Your SlideShare is downloading. ×
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Venkatesh2003
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Venkatesh2003

409

Published on

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
409
On Slideshare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3, pp. 425-478/September 2003 425RESEARCH ARTICLEUSER ACCEPTANCE OF INFORMATIONTECHNOLOGY: TOWARD A UNIFIED VIEW1By: Viswanath VenkateshRobert H. Smith School of BusinessUniversity of MarylandVan Munching HallCollege Park, MD 20742U.S.A.vvenkate@rhsmith.umd.eduMichael G. MorrisMcIntire School of CommerceUniversity of VirginiaMonroe HallCharlottesville, VA 22903-2493U.S.A.mmorris@virginia.eduGordon B. DavisCarlson School of ManagementUniversity of Minnesota321 19thAvenue SouthMinneapolis, MN 55455U.S.A.gdavis@csom.umn.eduFred D. DavisSam M. Walton College of BusinessUniversity of ArkansasFayetteville, AR 72701-1201U.S.A.fdavis@walton.uark.edu1Cynthia Beath was the accepting senior editor for thispaper.AbstractInformation technology (IT) acceptance researchhas yielded many competing models, each withdifferent sets of acceptance determinants. In thispaper, we (1) review user acceptance literatureand discuss eight prominent models, (2) empiri-cally compare the eight models and their exten-sions, (3) formulate a unified model that integrateselements across the eight models, and (4) empiri-cally validate the unified model. The eight modelsreviewed are the theory of reasoned action, thetechnology acceptance model, the motivationalmodel, the theory of planned behavior, a modelcombining the technology acceptance model andthe theory of planned behavior, the model of PCutilization, the innovation diffusion theory, and thesocial cognitive theory. Using data from fourorganizations over a six-month period with threepoints of measurement, the eight models ex-plained between 17 percent and 53 percent of thevariance in user intentions to use informationtechnology. Next, a unified model, called theUnified Theory of Acceptance and Use of Tech-nology (UTAUT), was formulated, with four coredeterminants of intention and usage, and up tofour moderators of key relationships. UTAUT wasthen tested using the original data and found tooutperform the eight individual models (adjustedR2of 69 percent). UTAUT was then confirmedwith data from two new organizations with similarresults (adjusted R2of 70 percent). UTAUT thusprovides a useful tool for managers needing to
  • 2. Venkatesh et al./User Acceptance of IT426 MIS Quarterly Vol. 27 No. 3/September 2003assess the likelihood of success for new techno-logy introductions and helps them understand thedrivers of acceptance in order to proactively de-sign interventions (including training, marketing,etc.) targeted at populations of users that may beless inclined to adopt and use new systems. Thepaper also makes several recommendations forfuture research including developing a deeperunderstanding of the dynamic influences studiedhere, refining measurement of the core constructsused in UTAUT, and understanding the organiza-tional outcomes associated with new technologyuse.Keywords: Theory of planned behavior, inno-vation characteristics, technology acceptancemodel, social cognitive theory, unified model,integrated modelIntroductionThe presence of computer and information tech-nologies in today’s organizations has expandeddramatically. Some estimates indicate that, sincethe 1980s, about 50 percent of all new capitalinvestment in organizations has been in informa-tion technology (Westland and Clark 2000). Yet,for technologies to improve productivity, they mustbe accepted and used by employees in organi-zations. Explaining user acceptance of new tech-nology is often described as one of the mostmature research areas in the contemporary infor-mation systems (IS) literature (e.g., Hu et al.1999). Research in this area has resulted inseveral theoretical models, with roots in informa-tion systems, psychology, and sociology, thatroutinely explain over 40 percent of the variance inindividual intention to use technology (e.g., Daviset al. 1989; Taylor and Todd 1995b; Venkateshand Davis 2000). Researchers are confrontedwith a choice among a multitude of models andfind that they must “pick and choose” constructsacross the models, or choose a “favored model”and largely ignore the contributions fromalternative models. Thus, there is a need for areview and synthesis in order to progress towarda unified view of user acceptance.The current work has the following objectives:(1) To review the extant user acceptancemodels: The primary purpose of this reviewis to assess the current state of knowledgewith respect to understanding individualacceptance of new information technologies.This review identifies eight prominent modelsand discusses their similarities and dif-ferences. Some authors have previously ob-served some of the similarities acrossmodels.2However, our review is the first toassess similarities and differences across alleight models, a necessary first step towardthe ultimate goal of the paper: the develop-ment of a unified theory of individual accep-tance of technology. The review is presentedin the following section.(2) To empirically compare the eight models:We conduct a within-subjects, longitudinalvalidation and comparison of the eightmodels using data from four organizations.This provides a baseline assessment of therelative explanatory power of the individualmodels against which the unified model canbe compared. The empirical model compari-son is presented in the third section.(3) To formulate the Unified Theory of Accep-tance and Use of Technology (UTAUT):Based upon conceptual and empirical simi-larities across models, we formulate a unifiedmodel. The formulation of UTAUT is pre-sented in the fourth section.(4) To empirically validate UTAUT: An empiricaltest of UTAUT on the original data providespreliminary support for our contention thatUTAUT outperforms each of the eight originalmodels. UTAUTis then cross-validated usingdata from two new organizations. The empiri-cal validation of UTAUT is presented in thefifth section.2For example, Moore and Benbasat (1991) adapted theperceived usefulness and ease of use items from Daviset al.’s (1989) TAM to measure relative advantage andcomplexity, respectively, in their innovation diffusionmodel.
  • 3. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 427Individual reactions tousing informationtechnologyIntentions to useinformationtechnologyActual use ofinformationtechnologyIndividual reactions tousing informationtechnologyIntentions to useinformationtechnologyActual use ofinformationtechnologyFigure 1. Basic Concept Underlying User Acceptance ModelsReview of Extant UserAcceptance ModelsDescription of Modelsand ConstructsIS research has long studied how and why indivi-duals adopt new information technologies. Withinthis broad area of inquiry, there have been severalstreams of research. One stream of researchfocuses on individual acceptance of technology byusing intention or usage as a dependent variable(e.g., Compeau and Higgins 1995b; Davis et al.1989). Other streams have focused onimplementation success atthe organizational level(Leonard-Barton and Deschamps 1988) and task-technology fit (Goodhue 1995; Goodhue andThompson 1995), among others. While each ofthese streams makes important and uniquecontributions to the literature on user acceptanceof information technology, the theoretical modelsto be included in the present review, comparison,and synthesis employ intention and/or usage asthe key dependent variable. The goal here is tounderstand usage as the dependent variable. Therole of intention as a predictor of behavior (e.g.,usage) is critical and has been well-established inIS and the reference disciplines (see Ajzen 1991;Sheppard et al. 1988; Taylor and Todd 1995b).Figure 1 presents the basic conceptual frameworkunderlying the class of models explaining indivi-dual acceptance of information technology thatforms the basis of this research. Our review re-sulted in the identification of eight key competingtheoretical models. Table 1 describes the eightmodels and defines their theorized determinantsof intention and/or usage. The models hypo-thesize between two and seven determinants ofacceptance, for a total of 32 constructs acrossthe eight models. Table 2 identifies four keymoderating variables (experience, voluntariness,gender, and age) that have been found to besignificant in conjunction with these models.Prior Model Tests andModel ComparisonsThere have been many tests of the eight modelsbut there have only been four studies reportingempirically-based comparisons of two or more ofthe eight models published in the major informa-tion systems journals. Table 3 provides a briefoverview of each of the model comparisonstudies. Despite the apparent maturity of the re-search stream, a comprehensive comparison ofthe key competing models has not been con-ducted in a single study. Below, we identify fivelimitations of these prior model tests and compari-sons, and how we address these limitations in ourwork.• Technology studied: The technologies thathave been studied in many of the modeldevelopment and comparison studies havebeen relatively simple, individual-orientedinformation technologies as opposed to morecomplex and sophisticated organizationaltechnologies that are the focus of managerialconcern and of this study.
  • 4. Table 1. Models and Theories of Individual AcceptanceTheory of Reasoned Action (TRA) Core Constructs DefinitionsDrawn from social psychology, TRA is one of the mostfundamental and influential theories of human behavior.It has been used to predict a wide range of behaviors(see Sheppard et al. 1988 for a review). Davis et al.(1989) applied TRA to individual acceptance of techno-logy and found that the variance explained was largelyconsistent with studies that had employed TRA in thecontext of other behaviors.Attitude TowardBehavior“an individual’s positive or negative feelings (evaluativeaffect) about performing the target behavior” (Fishbein andAjzen 1975, p. 216).Subjective Norm“the person’s perception that most people who areimportant to him think he should or should not perform thebehavior in question” (Fishbein and Ajzen 1975, p. 302).Technology Acceptance Model (TAM)TAM is tailored to IS contexts, and was designed to pre-dict information technology acceptance and usage onthe job. Unlike TRA, the final conceptualization of TAMexcludes the attitude construct in order to better explainintention parsimoniously. TAM2 extended TAM by in-cluding subjective norm as an additional predictor ofintention in the case of mandatory settings (Venkateshand Davis 2000). TAM has been widely applied to adiverse set of technologies and users.Perceived Usefulness“the degree to which a person believes that using aparticular system would enhance his or her jobperformance” (Davis 1989, p. 320).Perceived Ease ofUse“the degree to which a person believes that using aparticular system would be free of effort” (Davis 1989, p.320).Subjective Norm Adapted from TRA/TPB. Included in TAM2 only.Motivational Model (MM)A significant body of research in psychology has sup-ported general motivation theory as an explanation forbehavior. Several studies have examined motivationaltheory and adapted it for specific contexts. Vallerand(1997) presents an excellent review of the fundamentaltenets of this theoretical base. Within the informationsystems domain, Davis et al. (1992) applied motiva-tional theory to understand new technology adoptionand use (see also Venkatesh and Speier 1999).Extrinsic MotivationThe perception that users will want to perform an activity“because it is perceived to be instrumental in achievingvalued outcomes that are distinct from the activity itself,such as improved job performance, pay, or promotions”(Davis et al. 1992, p. 1112).Intrinsic MotivationThe perception that users will want to perform an activity“for no apparent reinforcement other than the process ofperforming the activity per se” (Davis et al. 1992, p. 1112).
  • 5. Table 1. Models and Theories of Individual Acceptance (Continued)Theory of Planned Behavior (TPB) Core Constructs DefinitionsTPB extended TRA by adding the construct of perceivedbehavioral control. In TPB, perceived behavioral controlis theorized to be an additional determinant of intentionand behavior. Ajzen (1991) presented a review ofseveral studies that successfully used TPB to predictintention and behavior in a wide variety of settings. TPBhas been successfully applied to the understanding ofindividual acceptance and usage of many different tech-nologies (Harrison et al. 1997; Mathieson 1991; Taylorand Todd 1995b). A related model is the DecomposedTheory of Planned Behavior (DTPB). In terms of pre-dicting intention, DTPB is identical to TPB. In contrastto TPB but similar to TAM, DTPB “decomposes” atti-tude, subjective norm, and perceived behavioral controlinto its the underlying belief structure within technologyadoption contexts.Attitude TowardBehaviorAdapted from TRA.Subjective Norm Adapted from TRA.Perceived BehavioralControl“the perceived ease or difficulty of performing thebehavior” (Ajzen 1991, p. 188). In the context of ISresearch, “perceptions of internal and external constraintson behavior” (Taylor and Todd 1995b, p. 149).Combined TAM and TPB (C-TAM-TPB)This model combines the predictors of TPB withperceived usefulness from TAM to provide a hybridmodel (Taylor and Todd 1995a).Attitude TowardBehaviorAdapted from TRA/TPB.Subjective Norm Adapted from TRA/TPB.Perceived BehavioralControlAdapted from TRA/TPB.Perceived Usefulness Adapted from TAM.
  • 6. Table 1. Models and Theories of Individual Acceptance (Continued)Model of PC Utilization (MPCU) Core Constructs DefinitionsDerived largely from Triandis’ (1977) theory of humanbehavior, this model presents a competing perspectiveto that proposed by TRA and TPB. Thompson et al.(1991) adapted and refined Triandis’ model for IS con-texts and used the model to predict PC utilization.However, the nature of the model makes it particularlysuited to predict individual acceptance and use of arange of information technologies. Thompson et al.(1991) sought to predict usage behavior rather thanintention; however, in keeping with the theory’s roots,the current research will examine the effect of thesedeterminants on intention. Also, such an examination isimportant to ensure a fair comparison of the differentmodels.Job-fit“the extent to which an individual believes that using [atechnology] can enhance the performance of his or herjob” (Thompson et al. 1991, p. 129).ComplexityBased on Rogers and Shoemaker (1971), “the degree towhich an innovation is perceived as relatively difficult tounderstand and use” (Thompson et al. 1991, p. 128).Long-termConsequences“Outcomes that have a pay-off in the future” (Thompson etal. 1991, p. 129).Affect Towards UseBased on Triandis, affect toward use is “feelings of joy,elation, or pleasure, or depression, disgust, displeasure,or hate associated by an individual with a particular act”(Thompson et al. 1991, p. 127).Social FactorsDerived from Triandis, social factors are “the individual’sinternalization of the reference group’s subjective culture,and specific interpersonal agreements that the individualhas made with others, in specific social situations”(Thompson et al. 1991, p. 126).Facilitating ConditionsObjective factors in the environment that observers agreemake an act easy to accomplish. For example, returningitems purchased online is facilitated when no fee ischarged to return the item. In an IS context, “provision ofsupport for users of PCs may be one type of facilitatingcondition that can influence system utilization” (Thompsonet al. 1991, p. 129).
  • 7. Table 1. Models and Theories of Individual Acceptance (Continued)Innovation Diffusion Theory (IDT) Core Constructs DefinitionsGrounded in sociology, IDT (Rogers 1995) has beenused since the 1960s to study a variety of innovations,ranging from agricultural tools to organizationalinnovation (Tornatzky and Klein 1982). Withininformation systems, Moore and Benbasat (1991)adapted the characteristics of innovations presented inRogers and refined a set of constructs that could beused to study individual technology acceptance. Mooreand Benbasat (1996) found support for the predictivevalidity of these innovation characteristics (see alsoAgarwal and Prasad 1997, 1998; Karahanna et al. 1999;Plouffe et al. 2001).Relative Advantage“the degree to which an innovation is perceived as beingbetter than its precursor” (Moore and Benbasat 1991, p.195).Ease of Use“the degree to which an innovation is perceived as beingdifficult to use” (Moore and Benbasat 1991, p. 195).Image“The degree to which use of an innovation is perceived toenhance one’s image or status in one’s social system”(Moore and Benbasat 1991, p. 195).VisibilityThe degree to which one can see others using the systemin the organization (adapted from Moore and Benbasat1991).Compatibility“the degree to which an innovation is perceived as beingconsistent with the existing values, needs, and pastexperiences of potential adopters” (Moore and Benbasat1991, p. 195).Results Demon-strability“the tangibility of the results of using the innovation,including their observability and communicability” (Mooreand Benbasat 1991, p. 203).Voluntariness of Use“the degree to which use of the innovation is perceived asbeing voluntary, or of free will” (Moore and Benbasat1991, p. 195).
  • 8. Table 1. Models and Theories of Individual Acceptance (Continued)Social Cognitive Theory (SCT) Core Constructs DefinitionsOne of the most powerful theories of human behavior issocial cognitive theory (see Bandura 1986). Compeauand Higgins (1995b) applied and extended SCT to thecontext of computer utilization (see also Compeau et al.1999); while Compeau and Higgins (1995a) also em-ployed SCT, it was to study performance and thus isoutside the goal of the current research. Compeau andHiggins’ (1995b) model studied computer use but thenature of the model and the underlying theory allow it tobe extended to acceptance and use of informationtechnology in general. The original model of Compeauand Higgins (1995b) used usage as a dependentvariable but in keeping with the spirit of predictingindividual acceptance, we will examine the predictivevalidity of the model in the context of intention andusage to allow a fair comparison of the models.Outcome Expec-tations—PerformanceThe performance-related consequences of the behavior.Specifically, performance expectations deal with job-related outcomes (Compeau and Higgins 1995b).OutcomeExpectations—PersonalThe personal consequences of the behavior. Specifically,personal expectations deal with the individual esteem andsense of accomplishment (Compeau and Higgins 1995b).Self-efficacyJudgment of one’s ability to use a technology (e.g.,computer) to accomplish a particular job or task.AffectAn individual’s liking for a particular behavior (e.g.,computer use).AnxietyEvoking anxious or emotional reactions when it comes toperforming a behavior (e.g., using a computer).
