|
...extensions, (3) formulate unified model that integrates elements across the eight models, and (4) empirically validate the unified model. The eight models reviewed are the theory of reasoned action, the technology acceptance model, the motivational model, the theory of planned behavior, a model combining the technology acceptance model and the theory of planned behavior, the model of PC utilization, the innovation diffusion theory, and the social cognitive theory. Using data from four organizations over a six-month period with three points of measurement, the eight models explained between 17 percent and 53 percent of the variance in user intentions to use information technology. Next, a unified model, called the Unified Theory of Acceptance and Use of Technology (UTAUT), was formulated, with four core determinants of intention and usage, and up to four moderators of key relationships. UTAUT was then tested using the original data and found to outperform the eight individual models (adjusted [R.sup.2] of 69 percent). UTAUT was then confirmed with data from two new organizations with similar results (adjusted [R.sup.2] of 70 percent). UTAUT thus provides a useful tool for managers needing to assess the likelihood of success for new technology introductions and helps them understand the drivers of acceptance in order to proactively design interventions (including training, marketing, etc.) targeted at populations of users that may be less inclined to adopt and use new systems. The paper also makes several recommendations for future research including developing a deeper understanding of the dynamic influences studied here, refining measurement of the core constructs used in UTAUT, and understanding the organizational outcomes associated with new technology use.
Keywords: Theory of planned behavior, innovation characteristics, technology acceptance model, social cognitive theory, unified model, integrated model
Introduction
The presence of computer and information technologies in today's organizations has expanded dramatically. Some estimates indicate that, since the 1980s, about 50 percent of all new capital investment in organizations has been in information technology (Westland and Clark 2000). Yet, for technologies to improve productivity, they must be accepted and used by employees in organizations. Explaining user acceptance of new technology is often described as one of the most mature research areas in the contemporary information systems (IS) literature (e.g., Hu et al. 1999). Research in this area has resulted in several theoretical models, with roots in information systems, psychology, and sociology, that routinely explain over 40 percent of the variance in individual intention to use technology (e.g., Davis et al. 1989; Taylor end Todd 1995b; Venkatesh and Davis 2000). Researchers are confronted with a choice among a multitude of models and find that they must "pick and choose" constructs across the models, or choose a "favored model" and largely ignore the contributions from alternative models. Thus, there is a need for a review and synthesis in order to progress toward a unified view of user acceptance.
The current work has the following objectives:
(1) To review the extant user acceptance models: The primary purpose of this review is to assess the current state of knowledge with respect to understanding individual acceptance of new information technologies. This review identifies eight prominent models and discusses their similarities and differences. Some authors have previously observed some of the similarities across models. (2) However, our review is the first to assess similarities and differences across all eight models, a necessary first step toward the ultimate goal of the paper: the development of a unified theory of individual acceptance of technology. The review is presented in the following section.
(2) To empirically compare the eight models: We conduct a within-subjects, longitudinal validation and comparison of the eight models using data from four organizations. This provides a baseline assessment of the relative explanatory power of the individual models against which the unified model can be compared. The empirical model comparison is presented in the third section.
(3) To formulate the Unified Theory of Acceptance and Use of Technology (UTAUT): Based upon conceptual and empirical similarities across models, we formulate a unified model. The formulation of UTAUT is presented in the fourth section.
(4) To empirically validate UTAUT: An empirical test of UTAUT on the original data provides preliminary support for our contention that UTAUT outperforms each of the eight original models. UTAUT is then cross-validated using data from two new organizations. The empirical validation of UTAUT is presented in the fifth section.
[FIGURE 1 OMITTED]
Review of Extant User Acceptance Models
Description of Models and Constructs
IS research has long studied how and why individuals adopt new information technologies. Within this broad area of inquiry, there have been several streams of research. One stream of research focuses on individual acceptance of technology by using intention or usage as a dependent variable (e.g., Compeau and Higgins 1995b; Davis et al. 1989). Other streams have focused on implementation success at the organizational level (Leonard-Barton and Deschamps 1988) and task-technology fit (Goodhue 1995; Goodhue and Thompson 1995), among others. While each of these streams makes important and unique contributions to the literature on user acceptance of information technology, the theoretical models to be included in the present review, comparison, and synthesis employ intention and/or usage as the key dependent variable. The goal here is to understand usage as the dependent variable. The role of intention as a predictor of behavior (e.g., usage) is critical and has been well-established in IS and the reference disciplines (see Ajzen 1991 ; Sheppard et al. 1988; Taylor and Todd 1995b). Figure 1 presents the basic conceptual framework underlying the class of models explaining individual acceptance of information technology that forms the basis of this research. Our review resulted in the identification of eight key competing theoretical models. Table 1 describes the eight models and defines their theorized determinants of intention and/or usage. The models hypothesize between two and seven determinants of acceptance, for a total of 32 constructs across the eight models. Table 2 identifies four key moderating variables (experience, voluntariness, gender, and age) that have been found to be significant in conjunction with these models.
