Davis 1989 TAM
Davis 1989 TAM
Davis 1989 TAM
discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/200085965
CITATIONS
DOWNLOADS
VIEWS
8,644
5,543
1,119
2 AUTHORS, INCLUDING:
Fred D. Davis
University of Arkansas
69 PUBLICATIONS 32,530 CITATIONS
SEE PROFILE
Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology
Author(s): Fred D. Davis
Source: MIS Quarterly, Vol. 13, No. 3 (Sep., 1989), pp. 319-340
Published by: Management Information Systems Research Center, University of Minnesota
Stable URL: http://www.jstor.org/stable/249008 .
Accessed: 06/02/2014 14:35
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 [email protected].
Management Information Systems Research Center, University of Minnesota is collaborating with JSTOR to
digitize, preserve and extend access to MIS Quarterly.
http://www.jstor.org
Perceived
Perceived
and
Use,
Usefulness,
of
Ease
User
Acceptance
of
Information
Technology
Introduction
By: Fred D. Davis
Computer and Information Systems
Graduate School of Business
Administration
University of Michigan
Ann Arbor, Michigan 48109
Abstract
Validmeasurement scales for predicting user
acceptance of computers are in short supply.
Most subjective measures used in practice are
unvalidated, and their relationship to system
usage is unknown. The present research develops and validates new scales for two specific variables, perceived usefulness and perceived ease of use, which are hypothesized to
be fundamental determinants of user acceptance. Definitionsfor these two variables were
used to develop scale items thatwere pretested
for content validityand then tested for reliability
and construct validityin two studies involving
a total of 152 users and four applicationprograms. Themeasures were refinedand streamlined, resultingin two six-item scales with reliabilities of .98 for usefulness and .94 for ease
of use. The scales exhibited high convergent,
and factorialvalidity.Perceivedusediscriminant,
fulness was significantlycorrelatedwithbothselfreported current usage (r=.63, Study 1) and
self-predictedfutureusage (r= .85, Study2). Perceived ease of use was also significantlycorrelated with currentusage (r=.45, Study 1) and
futureusage (r=.59, Study 2). In both studies,
usefulness had a significantlygreater correlation with usage behavior than did ease of use.
Regression analyses suggest that perceived
ease of use may actually be a causal antece-
Information
technologyoffersthe potentialforsubstantially improving white collar performance
(Curley, 1984; Edelman, 1981; Sharda, et al.,
1988). But performance gains are often obstructed by users' unwillingnessto accept and
use available systems (Bowen, 1986; Young,
1984). Because of the persistence and importance of this problem, explaining user acceptance has been a long-standingissue in MIS
research (Swanson, 1974; Lucas, 1975; Schultz
and Slevin, 1975; Robey, 1979; Ginzberg,1981;
Swanson, 1987). Althoughnumerousindividual,
organizational,and technologicalvariableshave
been investigated(Benbasat and Dexter, 1986;
Franz and Robey, 1986; Markus and BjornAnderson, 1987; Robey and Farrow,1982), research has been constrained by the shortage
of high-qualitymeasures for key determinants
of user acceptance. Past research indicatesthat
many measures do not correlate highly with
system use (DeSanctis, 1983; Ginzberg, 1981;
Schewe, 1976; Srinivasan, 1985), and the size
of the usage correlationvaries greatlyfromone
study to the next depending on the particular
measures used (Baroudi,et al., 1986; Barkiand
Huff,1985; Robey, 1979; Swanson, 1982, 1987).
The developmentof improvedmeasures for key
theoreticalconstructs is a research priorityfor
the informationsystems field.
Aside fromtheir theoreticalvalue, better measures for predictingand explaining system use
would have great practicalvalue, both for vendors who would like to assess user demand for
new design ideas, and for informationsystems
managers withinuser organizationswho would
like to evaluate these vendor offerings.
Unvalidatedmeasures are routinelyused in practice today throughout the entire spectrum of
design, selection, implementationand evaluation
activities.For example: designers withinvendor
organizationssuch as IBM(Gould,et al., 1983),
Xerox(Brewley,et al., 1983), and DigitalEquip-
Theoretical Foundations
The theoreticalimportanceof perceived usefulness and perceivedease of use as determinants
of user behavioris indicatedby several diverse
lines of research. The impactof perceived usefulness on system utilizationwas suggested by
the workof Schultzand Slevin (1975) and Robey
(1979). Schultz and Slevin (1975) conductedan
exploratoryfactor analysis of 67 questionnaire
items, which yielded seven dimensions. Of
these, the "performance"
dimension,interpreted
by the authors as the perceived "effect of the
model on the manager'sjob performance,"was
most highlycorrelatedwithself-predicteduse of
a decision model (r=.61). Usingthe Schultzand
Slevinquestionnaire,Robey (1979) findsthe performance dimensionto be most correlatedwith
two objectivemeasures of system usage (r=.79
and .76). Buildingon Vertinsky,et al.'s (1975)
expectancy model, Robey (1979) theorizes that:
"A system that does not help people perform
their jobs is not likelyto be received favorably
Self-efficacytheory
The importanceof perceived ease of use is supported by Bandura's(1982) extensive research
on self-efficacy, defined as "judgmentsof how
well one can execute courses of action required
to deal withprospectivesituations"(p. 122). Selfefficacy is similarto perceived ease of use as
definedabove. Self-efficacybeliefs are theorized
to functionas proximaldeterminantsof behavior. Bandura'stheory distinguishes self-efficacy
judgments from outcome judgments, the latter
being concerned with the extent to which a behavior,once successfully executed, is believed
to be linkedto valued outcomes. Bandura's"outcome judgment"variableis similarto perceived
usefulness. Bandura argues that self-efficacy
and outcome beliefs have differingantecedents
and that, "Inany given instance, behaviorwould
be best predicted by considering both selfefficacy and outcome beliefs" (p. 140).