  • 9. Table 2. Role of Moderators in Existing ModelsModel Experience Voluntariness Gender AgeTheory ofReasonedActionExperience was not explicitly included inthe original TRA. However, the role ofexperience was empirically examinedusing a cross-sectional analysis by Daviset al. (1989). No change in the salience ofdeterminants was found. In contrast,Karahanna et al. (1999) found that attitudewas more important with increasingexperience, while subjective norm becameless important with increasing experience.Voluntariness was notincluded in the originalTRA. Although nottested, Hartwick andBarki (1994) suggestedthat subjective normwas more importantwhen system use wasperceived to be lessvoluntary.N/A N/ATechnologyAcceptanceModel (andTAM2)Experience was not explicitly included inthe original TAM. Davis et al. (1989) andSzajna (1996), among others, have pro-vided empirical evidence showing thatease of use becomes nonsignificant withincreased experience.Voluntariness was notexplicitly included in theoriginal TAM. WithinTAM2, subjective normwas salient only inmandatory settings andeven then only in casesof limited experiencewith the system (i.e., athree-way interaction).Gender was not in-cluded in the originalTAM. Empirical evi-dence demonstratedthat perceived useful-ness was more salientfor men while perceivedease of use was moresalient for women(Venkatesh and Morris2000). The effect ofsubjective norm wasmore salient for womenin the early stages ofexperience (i.e., athree-way interaction).N/AMotivationalModelN/A N/A N/A N/A
  • 10. Table 2. Role of Moderators in Existing Models (Continued)Model Experience Voluntariness Gender AgeTheory ofPlannedBehaviorExperience was not explicitly included inthe original TPB or DTPB. It has beenincorporated into TPB via follow-on studies(e.g., Morris and Venkatesh 2000).Empirical evidence has demonstrated thatexperience moderates the relationshipbetween subjective norm and behavioralintention, such that subjective normbecomes less important with increasinglevels of experience. This is similar to thesuggestion of Karahanna et al. (1999) inthe context of TRA.Voluntariness was notincluded in the originalTPB or DTPB. Asnoted in the discussionregarding TRA,although not tested,subjective norm wassuggested to be moreimportant when systemuse was perceived tobe less voluntary (Hart-wick and Barki 1994).Venkatesh et al. (2000)found that attitude wasmore salient for men.Both subjective normand perceived beha-vioral control were moresalient for women inearly stages of exper-ience (i.e., three-wayinteractions).Morris and Venkatesh(2000) found that atti-tude was more salientfor younger workerswhile perceivedbehavioral control wasmore salient for olderworkers. Subjectivenorm was more salientto older women (i.e., athree-way interaction).CombinedTAM-TPBExperience was incorporated into thismodel in a between-subjects design(experienced and inexperienced users).Perceived usefulness, attitude towardbehavior, and perceived behavioral controlwere all more salient with increasingexperience while subjective norm becameless salient with increasing experience(Taylor and Todd 1995a).N/A N/A N/A
  • 11. Table 2. Role of Moderators in Existing Models (Continued)Model Experience Voluntariness Gender AgeModel of PCUtilizationThompson et al. (1994) found that com-plexity, affect toward use, social factors,and facilitating conditions were all moresalient with less experience. On the otherhand, concern about long-term conse-quences became increasingly importantwith increasing levels of experience.N/A N/A N/AInnovationDiffusionTheoryKarahanna et al. (1999) conducted abetween-subjects comparison to study theimpact of innovation characteristics onadoption (no/low experience) and usagebehavior (greater experience) and founddifferences in the predictors of adoptionvs. usage behavior. The results showedthat for adoption, the significant predictorswere relative advantage, ease of use, trial-ability, results demonstrability, and visi-bility. In contrast, for usage, only relativeadvantage and image were significant.Voluntariness was nottested as a moderator,but was shown to havea direct effect onintention.N/A N/ASocialCognitiveTheoryN/A N/A N/A N/A
  • 12. Table 3. Review of Prior Model ComparisonsModelComparisonStudiesTheories/ModelsComparedContext of Study(Incl.Technology) ParticipantsNewness ofTechnologyStudiedNumber of Pointsof MeasurementCross-Sectional orLongitudinalAnalysis FindingsDavis et al.(1989)TRA, TAM Within-subjectsmodel compari-son of intentionand use of a wordprocessor107 students Participantswere new tothe technologyTwo; 14 weeksapartCross-sectionalanalysis at thetwo points intimeThe variance in intentionand use explained by TRAwas 32% and 26%, andTAM was 47% and 51%,respectively.Mathieson(1991)TAM, TPB Between-subjectsmodel compari-son of intention touse a spread-sheet andcalculator262 students Some famil-iarity with thetechnology aseach partici-pant had tochoose a tech-nology to per-form a taskOne Cross-sectionalThe variance in intentionexplained by TAM was70% and TPB was 62%Taylor andTodd (1995b)TAM,TPB/DTPBWithin-subjectsmodel compari-son of intention touse a computingresource center786 students Many studentswere alreadyfamiliar with thecenterFor a three-monthperiod, all studentsvisiting the centerwere surveyed—i.e., multiple mea-sures per student.Cross-sectionalThe variance in intentionexplained by TAM was52%, TPB was 57%, andDTPB was 60%Plouffe et al.(2001)TAM, IDT Within-subjectsmodel compari-son of behavioralintention to useand use in thecontext of a mar-ket trial of anelectronic pay-ment systemusing smart card.176merchantsSurveyadministeredafter 10 monthsof useOne Cross-sectionalThe variance in intentionexplained by TAM was33% and IDT was 45%
  • 13. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 437• Participants: While there have been sometests of each model in organizational settings,the participants in three of the four modelcomparison studies have been students—only Plouffe et al. (2001) conducted theirresearch in a nonacademic setting. Thisresearch is conducted using data collectedfrom employees in organizations.• Timing of measurement: In general, most ofthe tests of the eight models were conductedwell after the participants’ acceptance orrejection decision rather than during theactive adoption decision-making process.Because behavior has become routinized,individual reactions reported in those studiesare retrospective (see Fiske and Taylor 1991;Venkatesh et al. 2000). With the exception ofDavis et al. (1989), the model comparisonsexamined technologies that were alreadyfamiliar to the individuals at the time of mea-surement. In this paper, we examine techno-logies from the time of their initial introductionto stages of greater experience.• Nature of measurement: Even studies thathave examined experience have typicallyemployed cross-sectional and/or between-subjects comparisons (e.g., Davis et al. 1989;Karahanna et al. 1999; Szajna 1996; Taylorand Todd 1995a; Thompson et al. 1994).This limitation applies to model comparisonstudies also. Our work tracks participantsthrough various stages of experience with anew technology and compares all models onall participants.• Voluntary vs. mandatory contexts: Most ofthe model tests and all four model com-parisons were conducted in voluntary usagecontexts.3Therefore, one must use cautionwhen generalizing those results to themandatory settings that are possibly of moreinterest to practicing managers. This re-search examines both voluntary and man-datory implementation contexts.Empirical Comparison of theEight ModelsSettings and ParticipantsLongitudinal field studies were conducted at fourorganizations among individuals being introducedto a new technology in the workplace. To helpensure our results would be robust acrosscontexts, we sampled for heterogeneity acrosstechnologies, organizations, industries, businessfunctions, and nature of use (voluntary vs.mandatory). In addition, we captured perceptionsas the users’ experience with the technologyincreased. At each firm, we were able to time ourdata collection in conjunction with a trainingprogram associated with the new technologyintroduction. This approach is consistent withprior training and individual acceptance researchwhere individual reactions to a new technologywere studied (e.g., Davis et al. 1989; Olfman andMandviwalla 1994; Venkatesh and Davis 2000).A pretested questionnaire containing items mea-suring constructs from all eight models wasadministered at three different points in time:post-training (T1),onemonth after implementation(T2), and three months after implementation (T3).Actual usage behavior was measured over the six-month post-training period. Table 4 summarizeskey characteristics of the organizational settings.Figure 2 presents the longitudinal data collectionschedule.MeasurementA questionnaire was created with items validatedin prior research adapted to the technologies andorganizations studied. TRA scales were adaptedfrom Davis et al. (1989); TAM scales wereadapted from Davis (1989), Davis et al. (1989),3Notable exceptions are TRA (Hartwick and Barki 1994)and TAM2 (Venkatesh and Davis 2000) as well asstudies that have incorporated voluntariness as a directeffect (on intention) in order to account for perceivednonvoluntary adoption (e.g., Agarwal and Prasad 1997;Karahanna et al. 1999; Moore and Benbasat 1991).
  • 14. Venkatesh et al./User Acceptance of IT438 MIS Quarterly Vol. 27 No. 3/September 2003X O X O X O X OTraining User System User System User System UsageReactions Use Reactions/ Use Reactions/ Use MeasurementUsage UsageMeasurement Measurement1 week 1 month 3 months 6 monthsX O X O X O X OTraining User System User System User System UsageReactions Use Reactions/ Use Reactions/ Use MeasurementUsage UsageMeasurement Measurement1 week 1 month 3 months 6 monthsFigure 2. Longitudinal Data Collection ScheduleTable 4. Description of StudiesStudy IndustryFunctionalAreaSampleSize System DescriptionVoluntary Use1a EntertainmentProductDevelopment54Online meeting manager that could beused to conduct Web-enabled video oraudio conferences in lieu of face-to-faceor traditional phone conferences1bTelecommServicesSales 65Database application that could be usedto access industry standards for particularproducts in lieu of other resources (e.g.,technical manuals, Web sites)Mandatory Use2a BankingBusinessAccountManagement58Portfolio analyzer that analysts wererequired to use in evaluating existing andpotential accounts2bPublicAdministrationAccounting 38Proprietary accounting systems on a PCplatform that accountants were requiredto use for organizational bookkeepingand Venkatesh and Davis (2000); MM scales wereadapted from Davis et al. (1992); TPB/DTPBscales were adapted from Taylor and Todd(1995a, 1995b); MPCU scales were adapted fromThompson et al. (1991); IDT scales were adaptedfrom Moore and Benbasat (1991); and SCT scaleswere adapted from Compeau and Higgins (1995a,1995b) and Compeau et al. (1999). Behavioralintention to use the system was measured usinga three-item scale adapted from Davis et al.(1989) and extensively used in much of theprevious individual acceptance research. Seven-point scales were used for all of the aforemen-tioned constructs’ measurement, with 1 being thenegative end of the scale and 7 being the positiveend of the scale. In addition to these measures,perceived voluntariness was measured as amanipulation check per the scale of Moore andBenbasat (1991), where 1 was nonvoluntary and7 was completely voluntary. The tense of theverbs in the various scales reflected the timing ofmeasurement: future tense was employed at T1,present tense was employed at T2 and T3 (seeKarahanna et al. 1999). The scales used to mea-sure the key constructs are discussed in a latersection where we perform a detailed comparison(Tables 9 through 13). A focus group of fivebusiness professionals evaluated the question-naire, following which minor wording changeswere made. Actual usage behavior was mea-sured as duration of use via system logs. Due tothe sensitivity of usage measures to network
  • 15. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 439availability, in all organizations studied, the systemautomatically logged off inactive users after aperiod of 5 to 15 minutes, eliminating most idletime from the usage logs.ResultsThe perceptions of voluntariness were very high instudies 1a and 1b (1a: M = 6.50, SD = 0.22; 1b:M = 6.51, SD = 0.20) and very low in studies 2aand 2b (1a: M = 1.50, SD = 0.19; 1b: M = 1.49,SD = 0.18). Given this bi-modal distribution in thedata (voluntary vs. mandatory), we created twodata sets: (1) studies 1a and 1b, and (2) studies2a and 2b. This is consistent with Venkatesh andDavis (2000).Partial least squares (PLS Graph, Version2.91.03.04) was used to examine the reliabilityand validity of the measures. Specifically, 48separate validity tests (two studies, eight models,three time periods each) were run to examineconvergent and discriminant validity. In testingthe various models, only the direct effects onintention were modeled as the goal was toexamine the prediction of intention rather thaninterrelationships among determinants of inten-tion; further, the explained variance (R2) is notaffected by indirect paths. The loading patternwas found to be acceptable with most loadingsbeing .70 or higher. All internal consistencyreliabilities were greater than .70. The patterns ofresults found in the current work are highly con-sistent with the results of previous research.PLS was used to test all eight models at the threepoints of measurement in each of the two datasets. In all cases, we employed a bootstrappingmethod (500 times) that used randomly selectedsubsamples to test the PLS model.4Tables 5 and6 present the model validation results at each ofthe points of measurement. The tables report thevariance explained and the beta coefficients. Keyfindings emerged from these analyses. First, alleight models explained individual acceptance,with variance in intention explained ranging from17 percent to 42 percent. Also, a key differenceacross studies stemmed from the voluntary vs.mandatory settings—in mandatory settings (study2), constructs related to social influence weresignificant whereas in the voluntary settings (study1), they were not significant. Finally, the deter-minants of intention varied over time, with somedeterminants going from significant to nonsigni-ficant with increasing experience.Following the test of the baseline/original specifi-cations of the eight models (Tables 5 and 6), weexamined the moderating influences suggested(either explicitly or implicitly) in the literature—i.e.,experience, voluntariness, gender, and age(Table 2). In order to test these moderating influ-ences, stay true to the model extensions(Table 2), and conduct a complete test of theexisting models and their extensions, the datawere pooled across studies and time periods.Voluntariness was a dummy variable used toseparate the situational contexts (study 1 vs.study 2); this approach is consistent with previousresearch (Venkatesh and Davis 2000). Genderwas coded as a 0/1 dummy variable consistentwith previous research (Venkatesh and Morris2000) and age was coded as a continuous vari-able, consistent with prior research (Morris andVenkatesh 2000). Experience was operationa-lized via a dummy variable that took ordinal valuesof 0, 1, or 2 to capture increasing levels of userexperience with the system (T1, T2, and T3).Using an ordinal dummy variable, rather thancategorical variables, is consistent with recentresearch (e.g., Venkatesh and Davis 2000).Pooling the data across the three points of mea-surement resulted in a sample of 645 (215 × 3).The results of the pooled analysis are shown inTable 7.Because pooling across time periods allows theexplicit modeling of the moderating role of exper-ience, there is an increase in the variance ex-plained in the case of TAM2 (Table 7) comparedto a main effects-only model reported earlier(Tables 5 and 6). One of the limitations of poolingis that there are repeated measures from the4The interested reader is referred to a more detailedexposition of bootstrapping and how it compares to othertechniques of resampling such as jackknifing (see Chin1998; Efron and Gong 1983).
  • 16. Venkatesh et al./User Acceptance of IT440 MIS Quarterly Vol. 27 No. 3/September 2003Table 5. Study 1: Predicting Intention in Voluntary SettingsTime 1 (N = 119) Time 2 (N = 119) Time 3 (N = 119)Models Independent variables R2Beta R2Beta R2BetaTRA Attitude toward using tech. .30 .55*** .26 .51*** .19 .43***Subjective norm .06 .07 .08TAM/TAM2Perceived usefulness .38 .55*** .36 .60*** .37 .61***Perceived ease of use .22** .03 .05Subjective norm .02 .06 .06MM Extrinsic motivation .37 .50*** .36 .47*** .37 .49***Intrinsic motivation .22** .22** .24***TPB/DTPBAttitude toward using tech. .37 .52*** .25 .50*** .21 .44***Subjective norm .05 .04 .05Perceived behavioral control .24*** .03 .02C-TAM-TPBPerceived usefulness .39 .56*** .36 .60*** .39 .63***Attitude toward using tech. .04 .03 .05Subjective norm .06 .04 .03Perceived behavioral control .25*** .02 .03MPCU Job-fit .37 .54*** .36 .60*** .38 .62***Complexity (reversed) .23*** .04 .04Long-term consequences .06 .04 .07Affect toward use .05 .05 .04Social factors .04 .07 .06Facilitating conditions .05 .06 .04IDT Relative advantage .38 .54*** .37 .61*** .39 .63***Ease of use .26** .02 .07Result demonstrability .03 .04 .06Trialability .04 .09 .08Visibility .06 .03 .06Image .06 .05 .07Compatibility .05 .02 .04Voluntariness .03 .04 .03SCT Outcome expectations .37 .47*** .36 .60*** .36 .60***Self-efficacy .20*** .03 .01Affect .05 .03 .04Anxiety -.17* .04 .06Notes: 1. *p < .05; **p < .01; ***p < .001.2. When the data were analyzed separately for studies 2a and 2b, the pattern of results wasvery similar.