Prior Model Tests and Model Comparisons
There have been many tests of the eight models but there have only been four studies reporting empirically-based comparisons of two or more of the eight models published in the major information systems journals. Table 3 provides a brief overview of each of the model comparison studies. Despite the apparent maturity of the research stream, a comprehensive comparison of the key competing models has not been conducted in a single study. Below, we identify five limitations of these prior model tests and comparisons, and how we address these limitations in our work.
* Technology studied: The technologies that have been studied in many of the model development and comparison studies have been relatively simple, individual-oriented information technologies as opposed to more complex and sophisticated organizational technologies that are the focus of managerial concern and of this study.
* Participants: While there have been some tests of each model in organizational settings, the participants in three of the four model comparison studies have been students--only Plouffe et al. (2001) conducted their research in a nonacademic setting. This research is conducted using data collected from employees in organizations.
* Timing of measurement: In general, most of the tests of the eight models were conducted well after the participants' acceptance or rejection decision rather than during the active adoption decision-making process. Because behavior has become routinized, individual reactions reported in those studies are retrospective (see Fiske and Taylor 1991; Venkatesh et al. 2000). With the exception of Davis et al. (1989), the model comparisons examined technologies that were already familiar to the individuals at the time of measurement. In this paper, we examine technologies from the time of their initial introduction to stages of greater experience.
* Nature of measurement: Even studies that have examined experience have typically employed cross-sectional and/or between-subjects comparisons (e.g., Davis et al. 1989; Karahanna et al. 1999; Szajna 1996; Taylor and Todd 1995a; Thompson et al. 1994). This limitation applies to model comparison studies also. Our work tracks participants through various stages of experience with a new technology and compares all models on all participants.
* Voluntary vs. mandatory contexts: Most of the model tests and all four model comparisons were conducted in voluntary usage contexts.(3) Therefore, one must use caution when generalizing those results to the mandatory settings that are possibly of more interest to practicing managers. This research examines both voluntary and mandatory implementation contexts.
Empirical Comparison of the Eight Models
Settings and Participants
Longitudinal field studies were conducted at four organizations among individuals being introduced to a new technology in the workplace. To help ensure our results would be robust across contexts, we sampled for heterogeneity across technologies, organizations, industries, business functions, and nature of use (voluntary vs. mandatory). In addition, we captured perceptions as the users' experience with the technology increased. At each firm, we were able to time our data collection in conjunction with a training program associated with the new technology introduction. This approach is consistent with prior training and individual acceptance research where individual reactions to a new technology were studied (e.g., Davis et al. 1989; Olfman and Mandviwalla 1994; Venkatesh and Davis 2000). A pretested questionnaire containing items measuring constructs from all eight models was administered at three different points in time: post-training (T1), one month after implementation (T2), and three months after implementation (T3). Actual usage behavior was measured over the six-month post-training period. Table 4 summarizes key characteristics of the organizational settings. Figure 2 presents the longitudinal data collection schedule.
Measurement
A questionnaire was created with items validated in prior research adapted to the technologies end organizations studied. TRA scales were adapted from Davis et al. (1989); TAM scales were adapted from Davis (1989), Davis et al. (1989), and Venkatesh and Davis (2000); MM scales were adapted from Davis et al. (1992); TPB/DTPB scales were adapted from Taylor and Todd (1995a, 1995b); MPCU scales were adapted from Thompson et al. (1991); IDT scales were adapted from Moore and Benbasat (1991); and SCT scales were adapted from Compeau and Higgins (1995a, 1995b) and Compeau et al. (1999). Behavioral intention to use the system was measured using a three-item scale adapted from Davis et al. (1989) and extensively used in much of the previous individual acceptance research. Seven-point scales were used for all of the aforementioned constructs' measurement, with 1 being the negative end of the scale and 7 being the positive end of the scale. In addition to these measures, perceived voluntariness was measured as a manipulation check per the scale of Moore and Benbasat (1991), where 1 was nonvoluntary and 7 was completely voluntary. The tense of the verbs in the various scales reflected the timing of measurement: future tense was employed at T1, present tense was employed at T2 and T3 (see Karahanna et al. 1999). The scales used to measure the key constructs are discussed in a later section where we perform a detailed comparison (Tables 9 through 13). A focus group of five business professionals evaluated the questionnaire, following which minor wording changes were made. Actual usage behavior was measured as duration of use via system logs. Due to the sensitivity of usage measures to network availability, in all organizations studied, the system automatically logged off inactive users after a period of 5 to 15 minutes, eliminating most idle time from the usage logs.