Hill,et al. (1987) findthat both self-efficacyand
outcome beliefs exert an influenceon decisions
Cost-benefitparadigm
The cost-benefitparadigmfrombehavioraldecision theory (Beach and Mitchell,1978; Johnson
and Payne, 1985; Payne, 1982) is also relevant
to perceived usefulness and ease of use. This
research explains people's choice among various decision-makingstrategies (such as linear
compensatory,conjunctive,disjunctiveand elmination-by-aspects)in terms of a cognitivetradeoff betweenthe effortrequiredto employthe strategy and the quality (accuracy) of the resulting
decision. This approach has been effective for
explainingwhy decision makersaltertheirchoice
strategies in response to changes in task complexity.Althoughthe cost-benefit approach has
mainly concerned itself with unaided decision
making, recent work has begun to apply the
same form of analysis to the effectiveness of
informationdisplay formats (Jarvenpaa, 1989;
Kleinmuntzand Schkade, 1988).
Cost-benefitresearch has primarilyused objective measures of accuracyand effortin research
studies, downplayingthe distinctionbetween objective and subjective accuracy and effort. Increased emphasison subjectiveconstructsis warranted, however, since (1) a decision maker's
choice of strategy is theorized to be based on
subjectiveas opposed to objectiveaccuracyand
effort(Beach and Mitchell,1978), and (2) other
research suggests that subjectivemeasures are
often in disagreementwiththeirojbective counterparts(Abelson and Levi, 1985; Adelbrattand
Montgomery,1980; Wright,1975). Introducing
measures of the decision maker'sown perceived
costs and benefits, independentof the decision
actually made, has been suggested as a way
of mitigating
criticismsthatthe cost/benefit framework is tautological(Abelson and Levi, 1985).
The distinctionmade herein between perceived
usefulness and perceived ease of use is similar
to the distinctionbetween subjective decisionmakingperformanceand effort.
321
Adoptionof innovations
Research on the adoption of innovations also
suggests a prominentrole for perceived ease
of use. Intheir meta-analysisof the relationship
between the characteristicsof an innovationand
its adoption,Tornatzkyand Klein(1982) findthat
compatibility,relative advantage, and complexity have the most consistent significantrelationships across a broad range of innovationtypes.
Complexity,defined by Rogers and Shoemaker
(1971) as "the degree to which an innovation
is perceived as relativelydifficultto understand
and use" (p. 154), parallels perceived ease of
use quite closely. As Tornatzkyand Klein(1982)
pointout, however,compatibilityand relativeadvantage have both been dealt with so broadly
and inconsistentlyin the literatureas to be difficult to interpret.
Evaluationof information
reports
Past research within MIS on the evaluation of
informationreports echoes the distinctionbetween usefulness and ease of use made herein.
Larckerand Lessig (1980) factor analyzed six
items used to rate fourinformation
reports.Three
items load on each of two distinct factors: (1)
perceived importance,whichLarckerand Lessig
define as "the qualitythat causes a particular
informationset to acquire relevance to a decision maker,"and the extent to which the informationelements are "a necessary inputfortask
accomplishment," and (2) perceived usableness, which is defined as the degree to which
"the informationformat is unambiguous, clear
or readable"(p. 123). These two dimensionsare
similarto perceived usefulness and perceived
ease of use as defined above, repsectively,although Larckerand Lessig refer to the two dimensions collectivelyas "perceivedusefulness."
Reliabilitiesfor the two dimensions fall in the
range of .64-.77, short of the .80 minimallevel
recommended for basic research. Correlations
with actual use of informationreportswere not
addressed in their study.
Channeldispositionmodel
Swanson (1982, 1987) introducedand tested a
model of "channeldisposition"forexplainingthe
choice and use of informationreports.The concept of channel dispositionis defined as having
Non-MISstudies
Outside the MIS domain, a marketingstudy by
Hauserand Simmie (1981) concerninguser perceptions of alternativecommunicationtechnologies similarlyderivedtwo underlyingdimensions:
ease of use and effectiveness, the latterbeing
similarto the perceived usefulness constructdefined above. Bothease of use and effectiveness
were influentialin the formationof user preferences regardinga set of alternativecommunication technologies. The human-computerinteraction (HCI) research community has heavily
emphasized ease of use in design (Branscomb
and Thomas, 1984; Card, et al., 1983; Gould
and Lewis, 1985). For the most part, however,
these studies have focused on objective measures of ease of use, such as task completion
time and errorrates. In many vendor organizations, usabilitytesting has become a standard
phase in the product development cycle, with
Convergenceof findings
There is a strikingconvergence among the wide
range of theoretical perspectives and research
studies discussed above. Although Hill, et al.