  • 17. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 441Table 6. Study 2: Predicting Intention in Mandatory SettingsTime 1 (N = 96) Time 2 (N = 96) Time 3 (N = 96)Models Independent variables R2Beta R2Beta R2BetaTRA Attitude toward using tech. .26 .27*** .26 .28*** .17 .40***Subjective norm .20** .21** .05TAM/TAM2Perceived usefulness .39 .42*** .41 .50*** .36 .60***Perceived ease of use .21* .23** .03Subjective norm .20* .03 .04MM Extrinsic motivation .38 .47*** .40 .49*** .35 .44***Intrinsic motivation .21** .24** .19**TPB/DTPBAttitude toward using tech. .34 .22* .28 .36*** .18 .43***Subjective norm .25*** .26** .05Perceived behavioral control .19* .03 .08C-TAM-TPBPerceived usefulness .36 .42*** .35 .51*** .35 .60***Attitude toward using tech. .07 .08 .04Subjective norm .20* .23** .03Perceived behavioral control .19* .11 .09MPCU Job-fit .37 .42*** .40 .50*** .37 .61***Complexity (reversed) .20* .02 .04Long-term consequences .07 .07 .07Affect toward use .01 .05 .04Social factors .18* .23** .02Facilitating conditions .05 .07 .07IDT Relative advantage .38 .47*** .42 .52*** .37 .61***Ease of use .20* .04 .04Result demonstrability .03 .07 .04Trialability .05 .04 .04Visibility .04 .04 .01Image .18* .27** .05Compatibility .06 .02 .04Voluntariness .02 .06 .03SCT Outcome expectations .38 .46*** .39 .44*** .36 .60***Self-efficacy .19** .21*** .03Affect .06 .04 .05Anxiety -.18* -.16* .02Notes: 1. *p < .05; **p < .01; ***p < .001.2. When the data were analyzed separately for studies 2a and 2b, the pattern of results wasvery similar.
  • 18. Venkatesh et al./User Acceptance of IT442 MIS Quarterly Vol. 27 No. 3/September 2003Table 7. Predicting Intention—Model Comparison Including Moderators: DataPooled Across Studies (N = 645)Model Version Independent Variables R2Beta ExplanationTRA 1 Attitude (A) .36 .41*** Direct effectSubjective norm (SN) .11Experience (EXP) .09Voluntariness (VOL) .04A × EXP .03SN × EXP -.17* Effect decreases with increasingexperienceSN × VOL .17* Effect present only in mandatorysettingsTAM 2aTAM2Perceived usefulness (U) .53 .48*** Direct effectPerceived ease of use (EOU) .11Subjective norm (SN) .09Experience (EXP) .06Voluntariness (VOL) .10EOU × EXP -.20** Effect decreases with increasingexperienceSN × EXP -.15SN × VOL -.16* Cannot be interpreted due topresence of higher-order termEXP × VOL .07SN × EXP × VOL -.18** Effect exists only in mandatorysettings but decreases withincreasing experience2bTAMincl.genderPerceived usefulness (U) .52 .14* Cannot be interpreted due topresence of interaction termPercd. ease of use (EOU) .08Subjective norm (SN) .02Gender (GDR) .11Experience (EXP) .07U × GDR .31*** Effect is greater for menEOU × GDR -.20** Effect is greater for womenSN × GDR .11SN × EXP .02EXP × GDR .09SN × GDR × EXP .17** Effect is greater for women butdecreases with increasingexperience
  • 19. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 443Table 7. Predicting Intention—Model Comparison Including Moderators: DataPooled Across Studies (N = 645) (Continued)Model Version Independent Variables R2Beta ExplanationMM 3 Extrinsic motivation .38 .50*** Direct effectIntrinsic motivation .20*** Direct effectTPB/DTPB4aTPBincl. volAttitude (A) .36 .40*** Direct effectSubjective norm (SN) .09Percd. behrl. control (PBC) .13Experience (EXP) .10Voluntariness (VOL) .05SN × EXP -.17** Effect decreases with increasingexperienceSN × VOL .17** Effect present only in mandatorysettings4bTPBincl.genderAttitude (A) .46 .17*** Cannot be interpreted due topresence of interaction termSubjective norm (SN) .02Percd. behrl. control (PBC) .10Gender (GDR) .01Experience (EXP) .02A × GDR .22*** Effect is greater for menSN × EXP -.12SN × GDR .10PBC × GDR .07PBC × EXP .04GDR × EXP .15* Term included to test higher-order interactions belowSN × GDR × EXP -.18** Both SN and PBC effects arehigher for women, but the effectsdecrease with increasingexperiencePBC × GDR × EXP -.16*4cTPBincl.ageAttitude (A) .47 .17*** Cannot be interpreted due topresence of interaction termSubjective norm (SN) .02Percd. behrl. control (PBC) .10Age (AGE) .01Experience (EXP) .02A × AGE -.26*** Effect is greater for youngerworkersSN × EXP -.03SN × AGE .11PBC × AGE .21** Effect is greater for olderworkers
  • 20. Venkatesh et al./User Acceptance of IT444 MIS Quarterly Vol. 27 No. 3/September 2003Table 7. Predicting Intention—Model Comparison Including Moderators: DataPooled Across Studies (N = 645) (Continued)Model Version Independent Variables R2Beta ExplanationAGE × EXP .15* Term included to test higher-order interaction belowSN × AGE × EXP -.18** Effect is greater for olderworkers, but the effectdecreases with increasingexperienceC-TAM-TPB5 Perceived usefulness (U) .39 .40*** Direct effectAttitude (A) .09Subjective norm (SN) .08Perceived beholder control(PBC).16* Cannot be interpreted due topresence of interaction termExperience (EXP) .11U × EXP .01A × EXP .08SN × EXP -.17* Effect decreases with increasingexperiencePBC × EXP -.19** Effect decreases with increasingexperienceMPCU 6 Job-fit (JF) .47 .40*** Direct effectComplexity (CO) (reversed) .07Long-term consequences (LTC) .02Affect toward use (ATU) .05Social factors (SF) .10Facilitating condns. (FC) .07Experience (EXP) .08CO × EXP (CO—reversed) -.17* Effect decreases with increasingexperienceLTC × EXP .02ATU × EXP .01SF × EXP -.20** Effect decreases with increasingexperienceFC × EXP .05IDT 7 Relative advantage (RA) .40 .49*** Direct effectEase of use (EU) .05Result demonstrability (RD) .02Trialability (T) .04Visibility (V) .03Image (I) .01Compatibility (COMPAT) .06Voluntariness of use (VOL) .11
  • 21. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 445Table 7. Predicting Intention—Model Comparison Including Moderators: DataPooled Across Studies (N = 645) (Continued)Model Version Independent Variables R2Beta ExplanationExperience (EXP) .03EU × EXP -.16* Effect decreases with increasingexperienceRD × EXP .03V × EXP -.14* Effect decreases with increasingexperienceI × EXP -.14* Effect decreases with increasingexperienceVoluntariness of use (VOL) .05SCT 8 Outcome expectations .36 .44*** Direct effectSelf-efficacy .18* Direct effectAffect .01Anxiety -.15* Direct effectNotes: 1. *p < .05; **p < .01; ***p < .001.2. The significance of the interaction terms was also verified using Chow’s test.same individuals, resulting in measurement errorsthat are potentially correlated across time.However, cross-sectional analysis using Chow’s(1960) test of beta differences (p < .05) from eachtime period (not shown here) confirmed thepattern of results shown in Table 7. Those betadifferences with a significance of p < .05 or better(when using Chow’s test) are discussed in the“Explanation” column in Table 7. The interactionterms were modeled as suggested by Chin et al.(1996) by creating interaction terms that were atthe level of the indicators. For example, if latentvariable A is measured by four indicators (A1, A2,A3, and A4) and latent variable B is measured bythree indicators (B1, B2, and B3), the interactionterm A × B is specified by 12 indicators, each onea product term—i.e., A1 × B1, A1 × B2, A1 × B3,A2 × B1, etc.With the exception of MM and SCT, the predictivevalidity of the models increased after including themoderating variables. For instance, the varianceexplained by TAM2 increased to 53 percent andTAM including gender increased to 52 percentwhen compared to approximately 35 percent incross-sectional tests of TAM(without moderators).The explained variance of TRA, TPB/DTPB,MPCU, and IDT also improved. For each model,we have only included moderators previouslytested in the literature. For example, in the caseof TAM and its variations, the extensive priorempirical work has suggested a larger number ofmoderators when compared to moderators sug-gested for other models. This in turn may haveunintentionally biased the results and contributedto the high variance explained in TAM-relatedmodels when compared to the other models.Regardless, it is clear that the extensions to thevarious models identified in previous researchmostly enhance the predictive validity of thevarious models beyond the original specifications.In looking at technology use as the dependentvariable, in addition to intention as a key predictor,TPB and DTPB employ perceived behavioralcontrol as an additional predictor. MPCU employsfacilitating conditions, a construct similar to per-ceived behavioral control, to predict behavior.Thus, intention and perceived behavioral controlwere used to predict behavior in the subsequent
  • 22. Venkatesh et al./User Acceptance of IT446 MIS Quarterly Vol. 27 No. 3/September 2003Table 8. Predicting Usage BehaviorUse12 Use23 Use34Independent Variables R2Beta R2Beta R2BetaStudies1a and 1b(voluntary)(N=119)Behavioral intention to use (BI) .37 .61*** .36 .60*** .39 .58***Perceived behavioral control (PBC) .04 .06 .17*Studies2a and 2b(mandatory)(N = 96)Behavioral intention to use (BI) .35 .58*** .37 .61*** .39 .56***Perceived behavioral control (PBC) .07 .07 .20*Notes: 1. BI, PBC measured at T1 were used to predict usage between time periods 1 and 2 (denotedUse12); BI, PBC measured at T2 were used to predict usage between time periods 2 and 3(Use23); BI, PBC measured at T3 were used to predict usage between time periods 3 and 4(Use34).2. *p < .05; **p < .01; ***p < .001.time period: intention from T1 was used to predictusage behavior measured between T1 and T2 andso on (see Table 8). Since intention was used topredict actual behavior, concerns associated withthe employment of subjective measures of usagedo not apply here (see Straub et al. 1995). Inaddition to intention being a predictor of use, per-ceived behavioral control became a significantdirect determinant of use over and above intentionwith increasing experience (at T3) indicating thatcontinued use could be directly hindered orfostered by resources and opportunities. A nearlyidentical pattern of results was found when thedata were analyzed using facilitating conditions(from MPCU) in place of perceived behavioralcontrol (the specific results are not shown here).Having reviewed and empirically compared theeight competing models, we now formulate aunified theory of acceptance and use of techno-logy (UTAUT). Toward this end, we examine com-monalities across models as a first step. Tables5, 6, 7, and 8 presented cross-sectional tests ofthe baseline models and their extensions. Severalconsistent findings emerged. First, for everymodel, there was at least one construct that wassignificant in all time periods and that constructalso had the strongest influence—e.g., attitude inTRA and TPB/DTPB, perceived usefulness inTAM/TAM2 and C-TAM-TPB, extrinsic motivationin MM, job-fit in MPCU, relative advantage in IDT,and outcome expectations in SCT. Second,several other constructs were initially significant,but then became nonsignificant over time, in-cluding perceived behavioral control in TPB/DTPBand C-TAM-TPB, perceived ease of use in TAM/TAM2, complexity in MPCU, ease of use in IDT,and self-efficacy and anxiety in SCT. Finally, thevoluntary vs. mandatory context did have aninfluence on the significance of constructs relatedto social influence: subjective norm (TPB/DTPB,C-TAM-TPB and TAM2), social factors (MPCU),and image (IDT) were only significant inmandatory implementations.Formulation of the UnifiedTheory of Acceptance andUse of Technology(UTAUT)Seven constructs appeared to be significant directdeterminants of intention or usage in one or moreof the individual models (Tables 5 and 6). Ofthese, we theorize that four constructs will play a
  • 23. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 447PerformanceExpectancyEffortExpectancySocialInfluenceFacilitatingConditionsVoluntarinessof UseExperienceAgeGenderBehavioralIntentionUseBehaviorPerformanceExpectancyEffortExpectancySocialInfluenceFacilitatingConditionsVoluntarinessof UseExperienceAgeGenderBehavioralIntentionUseBehaviorFigure 3. Research Modelsignificant role as direct determinants of useracceptance and usage behavior: performanceexpectancy, effort expectancy, social influence,and facilitating conditions. As will be explainedbelow, attitude toward using technology, self-efficacy, and anxiety are theorized not to be directdeterminants of intention. The labels used for theconstructs describe the essence of the constructand are meant to be independent of any particulartheoretical perspective. In the remainder of thissection, we define each of the determinants,specify the role of key moderators (gender, age,voluntariness, and experience), and provide thetheoretical justification for the hypotheses.Figure 3 presents the research model.Performance ExpectancyPerformance expectancy is defined as the degreeto which an individual believes that using the sys-tem will help him or her to attain gains in jobperformance. The five constructs from the dif-ferent models that pertain to performanceexpectancy are perceived usefulness (TAM/TAM2and C-TAM-TPB), extrinsic motivation (MM), job-fit(MPCU), relative advantage (IDT), and outcomeexpectations (SCT). Even as these constructsevolved in the literature, some authors acknowl-edged their similarities: usefulness and extrinsicmotivation (Davis et al. 1989, 1992), usefulnessand job-fit (Thompson et al. 1991), usefulness andrelative advantage (Davis et al. 1989; Moore andBenbasat 1991; Plouffe et al. 2001), usefulnessand outcome expectations (Compeau and Higgins1995b; Davis et al. 1989), and job-fit and outcomeexpectations (Compeau and Higgins 1995b).The performance expectancy construct withineach individual model (Table 9) is the strongestpredictor of intention and remains significant at allpoints of measurement in both voluntary and man-datory settings (Tables 5, 6, and 7), consistentwith previous model tests (Agarwal and Prasad
  • 24. Venkatesh et al./User Acceptance of IT448 MIS Quarterly Vol. 27 No. 3/September 2003Table 9. Performance Expectancy: Root Constructs, Definitions, and ScalesConstruct Definition ItemsPerceivedUsefulness(Davis 1989; Davis etal. 1989)The degree to which aperson believes that usinga particular system wouldenhance his or her jobperformance.1. Using the system in my job wouldenable me to accomplish tasks morequickly.2. Using the system would improve my jobperformance.3. Using the system in my job wouldincrease my productivity.4. Using the system would enhance myeffectiveness on the job.5. Using the system would make it easierto do my job.6. I would find the system useful in my job.Extrinsic Motivation(Davis et al. 1992)The perception that userswill want to perform anactivity because it is per-ceived to be instrumentalin achieving valued out-comes that are distinctfrom the activity itself, suchas improved job perfor-mance, pay, or promotionsExtrinsic motivation is operationalized usingthe same items as perceived usefulnessfrom TAM (items 1 through 6 above).Job-fit(Thompson et al.1991)How the capabilities of asystem enhance an indi-vidual’s job performance.1. Use of the system will have no effect onthe performance of my job (reversescored).2. Use of the system can decrease thetime needed for my important jobresponsibilities.3. Use of the system can significantlyincrease the quality of output on my job.4. Use of the system can increase theeffectiveness of performing job tasks.5. Use can increase the quantity of outputfor the same amount of effort.6. Considering all tasks, the general extentto which use of the system could assiston the job. (different scale used for thisitem).