Results
The perceptions of voluntariness were very high in studies 1a and 1b (1a: M = 6.50, SD = 0.22; 1b: M = 6.51, SD = 0.20) and very low in studies 2a and 2b (1a: M= 1.50, SD= 0.19; 1b: M= 1.49, SD = 0.18). Given this bi-modal distribution in the data (voluntary vs. mandatory), we created two data sets: (1) studies 1a and 1b, and (2) studies 2a and 2b. This is consistent with Venkatesh and Davis (2000).
Partial least squares (PLS Graph, Version 2.91.03.04) was used to examine the reliability and validity of the measures. Specifically, 48 separate validity tests (two studies, eight models, three time periods each) were run to examine convergent and discriminant validity. In testing the various models, only the direct effects on intention were modeled as the goal was to examine the prediction of intention rather than interrelationships among determinants of intention; further, the explained variance ([R.sup.2]) is not affected by indirect paths. The loading pattern was found to be acceptable with most loadings being .70 or higher. All internal consistency reliabilities were greater than .70. The patterns of results found in the current work are highly consistent with the results of previous research.
PLS was used to test all eight models at the three points of measurement in each of the two data sets. In all cases, we employed a bootstrapping method (500 times) that used randomly selected subsamples to test the PLS model. (4) Tables 5 and 6 present the model validation results at each of the points of measurement. The tables report the variance explained and the beta coefficients. Key findings emerged from these analyses. First, all eight models explained individual acceptance, with variance in intention explained ranging from 17 percent to 42 percent. Also, a key difference across studies stemmed from the voluntary vs. mandatory settings--in mandatory settings (study 2), constructs related to social influence were significant whereas in the voluntary settings (study 1), they were not significant. Finally, the determinants of intention varied over time, with some determinants going from significant to nonsignificant with increasing experience.
Following the test of the baseline/original specifications of the eight models (Tables 5 and 6), we examined 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 influences, stay true to the model extensions (Table 2), and conduct a complete test of the existing models and their extensions, the data were pooled across studies and time periods. Voluntariness was a dummy variable used to separate the situational contexts (study 1 vs. study 2); this approach is consistent with previous research (Venkatesh and Davis 2000). Gender was coded as a 0/1 dummy variable consistent with previous research (Venkatesh and Morris 2000) and age was coded as a continuous variable, consistent with prior research (Morris and Venkatesh 2000). Experience was operationalized via a dummy variable that took ordinal values of 0, 1, or 2 to capture increasing levels of user experience with the system (T1, T2, and T3). Using an ordinal dummy variable, rather than categorical variables, is consistent with recent research (e.g., Venkatesh and Davis 2000). Pooling the data across the three points of measurement resulted in a sample of 645 (215 x 3). The results of the pooled analysis are shown in Table 7.
Because pooling across time periods allows the explicit modeling of the moderating role of experience, there is an increase in the variance explained in the case of TAM2 (Table 7) compared to a main effects-only model reported earlier (Tables 5 and 6). One of the limitations of pooling is that there are repeated measures from the same individuals, resulting in measurement errors that are potentially correlated across time. However, cross-sectional analysis using Chow's (1960) test of beta differences (p < .05) from each time period (not shown here) confirmed the pattern of results shown in Table 7. Those beta differences with a significance of p < .05 or better (when using Chow's test) are discussed in the "Explanation" column in Table 7. The interaction terms were modeled as suggested by Chin et al. (1996) by creating interaction terms that were at the level of the indicators. For example, if latent variable A is measured by four indicators (A1, A2, A3, and A4) and latent variable B is measured by three indicators (B1, B2, and B3), the interaction term A x B is specified by 12 indicators, each one a product term--i.e., A1 x B1, A1 x B2, A1 x B3, A2 x B1, etc.
With the exception of MM and SCT, the predictive validity of the models increased after including the moderating variables. For instance, the variance explained by TAM2 increased to 53 percent and TAM including gender increased to 52 percent when compared to approximately 35 percent in cross-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 previously tested in the literature. For example, in the case of TAM and its variations, the extensive prior empirical work has suggested a larger number of moderators when compared to moderators suggested for other models. This in turn may have unintentionally biased the results and contributed to the high variance explained in TAM-related models when compared to the other models. Regardless, it is clear that the extensions to the various models identified in previous research mostly enhance the predictive validity of the various models beyond the original specifications.