(1987) examined learninga computerlanguage,
Larckerand Lessig (1980) and Swanson (1982,
1987) dealt with evaluating informationreports,
and Hauser and Simmie (1981) studied communicationtechnologies,all are supportiveof the
conceptualand empiricaldistinctionbetweenusefulness and ease of use. The accumulatedbody
of knowledgeregardingself-efficacy,contingent
decision behavior and adoption of innovations
provides theoreticalsupport for perceived usefulness and ease of use as key determinants
of behavior.
From multipledisciplinaryvantage points, perceived usefulness and perceived ease of use
are indicatedas fundamentaland distinctconstructsthat are influentialin decisions to use informationtechnology. Althoughcertainlynot the
only variables of interest in explaininguser behavior (for other variables, see Cheney, et al.,
1986; Davis, et al., 1989; Swanson, 1988), they
do appear likelyto play a centralrole. Improved
measures are needed to gain furtherinsightinto
the nature of perceived usefulness and perceived ease of use, and their roles as determinants of computer use.
others. By asking individualsto rate the similarity of items to one another, we can performa
clusteranalysis to determinethe structureof the
substrata,remove items where excess coverage
is suggested, and add items where inadequate
coverage is indicated.
Pretest participantsconsisted of a sample of 15
experienced computer users from the Sloan
School of Management,MIT,includingfive secretaries, five graduate students and five members of the professionalstaff. In face-to-face interviews,participantswere asked to performtwo
tasks, prioritizationand categorization, which
were done separately for usefulness and ease
of use. For prioritization,they were first given
a card containingthe definitionof the targetconstructand asked to read it. Next,they were given
13 index cards each having one of the items
forthat constructwrittenon it. The 14thor "overall" item for each construct was omitted since
its wordingwas almost identicalto the label on
the definitioncard (see Tables 1 and 2). Participants were asked to rankthe 13 cards according to how well the meaning of each statement
matched the given definitionof ease of use or
usefulness.
For the categorization task, participantswere
asked to put the 13 cards intothree to five categories so that the statements withina category
were most similarin meaning to each other and
dissimilarin meaning from those in other categories. This was an adaptationof the "owncategories" procedure of Sherif and Sherif (1967).
Categorizationprovidesa simple indicantof similaritythat requiresless time and effortto obtain
than other similaritymeasurement procedures
such as paid comparisons. The similaritydata
was cluster analyzed by assigning to the same
clusteritems thatseven or more subjects placed
in the same category. The clusters are considered to be a reflectionof the domain substrata
for each constructand serve as a basis of assessing coverage, or representativeness,of the
item pools.
The resultingrankand cluster data are summarized in Tables 3 (usefulness) and 4 (ease of
use). Forperceivedusefulness, noticethat items
fall into three main clusters. The firstcluster relates to job effectiveness, the second to productivityand time savings, and the thirdto the importance of the system to one's job. If we
eliminate the lowest-rankeditems (items 1, 4,
5 and 9), we see that the three majorclusters
each have at least two items. Item 2, "control
Item
Job DifficultWithout
ControlOver Work
Job Performance
Addresses My Needs
Saves Me Time
WorkMoreQuickly
Criticalto MyJob
AccomplishMoreWork
Cut UnproductiveTime
Effectiveness
Qualityof Work
Increase Productivity
Makes Job Easier
Useful
New
Item #
Rank
13
9
2
12
11
7
5
6
10
1
3
4
8
NA
2
6
3
4
7
8
1
5
9
10
Cluster
C
A
C
B
B
C
B
B
A
A
B
C
NA
Item
Confusing
ErrorProne
Frustrating
Dependence on Manual
MentalEffort
ErrorRecovery
Rigid& Inflexible
Controllable
Unexpected Behavior
Cumbersome
Understandable
Ease of Remembering
Provides Guidance
Easy to Use
Ease of Learning
Effortto Become Skillful
Study 1
A field study was conducted to assess the reliability,convergent validity,discriminantvalidity,
and factorialvalidityof the 10-item scales resultingfromthe pretest. A sample of 120 users
withinIBMCanada'sTorontoDevelopmentLaboratorywere given a questionnaireasking them
to rate the usefulness and ease of use of two
systems availablethere: PROFS electronicmail
and the XEDITfile editor. The computingenvironmentconsisted of IBMmainframesaccessible through327X terminals.The PROFS electronic mail system is a simple but limited
messaging facility for brief messages. (See
Panko, 1988.) The XEDITeditoris widely avail-
Rank
7
13
3
9
5
10
6
1
11
2
4
8
12
NA
NA
NA
New
Item #
Cluster
B
3
(replace)
7
B
C
B
5
4
A
A
1
8
6
(replace)
10
2
9
A
B
C
C
NA
NA
NA
the sample, 10 percent were managers, 35 percent were administrativestaff, and 55 percent
were professionalstaff (whichincludeda broad
mixof marketanalysts,productdevelopmentanalysts, programmers,financial analysts and research scientists).