  • 25. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 449Table 9. Performance Expectancy: Root Constructs, Definitions, and Scales(Continued)Construct Definition ItemsRelative Advantage(Moore and Benbasat1991)The degree to which usingan innovation is perceivedas being better than usingits precursor.1. Using the system enables me toaccomplish tasks more quickly.2. Using the system improves the quality ofthe work I do.3. Using the system makes it easier to domy job.4. Using the system enhances myeffectiveness on the job.5. Using the system increases myproductivity.OutcomeExpectations(Compeau andHiggins 1995b;Compeau et al. 1999)Outcome expectationsrelate to the consequencesof the behavior. Based onempirical evidence, theywere separated into per-formance expectations(job-related) and personalexpectations (individualgoals). For pragmaticreasons, four of the highestloading items from theperformance expectationsand three of the highestloading items from thepersonal expectationswere chosen from Com-peau and Higgins (1995b)and Compeau et al. (1999)for inclusion in the currentresearch. However, ourfactor analysis showed thetwo dimensions to load ona single factor.If I use the system…1. I will increase my effectiveness on thejob.2. I will spend less time on routine jobtasks.3. I will increase the quality of output of myjob.4. I will increase the quantity of output forthe same amount of effort.5. My coworkers will perceive me ascompetent.6. I will increase my chances of obtaining apromotion.7. I will increase my chances of getting araise.1998; Compeau and Higgins 1995b; Davis et al.1992; Taylor and Todd 1995a; Thompson et al.1991; Venkatesh and Davis 2000). However, froma theoretical point of view, there is reason toexpect that the relationship between performanceexpectancy and intention will be moderated bygender and age. Research on gender differencesindicates that men tend to be highly task-oriented(Minton and Schneider 1980) and, therefore, per-formance expectancies, which focus on taskaccomplishment, are likely to be especially salientto men. Gender schema theory suggests thatsuch differences stem from gender roles andsocialization processes reinforced from birthrather than biological gender per se (Bem 1981;Bem and Allen 1974; Kirchmeyer 1997; Lubinskiet al. 1983; Lynott and McCandless 2000; Moto-widlo 1982). Recent empirical studies outside the
  • 26. Venkatesh et al./User Acceptance of IT450 MIS Quarterly Vol. 27 No. 3/September 2003IT context (e.g., Kirchmeyer 2002; Twenge 1997)have shown that gender roles have a strongpsychological basis and are relatively enduring,yet open to change over time (see also Ashmore1990; Eichinger et al. 1991; Feldman and Aschen-brenner 1983; Helson and Moane 1987).Similar to gender, age is theorized to play amoderating role. Research on job-related atti-tudes (e.g., Hall and Mansfield 1975; Porter 1963)suggests that younger workers may place moreimportance on extrinsic rewards. Gender and agedifferences have been shown to exist in techno-logy adoption contexts also (Morris and Venkatesh2000; Venkatesh and Morris 2000). In looking atgender and age effects, it is interesting to notethat Levy (1988) suggests that studies of genderdifferences can be misleading without reference toage. For example, given traditional societal gen-der roles, the importance of job-related factorsmay change significantly (e.g., become sup-planted by family-oriented responsibilities) forworking women between the time that they enterthe labor force and the time they reach child-rearing years (e.g., Barnett and Marshall 1991).Thus, we expect that the influence of performanceexpectancy will be moderated by both gender andage.H1: The influence of performance ex-pectancy on behavioral intention willbe moderated by gender and age,such that the effect will be strongerfor men and particularly for youngermen.Effort ExpectancyEffort expectancy is defined as the degree of easeassociated with the use of the system. Three con-structs from the existing models capture theconcept of effort expectancy: perceived ease ofuse (TAM/TAM2), complexity (MPCU), and easeof use (IDT). As can be seen in Table 10, there issubstantial similarity among the construct defini-tions and measurement scales. The similaritiesamong these constructs have been noted in priorresearch (Davis et al. 1989; Moore and Benbasat1991; Plouffe et al. 2001; Thompson et al. 1991).The effort expectancy construct within each model(Table 10) is significant in both voluntary andmandatory usage contexts; however, each one issignificant only during the first time period (post-training, T1), becoming nonsignificant overperiods of extended and sustained usage (seeTables 5, 6, and 7), consistent with previousresearch (e.g., Agarwal and Prasad 1997, 1998;Davis et al. 1989; Thompson et al. 1991, 1994).Effort-oriented constructs are expected to be bemore salient in the early stages of a new behavior,when process issues represent hurdles to beovercome, and later become overshadowed byinstrumentality concerns (Davis et al. 1989;Szajna 1996; Venkatesh 1999).Venkatesh and Morris (2000), drawing upon otherresearch (e.g., Bem and Allen 1974; Bozionelos1996), suggest that effort expectancy is moresalient for women than for men. As noted earlier,the gender differences predicted here could bedriven by cognitions related to gender roles (e.g.,Lynott and McCandless 2000; Motowidlo 1982;Wong et al. 1985). Increased age has beenshown to be associated with difficulty in pro-cessing complexstimuli and allocating attention toinformation on the job (Plude and Hoyer 1985),both of which may be necessary when usingsoftware systems. Prior research supports thenotion that constructs related to effort expectancywill be stronger determinants of individuals’ inten-tion for women (Venkatesh and Morris 2000;Venkatesh et al. 2000) and for older workers(Morris and Venkatesh 2000). Drawing from thearguments made in the context of performanceexpectancy, we expect gender, age, and exper-ience to work in concert (see Levy 1988). Thus,we propose that effort expectancy will be mostsalient for women, particularly those who are olderand with relatively little experience with thesystem.H2: The influence of effort expectancyon behavioral intention will bemoderated by gender, age, andexperience, such that the effect willbe stronger for women, particularlyyounger women, and particularly atearly stages of experience.
  • 27. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 451Table 10. Effort Expectancy: Root Constructs, Definitions, and ScalesConstruct Definition ItemsPerceived Ease of Use(Davis 1989; Davis etal. 1989)The degree to which aperson believes thatusing a system would befree of effort.1. Learning to operate the system would beeasy for me.2. I would find it easy to get the system todo what I want it to do.3. My interaction with the system would beclear and understandable.4. I would find the system to be flexible tointeract with.5. It would be easy for me to becomeskillful at using the system.6. I would find the system easy to use.Complexity(Thompson et al. 1991)The degree to which asystem is perceived asrelatively difficult tounderstand and use.1. Using the system takes too much timefrom my normal duties.2. Working with the system is socomplicated, it is difficult to understandwhat is going on.3. Using the system involves too much timedoing mechanical operations (e.g., datainput).4. It takes too long to learn how to use thesystem to make it worth the effort.Ease of Use(Moore and Benbasat1991)The degree to whichusing an innovation isperceived as beingdifficult to use.1. My interaction with the system is clearand understandable.2. I believe that it is easy to get the systemto do what I want it to do.3. Overall, I believe that the system is easyto use.4. Learning to operate the system is easyfor me.Social InfluenceSocial influence is defined as the degree to whichan individual perceives that important othersbelieve he or she should use the new system.Social influence as a direct determinant of behav-ioral intention is represented as subjective norm inTRA, TAM2, TPB/DTPB and C-TAM-TPB, socialfactors in MPCU, and image in IDT. Thompson etal. (1991) used the term social norms in definingtheir construct, and acknowledge its similarity tosubjective norm within TRA. While they havedifferent labels, each of these constructs containsthe explicit or implicit notion that the individual’sbehavior is influenced by the way in which theybelieve others will view them as a result of havingused the technology. Table 11 presents the threeconstructs related to social influence: subjectivenorm (TRA, TAM2, TPB/ DTPB, and C-TAM-TPB),social factors (MPCU), and image (IDT).The current model comparison (Tables 5, 6, and7) found that the social influence constructs listedabove behave similarly. None of the social influ-ence constructs are significant in voluntary con-texts; however, each becomes significant when
  • 28. Venkatesh et al./User Acceptance of IT452 MIS Quarterly Vol. 27 No. 3/September 2003Table 11. Social Influence: Root Constructs, Definitions, and ScalesConstruct Definition ItemsSubjective Norm(Ajzen 1991; Davis et al.1989; Fishbein and Azjen1975; Mathieson 1991;Taylor and Todd 1995a,1995b)The person’s perceptionthat most people who areimportant to him think heshould or should notperform the behavior inquestion.1. People who influence mybehavior think that I should usethe system.2. People who are important to methink that I should use thesystem.Social Factors(Thompson et al. 1991)The individual’s inter-nalization of the referencegroup’s subjective culture,and specific interpersonalagreements that the indivi-dual has made with others,in specific social situations.1. I use the system because of theproportion of coworkers who usethe system.2. The senior management of thisbusiness has been helpful in theuse of the system.3. My supervisor is very supportiveof the use of the system for myjob.4. In general, the organization hassupported the use of the system.Image(Moore and Benbasat 1991)The degree to which use ofan innovation is perceivedto enhance one’s image orstatus in one’s socialsystem.1. People in my organization whouse the system have moreprestige than those who do not.2. People in my organization whouse the system have a highprofile.3. Having the system is a statussymbol in my organization.use is mandated. Venkatesh and Davis (2000)suggested that such effects could be attributedto compliance in mandatory contexts thatcauses social influences to have a direct effecton intention; in contrast, social influence involuntary contexts operates by influencing per-ceptions about the technology—the mech-anisms at play here are internalization andidentification. In mandatory settings, socialinfluence appears to be important only in theearly stages of individual experience with thetechnology, with its role eroding over time andeventually becoming nonsignificant with sus-tained usage (T3), a pattern consistent with theobservations of Venkatesh and Davis (2000).The role of social influence in technologyacceptance decisions is complex and subject toa wide range of contingent influences. Socialinfluence has an impact on individual behaviorthrough three mechanisms: compliance, inter-nalization, and identification (see Venkatesh andDavis 2000; Warshaw 1980). While the lattertwo relate to altering an individual’s belief struc-ture and/or causing an individual to respond topotential social status gains, the compliancemechanism causes an individual to simply alterhis or her intention in response to the socialpressure—i.e., the individual intends to complywith the social influence. Prior research sug-gests that individuals are more likely to comply
  • 29. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 453with others’ expectations when those referentothers have the ability to reward the desiredbehavior or punish nonbehavior (e.g., Frenchand Raven 1959; Warshaw 1980). This view ofcompliance is consistent with results in thetechnology acceptance literature indicating thatreliance on others’ opinions is significant only inmandatory settings (Hartwick and Barki 1994),particularly in the early stages of experience,when an individual’s opinions are relatively ill-informed (Agarwal and Prasad 1997; Hartwickand Barki 1994; Karahanna et al. 1999; Taylorand Todd 1995a; Thompson et al. 1994; Venka-tesh and Davis 2000). This normative pressurewill attenuate over time as increasing exper-ience provides a more instrumental (rather thansocial) basis for individual intention to use thesystem.Theory suggests that women tend to be moresensitive to others’ opinions and therefore findsocial influence to be more salient when formingan intention to use new technology (Miller 1976;Venkatesh et al. 2000), with the effect decliningwith experience (Venkatesh and Morris 2000).As in the case of performance and effort expec-tancies, gender effects may be driven by psy-chological phenomena embodied within socially-constructed gender roles (e.g., Lubinski et al.1983). Rhodes’ (1983) meta-analytic review ofage effects concluded that affiliation needs in-crease with age, suggesting that older workersare more likely to place increased salience onsocial influences, with the effect declining withexperience (Morris and Venkatesh 2000).Therefore, we expect a complex interaction withthese moderating variables simultaneously influ-encing the social influence-intention relation-ship.H3: The influence of social influenceon behavioral intention will bemoderated by gender, age,voluntariness, and experience,such that the effect will bestronger for women, particularlyolder women, particularly in man-datory settings in the early stagesof experience.Facilitating ConditionsFacilitating conditions are defined as the degreeto which an individual believes that an organi-zational and technical infrastructure exists tosupport use of the system. This definition cap-tures concepts embodied by three different con-structs: perceived behavioral control (TPB/DTPB, C-TAM-TPB), facilitating conditions(MPCU), and compatibility (IDT). Each of theseconstructs is operationalized to include aspectsof the technological and/or organizational envi-ronment that are designed to remove barriers touse (see Table 12). Taylor and Todd (1995b)acknowledged the theoretical overlap bymodeling facilitating conditions as a core com-ponent of perceived behavioral control inTPB/DTPB. The compatibility construct fromIDT incorporates items that tap the fit betweenthe individual’s work style and the use of thesystem in the organization.The empirical evidence presented in Tables 5,6, 7, and 8 suggests that the relationshipsbetween each of the constructs (perceivedbehavioral control, facilitating conditions, andcompatibility) and intention are similar. Specifi-cally, one can see that perceived behavioralcontrol is significant in both voluntary and man-datory settings immediately following training(T1), but that the construct’s influence onintention disappears by T2. It has beendemonstrated that issues related to the supportinfrastructure—a core concept within the facili-tating conditions construct—are largely capturedwithin the effort expectancy construct which tapsthe ease with which that tool can be applied(e.g., Venkatesh 2000). Venkatesh (2000)found support for full mediation of the effect offacilitating conditions on intention by effortexpectancy. Obviously, if effort expectancy isnot present in the model (as is the case withTPB/DTPB), then one would expect facilitatingconditions to become predictive of intention.Our empirical results are consistent with thesearguments. For example, in TPB/DTPB, theconstruct is significant in predicting intention;however, in other cases (MPCU and IDT), it isnonsignificant in predicting intention. In short,
  • 30. Venkatesh et al./User Acceptance of IT454 MIS Quarterly Vol. 27 No. 3/September 2003Table 12. Facilitating Conditions: Root Constructs, Definitions, and ScalesConstruct Definition ItemsPerceived BehavioralControl(Ajzen 1991; Taylor andTodd 1995a, 1995b)Reflects perceptions ofinternal and externalconstraints on behaviorand encompasses self-efficacy, resource facili-tating conditions, andtechnology facilitatingconditions.1. I have control over using the system.2. I have the resources necessary to usethe system.3. I have the knowledge necessary touse the system.4. Given the resources, opportunitiesand knowledge it takes to use thesystem, it would be easy for me to usethe system.5. The system is not compatible withother systems I use.Facilitating Conditions(Thompson et al. 1991)Objective factors in theenvironment thatobservers agree makean act easy to do,including the provisionof computer support.1. Guidance was available to me in theselection of the system.2. Specialized instruction concerning thesystem was available to me.3. A specific person (or group) isavailable for assistance with systemdifficulties.Compatibility(Moore and Benbasat1991)The degree to which aninnovation is perceivedas being consistent withexisting values, needs,and experiences ofpotential adopters.1. Using the system is compatible withall aspects of my work.2. I think that using the system fits wellwith the way I like to work.3. Using the system fits into my workstyle.when both performance expectancy constructsand effort expectancy constructs are present,facilitating conditions becomes nonsignificant inpredicting intention.H4a: Facilitating conditions will nothave a significant influence onbehavioral intention.5The empirical results also indicate that facilitatingconditions do have a direct influence on usagebeyond that explained by behavioral intentionsalone (see Table 8). Consistent with TPB/DTPB,facilitating conditions are also modeled as a directantecedent of usage (i.e., not fully mediated byintention). In fact, the effect is expected toincrease with experience as users of technologyfind multiple avenues for help and supportthroughout the organization, thereby removingimpediments to sustained usage (Bergeron et al.1990). Organizational psychologists have notedthat older workers attach more importance toreceiving help and assistance on the job (e.g., Halland Mansfield 1975). This is further underscoredin the context of complex IT use given theincreasing cognitive and physical limitations asso-ciated with age. These arguments are in line withempirical evidence from Morris and Venkatesh(2000). Thus, when moderated by experience andage, facilitating conditions will have a significantinfluence on usage behavior.H4b: The influence of facilitating con-ditions on usage will be mode-5To test the nonsignificant relationship, we perform apower analysis in the results section.
  • 31. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 455rated by age and experience,such that the effect will bestronger for older workers, par-ticularly with increasing exper-ience.Constructs Theorized Not to BeDirect Determinants of IntentionAlthough self-efficacy and anxiety appeared to besignificant direct determinants of intention in SCT(see Tables 5 and 6), UTAUT does not includethem as direct determinants. Previous research(Venkatesh 2000) has shown self-efficacy andanxiety to be conceptually and empirically distinctfrom effort expectancy (perceived ease of use).Self-efficacy and anxiety have been modeled asindirect determinants of intention fully mediated byperceived ease of use (Venkatesh 2000). Consis-tent with this, we found that self-efficacy andanxiety appear to be significant determinants ofintention in SCT—i.e., without controlling for theeffect of effort expectancy. We therefore expectself-efficacy and anxiety to behave similarly, thatis, to be distinct from effort expectancy and tohave no direct effect on intention above andbeyond effort expectancy.H5a: Computer self-efficacy will nothave a significant influence onbehavioral intention.6H5b: Compute anxiety will not havea significant influence on beha-vioral intention.7Attitude toward using technology is defined as anindividual’s overall affective reaction to using asystem. Four constructs from the existing modelsalign closely with this definition: attitude towardbehavior (TRA, TPB/DTPB, C-TAM-TPB), intrinsicmotivation8(MM), affect toward use (MPCU), andaffect (SCT). Table 13 presents the definitionsand associated scale items for each construct.Each construct has a component associated withgeneralized feeling/affect associated with a givenbehavior (in this case, using technology). Inexamining these four constructs, it is evident thatthey all tap into an individual’s liking, enjoyment,joy, and pleasure associated with technology use.Empirically, the attitude constructs present aninteresting case (see Tables 5, 6, and 7). In somecases (e.g., TRA, TPB/DTPB, and MM), theattitude construct is significant across all threetime periods and is also the strongest predictor ofbehavioral intention. However, in other cases (C-TAM-TPB, MPCU, and SCT), the construct wasnot significant. Upon closer examination, theattitudinal constructs are significant only whenspecific cognitions—in this case, constructsrelated to performance and effort expectancies—are not included in the model. There is empiricalevidence to suggest that affective reactions (e.g.,intrinsic motivation) may operate through effortexpectancy (see Venkatesh 2000). Therefore, weconsider any observed relationship betweenattitude and intention to be spurious and resultingfrom the omission of the other key predictors(specifically, performance and effort expec-tancies). This spurious relationship likely stemsfrom the effect of performance and effort expec-tancies on attitude (see Davis et al. 1989). Thenon-significance of attitude in the presence ofsuch other constructs has been reported in pre-vious model tests (e.g., Taylor and Todd 1995a;Thompson et al. 1991), despite the fact that thisfinding is counter to what is theorized in TRA andTPB/DTPB. Given that we expect strong relation-ships in UTAUTbetween performance expectancyand intention, and between effort expectancy andintention, we believe that, consistent with the logicdeveloped here, attitude toward using technology6To test the nonsignificant relationship, we perform apower analysis in the results section.7To test the nonsignificant relationship, we perform apower analysis in the results section.8Some perspectives differ on the role of intrinsicmotivation. For example, Venkatesh (2000) models it asa determinant of perceived ease of use (effortexpectancy). However, in the motivational model, it isshown as a direct effect on intention and is shown assuch here.