In looking at technology use as the dependent variable, in addition to intention as a key predictor, TPB and DTPB employ perceived behavioral control as an additional predictor. MPCU employs facilitating conditions, a construct similar to perceived behavioral control, to predict behavior. Thus, intention and perceived behavioral control were used to predict behavior in the subsequent time period: intention from T1 was used to predict usage behavior measured between T1 and T2 and so on (see Table 8). Since intention was used to predict actual behavior, concerns associated with the employment of subjective measures of usage do not apply here (see Straub et al. 1995). In addition to intention being a predictor of use, perceived behavioral control became a significant direct determinant of use over and above intention with increasing experience (at T3) indicating that continued use could be directly hindered or fostered by resources and opportunities. A nearly identical pattern of results was found when the data were analyzed using facilitating conditions (from MPCU) in place of perceived behavioral control (the specific results are not shown here).
Having reviewed and empirically compared the eight competing models, we now formulate a unified theory of acceptance and use of technology (UTAUT). Toward this end, we examine commonalities across models as a first step. Tables 5, 6, 7, and 8 presented cross-sectional tests of the baseline models and their extensions. Several consistent findings emerged. First, for every model, there was at least one construct that was significant in all time periods and that construct also had the strongest influence--e.g., attitude in TRA and TPB/DTPB, perceived usefulness in TAM/TAM2 and C-TAM-TPB, extrinsic motivation in 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, including perceived behavioral control in TPB/DTPB and 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, the voluntary vs. mandatory context did have an influence on the significance of constructs related to social influence: subjective norm (TPB/DTPB, C-TAM-TPB and TAM2), social factors (MPCU), and image (IDT) were only significant in mandatory implementations.
Formulation of the Unified Theory of Acceptance and Use of Technology (UTAUT)
Seven constructs appeared to be significant direct determinants of intention or usage in one or more of the individual models (Tables 5 and 6). Of these, we theorize that four constructs will play a significant role as direct determinants of user acceptance and usage behavior: performance expectancy, effort expectancy, social influence, and facilitating conditions. As will be explained below, attitude toward using technology, self-efficacy, and anxiety are theorized not to be direct determinants of intention. The labels used for the constructs describe the essence of the construct and are meant to be independent of any particular theoretical perspective. In the remainder of this section, we define each of the determinants, specify the role of key moderators (gender, age, voluntariness, and experience), and provide the theoretical justification for the hypotheses. Figure 3 presents the research model.
Performance Expectancy
Performance expectancy is defined as the degree to which an individual believes that using the system will help him or her to attain gains in job performance. The five constructs from the different models that pertain to performance expectancy are perceived usefulness (TAM/TAM2 and C-TAM-TPB), extrinsic motivation (MM), job-fit (MPCU), relative advantage (IDT), and outcome expectations (SCT). Even as these constructs evolved in the literature, some authors acknowledged their similarities: usefulness and extrinsic motivation (Davis et al. 1989, 1992), usefulness and job-fit (Thompson et al. 1991), usefulness and relative advantage (Davis et al. 1989; Moore and Benbasat 1991; Plouffe et al. 2001), usefulness and outcome expectations (Compeau and Higgins 1995b; Davis et al. 1989), and job-fit and outcome expectations (Compeau and Higgins 1995b).
The performance expectancy construct within each individual model (Table 9) is the strongest predictor of intention and remains significant at all points of measurement in both voluntary and mandatory settings (Tables 5, 6, and 7), consistent with previous model tests (Agarwal and Prasad 1998; Compeau and Higgins 1995b; Davis et al. 1992; Taylor and Todd 1995a; Thompson et al. 1991; Venkatesh and Davis 2000). However, from a theoretical point of view, there is reason to expect that the relationship between performance expectancy and intention will be moderated by gender and age. Research on gender differences indicates that men tend to be highly task-oriented...
NOTE: All illustrations and photos
have been removed from this article.
More articles from MIS Quarterly
In pursuit of moderation: nine common errors and their solutions (1).(..., September 01, 2003
Looking for additional articles?
Search our database of over 3 million articles.
Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication
name or publication date.
About Goliath
Whether you're looking for sales prospects, competitive information, company
analysis or best practices in managing your organization,
Goliath can help you meet your business needs.
Our extensive business information databases empower business
professionals with both the breadth and depth of credible,
authoritative information they need to support their business
goals. Whether it be strategic planning, sales prospecting,
company research or defining management best practices -
Goliath is your leading source for accurate information.
|
.(research%20article)%20%20&position=1)