Reliabilityand validity
The perceived usefulness scale attained Cronbach alpha reliabilityof .97 for both the electronicmailand XEDITsystems, while perceived
ease of use achieved a reliabilityof .86 for electronic mail and .93 for XEDIT.When observations were pooled for the two systems, alpha
was .97 for usefulness and .91 for ease of use.
Convergentand discriminantvaliditywere tested
(MTMM)analysis
using multitrait-multimethod
(Campbelland Fiske, 1959). The MTMMmatrix
of items (methods)
containsthe intercorrelations
appliedto the two differenttest systems (traits),
electronic mail and XEDIT.Convergentvalidity
refers to whether the items comprisinga scale
behave as if they are measuringa common underlyingconstruct.In orderto demonstrateconvergent validity,items that measure the same
trait should correlate highly with one another
(Campbelland Fiske, 1959). That is, the elements in the monotraittriangles (the submatrix
of intercorrelationsbetween items intended to
measure the same construct for the same
system) withinthe MTMMmatrices should be
large.Forperceivedusefulness,the 90 monotraitheteromethodcorrelationswere all significantat
the .05 level. For ease of use, 86 out of 90,
correor 95.6%, of the monotrait-heteromethod
lationswere significant.Thus, our data supports
the convergent validityof the two scales.
Discriminantvalidityis concerned with the ability of a measurement item to differentiatebetween objects being measured. For instance,
withinthe MTMMmatrix,a perceived usefulness
item appliedto electronicmail should not correlate too highly with the same item applied to
XEDIT.Failureto discriminatemay suggest the
presence of "commonmethod variance,"which
means that an item is measuringmethodological
artifactsunrelatedto the target construct(such
as individualdifferences in the style of respondingto questions (see Campbell,et al., 1967; Silk,
1971) ). The test for discriminantvalidityis that
an item should correlatemore highlywith other
items intended to measure the same traitthan
with either the same item used to measure a
Scale refinement
In applied testing situations, it is importantto
keep scales as brief as possible, particularly
when multiplesystems are going to be evaluated. The usefulness and ease of use scales
were refined and streamlinedbased on results
from Study 1 and then subjected to a second
roundof empiricalvalidationin Study2, reported
below. Applyingthe Spearman-Brownprophecy
formulato the .97 reliabilityobtained for perceived usefulness indicatesthat a six-itemscale
composed of items having comparable reliability would yield a scale reliabilityof .94. The five
positive ease of use items had a reliabilityof
.92. Taken together, these findingsfrom Study
1 suggest that six items would be adequate to
achieve reliabilitylevels above .9 while maintaining adequate validitylevels. Based on the
results of the field study, six of the 10 items for
each construct were selected to form modified
scales.
For the ease of use scale, the five negatively
worded items were eliminateddue to their apparentcommon method variance, leaving items
2, 4, 6, 8 and 10. Item 6 ("easy to remember
how to performtasks"), which the pretest indicated was concerned withease of learning,was
replaced by a reversal of item 9 ("easy to
become skillful"),which was specifically designed to more directly tap ease of learning.
These items include two from cluster C, one
each fromclusters A and B, and the overallitem.
(See Table 4.) In order to improverepresentative coverage of the content domain, an additional A item was added. Of the two remaining
A items (#1, Cumbersome, and #5, Rigid and
Inflexible),item5 is readilyreversedto form"flexible to interactwith."This item was added to
form the sixth item, and the order of items 5
and 8 was permutedin order to prevent items
fromthe same cluster (items 4 and 5) fromappearing next to one another.
In order to select six items to be used for the
usefulness scale, an item analysis was performed. Corrected item-totalcorrelationswere
computed for each item, separately for each
system studied. Average Z-scores of these correlationswere used to rankthe items. Items 3,
5, 6, 8, 9 and 10 were top-rankeditems. Referring to the cluster analysis (Table 3), we see
that this set is well-representativeof the content
domain, includingtwo items fromcluster A, two
from cluster B and one from cluster C, as well
as the overall item (#10). The items were permuted to prevent items from the same cluster
fromappearingnext to one another.The result-
Factor 1
(Usefulness)
.80
Factor 1
(Ease of Use)
.10
.86
-.03
3
4
5
WorkMoreQuickly
Criticalto MyJob
Increase Productivity
.79
.87
.87
.17
-.11
.10
Job Performance
.93
-.07
.91
.96
.80
.74
-.02
-.03
.16
.23
.00
.08
.02
.13
.09
.17
-.07
.29
-.25
.23
.73
.60
.65
.74
.54
.62
.76
.64
.88
.72
7
8
9
10
Ease
1
2
3
4
5
6
7
8
9
10
AccomplishMoreWork
Effectiveness
Makes Job Easier
Useful
of Use
Cubersome
Ease of Learning
Frustrating
Controllable
Rigid& Inflexible
Ease of Remembering
MentalEffort
Understandable
Effortto Be Skillful
Easy to Use
Relationshipto use
Participants were asked to self-report their
degree of currentusage of electronic mail and
XEDITon six-position categorical scales with
boxes labeled "Don'tuse at all,""Use less than
once each week," "Use aboutonce each week,"
"Use several times a week," "Use about once
each day," and "Use several times each day."
