Technology, Wuhan, China, in 1991, and his Ph.D. in Business
from the University of Kansas, USA, in 1995. His research interests include
electronic business, data quality, systems analysis and design, and uncertainty
reasoning in expert systems. His articles have appeared in Decision Support
Systems, European Journal of Operational Research, Information Systems Frontier,
International Journal of Approximate Reasoning, Journal of Risk and Uncertainty,
and others. He has served as a guest editor for International Journal of Intelligent
Systems. He has strong practical and teaching interests in Client/Server and Web-
Based Systems Design and Development using advanced DBMS, CASE, and RAD
tools. He has won several teaching awards.

Copyright © 2004, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group
Inc. is prohibited.

View publication stats