  • 32. Venkatesh et al./User Acceptance of IT456 MIS Quarterly Vol. 27 No. 3/September 2003Table 13. Attitude Toward Using Technology: Root Constructs, Definitions, andScalesConstruct Definition ItemsAttitude TowardBehavior(Davis et al. 1989;Fishbein and Ajzen1975; Taylor and Todd1995a, 1995b)An individual’s positive ornegative feelings aboutperforming the targetbehavior.1. Using the system is a bad/good idea.2. Using the system is a foolish/wise idea.3. I dislike/like the idea of using thesystem.4. Using the system is unpleasant/pleasant.Intrinsic Motivation(Davis et al. 1992)The perception that userswill want to perform anactivity for no apparentreinforcement other thanthe process of performingthe activity per se.1. I find using the system to be enjoyable2. The actual process of using the systemis pleasant.3. I have fun using the system.Affect Toward Use(Thompson et al. 1991)Feelings of joy, elation, orpleasure; or depression,disgust, displeasure, orhate associated by anindividual with a particularact.1. The system makes work moreinteresting.2. Working with the system is fun.3. The system is okay for some jobs, butnot the kind of job I want. (R)Affect(Compeau and Higgins1995b; Compeau et al.1999)An individual’s liking ofthe behavior.1. I like working with the system.2. I look forward to those aspects of myjob that require me to use the system.3. Using the system is frustrating for me.(R)4. Once I start working on the system, Ifind it hard to stop.5. I get bored quickly when using thesystem. (R)will not have a direct or interactive influence onintention.H5c: Attitude toward using techno-logy will not have a significantinfluence on behavioralintention.9Behavioral IntentionConsistent with the underlying theory for all of theintention models10discussed in this paper, weexpect that behavioral intention will have asignificant positive influence on technology usage.H6: Behavioral intention will have asignificant positive influence onusage.9To test the absence of a relationship, we perform apower analysis in the results section.10For example, see Sheppard et al. (1988) for anextended review of the intention-behavior relationship.
  • 33. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 457Empirical Validation of UTAUTPreliminary Test of UTAUTUsing the post-training data (T1) pooled acrossstudies (N = 215), a measurement model of theseven direct determinants of intention (using allitems that related to each of the constructs) wasestimated. All constructs, with the exception ofuse, were modeled using reflective indicators. Allinternal consistency reliabilities (ICRs) weregreater than .70. The square roots of the sharedvariance between the constructs and their mea-sures were higher than the correlations acrossconstructs, supporting convergent and discri-minant validity—see Table 14(a). The reverse-coded affect items of Compeau and Higgins(1995b) had loadings lower than .60 and weredropped and the model was reestimated. Withthe exception of eight loadings, all others weregreater than .70, the level that is generally con-sidered acceptable (Fornell and Larcker 1981; seealso Compeau and Higgins 1995a, 1995b; Com-peau et al. 1999)—see Table 15. Inter-item corre-lation matrices (details not shown here due tospace constraints) confirmed that intra-constructitem correlations were very high while inter-con-struct item correlations were low. Results ofsimilar analyses from subsequent time periods (T2and T3) also indicated an identical pattern and areshown in Tables 14(b) and 14(c).Although the structural model was tested on allthe items, the sample size poses a limitation herebecause of the number of latent variables andassociated items. Therefore, we reanalyzed thedata using only four of the highest loading itemsfrom the measurement model for each of thedeterminants; intention only employs three itemsand use is measured via a single indicator.UTAUT was estimated using data pooled acrossstudies at each time period (N = 215). As with themodel comparison tests, the bootstrappingmethod was used here to test the PLS models.An examination of the measurement model fromthe analysis using the reduced set of items wassimilar to that reported in Tables 14 and 15 interms of reliability, convergent validity, discrimi-nant validity, means, standard deviations, andcorrelations. The results from the measurementmodel estimations (with reduced items) are notshown here in the interest of space.An examination of these highest loading itemssuggested that they adequately represented theconceptual underpinnings of the constructs—thispreliminary content validity notwithstanding, wewill return to this issue later in our discussion ofthe limitations of this work. Selection based onitem loadings or corrected item-total correlationsare often recommended in the psychometricliterature (e.g., Nunnally and Bernstein 1994).This approach favors building a homogenousinstrument with high internal consistency, butcould sacrifice content validity by narrowingdomain coverage.11The items selected for furtheranalysis are indicated via an asterisk in Table 15,and the actual items are shown in Table 16.Tables 17(a) and 17(b) show the detailed modeltest results at each time period for intention andusage, respectively, including all lower-levelinteraction terms. Tables 17(a) and 17(b) alsoshow the model with direct effects only so thereader can compare that to a model that includesthe moderating influences. The variance ex-plained at various points in time by a directeffects-only model and the full model includinginteraction terms are shown in Tables 17(a) and17(b) for intention and usage behavior, respec-tively.12We pooled the data across the different11Bagozzi and Edwards (1998) discuss promising newalternatives to this approach for coping with the inherenttension between sample size requirements and thenumber of items, such as full or partial aggregation ofitems.12Since PLS does not produce adjusted R2, we used analternative procedure to estimate an adjusted R2. PLSestimates a measurement model that in turn is used togenerate latent variable observations based on theloadings; these latent variable observations are used toestimate the structural model using OLS. Therefore, inorder to estimate the adjusted R2, we used the latentvariable observations generated from PLS and analyzedthe data using hierarchical regressions in SPSS. Wereport the R2and adjusted R2from the hierarchicalregressions. This allows for a direct comparison ofvariance explained from PLS with variance explained(both R2and adjusted R2) from traditional OLSregressions and allows the reader to evaluate thevariance explained-parsimony trade-off.
  • 34. Venkatesh et al./User Acceptance of IT458 MIS Quarterly Vol. 27 No. 3/September 2003Table 14. Measurement Model Estimation for the Preliminary Test of UTAUT(a) T1 Results (N = 215)ICR Mean S Dev PE EE ATUT SI FC SE ANX BIPE .92 5.12 1.13 .94EE .91 4.56 1.40 .31*** .91ATUT .84 4.82 1.16 .29*** .21** .86SI .88 4.40 1.04 .30*** -.16* .21** .88FC .87 4.17 1.02 .18* .31*** .17* .21** .89SE .89 5.01 1.08 .14 .33*** .16* .18** .33*** .87ANX .83 3.11 1.14 -.10 -.38*** -.40*** -.20** -.18** -.36*** .84BI .92 4.07 1.44 .38*** .34*** .25*** .35*** .19** .16* -.23** .84(b) T2 Results (N = 215)ICR Mean S Dev PE EE ATUT SI FC SE ANX BIPE .91 4.71 1.11 .92EE .90 5.72 0.77 .30*** .90ATUT .77 5.01 1.42 .25** .20** .86SI .94 3.88 1.08 .27*** -.19* .21** .88FC .83 3.79 1.17 .19* .31*** .18* .20** .86SE .89 5.07 1.14 .24** .35*** .19** .21** .33*** .75ANX .79 3.07 1.45 -.07 -.32*** -.35*** -.21** -.17* -.35*** .82BI .90 4.19 0.98 .41*** .27*** .23** .21** .16* .16* -.17* .87(c) T3 Results (N = 215)ICR Mean S Dev PE EE ATUT SI FC SE ANX BIPE .91 4.88 1.17 .94EE .94 5.88 0.62 .34*** .91ATUT .81 5.17 1.08 .21** .24** .79SI .92 3.86 1.60 .27*** -.15 .20** .93FC .85 3.50 1.12 .19* .28*** .18* .22** .84SE .90 5.19 1.07 .14* .30*** .22** .20** .33*** .77ANX .82 2.99 1.03 -.11 -.30*** -.30*** -.20** -.24** -.32*** .82BI .90 4.24 1.07 .44*** .24** .20** .16* .16* .16* -.14 .89Notes: 1. ICR: Internal consistency reliability.2. Diagonal elements are the square root of the shared variance between the constructs andtheir measures; off-diagonal elements are correlations between constructs.3. PE: Performance expectancy; EE: Effort expectancy; ATUT: Attitude toward usingtechnology; SI: Social influence; FC: Facilitating conditions; SE: Self-efficacy; ANX: Anxiety;BI: Behavioral intention to use the system.
  • 35. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 459Table 15. Item Loadings from PLS (N = 215 at Each Time Period)Items T1 T2 T3PerformanceExpectancy(PE)U1 .82 .81 .80U2 .84 .80 .81U3 .81 .84 .84U4 .80 .80 .84U5 .81 .78 .84*U6 .88 .88 .90*RA1 .87 .90 .90RA2 .73 .70 .79RA3 .70 .69 .83RA4 .71 .74 .74*RA5 .86 .88 .94JF1 .70 .71 .69JF2 .67 .73 .64JF3 .74 .70 .79JF4 .73 .79 .71JF5 .77 .71 .73JF6 .81 .78 .81OE1 .72 .80 .75OE2 .71 .68 .77OE3 .75 .70 .70OE4 .70 .72 .67OE5 .72 .72 .70OE6 .69 .79 .74*OE7 .86 .87 .90EffortExpectancy(EE)EOU1 .90 .89 .83EOU2 .90 .89 .88*EOU3 .94 .96 .91EOU4 .81 .84 .88*EOU5 .91 .90 .90*EOU6 .92 .92 .93EU1 .84 .80 .84EU2 .83 .88 .85EU3 .89 .84 .80*EU4 .91 .91 .92CO1 .83 .82 .81CO2 .83 .78 .80CO3 .81 .84 .80CO4 .75 .73 .78AttitudeTowardUsingTechnology(ATUT)*A1 .80 .83 .85A2 .67 .64 .65A3 .64 .64 .71A4 .72 .71 .64IM1 .70 .78 .72IM2 .72 .72 .78IM3 .73 .75 .81*AF1 .79 .77 .84*AF2 .84 .83 .84AF3 .71 .70 .69*Affect1 .82 .85 .82Affect2 .67 .70 .70Affect3 .62 .68 .64Items TI T2 T3SocialInfluence(SI)*SN1 .82 .85 .90*SN2 .83 .85 .84SF1 .71 .69 .76*SF2 .84 .80 .90SF3 .72 .74 .77*SF4 .80 .82 .84I1 .69 .72 .72I2 .65 .75 .70I3 .71 .72 .69FacilitatingConditions(FC)PBC1 .72 .66 .62*PBC2 .84 .81 .80*PBC3 .81 .81 .82PBC4 .71 .69 .70*PBC5 .80 .82 .80FC1 .74 .73 .69FC2 .78 .77 .64*FC3 .80 .80 .82C1 .72 .72 .70C2 .71 .74 .74C3 .78 .77 .70Self-Efficacy(SE)*SE1 .80 .83 .84SE2 .78 .80 .80SE3 .72 .79 .74*SE4 .80 .84 .84SE5 .77 .74 .69*SE6 .81 .82 .82*SE7 .87 .85 .86SE8 .70 .69 .72Anxiety(ANX)*ANX1 .78 .74 .69*ANX2 .71 .70 .72*ANX3 .72 .69 .73*ANX4 .74 .72 .77Inten-tion(BI)*BI1 .88 .84 .88*BI2 .82 .86 .88*BI3 .84 .88 .87Notes: 1. The loadings at T1, T2, and T3 respectively are from separate measurement model tests and relate toTables 14(a), 14(b), and 14(c) respectively.2. Extrinsic motivation (EM) was measured using the same scale as perceived usefulness (U).3. Items denoted with an asterisk are those that were selected for inclusion in the test of UTAUT.
  • 36. Venkatesh et al./User Acceptance of IT460 MIS Quarterly Vol. 27 No. 3/September 2003Table 16. Items Used in Estimating UTAUTPerformance expectancyU6: I would find the system useful in my job.RA1: Using the system enables me to accomplish tasks more quickly.RA5: Using the system increases my productivity.OE7: If I use the system, I will increase my chances of getting a raise.Effort expectancyEOU3: My interaction with the system would be clear and understandable.EOU5: It would be easy for me to become skillful at using the system.EOU6: I would find the system easy to use.EU4: Learning to operate the system is easy for me.Attitude toward using technologyA1: Using the system is a bad/good idea.AF1: The system makes work more interesting.AF2: Working with the system is fun.Affect1: I like working with the system.Social influenceSN1: People who influence my behavior think that I should use the system.SN2: People who are important to me think that I should use the system.SF2: The senior management of this business has been helpful in the use of the system.SF4: In general, the organization has supported the use of the system.Facilitating conditionsPBC2: I have the resources necessary to use the system.PBC3: I have the knowledge necessary to use the system.PBC5: The system is not compatible with other systems I use.FC3: A specific person (or group) is available for assistance with system difficulties.Self-efficacyI could complete a job or task using the system…SE1: If there was no one around to tell me what to do as I go.SE4: If I could call someone for help if I got stuck.SE6: If I had a lot of time to complete the job for which the software was provided.SE7: If I had just the built-in help facility for assistance.AnxietyANX1: I feel apprehensive about using the system.ANX2: It scares me to think that I could lose a lot of information using the system by hittingthe wrong key.ANX3: I hesitate to use the system for fear of making mistakes I cannot correct.ANX4: The system is somewhat intimidating to me.Behavioral intention to use the systemBI1: I intend to use the system in the next <n> months.BI2: I predict I would use the system in the next <n> months.BI3: I plan to use the system in the next <n> months.
  • 37. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 461points of measurement by converting time period(experience) into a moderator. The column titled“Pooled Analysis” reports the results of thisanalysis. Caution is necessary when conductingsuch analyses and the reader is referred toAppendix A for a discussion of the potential con-straints of pooling. Also reported in Appendix Aare the statistical tests that we conducted prior topooling the data for the analysis in Table 17.As expected, the effect of performance expec-tancy was in the form of a three-way inter-action—the effect was moderated by gender andage such that it was more salient to youngerworkers, particularly men, thus supporting H1.Note that a direct effect for performance expec-tancy on intention was observed; however, thesemain effects are not interpretable due to thepresence of interaction terms (e.g., Aiken andWest 1991). The effect of effort expectancy wasvia a three-way interaction—the effect was moder-ated by gender and age (more salient to womenand more so to older women). Based on Chow’stest of beta differences (p < .05), effort expectancywas more significant with limited exposure to thetechnology (effect decreasing with experience),thus supporting H2. The effect of social influencewas via a four-way interaction—with its role beingmore important in the context of mandatory use,more so among women, and even more so amongolder women. The Chow’s test of beta differences(p < .05) indicated that social influence was evenmore significant in the early stages of individualexperience with the technology, thus supportingH3. Facilitating conditions was nonsignificant asa determinant of intention, thus supporting H4a.13As expected self-efficacy, anxiety, and attitudedid not have any direct effect on intention, thussupporting H5a, H5b, and H5c. Three of the fournonsignificant determinants (i.e., self-efficacy,anxiety, and attitude) were dropped from themodel and the model was reestimated; facilitatingconditions was not dropped from the modelbecause of its role in predicting use.14The re-estimated model results are shown in Table 17.In predicting usage behavior (Table 17b), bothbehavioral intention (H6)and facilitating conditionswere significant, with the latter’s effect beingmoderated by age (the effect being more impor-tant to older workers); and, based on Chow’s testof the beta differences (p < .05), the effect wasstronger with increasing experience with thetechnology (H4b). Since H4a, H5a, H5b, and H5care hypothesized such that the predictor is notexpected to have an effect on the dependentvariable, a power analysis was conducted toexamine the potential for type II error. The likeli-hood of detecting medium effects was over 95percent for an alpha level of .05 and the likelihoodof detecting small effects was under 50 pecent.Cross-Validation of UTAUTData were gathered from two additional organi-zations to further validate UTAUT and addexternal validity to the preliminary test. The majordetails regarding the two participating organi-zations are provided in Table 18. The data werecollected on the same timeline as studies 1 and 2(see Figure 2). The data analysis procedureswere the same as the previous studies. Theresults were consistent with studies 1 and 2. Theitems used in the preliminary test of UTAUT (listedin Table 16) were used to estimate the mea-surement and structural models in the new data.This helped ensure the test of the same model inboth the preliminary test and this validation, thuslimiting any variation due to the changing of items.UTAUT was then tested separately for each time13Since these two hypotheses were about nonsignificantrelationships, the supportive results should be inter-preted with caution, bearing in mind the power analysisreported in the text. In this case, the concern about thelack of a relationship is somewhat alleviated as someprevious research in this area has found attitude andfacilitating conditions to not be predictors of intention inthe presence of performance expectancy and effortexpectancy constructs (see Taylor and Todd 1995b;Venkatesh 2000).14Since these two hypotheses were about nonsignificantrelationships, the supportive results should be inter-preted with caution after consideration of the poweranalyses reported in the text. The results are presentedhere for model completeness and to allow the reader tolink the current research with existing theoreticalperspectives in this domain.