Usage was significantlycorrelatedwithboth perceived usefulness and perceived ease of use
for both PROFS mail and XEDIT.PROFS mail
usage correlated.56 with perceived usefulness
and .32 with perceived ease of use. XEDIT
usage correlated .68 with usefulness and .48
withease of use. Whendata were pooled across
systems, usage correlated .63 with usefulness
and .45 withease of use. The overallusefulnessuse correlationwas significantlygreaterthan the
ease of use-use correlationas indicated by a
test of dependent correlations (t181=3.69,
p<.001) (Cohen and Cohen, 1975). Usefulness
and ease of use were significantlycorrelated
with each other for electronicmail (.56), XEDIT
Study 2
A lab study was performedto evaluate the sixitem usefulness and ease of use scales resulting from scale refinementin Study 1. Study 2
was designed to approximateapplied prototype
testing or system selection situations,an important class of situations where measures of this
kind are likelyto be used in practice. In prototype testing and system selection contexts, prospective users are typicallygiven a briefhandson demonstrationinvolvingless than an hourof
actually interactingwith the candidate system.
Thus, representativeusers are asked to rate the
future usefulness and ease of use they would
expect based on relativelylittleexperience with
the systems being rated. We are especially interested in the propertiesof the usefulness and
ease of use scales when they are worded in
a prospective sense and are based on limited
experience with the target systems. Favorable
psychometric properties under these circumstances would be encouraging relative to their
use as early warning indicants of user acceptance (Ginzberg,1981).
The lab study involved40 voluntaryparticipants
who were evening MBAstudents at Boston University. They were paid $25 for participatingin
the study. They had an average of five years'
workexperience and were employed full-timein
several industries,includingeducation (10 percent), government(10 percent),financial(28 percent), health (18 percent),and manufacturing(8
percent). They had a range of priorexperience
with computers in general (35 percent none or
limited;48 percent moderate; and 17 percent
extensive) and personal computers in particular
(35 percent none or limited;48 percent moderate; and 15 percent extensive) but were unfamiliarwith the two systems used in the study.
330
The study involved evaluating two IBM PCbased graphics systems: Chart-Master(by Decision Resources, Inc.of Westport,CN) and Pendraw (by Pencept, Inc. of Waltham,MA).ChartMaster is a menu-drivenpackage that creates
numericalbusiness graphs, such as bar charts,
line charts, and pie charts based on parameters
defined by the user. Throughthe keyboardand
menus, the user inputsthe data for, and defines
the desired characteristicsof, the chart to be
made. The user can specify a wide variety of
options relatingto titlefonts, colors, plotorientation, cross-hatching pattern, chart format, and
so on. The chart can then be previewedon the
screen, saved, and printed.Chart-Masteris a
successful commercialproductthat typifies the
categoryof numericbusiness chartingprograms.
Pendrawis quite differentfromthe typicalbusiness chartingprogram.Ituses bit-mappedgraphics and a "directmanipulation"interfacewhere
users draw desired shapes using a digitizer
tablet and an electronic "pen"as a stylus. The
digitizertablet supplants the keyboard as the
inputmedium. By drawingon a tablet, the user
manipulatesthe image, which is visible on the
screen as it is being created. Pendraw offers
capabilities typical of PC-based, bit-mapped
"paint"programs (see Panko, 1988), allowing
the user to performfreehanddrawingand select
from among geometric shapes, such as boxes,
lines, and circles. A varietyof line widths, color
selections and title fonts are available. The
digitizeris also capable of performingcharacter
recognition,converting hand-printercharacters
into various fonts (Ward and Blesser, 1985).
Pencept had positionedthe Pendrawproductto
complete with business chartingprograms.The
manualintroducesPendrawby guidingthe user
throughthe process of creating a numericbar
chart. Thus, a key marketing issue was the
extent to whichthe new productwouldcompete
favorablywithestablishedbrands,such as ChartMaster.
Participantswere given one hour of hands-on
experience with Chart-Masterand Pendraw,
using workbooksthat were designed to follow
the same instructionalsequence as the user
manuals for the two products,while equalizing
the style of writingand eliminatingvalue statements (e.g., "See how easy that was to do?").
Half of the participantstried Chart-Masterfirst
and half tried Pendraw first. After using each
package, a questionnairewas completed.
Reliabilityand validity
Cronbachalpha was .98 for perceived usefulness and .94 for perceived ease of use. Convergent validitywas supported,withonly two of
72 monotrait-heteromethodcorrelations falling
below significance. Ease of use item 4 (flexibilwas not significantly
ity),appliedto Chart-Master,
correlatedwith either items 3 (clear and understandable) or 5 (easy to become skillful).This
suggests that, contraryto conventionalwisdom,
flexibilityis not always associated with ease of
use. As Goodwin(1987) pointsout, flexibilitycan
actually impair ease of use, particularly for
novice users. With item 4 omitted, Cronbach
alpha for ease of use would increase from .94
to .95. Despite the two departures to convergent validityrelated to ease of use item 4, no
exceptions to the discriminantvaliditycriteriaoccurred across a total of 720 comparisons (360
for each scale).
Factorialvaliditywas assessed by factor analyzingthe 12 scale items using principalcomponents extraction and oblique rotation.The resultingtwo-factorsolutionis very consistent with
distinct,unidimensionalusefulness and each of
use scales (Table7). Thus, as in Study 1, Study
2 reflects favorablyon the convergent, discriminant, and factorialvalidityof the usefulness and
ease of use scales.
Relationshipto use
Participants were asked to self-predict their
future use of Chart-Masterand Pendraw. The
Factor 1
(Usefulness)
.91
Factor 2
(Ease of Use)
.01
Job Performance
.98
-.03
3
4
Increase Productivity
Effectiveness
.98
.94
-.03
.04
.95
-.01
6
Ease
1
2
3
4
5
6
Useful
of Use
Easy to Learn
Controllable
Clear & Understandable
Flexible
Easy to Become Skillful
Easy to Use
.88
.11
-.20
.19
-.04
.13
.07
.09
.97
.83
.89
.63
.91
.91
331
*** p<.001
.56***
.68***
.63***
.32***
.48***
.45***
.56***
.69***
.64***
.71***
.59***
.85***
.25
.47***
.59***
.25
.38**
.56***
.65***
.70***
.27**
.12
.10
.23**
* p<.05
**
p<.01
R2
.55***
.69***
.57***
.01
.02
.07
.31
.46
.38
.69***
.76***
.75***
.08
.17
.17*
.51
.71
.74
.62***
.71**
.20***
-.06
.45
.49
p<.05
332
Ease of Use
& Usefulness
throughusefulness. Partialcorrelationanalysis
indicates that the variance in usage explained
by ease of use drops by 91% when usefulness
is controlledfor. Consistent with Study 1, these
regressionand partialcorrelationresultssuggest
that usefulness mediates the effect of ease of
use on usage. The implicationsof this are addressed in the followingdiscussion.
Discussion
The purposeof this investigationwas to develop
and validate new measurement scales for perceived usefulness and perceived ease of use,
two distinctvariables hypothesizedto be deter-
minantsof computerusage. This effortwas successful inseveral respects. The new scales were
found to have strong psychometric properties
and to exhibitsignificantempiricalrelationships
with self-reportedmeasures of usage behavior.
Also, several new insightswere generated about
the natureof perceived usefulness and ease of
use, and their roles as determinants of user
acceptance.
The new scales were developed, refined, and
streamlined in a several-step process. Explicit
definitionswere stated, followedby a theoretical
analysis from a variety of perspectives, including: expectancy theory;self-efficacytheory;behavioraldecision theory;diffusionof innovations;
marketing;and human-computerinteraction,regardingwhy usefulness and ease of use are hypothesized as importantdeterminantsof system
use. Based on the stated definitions,initialscale
items were generated. To enhance content validity,these were pretested in a small pilotstudy,
and several items were eliminated.The remaining items, 10 foreach of the two constructs,were
tested for validityand reliabilityin Study 1, a
field study of 112 users and two systems (the
PROFS electronic mail system and the XEDIT
file editor).Itemanalysis was performedto eliminate moreitemsand refineothers,furtherstreamliningand purifyingthe scales. The resultingsixitem scales were subjected to furtherconstruct
validationin Study 2, a lab study of 40 users
and two systems: Chart-Master(a menu-driven
business chartingprogram)and Pendraw(a bitmapped paint programwith a digitizertablet as
its inputdevice).
The new scales exhibitedexcellent psychometric characteristics.Convergentand discriminant
validity were strongly supported by multitraitmultimethodanalyses in both validationstudies.
These two data sets also providedstrong supportforfactorialvalidity:the patternof factorloadings confirmedthat a prioristructureof the two
instruments,withusefulness items loadinghighly
on one factor, ease of use items loading highly
on the other factor, and small cross-factorloadfor perceiveduseings. Cronbachalphareliability
fulness was .97 in Study 1 and .98 in Study 2.
Reliabilityfor ease of use was .91 in Study 1
and .94 in Study 2. These findingsmutuallyconfirmthe psychometricstrengthof the new measurementscales.