  • 38. Venkatesh et al./User Acceptance of IT462 MIS Quarterly Vol. 27 No. 3/September 2003Table 17. Preliminary Test of UTAUT(a) Dependent Variable: IntentionT1(N = 215)T2(N = 215)T3(N = 215)Pooled(N = 645)D ONLY D + I D ONLY D + I D ONLY D + I D ONLY D + IR2(PLS) .40 .51 .41 .52 .42 .50 .31 .76R2(hierarchical regrn.) .39 .51 .41 .51 .42 .50 .31 .77Adjusted R2(hierarchical regrn.) .35 .46 .38 .46 .36 .45 .27 .69Performance expectancy (PE) .46*** .17* .57*** .15* .59*** .16* .53*** .18*Effort expectancy (EE) .20* -.12 .08 .02 .09 .11 .10 .04Social influence (SI) .13 .10 .10 .07 .07 .04 .11 .01Facilitating conditions (FC) .03 .04 .02 .01 .01 .01 .09 .04Gender (GDR) .04 .02 .04 .01 .02 .01 .03 .01Age (AGE) .08 .02 .09 .08 .01 -.08 .06 .00Voluntariness (VOL) .01 .04 .03 .02 .04 -.04 .02 .00Experience (EXP) .04 .00PE × GDR .07 .17* .06 .02PE × AGE .13 .04 .10 .01GDR × AGE .07 .02 .02 .06PE × GDR × AGE .52*** .55*** .57*** .55***EE × GDR .17* .08 .09 .02EE × AGE .08 .04 .02 .04EE × EXP .02GDR × AGE (included earlier) Earlier Earlier Earlier EarlierGDR × EXP .02AGE × EXP .01EE × GDR × AGE .22** .20*** .18* .01EE × GDR × EXP -.10EE × AGE × EXP -.02GDR × AGE × EXP -.06EE × GDR × AGE × EXP -.27***SI × GDR .11 .00 .02 .02SI × AGE .01 .06 .01 .02SI × VOL .02 .01 .02 .06SI × EXP .04GDR × AGE (included earlier) Earlier Earlier Earlier EarlierGDR × VOL .01 .04 .02 .01GDR × EXP (included earlier) EarlierAGE × VOL .00 .02 .06 .02AGE × EXP (included earlier) EarlierVOL × EXP .02SI × GDR × AGE -.10 .02 .04 .04SI × GDR × VOL -.01 .03 .02 .01SI × GDR × EXP .01SI × AGE × VOL -.17* .02 .06 .06SI × AGE × EXP .01SI × VOL × EXP .00
  • 39. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 463Table 17. Preliminary Test of UTAUT (Continued)T1(N = 215)T2(N = 215)T3(N = 215)Pooled(N = 645)D ONLY D + I D ONLY D + I D ONLY D + I D ONLY D + IGDR × AGE × VOL .02 .02 .01 .00GDR × AGE × EXP (includedearlier)EarlierGDR × VOL × EXP .00AGE × VOL × EXP .01SI × GDR × AGE × VOL .25** .23** .20* .04GDR × AGE × VOL × EXP .02SI × GDR × AGE × VOL × EXP -.28***(b) Dependent Variable: Usage BehaviorR2(PLS) .37 .43 .36 .43 .39 .44 .38 .53R2(hierarchical regrn.) .37 .43 .36 .43 .39 .43 .38 .52Adjusted R2(hierarchical regrn.) .36 .41 .35 .40 .38 .41 .37 .47Behavioral intention (BI) .61*** .57*** .60*** .58*** .58*** .59*** .59*** .52***Facilitating conditions (FC) .05 .07 .06 .07 .18* .07 .10 .11Age (AGE) .02 .02 .01 .02 .04 .13 .04 .08Experience (EXP) .06FC × AGE .22* .24* .27** .02FC × EXP .00AGE × EXP .01FC × AGE × EXP .23**Notes: 1. D ONLY: Direct effects only; D + I: Direct effects and interaction terms.2. “Included earlier” indicates that the term has been listed earlier in the table, but is included again forcompleteness as it relates to higher-order interaction terms being computed. Grayed out cells are notapplicable for the specific column.Table 18. Description of StudiesStudy IndustryFunctionalAreaSampleSize System DescriptionVoluntary Use3 FinancialServicesResearch 80 This software application was one of theresources available to analysts to conductresearch on financial investment opportunitiesand IPOsMandatory Use4 RetailElectronicsCustomerService53 Application that customer service representa-tives were required to use to document andmanage service contracts
  • 40. Venkatesh et al./User Acceptance of IT464 MIS Quarterly Vol. 27 No. 3/September 2003Table 19. Measurement Model Estimation for the Cross-Validation of UTAUT(a) T1 Results (N = 133)ICR Mean S Dev PE EE ATUT SI FC SE ANX BIPE .90 4.87 1.20 .88EE .90 3.17 1.09 .30*** .93ATUT .80 2.67 0.87 .28*** .16* .80SI .91 4.01 1.07 .34*** -.21* .20* .89FC .85 3.12 1.11 .18* .30*** .15* .14 .84SE .85 4.12 1.08 .19** .37*** .17* .22** .41*** .82ANX .82 2.87 0.89 -.17* -.41*** -.41*** -.22** -.25** -.35*** .80BI .89 4.02 1.19 .43*** .31*** .23*** .31*** .22* .24** -.21** .85(b) T2 Results (N=133)ICR Mean S Dev PE EE ATUT SI FC SE ANX BIPE .90 4.79 1.22 .91EE .92 4.12 1.17 .34*** .89ATUT .84 3.14 0.79 .32*** .18** .78SI .92 4.11 1.08 .30*** -.20* .21** .91FC .88 3.77 1.08 .19* .31*** .13 .11 .86SE .87 4.13 1.01 .21** .38*** .25** .20** .35*** .80ANX .83 3.00 0.82 -.17* -.42*** -.32*** -.24** -.26* -.36*** .81BI .88 4.18 0.99 .40*** .24** .21** .24** .20* .21** -.22** .88(c) T3 Results (N = 133)ICR Mean S Dev PE EE ATUT SI FC SE ANX BIPE .94 5.01 1.17 .92EE .92 4.89 0.88 .34*** .90ATUT .83 3.52 1.10 .30*** .19** .80SI .92 4.02 1.01 .29*** -.21* .21** .84FC .88 3.89 1.00 .18* .31*** .14 .15 .81SE .87 4.17 1.06 .18** .32*** .21* .20* .35*** .84ANX .85 3.02 0.87 -.15* -.31*** -.28*** -.22* -.25* -.38*** .77BI .90 4.07 1.02 .41*** .20** .21* .19* .20* .20* -.21** .84Notes: 1. ICR: Internal consistency reliability.2. Diagonal elements are the square root of the shared variance between the constructs andtheir measures; off-diagonal elements are correlations between constructs.3. PE: Performance expectancy; EE: Effort expectancy; ATUT: Attitude toward usingtechnology; SI: Social influence; FC: Facilitating conditions; SE: Self-efficacy; ANX: Anxiety;BI: Behavioral intention to use the system.
  • 41. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 465Table 20. Item Loadings from PLS (N=133 at each time period)Items T1 T2 T3 Items T1 T2 T3PerformanceExpectancy(PE)U6 .91 .92 .91FacilitatingConditions(FC)PBC2 .84 .88 .85RA1 .90 .89 .88 PBC3 .88 .89 .88RA5 .94 .89 .90 PBC5 .86 .89 .84OE7 .89 .90 .91 FC3 .87 .78 .81EffortExpectancy(EE)EOU3 .91 .90 .94Self-Efficacy(SE)SE1 .90 .84 .88EOU5 .92 .91 .90 SE4 .88 .82 .81EOU6 .93 .90 .89 SE6 .80 .85 .79EU4 .87 .87 .90 SE7 .81 .77 .75AttitudeTowardUsingTechnology(ATUT)A1 .84 .80 .86Anxiety(ANX)ANX1 .80 .84 .80AF1 .82 .83 .77 ANX2 .84 .84 .82AF2 .80 .80 .76 ANX3 .83 .80 .83Affect1 .87 .84 .76 ANX4 .84 .77 .83SocialInfluence(SI)SN1 .94 .90 .90Intention(BI)BI1 .92 .90 .91SN2 .90 .93 .88 BI2 .90 .90 .91SF2 .89 .92 .94 BI3 .90 .92 .92SF4 .92 .81 .79Note: The loadings at T1, T2, and T3 respectively are from separate measurement model tests andrelate to Tables 18(a), 18(b), and 18(c) respectively.Table 21. Cross-Validation of UTAUT(a) Dependent Variable: IntentionT1(N = 133)T2(N = 133)T3(N = 133)Pooled(N = 399)D ONLY D + I D ONLY D + I D ONLY D + I D ONLY D + IR2(PLS) .42 .52 .41 .52 .42 .51 .36 .77R2(hierarchical regrn.) .41 .52 .41 .52 .42 .51 .36 .77Adjusted R2(hierarchical regrn.) .37 .48 .36 .47 .36 .46 .30 .70Performance expectancy (PE) .45*** .15 .59*** .16* .59*** .15* .53*** .14Effort expectancy (EE) .22** .02 .06 .06 .04 .01 .10 .02Social influence (SI) .02 .04 .04 .04 .02 .01 .02 .02Facilitating conditions (FC) .07 .01 .00 .08 .01 .00 .07 .01Gender (GDR) .02 .06 .01 .04 .07 .02 .03 .03Age (AGE) .01 .00 .07 .02 .02 .01 .07 .01Voluntariness (VOL) .00 .00 .01 .06 .02 .04 .00 .01Experience (EXP) .08 .00PE × GDR .14 .17* .18* .01PE × AGE .06 .01 .02 -.04GDR × AGE .02 .04 .04 -.02EE × GDR -.06 .02 .04 -.09EE × AGE -.02 .01 .02 -.04EE × EXP .01
  • 42. Venkatesh et al./User Acceptance of IT466 MIS Quarterly Vol. 27 No. 3/September 2003Table 21. Cross-Validation of UTAUT (Continued)T1(N = 133)T2(N = 133)T3(N = 133)Pooled(N = 399)D ONLY D + I D ONLY D + I D ONLY D + I D ONLY D + IGDR × AGE (included earlier) Earlier Earlier Earlier EarlierGDR × EXP .02AGE × EXP .06EE × GDR × AGE .21** .18* .16* .04EE × GDR × EXP .00EE × AGE × EXP .00GDR × AGE × EXP .01EE × GDR × AGE × EXP -.25***SI × GDR -.06 .00 .02 -.07SI × AGE .02 .01 .04 .02SI × VOL .01 .04 .00 .00SI × EXP .02GDR × AGE (included earlier) Earlier Earlier Earlier EarlierGDR × VOL .04 .02 -.03 .02GDR × EXP (included earlier) EarlierAGE × VOL .00 .01 .00 .07AGE × EXP (included earlier) EarlierVOL × EXP .02SI × GDR × AGE -.03 -.01 -.07 .00SI × GDR × VOL .04 -.03 .04 .00SI × GDR × EXP .00SI × AGE × VOL .06 .02 .00 .01SI × AGE × EXP .07SI × VOL × EXP .02GDR × AGE × VOL .01 .06 .01 .04GDR × AGE × EXP (includedearlier)EarlierGDR × VOL × EXP .01AGE × VOL × EXP .00SI × GDR × AGE × VOL .27*** .21** .16* .01GDR × AGE × VOL × EXP .00SI × GDR × AGE × VOL × EXP -.29***(b) Dependent Variable: Usage BehaviorR2(PLS) .37 .44 .36 .41 .36 .44 .38 .52R2(hierarchical regrn.) .37 .43 .36 .41 .36 .44 .37 .52Adjusted R2(hierarchical regrn.) .36 .41 .35 .38 .35 .41 .36 .48Behavioral intention (BI) .60*** .56*** .59*** .50*** .59*** .56*** .59*** .51***Facilitating conditions (FC) .04 .11 .01 .01 .02 .06 .14* .08Age (AGE) .06 .02 .06 -.03 -.03 -.01 .06 .02Experience (EXP) .10FC × AGE .17* .21** .24** .01FC × EXP -.06AGE × EXP -.07FC × AGE × EXP .22**Notes: 1. D ONLY: Direct effects only; D + I: Direct effects and interaction terms.2. “Included earlier” indicates that the term has been listed earlier in the table, but is included again forcompleteness as it relates to higher-order interaction terms being computed. Grayed out cells are notapplicable for the specific column.