As theorized, both perceived usefulness and
ease of use were significantly
correlatedwithselfreportedindicantsof system use. Perceiveduse-
fulness was correlated.63 withself-reportedcurrent use in Study 1 and .85 with self-predicted
use in Study 2. Perceived ease of use was correlated .45 withuse in Study 1 and .69 in Study
2. The same pattern of correlations is found
when correlationsare calculated separately for
each of the two systems in each study (Table
8). These correlations,especiallythe usefulnessuse link,compare favorablywith other correlations between subjective measures and selfreporteduse found in the MIS literature.Swanson's (1987) "value"dimension correlated .20
withuse, while his "accessibility"dimensioncorrelated .13 with self-reporteduse. Correlations
between "userinformationsatisfaction"and selfreporteduse of .39 (Barkiand Huff, 1985) and
.28 (Baroudi,et al., 1986) have been reported.
"Realismof expectations"has been foundto be
correlated .22 with objectively measured use
(Ginzberg,1981) and .43 withself-reporteduse
force"was
(Barkiand Huff,1985). "Motiviational
correlated.25 withsystem use, objectivelymeasured (DeSanctis, 1983). Among the usage correlationsreportedin the literature,the .79 correlationbetween "performance"and use reported
by Robey (1979) stands out. Recallthat Robey's
expectancy model was a key underpinningfor
the definitionof perceived usefulness stated in
this article.
One of the most significantfindings is the relative strength of the usefulness-usage relationship compared to the ease of use-usage relationship. In both studies, usefulness was
significantlymore stronglylinkedto usage than
was ease of use. Examiningthe jointdirecteffect
of the two variableson use in regression analyses, this differencewas even more pronounced:
the usefulness-usage relationship remained
large, while the ease of use-usage relationship
was diminished substantially(Table 8). Multicollinearityhas been ruled out as an explanation for the results using specific tests for the
presence of multicollinearity.In hindsight, the
prominence of perceived usefulness makes
sense conceptually:users are driven to adopt
an applicationprimarilybecause of the functions
it performsfor them, and secondarily for how
easy or hard it is to get the system to perform
those functions. For instance, users are often
willingto cope with some difficultyof use in a
system that provides criticallyneeded functionality. Althoughdifficultyof use can discourage
adoption of an otherwise useful system, no
amount of ease of use can compensate for a
334
Research implications
Futureresearch is needed to address how other
variables relate to usefulness, ease of use, and
acceptance. Intrinsicmotivation,for example,
has received inadequate attentionin MIStheories. Whereas perceived usefulness is concerned withperformanceas a consequence use,
intrinsicmotivationis concerned with the reinforcementand enjoymentrelatedto the process
of performinga behaviorper se, irrespectiveof
whatever external outcomes are generated by
such behavior (Deci, 1975). Although intrinsic
motivationhas been studiedin the design of computergames (e.g., Malone, 1981), it is just beginningto be recognized as a potentialmechanism underlyinguser acceptance of end-user
Acknowledgements
This research was supportedby grantsfromthe
MITSloan School of Management,IBMCanada
Ltd., and The Universityof MichiganBusiness
School. The author is indebted to the anonymous associate editor and reviewers for their
many helpfulsuggestions.
References
Abelson, R.P. and Levi,A. "DecisionMakingand
Decision Theory,"in The Handbookof Social
Psychology, thirdedition, G. Lindsayand E.
Aronson (eds.), Knopf,New York,NY, 1985,
pp. 231-309.
MultitmethodMatrix,"Psychological Bulletin
(56:9), March1959, pp. 81-105.
Campbell,D.T., Siegman, C.R. and Rees, M.B.
Effects in the Relation"Direction-of-Wording
ships Between Scales," Psychological Bulletin (68:5), November 1967, pp. 293-303.
Card,S.K., Moran,T.P. and Newell,A. The Psychology of Human-Computer Interaction,
Erlbaum,Hillsdale,NJ, 1984.
Wheels
Carroll,J.M.and Carrithers,C. "Training
in a User Interface,"Communicationsof the
ACM (27:8), August 1984, pp. 800-806.
Carroll, J.M. and McKendree, J. "Interface
Design Issues for Advice-GivingExpertSystems," Communicationsof the ACM (30:1,
January 1987, pp. 14-31.
Carroll,J.M., Mack,R.L.,Lewis,C.H.,Grishkowsky, N.L. and Robertson,S.R. "ExploringExploringa WordProcessor,"Human-Computer
Interaction(1), 1985, pp. 283-307.
Carroll,J.M. and Thomas, J.C. "Fun,"SIGCHI
Bulletin(19:3), January 1988, pp. 21-24.
Cats-Baril,W.L.and Huber,G.P. "DecisionSupport Systems for Ill-StructuredProblems:An
EmpiricalStudy," Decision Sciences (18:3),
Summer 1987, pp. 352-372.
Cheney,P.H., Mann,R.I.and Amoroso,D.L."OrganizationalFactors Affectingthe Success of
End-User Computing,"Journal of Management Information Systems (3:1), Summer
1986, pp. 65-80.
Chin, J.P., Diehl, V.A. and Norman, K.L. "Development of an Instrumentfor Measuring
InUser Satisfactionof the Human-Computer
terface," CHI'88Human Factors in Computing Systems, Washington,D.C., May 15-19,
1988, ACM,New York,NY, pp. 213-218.
Cohen, J. and Cohen, P. Applied MultipleRegression/ CorrelationAnalysis for the Behavioral Sciences, Erlbaum,Hillsdale,NJ, 1975.
Curley,K.F. "AreThere any Real Benefits from
Office Automation?"Business Horizons (4),
July-August1984, pp. 37-42.
Davis, F.D. "A Technology Acceptance Model
for EmpiricallyTesting New End-User Information Systems: Theory and Results," doctoral dissertation,MITSloan School of Management, Cambridge,MA, 1986.
Davis, F.D., Bagozzi, R.P. and Warshaw, P.R.
User Acceptance of ComputerTechnology:A
Comparisonof Two TheoreticalModels,"Management Science (35:8), August 1989, pp.
982-1003.
Davis, J.A. The Logic of Causal Order, Sage,
Beverly Hills,CA, 1985.
Deci, E.L. Intrinsic Motivation, Plenum, New
York,NY, 1975.
DeSanctis, G. "ExpectancyTheoryas an Explanationof VoluntaryUse of a Decision Support
System," Psychological Reports (52), 1983,
pp. 247-260.
Dickson,G.W., DeSanctis, G. and McBride,D.J.
the Effectivenessof Computer
"Understanding
Graphicsfor Decision Support:A Cumulative
ExperimentalApproach,"Communicationsof
the ACM (29:1), January 1986, pp. 40-47.
Edelmann, F. "Managers,ComputerSystems,
and Productivity,"MIS Quarterly(5:3), September 1981, pp. 1-19.
Fishbein, M. and Ajzen, I. "Belief, Attitude,Intention and Behavior: An Introduction to
Theoryand Research,"Addison-Wesley,Reading, MA 1975.
ConFranz,C.R. and Robey, D. "Organizational
text, User Involvement,and the Usefulness of
InformationSystems," Decision Sciences
(17:3), Summer 1986, pp. 329-356.
Gallupe,R.B., DeSanctis, G. and Dickson,G.W.
"Computer-BasedSupportfor GroupProblem
MISQuarFinding:An EmpiricalInvestigation,"
terly (12:2), June 1988, pp. 277-296.
Ginzberg,M.J."EarlyDiagnosisof MISImplementation Failure:Promising Results and Unanswered Questions," Management Science
(27:4), April1981, pp. 459-478.
Good, M.,Spine, T.M.,Whiteside,J. and George
P. "User-DerivedImpactAnalysis as a Tool
for UsabilityEngineering,"CHI'86HumanFactors in ComputingSystems, Boston, April1317, 1986, ACM,New York,New Yorkpp. 241246.
and Usability,"
ComGoodwin,N.C."Functionality
municationsof the ACM (30:3), March1987,
pp. 229-233.
Goslar, M.D. "CapabilityCriteriafor Marketing
Decision Support Systems," Journalof Management InformationSystems (3:1), Summer
1986, pp. 81-95.
Gould, J., Conti, J. and Hovanyecz, T. "Composing letterswitha SimulatedListeningTypewriter,"Communicationsof the ACM (26:4),
April1983, pp. 295-308.
Gould,J.D. and Lewis C. "Designingfor Usability:Key Principlesand WhatDesignersThink,"
Communicationsof the ACM (28:3), March
1985, pp. 300-311.
Greenberg,K. "ExecutivesRate TheirPCs,"PC
World,September 1984, pp. 286-292.
Hauser, J.R. and Simmie, P. "ProfitMaximizing
PerceptualPositions:An IntegratedTheoryfor
the Selection of ProductFeatures and Price,"
Appendix
likely
likely
I
I
I
I
I
1neither
quite extremely
slightly
extremely quite
slightly
Using CHART-MASTER in my job would increase my productivity.
I
I
I
I
I
neither
slightly
quite extremely
slightly
extremely quite
would
enhance
effectiveness
CHART-MASTER
on the job.
my
Using
I
extremely
likely
neither
quite
slightly
quite extremely
slightly
Using CHART-MASTERwould make it easier to do my job.
I
I
I
I
I
I
II
neither
quite
slightly
slightly
extremely quite
extremely
I would find CHART-MASTER useful in my job.
likely
II
extremely
unlikely
unlikely
unlikely
unlikely
I
extremely
unlikely
unlikely
neither
slightly
quite
quite
slightly
likely
I
I
I
I
I
I
II
neither
quite
slightly
slightly
extremely quite
extremely
My interaction with CHART-MASTERwould be clear and understandable.
unlikely
likely
I
I
I
I
I
I
I
neither
quite extremely
slightly
slightly
extremely quite
I would find CHART-MASTERto be flexible to interact with.
unlikely
likely
I
I
I
I
I
I
I
neither
quite
slightly
slightly
extremely
extremely quite
It would be easy for me to become skillful at using CHART-MASTER.
I
I
I
I
I
I
I
I
neither
quite
slightly
slightly
extremely quite
extremely
unlikely
likely
unlikely
340
I
extremely
I
quite
slightly
I
neither
I
slightly
I
quite
I
extremely
unlikely