  • 43. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 467period (N = 133 at each time period). The mea-surement models are shown in Tables 19 and 20.The pattern of results in this validation (Tables21(a) and 21(b)) mirrors what was found in thepreliminary test (Table 17). The last column ofTables 21(a) and 21(b) reports observations fromthe pooled analysis as before. Appendix B reportsthe statistical tests we conducted prior to poolingthe data for the cross-validation test, consistentwith the approach taken in the preliminary test.Insofar as the no-relationship hypotheses wereconcerned, the power analysis revealed a highlikelihood (over 95 percent) of detecting mediumeffects. The variance explained was quite com-parable to that found in the preliminary test ofUTAUT.DiscussionThe present research set out to integrate thefragmented theory and research on individualacceptance of information technology into a uni-fied theoretical model that captures the essentialelements of eight previously established models.First, we identified and discussed the eightspecific models of the determinants of intentionand usage of information technology. Second,these models were empirically compared usingwithin-subjects, longitudinal data from four organi-zations. Third, conceptual and empirical simi-larities across the eight models were used toformulate the Unified Theory of Acceptance andUse of Technology (UTAUT). Fourth, the UTAUTwas empirically tested using the original data fromthe four organizations and then cross-validatedusing new data from an additional two organi-zations. These tests provided strong empiricalsupport for UTAUT, which posits three directdeterminants of intention to use (performanceexpectancy, effort expectancy, and social influ-ence) and two direct determinants of usagebehavior (intention and facilitating conditions).Significant moderating influences of experience,voluntariness, gender, and age were confirmed asintegral features of UTAUT. UTAUT was able toaccount for 70 percent of the variance (adjustedR2) in usage intention—a substantial improvementover any of the original eight models and theirextensions. Further, UTAUT was successful inintegrating key elements from among the initial setof 32 main effects and four moderators asdeterminants of intention and behavior collectivelyposited by eight alternate models into a model thatincorporated four main effects and fourmoderators.Thus, UTAUT is a definitive model that synthe-sizes what is known and provides a foundation toguide future research in this area. By encom-passing the combined explanatory power of theindividual models and key moderating influences,UTAUT advances cumulative theory while re-taining a parsimonious structure. Figure 3 pre-sents the model proposed and supported.Table 22 presents a summary of the findings. Itshould be noted that performance expectancyappears to be a determinant of intention in mostsituations: the strength of the relationship varieswith gender and age such that it is more signi-ficant for men and younger workers. The effect ofeffort expectancy on intention is also moderatedby gender and age such that it is more significantfor women and older workers, and those effectsdecrease with experience. The effect of socialinfluence on intention is contingent on all fourmoderators included here such that we found it tobe nonsignificant when the data were analyzedwithout the inclusion of moderators. Finally, theeffect of facilitating conditions on usage was onlysignificant when examined in conjunction with themoderating effects of age and experience—i.e.,they only matter for older workers in later stagesof experience.Prior to discussing the implications of this work, itis necessary to recognize some of its limitations.One limitation concerns the scales used to mea-sure the core constructs. For practical analyticalreasons, we operationalized each of the coreconstructs in UTAUT by using the highest-loadingitems from each of the respective scales. Thisapproach is consistent with recommendations inthe psychometric literature (e.g., Nunnally andBernstein 1994). Such pruning of the instrumentwas the only way to have the degrees of freedomnecessary to model the various interaction terms
  • 44. Venkatesh et al./User Acceptance of IT468 MIS Quarterly Vol. 27 No. 3/September 2003Table 22. Summary of FindingsHypothesisNumberDependentVariablesIndependentVariables Moderators ExplanationH1 BehavioralintentionPerformanceexpectancyGender, Age Effect stronger for men andyounger workersH2 BehavioralintentionEffortexpectancyGender, Age,ExperienceEffect stronger for women,older workers, and those withlimited experienceH3 BehavioralintentionSocialinfluenceGender, Age,Voluntariness,ExperienceEffect stronger for women,older workers, under conditionsof mandatory use, and withlimited experienceH4a BehavioralintentionFacilitatingconditionsNone Nonsignificant due to the effectbeing captured by effortexpectancyH4b Usage FacilitatingconditionsAge,ExperienceEffect stronger for olderworkers with increasingexperienceH5a BehavioralintentionComputerself-efficacyNone Nonsignificant due to the effectbeing captured by effortexpectancyH5b BehavioralintentionComputeranxietyNone Nonsignificant due to the effectbeing captured by effortexpectancyH5c BehavioralintentionAttitudetoward usingtech.None Nonsignificant to the effectbeing captured by processexpectancy and effortexpectancyH6 Usage BehavioralintentionNone Direct effectat the item level as recommended by Chin et al.(1996). However, one danger of this approach isthat facets of each construct can be eliminated,thus threatening content validity. Specifically, wefound that choosing the highest-loading itemsresulted in items from some of the models notbeing represented in some of the core constructs(e.g., items from MPCU were not represented inperformance expectancy). Therefore, the mea-sures for UTAUT should be viewed as preliminaryand future research should be targeted at morefully developing and validating appropriate scalesfor each of the constructs with an emphasis oncontent validity, and then revalidating the modelspecified herein (or extending it accordingly) withthe new measures. Our research employed stan-dard measures of intention, but future researchshould examine alternative measures of intentionand behavior in revalidating or extending theresearch presented here to other contexts.From a theoretical perspective, UTAUT providesa refined view of how the determinants of intentionand behavior evolve over time. It is important to
  • 45. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 469emphasize that most of the key relationships inthe model are moderated. For example, age hasreceived very little attention in the technologyacceptance research literature, yet our resultsindicate that it moderates all of the key relation-ships in the model. Gender, which has receivedsome recent attention, is also a key moderatinginfluence; however, consistent with findings in thesociology and social psychology literature (e.g.,Levy 1988), it appears to work in concert with age,a heretofore unexamined interaction. For exam-ple, prior research has suggested that effortexpectancy is more salient for women (e.g.,Venkatesh and Morris 2000). While this may betrue, our findings suggest this is particularly truefor the older generation of workers and those withrelatively little experience with a system. Whileexisting studies have contributed to our under-standing of gender and age influences indepen-dently, the present research illuminates the inter-play of these two key demographic variables andadds richness to our current understanding of thephenomenon. We interpret our findings to sug-gest that as the younger cohort of employees inthe workforce mature, gender differences in howeach perceives information technology maydisappear. This is a hopeful sign and suggeststhat oft-mentioned gender differences in the use ofinformation technology may be transitory, at leastas they relate to a younger generation of workersraised and educated in the Digital Age.The complex nature of the interactions observed,particularly for gender and age, raises severalinteresting issues to investigate in future research,especially given the interest in today’s societaland workplace environments to create equitablesettings for women and men of all ages. Futureresearch should focus on identifying the potential“magic number” for age where effects begin toappear (say for effort expectancy) or disappear(say for performance expectancy). While genderand age are the variables that reveal an inter-esting pattern of results, future research shouldidentify the underlying influential mechanisms—potential candidates here include computerliteracy and social or cultural background, amongothers. Finally, although gender moderates threekey relationships, it is imperative to understandthe importance of gender roles and the possibilitythat “psychological gender” is the root cause forthe effects observed. Empirical evidence hasdemonstrated that gender roles can have a pro-found impact on individual attitudes and behaviorsboth within and outside the workplace (e.g., Barilet al. 1989; Feldman and Aschenbrenner 1983;Jagacinski 1987;Keys 1985; Roberts 1997; Sachset al. 1992; Wong et al. 1985). Specifically, gen-der effects observed here could be a manifesta-tion of effects caused by masculinity, femininity,and androgyny rather than just “biological sex”(e.g., Lubinski et al. 1983). Future work might bedirected at more closely examining the importanceof gender roles and exploring the socio-psycho-logical basis for gender as a means for betterunderstanding its moderating role.As is evident from the literature, the role of socialinfluence constructs has been controversial.Some have argued for their inclusion in models ofadoption and use (e.g., Taylor and Todd 1995b;Thompson et al. 1991), while others have notincluded them (e.g., Davis et al. 1989). Previouswork has found social influence to be significantonly in mandatory settings (see Hartwick andBarki 1994; Venkatesh and Davis 2000). Otherwork has found social influence to be moresignificant among women in early stages ofexperience (e.g., Venkatesh and Morris 2000).Still other research has found social influence tobe more significant among older workers (e.g.,Morris and Venkatesh 2000). This research isamong the first to examine these moderating influ-ences in concert. Our results suggest that socialinfluences do matter; however, they are morelikely to be salient to older workers, particularlywomen, and even then during early stages ofexperience/adoption. This pattern mirrors that foreffort expectancy with the added caveat that socialinfluences are more likely to be important inmandatory usage settings. The contingenciesidentified here provide some insights into the wayin which social influences change over time andmay help explain some of the equivocal resultsreported in the literature. By helping to clarify thecontingent nature of social influences, this papersheds light on when social influence is likely toplay an important role in driving behavior andwhen it is less likely to do so.
  • 46. Venkatesh et al./User Acceptance of IT470 MIS Quarterly Vol. 27 No. 3/September 2003UTAUT underscores this point and highlights theimportance of contextual analysis in developingstrategies for technology implementation withinorganizations. While each of the existing modelsin the domain is quite successful in predictingtechnology usage behavior, it is only when oneconsiders the complex range of potential moder-ating influences that a more complete picture ofthe dynamic nature of individual perceptions abouttechnology begins to emerge. Despite the abilityof the existing models to predict intention andusage, current theoretical perspectives on indivi-dual acceptance are notably weak in providingprescriptive guidance to designers. For example,applying any of the models presented here mightinform a designer that some set of individualsmight find a new system difficult to use. Futureresearch should focus on integrating UTAUT withresearch that has identified causal antecedents ofthe constructs used within the model (e.g.,Karahanna and Straub 1999; Venkatesh 2000;Venkatesh and Davis 2000) in order to provide agreater understanding of howthe cognitive pheno-mena that were the focus of this research areformed. Examples of previously examined deter-minants of the core predictors include systemcharacteristics (Davis et al. 1989) and self-efficacy(Venkatesh 2000). Additional determinants thathave not been explicitly tied into this stream butmerit consideration in future work include task-technologyfit(Goodhue and Thompson 1994) andindividual ability constructs such as “g”—generalcognitive ability/intelligence (Colquitt et al. 2000).While the variance explained by UTAUT is quitehigh for behavioral research, further work shouldattempt to identify and test additional boundaryconditions of the model in an attempt to providean even richer understanding of technology adop-tion and usage behavior. This might take the formof additional theoretically motivated moderatinginfluences, different technologies (e.g., collabora-tive systems, e-commerce applications), differentuser groups (e.g., individuals in different functionalareas), and other organizational contexts (e.g.,public or government institutions). Results fromsuch studies will have the important benefit ofenhancing the overall generalizability of UTAUTand/or extending the existing work to account foradditional variance in behavior. Specifically, giventhe extensive moderating influences examinedhere, a research study that examines the general-izability of these findings with significant repre-sentation in each cell (total number of cells: 24;two levels of voluntariness, three levels of experi-ence [no, limited, more], two levels of gender, andat least two levels of age [young vs. old]) would bevaluable. Such a study would allow a pairwise,inter-cell comparison using the rigorous Chow’stest and provide a clear understanding of thenature of effects for each construct in each cell.Related to the predictive validity of this class ofmodels in general and UTAUT in particular is therole of intention as a key criterion in user accep-tance research—future research should investi-gate other potential constructs such as behavioralexpectation (Warshaw and Davis 1985) or habit(Venkatesh et al. 2000) in the nomological net-work. Employing behavioral expectation will helpaccount for anticipated changes in intention(Warshaw and Davis 1985) and thus shed lighteven in the early stages of the behavior about theactual likelihood of behavioral performance sinceintention only captures internal motivations toperform the behavior. Recent evidence suggeststhat sustained usage behavior may not be theresult of deliberated cognitions and are simplyroutinized or automatic responses to stimuli (seeVenkatesh et al. 2000).One of the most important directions for futureresearch is to tie this mature stream of researchinto other established streams of work. Forexample, little to no research has addressed thelink between user acceptance and individual ororganizational usage outcomes. Thus, while it isoften assumed that usage will result in positiveoutcomes, this remains to be tested. The unifiedmodel presented here might inform further inquiryinto the short- and long-term effects of informationtechnology implementation on job-related out-comes such as productivity, job satisfaction,organizational commitment, and other perfor-mance-oriented constructs. Future researchshould study the degree to which systems per-ceived as successful from an IT adoption per-spective (i.e., those that are liked and highly usedby users) are considered a success from anorganizational perspective.
  • 47. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 471ConclusionFollowing from the Unified Theory of Acceptanceand Use of Technology presented here, futureresearch should focus on identifying constructsthat can add to the prediction of intention andbehavior over and above what is already knownand understood. Given that UTAUT explains asmuch as 70 percent of the variance in intention, itis possible that we may be approaching thepractical limits of our ability to explain individualacceptance and usage decisions in organizations.In the study of information technology implemen-tations in organizations, there has been aproliferation of competing explanatory models ofindividual acceptance of information technology.The present work advances individual acceptanceresearch by unifying the theoretical perspectivescommon in the literature and incorporating fourmoderators to account for dynamic influencesincluding organizational context, user experience,and demographic characteristics.AcknowledgementsThe authors thank Cynthia Beath (the senioreditor), the associate editor, and the three anony-mous reviewers. We also thank Heather Adams,Wynne Chin, Deborah Compeau, Shawn Curley,Ruth Kanfer, Phillip Ackerman, Tracy Ann Sykes,Peter Todd, Shreevardhan Lele, and Paul Zantek.Last, but not least, we thank Jan DeGross for hertireless efforts in typesetting this rather longpaper.ReferencesAgarwal, R., and Prasad, J. “A Conceptual andOperational Definition of Personal Innovative-ness in the Domain of Information Technology,”Information Systems Research (9:2), 1998, pp.204-215.Agarwal, R., and Prasad, J. “The Role of Innova-tion Characteristics and Perceived Voluntari-ness in the Acceptance of Information Techno-logies,” Decision Sciences (28:3), 1997, pp.557-582.Aiken, L. S., and West, S. G. Multiple Regres-sion: Testing and Interpreting Interactions,Sage, London, 1991.Ajzen, I. “The Theory of Planned Behavior,”Organizational Behavior and Human DecisionProcesses (50:2), 1991, pp. 179-211.Ashmore, R. D. “Sex, Gender, and the Individual,”in Handbook of Personality, L. A. Pervin (ed.),Guilford Press, New York, 1980, pp. 486-526.Bagozzi, R. P., and Edwards, J. R. “A GeneralApproach to Construct Validation in Organi-zational Research: Application to the Measure-ment of Work Values,” Organizational Re-search Methods (1:1), 1998, pp. 45-87.Bandura, A. Social Foundations of Thought andAction: A Social Cognitive Theory, PrenticeHall, Englewood Cliffs, NJ, 1986.Baril, G. L., Elbert, N., Mahar-Potter, S., andReavy, G. C. “Are Androgynous ManagersReally More Effective?,” Group and Organiza-tion Studies (14:2), 1989, pp. 234-249.Barnett, R. C., and Marshall, N. L. “The Relation-ship between Women’s Work and Family Rolesand Their Subjective Well-Being and Psycho-logical Distress,” in Women, Work and Health:Stress and Opportunities, M. Frankenhaeuser,V. Lundberg, and M. A. Chesney (eds.),Plenum, New York, 1981, pp. 111-136.Bem, S. L. “The BSRI and Gender SchemaTheory: A Reply to Spence and Helmreich,”Psychological Review (88:4), 1981, pp. 369-371.Bem, D. J., and Allen, A. “On Predicting Some ofthe People Some of the Time: The Search forCross-Situational Consistencies in Behavior,”Psychological Review (81:6), 1974, pp. 506-520.Bergeron, F., Rivard, S., and De Serre, L. “Inves-tigating the Support Role of the InformationCenter,” MIS Quarterly (14:3), 1990, pp. 247-259.Bozionelos, N. “Psychology of Computer Use:Prevalence of Computer Anxiety in BritishManagers and Professionals,” PsychologicalReports (78:3), 1996, pp. 995-1002.Chin, W. W. “The Partial Least Squares Ap-proach for Structural Equation Modeling,” inModern Methods for Business Research,George A. Marcoulides (ed.), Lawrence
  • 48. Venkatesh et al./User Acceptance of IT472 MIS Quarterly Vol. 27 No. 3/September 2003Erlbaum Associates, New York, 1998, pp. 295-336.Chin, W. W., Marcolin, B. L., and Newsted, P. R.“A Partial Least Squares Latent VariableModeling Approach for Measuring InteractionEffects: Results from a Monte Carlo SimulationStudy and Voice Mail Emotion/Adoption Study,”in Proceedings of the International Conferenceon Information Systems, J. I. DeGross, A. Srini-vasan, and S. Jarvenpaa (eds.), Cleveland,OH, 1996, pp. 21-41.Chow, G. C. “Tests of Equality Between Sets ofCoefficients in Two Linear Regressions,”Econometrica (28:3), 1960, pp. 591-605.Colquitt, J. A., LePine, J. A., and Noe, R. A.“Toward an Integrative Theory of Training Moti-vation: A Meta-Analytic Path Analysis of 20Years of Training Research,” Journal of AppliedPsychology (85:5), 2000, pp. 678-707.Compeau, D. R., and Higgins, C. A. “Applicationof Social Cognitive Theory to Training for Com-puter Skills,” Information Systems Research(6:2), 1995a, pp. 118-143.Compeau, D. R., and Higgins, C. A. “ComputerSelf-Efficacy: Development of a Measure andInitial Test,” MIS Quarterly (19:2), 1995b, pp.189-211.Compeau, D. R., Higgins, C. A., and Huff, S.“Social Cognitive Theory and Individual Reac-tions to Computing Technology: A LongitudinalStudy,” MIS Quarterly (23:2), 1999, pp. 145-158.Davis, F. D. “Perceived Usefulness, PerceivedEase of Use, and User Acceptance of Informa-tion Technology,” MIS Quarterly (13:3), 1989,pp. 319-339.Davis, F. D., Bagozzi, R. P., and Warshaw, P. R.“Extrinsic and Intrinsic Motivation to Use Com-puters in the Workplace,” Journal of AppliedSocial Psychology (22:14), 1992, pp. 1111-1132.Davis, F. D., Bagozzi, R. P., and Warshaw, P. R.“User Acceptance of Computer Technology: AComparison of Two Theoretical Models,” Man-agement Science (35:8), 1989, pp. 982-1002.Efron, B., and Gong, G. “A Leisurely Look at theBootstrap, the Jackknife, and Cross-Validation,”The American Statistician (37:1), 1983, pp. 36-48.Eichinger, J., Heifetz, L.J., and Ingraham, C.“Situational Shifts in Sex Role Orientation:Correlates of Work Satisfaction and BurnoutAmong Women in Special Education,” SexRoles (25:7/8), 1991, pp. 425-430.Feldman, S. S., and Aschenbrenner, B. “Impactof Parenthood on Various Aspects of Mascu-linity and Femininity: A Short-Term Longitu-dinal Study,” Developmental Psychology (19:2),1983, pp. 278-289.Fiske, S. T., and Taylor, S. E. Social Cognition,McGraw-Hill, New York, 1991.Fishbein, M., and Ajzen, I. Belief, Attitude, Inten-tion and Behavior: An Introduction to Theoryand Research, Addison-Wesley, Reading, MA,1975.Fornell, C., and Larcker, D. F. “Evaluating Struc-tural Equation Models with UnobservableVariables and Measurement Error: Algebraand Statistics,” Journal of Marketing Research(18:3), 1981, pp. 382-388.French, J. R. P., and Raven, B. “The Bases ofSocial Power,” in Studies in Social Power, D.Cardwright (ed.), Institute for Social Research,Ann Arbor, MI, 1959, pp. 150-167.Goodhue, D. L. “Understanding User Evaluationsof Information Systems,” Management Science(41:12), 1995, pp. 1827-1844.Goodhue, D. L., and Thompson, R. L. “Task-Technology Fit and Individual Performance,”MIS Quarterly (19:2), 1995, pp. 213-236.Hall, D., and Mansfield, R. “Relationships of Ageand Seniority with Career Variables of Engi-neers and Scientists,” Journal of AppliedPsychology (60:2), 1995, pp. 201-210.Harrison, D. A., Mykytyn, P. P., and Riemen-schneider, C. K. “Executive Decisions AboutAdoption of Information Technology in SmallBusiness: Theory and Empirical Tests,” Infor-mation Systems Research (8:2), 1997, pp.171-195.Hartwick, J., and Barki, H. “Explaining the Role ofUser Participation in Information System Use,”Management Science (40:4), 1994, pp. 40-465.Helson, R., and Moane, G. “Personality Changein Women from College to Midlife,” Journal ofPersonality and Social Psychology (53:1),1987, pp. 176-186.
  • 49. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 473Hu, P. J., Chau, P. Y. K., Sheng, O. R. L., andTam, K. Y. “Examining the Technology Accep-tance Model Using Physician Acceptance ofTelemedicine Technology,” Journal of Manage-ment Information Systems (16:2), 1999, pp. 91-112.Jagacinski, C. M. “Androgyny in a Male-Domi-nated Field: The Relationship of Sex-TypedTraits to Performance and Satisfaction in Engi-neering, Sex Roles (17:X), 1987, pp. 529-547.Karahanna, E., and Straub, D. W. “The Psycho-logical Origins of Perceived Usefulness andEase of Use,” Information and Management(35:4), 1999, pp. 237-250.Karahanna, E., Straub, D. W., and Chervany, N.L. “Information Technology Adoption AcrossTime: A Cross-Sectional Comparison of Pre-Adoption and Post-Adoption Beliefs,” MISQuarterly (23:2), 1999, pp. 183-213.Keys, D. E. “Gender, Sex Role, and CareerDecision Making of Certified Management Ac-countants,” Sex Roles (13:1/2), 1985, pp. 33-46.Kirchmeyer, C. “Change and Stability in Mana-ger’s Gender Roles,” Journal of AppliedPsychology (87:5), 2002, pp. 929-939.Kirchmeyer, C. “Gender Roles in a TraditionallyFemale Occupation: A Study of Emergency,Operating, Intensive Care, and PsychiatricNurses, Journal of Vocational Behavior (50:1),1997, pp. 78-95.Leonard-Barton, D., and Deschamps, I. “Mana-gerial Influence in the Implementation of NewTechnology,” Management Science (34:10),1988, pp. 1252-1265.Levy, J. A. “Intersections of Gender and Aging,”The Sociological Quarterly (29:4), 1988, pp.479-486.Lubinski, D., Tellegen, A. and Butcher, J. N.“Masculinity, Femininity, and AndrogynyViewed and Assessed as Distinct Concepts,”Journal of Personality and Social Psychology(44:2), 1983, pp. 428-439.Lynott, P. P., and McCandless, N. J. “The Impactof Age vs. Life Experiences on the Gender RoleAttitudes of Women in Different Cohorts,” Jour-nal of Women and Aging (12:2), 2000, pp. 5-21.Mathieson, K. “Predicting User Intentions: Com-paring the Technology Acceptance Model withthe Theory of Planned Behavior,” InformationSystems Research (2:3), 1991, pp. 173-191.Miller, J. B. Toward a New Psychology ofWomen, Beacon Press, Boston, 1976.Minton, H. L., and Schneider, F. W. DifferentialPsychology, Waveland Press, ProspectHeights, IL, 1980.Moore, G. C., and Benbasat, I. “Development ofan Instrument to Measure the Perceptions ofAdopting an Information Technology Innova-tion,” Information Systems Research (2:3),1991, pp. 192-222.Moore, G. C., and Benbasat, I. “Integrating Diffu-sion of Innovations and Theory of ReasonedAction Models to Predict Utilization of Infor-mation Technology by End-Users,” in Diffusionand Adoption of Information Technology, K.Kautz and J. Pries-Hege (eds.), Chapman andHall, London, 1996, pp. 132-146.Morris, M. G., and Venkatesh, V. “Age Dif-ferences in Technology Adoption Decisions:Implications for a Changing Workforce,” Per-sonnel Psychology (53:2), 2000, pp. 375-403.Motowidlo, S. J. “Sex Role Orientation andBehavior in a Work Setting,” Journal of Per-sonality and Social Psychology (42:5), 1982,pp. 935-945.Nunnally, J. C., and Bernstein, I. H. PsychometricTheory (3rded.), McGraw-Hill, New York, 1994.Olfman, L., and Mandviwalla, M. “ConceptualVersus Procedural Software Training forGraphical User Interfaces: A Longitudinal FieldExperiment,” MIS Quarterly (18:4), 1994, pp.405-426.Plouffe, C. R., Hulland, J. S., and Vandenbosch,M. “Research Report: Richness Versus Parsi-mony in Modeling Technology Adoption Deci-sions—Understanding Merchant Adoption of aSmart Card-Based Payment System,” Infor-mation Systems Research (12:2), 2001, pp.208-222.Plude, D., and Hoyer, W. “Attention and Perfor-mance: Identifying and Localizing Age Defi-cits,” in Aging and Human Performance, N.Charness (ed.), John Wiley & Sons, New York,1985, pp. 47-99.Porter, L. “Job Attitudes in Management: Per-ceived Importance of Needs as a Function ofJob Level,” Journal of Applied Psychology(47:2), 1963, pp. 141-148.
  • 50. Venkatesh et al./User Acceptance of IT474 MIS Quarterly Vol. 27 No. 3/September 2003Rhodes, S. R. “Age-Related Differences in WorkAttitudes and Behavior: A Review and Concep-tual Analysis,” Psychological Bulletin (93:2),1983, pp. 328-367.Roberts, R. W. “Plaster or Plasticity: Are AdultWork Experiences Associated with PersonalityChange in Women?,” Journal of Personality(65:2), 1997, pp. 205-232.Rogers, E. Diffusion of Innovations, Free Press,New York, 1995.Rogers, E. M., and Shoemaker, F. F. Com-munication of Innovations: A Cross-CulturalApproach, Free Press, New York, 1971.Sachs, R., Chrisler, J. C., and Devlin, A. S.“Biographic and Personal Characteristics ofWomen in Management,” Journal of VocationalBehavior (41:1), 1992, pp. 89-100.Sheppard, B. H., Hartwick, J., and Warshaw, P. R.“The Theory of Reasoned Action: A Meta-Analysis of Past Research with Recommen-dations forModifications and Future Research,”Journal of Consumer Research (15:3), 1988,pp. 325-343.Straub, D., Limayem, M., and Karahanna, E.“Measuring System Usage: Implications for ISTheory Testing,” Management Science (41:8),1995, pp. 1328-1342.Szajna, B. “Empirical Evaluation of the RevisedTechnology Acceptance Model,” ManagementScience (42:1), 1996, pp. 85-92.Taylor, S., and Todd, P. A. “Assessing IT Usage:The Role of Prior Experience,” MIS Quarterly(19:2), 1995a, pp. 561-570.Taylor, S., and Todd, P. A. “Understanding Infor-mation Technology Usage: A Test of Com-peting Models,” Information Systems Research(6:4), 1995b, pp. 144-176.Thompson, R. L., Higgins, C. A., and Howell, J. M.“Influence of Experience on Personal ComputerUtilization: Testing a Conceptual Model,”Journal of Management Information Systems(11:1), 1994, pp. 167-187.Thompson, R. L., Higgins, C. A., and Howell, J. M.“Personal Computing: Toward a ConceptualModel of Utilization,” MIS Quarterly (15:1),1991, pp. 124-143.Tornatzky, L. G., and Klein, K. J. “InnovationCharacteristics and Innovation Adoption-Imple-mentation: A Meta-Analysis of Findings,” IEEETransactions on Engineering Management(29:1), 1982, pp. 28-45.Triandis, H. C. Interpersonal Behavior, Brooke/Cole, Monterey, CA, 1977.Twenge, J. M. “Changes in Masculine andFeminine Traits Over Time: A Meta-Analysis,”Sex Roles (35:5/6), 1997, pp. 305-325.Vallerand, R. J. “Toward a Hierarchical Model ofIntrinsic and Extrinsic Motivation,” in Advancesin Experimental Social Psychology (29), M.Zanna (ed.), Academic Press, New York, 1997,pp. 271-360.Venkatesh, V. “Creating Favorable User Percep-tions: Exploring the Role of Intrinsic Moti-vation,” MIS Quarterly (23:2), 1999, pp. 239-260.Venkatesh, V. “Determinants of Perceived Easeof Use: Integrating Perceived BehavioralControl, Computer Anxiety and Enjoyment intothe Technology Acceptance Model,” Informa-tion Systems Research (11:4), 2000, pp. 342-365.Venkatesh, V., and Davis, F. D. “A TheoreticalExtension of the Technology AcceptanceModel: Four Longitudinal Field Studies,” Man-agement Science (45:2), 2000, pp. 186-204.Venkatesh, V., and Morris, M. G. “Why Don’t MenEver Stop to Ask For Directions? Gender,Social Influence, and Their Role in TechnologyAcceptance and Usage Behavior,” MISQuarterly (24:1), 2000, pp. 115-139.Venkatesh, V., Morris, M. G., and Ackerman, P. L.“A Longitudinal Field Investigation of GenderDifferences in Individual Technology AdoptionDecision Making Processes,” OrganizationalBehavior and Human Decision Processes(83:1), 2000, pp. 33-60.Venkatesh, V., and Speier, C. “Computer Tech-nology Training in the Workplace: A Longi-tudinal Investigation of the Effect of the Mood,”Organizational Behavior and Human DecisionProcesses (79:1), 1999, pp. 1-28.Warshaw, P. R. “A New Model for PredictingBehavioral Intentions: An Alternative to Fish-bein,” Journal of Marketing Research (17:2),1980, pp. 153-172.Warshaw, P. R., and Davis, F. D. “DisentanglingBehavioral Intention and Behavioral Expec-tation,” Journal of Experimental Social Psych-ology (21:3), 1985, pp. 213-228.
  • 51. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 475West, S. G., and Hepworth, J. T., “StatisticalIssues in the Study of Temporal Data: DailyExperiences,” Journal of Personality (59:2),1991, pp. 609-662.Westland, J. C., and Clark, T. H. K. Global Elec-tronic Commerce: Theory and Case Studies,MIT Press, Cambridge, MA, 2000.Wong, P. T. P., Kettlewell, G., and Sproule, C. F.“On the Importance of Being Masculine: SexRole, Attribution, and Women’s CareerAchievement,” Sex Roles (12:7/8), 1985, pp.757-769.About the AuthorsViswanath Venkatesh is Tyser Fellow, associateprofessor, and director of MBA Consulting at theRobert H. Smith School of Business at the Univer-sity of Maryland, and has been on the faculty theresince 1997. He earned his Ph.D. at the Universityof Minnesota’s Carlson School of Management.His research interests are in implementation anduse of technologies in organizations and homes.His research has been published in many leadingjournals including MIS Quarterly, InformationSystems Research, Management Science, Com-munications of the ACM, Decision Sciences,Organizational Behavior and Human DecisionProcesses, Personnel Psychology. He is currentlyan associate editor at MIS Quarterly and Infor-mation Systems Research. He received MISQuarterly’s “Reviewer of the Year” award in 1999.Michael G. Morris is an assistant professor ofCommerce within the Information Technology areaat the McIntire School of Commerce, University ofVirginia. He received his Ph.D. in ManagementInformation Systems from Indiana University in1996. His research interests can broadly be clas-sified as socio-cognitive aspects of humanresponse to information technology, including useracceptance of information technology, usabilityengineering, and decision-making. His researchhas been published in MIS Quarterly, Organiza-tional Behavior and Human Decision Processes,and Personnel Psychology, among others.Gordon B. Davis, Honeywell Professor ofManagement Information Systems in the CarlsonSchool of Management at the University of Minne-sota, is one of the founders of the academicdiscipline of information systems. He has lecturedin 25 countries and has written 20 books and over200 articles, monographs, and book chapters. Heparticipated in and helped form the major aca-demic associations related to the field of man-agement information systems. He has beenhonored as an ACM Fellow and an AIS Fellow,and is a recipient of the AIS LEO award forlifetime achievement in the field of informationsystems. His research interests include concep-tual foundations of information systems, informa-tion system design and implementation, andmanagement of the IS function. He has a Ph.D.from Stanford University and honorary doctoratesfrom the University of Lyon, France, the Universityof Zurich, Switzerland and the Stockholm Schoolof Economics, Sweden.Fred D. Davis is David D. Glass Endowed ChairProfessor in Information Systems and Chair of theInformation Systems Department at the Sam M.Walton College of Business at the University ofArkansas. Dr. Davis earned his Ph.D. at MIT’sSloan School of Management, and has served onthe business school faculties at University ofMichigan, University of Minnesota, and Universityof Maryland. He has taught a wide range of infor-mation technology (IT) courses at the under-graduate, MBA, Ph.D., and executive levels. Dr.Davis has served as associate editor for thescholarly journals Management Science, MISQuarterly, and Information Systems Research. Hehas published extensively on the subject of useracceptance of ITand IT-assisted decision making.His research has appeared in such journals asMIS Quarterly, Information Systems Research,Management Science, Journal of ExperimentalSocial Psychology, Decision Sciences, Organiza-tional Behavior and Human Decision Processes,Communications of the AIS, and Journal of MIS.Current research interests include IT training andskill acquisition, management of emerging IT, andthe dynamics of organizational change.
  • 52. Venkatesh et al./User Acceptance of IT476 MIS Quarterly Vol. 27 No. 3/September 2003Appendix ACautions Related to Pooling Data and Associated StatisticalTests for Preliminary Test of UTAUT (Studies 1 and 2)The most critical concern when pooling repeated measures from the same subjects is the possibility ofcorrelated errors. West and Hepworth describe this problem as follows:A perusal of the empirical literature over the past 30 years reveals some of the statisticalapproaches to repeated measures over time that have been used by personalityresearchers. The first and perhaps most classic approach is to aggregate theobservations collected over time on each subject to increase the reliability ofmeasurement (1991, p. 610).However, in many cases, these observations will often violate the assumption of independence, requiringalternate analyses (see West and Hepworth for an extended discussion of alternate approaches).When error terms can be shown to be independent, West and Hepworth note that “ traditional statisticalanalyses such as ANOVA or MR [multiple regression] can be performed directly without any adjustmentof the data.” (pp. 612-613). The best way to determine whether it is appropriate to pool data is to conducta test for correlated errors. In order for it to be acceptable to pool the data, the error terms should beuncorrelated. A second approach uses bootstrapping to select subsamples to conduct a between-subjectsexamination of the within-subjects data. In order for it to be acceptable to pool the data, this second testshould yield identical results to the test of the model on the complete data set. Below we report the resultsfrom the specific findings from the correlated errors test and the pattern of findings from the second test.Correlated Errors TestWe computed the error terms associated with the prediction of intention at T1, T2, and T3 in studies 1(voluntary settings) and 2 (mandatory settings). Further, we also calculated the error terms when pooledacross both settings—i.e., for cross-sectional tests of the unified model at T1, T2, and T3 respectively.These computations were conducted both for the preliminary test and the cross-validation (reported below).The error term correlations across the intention predictions at various points in time are shown below. Notethat all error correlations are nonsignificant and, therefore, not significantly different from zero in allsituations.Between-Subjects Test of Within-Subjects DataWhile the results above are compelling, as a second check, we pooled the data across different levels ofexperience (T1, T2, and T3) and used PLS to conduct a between-subjects test using the within-subjectsdata. We used a DOS version of PLS to conduct the analyses (the reason for not using PLS-Graph as inthe primary analyses in the paper was because it did not allow the specification of filtering rules forselecting observations in bootstraps). Specifically for the between-subjects test, we applied a filtering rulethat selected any given respondent’s observation at only one of the three time periods. This approach en-
  • 53. Venkatesh et al./User Acceptance of ITMIS Quarterly Vol. 27 No. 3/September 2003 477Table A1. Correlations Between Error Terms of Intention Construct at Various TimePeriodsStudy 1 (Voluntary) T1 T2 T3T1T2 .04T3 .11 .09Study 2 (Mandatory) T1 T2 T3T1T2 .07T3 .08 .13Study 1 and 2 (Pooled) T1 T2 T3T1T2 .06T3 .09 .10sured that the data included to examine the interaction terms with experience did not include any potentialfor systematic correlated errors. Using 50 such random subsamples, the model was tested and the resultsderived supported the pattern observed when the entire data set was pooled across experience.Taken together, the analyses reported above support the pooling of data (see Table 17) across levels ofexperience and eliminate the potential statistical concerns noted by West and Hepworth in the analysis oftemporal data.
  • 54. Venkatesh et al./User Acceptance of IT478 MIS Quarterly Vol. 27 No. 3/September 2003Appendix BStatistical Tests Prior to Pooling of Data forCross-Validation of UTAUT (Studies 3 and 4)As with the test of the preliminary model, prior to pooling the data for the cross-validation studies (studies3 and 4), we conducted statistical tests to examine the independence of observations (as detailed inAppendix A). The table below presents the error term correlation matrices for intention for studies 3(voluntary) and 4 (mandatory) as well as pooled across both settings at T1, T2, and T3 respectively.As in the preliminary test of UTAUT, the error correlation matrices above suggest that there is no constraintto pooling in the cross-validation study of the model. As before, a between-subjects test of the within-subjects data was tested using PLS (as described in Appendix A), and the results of that test corroboratedthe independence of observations in this sample. In light of both sets of results, we proceeded with thepooled analysis as reported in the body of the paper (see Table 21).Table B1. Correlations Between Error Terms of Intention Construct at Various TimePeriodsStudy 3 (Voluntary) T1 T2 T3T1T2 .01T3 .07 .11Study 4 (Mandatory) T1 T2 T3T1T2 .04T3 .02 .08Study 3 and 4 (Pooled) T1 T2 T3T1T2 .03T3 .05 .10

×