Computers & Education 55 (2010) 155164

Contents lists available at ScienceDirect

Computers & Education

journal homepage: www.elsevier.com/locate/compedu

A study of student satisfaction in a blended e-learning system environment

Jen-Her Wu a, Robert D. Tennyson b,*, Tzyh-Lih Hsia c
a

Department of Information Management, National Sun Yat-Sen University, 70 Lien-Hai Road, Kaohsiung, 80424, Taiwan University of Minnesota, 56 East River Road, Minneapolis, Minnesota 55455, United States c Department of Information Management, Chinese Naval Academy, P.O. Box No. 90175 Tsoying, Kaohsiung 813, Taiwan
b

a r t i c l e

i n f o

a b s t r a c t
This study proposes a research model that examines the determinants of student learning satisfaction in a blended e-learning system (BELS) environment, based on social cognitive theory. The research model is tested using a questionnaire survey of 212 participants. Conrmatory factor analysis (CFA) was performed to test the reliability and validity of the measurements. The partial least squares (PLS) method was used to validate the measurement and hypotheses. The empirical ndings indicate that computer self-efcacy, performance expectations, system functionality, content feature, interaction, and learning climate are the primary determinants of student learning satisfaction with BELS. The results also show that learning climate and performance expectations signicantly affect learning satisfaction. Computer self-efcacy, system functionality, content feature and interaction signicantly affect performance expectations. Interaction has a signicant effect on learning climate. The ndings provide insight into those factors that are likely signicant antecedents for planning and implementing a blended e-learning system to enhance student learning satisfaction. 2010 Elsevier Ltd. All rights reserved.

Article history: Received 18 March 2009 Received in revised form 23 December 2009 Accepted 31 December 2009

Keywords: e-Learning Satisfaction Learner control Internet Teacher-directed Learner-directed Synchronous Asynchronous Face-to-face

1. Introduction Classroom learning typically occurs in a teacher-directed instructional context with face-to-face interaction in a live synchronous environment. In contrast to this form of instruction, is an approach that promotes learner-directed learning. With emerging Internet commercialization and the proliferation of information technologies, online or electronic learning (e-learning) environments offer the possibilities for communication, interaction and multimedia material delivery that enhance learner-directed learning (Wu, Tennyson, Hsia, & Liao, 2008). Although e-learning may increase access exibility, eliminate geographical barriers, improve convenience and effectiveness for individualized and collaborative learning, it suffers from some drawbacks such as lack of peer contact and social interaction, high initial costs for preparing multimedia content materials, substantial costs for system maintenance and updating, as well as the need for exible tutorial support (Kinshuk & Yang, 2003; Wu et al., 2008; Yang & Liu, 2007). Furthermore, students in virtual e-learning environments may experience feelings of isolation, frustration and confusion (Hara & Kling, 2000) or reduced interest in the subject matter (Maki, Maki, Patterson, & Whittaker, 2000). In addition, student satisfaction and effectiveness for e-learning has also been questioned (Piccoli, Ahmad, & Ives, 2001; Santhanam, Sasidharan, & Webster, 2008). With the concerns and dissatisfaction with e-learning, educators are searching for alternative instructional delivery solutions to relieve the above problems. The blended e-learning system (BELS) has been presented as a promising alternative learning approach (Graham, 2006). BELS refers to an instructional system that combines multiple learning delivery methods, including most often face-to-face classroom with asynchronous and/or synchronous online learning. It is characterized as maximizing the best advantages of face-to-face and online education. While BELS has been recognized as having a number of advantages (e.g., instructional richness, access to knowledge content, social interaction, personal agency, cost effectiveness, and ease of revision (Osguthorpe & Graham, 2003)), insufcient learning satisfaction is still an obstacle to the successful BELS adoption (So & Brush, 2008). In fact, research ndings from Bonk and colleagues have shown that learners had difculty adjusting to BELS environments due to the potential problems in computer and Internet access, learners abilities and

* Corresponding author. E-mail address: rtenny@umn.edu (R.D. Tennyson). 0360-1315/$ - see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.compedu.2009.12.012

156

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

beliefs in the use of technology, blended course design, participant interaction, and blended environments integration (Bonk, Olson, Wisher, & Orvis, 2002). These ndings imply that an effective BLES environment should consider the human and technology factors that affect learner satisfactions with BELS, such as individual attitudes, participant interaction, educational technologies, and course design (Wu et al., 2008). Thus, more careful analysis of learners, educational technologies, and social contexts in BELS environments are needed (ELDeghaidy & Nouby, 2008). The adoption of BELS in supporting learning has made it signicant to probe the crucial determinants that would entice learners to use BELS and enhance their learning satisfaction. The degree of student learning satisfaction with BELS courses plays an important role in evaluating the effectiveness of BELS adoption. Hence, comprehending the essentials of what determines student learning satisfaction can provide management insight into developing effective strategies that will allow educational institution administrators and instructors to create new educational benets and value for their students. Because BELS environments differ from typical classroom and virtual e-learning, a review of previous research in learning technology shows that there is a lack of studies that have examined the crucial factors that determine learning satisfaction with BELS, such as individual cognition, technological environments, and the social contexts, as stated above. There is a need for more in-depth research to understand what determines student learning satisfaction in a BELS environment and to investigate how the determinants inuence student perceptions of BELS contexts and their correlations. This study, therefore, proposes a research model, based on the social cognitive theory (Bandura, 1986), to investigate the primary determinants affecting student learning satisfaction in a BELS environment. We also empirically validate the proposed model and examine the relationships among those latent variables. 2. Basic concepts and theoretical foundation 2.1. Blended e-learning system Blended learning is described as a learning approach that combines different delivery methods and styles of learning. The blend could be between any form of instructional technology (e.g., videotape, CD-ROM, CAI, web-based learning) with classroom teaching. Recently there has been an increasing movement toward blending e-learning and face-to-face activities with students participating in collaborative learning and interaction with their instructors and classmates. This is called blended e-learning or blended e-learning system (Graham, 2006; Singh, 2003). Graham (2006) dened BELS as a mixing of instruction from two historically separate learning environments: classroom teaching and full e-learning. The term emphasizes the central role of computer-based technologies (e-learning systems) in blended learning, focusing on access and exibility, enhancing classroom teaching and learning activities, and transforming the way individuals learn. From a course design perspective, a BELS course can lie anywhere between the continuum anchored at opposite ends by full face-to-face and virtual e-learning approaches (Rovai & Jordan, 2004). Kerres and De Witt (2003) identied three critical components of BELS that considers the content of the learning materials, the communication between learners and tutors and between learners and their peers, and the construction of the learners sense of place and direction within the activities that denote the learning environment. This is an important distinction because it is certainly possible to enhance regular face-to-face courses with online resources without displacing classroom contact hours. Accordingly, we dened BELS as the combination of online and face-to-face instruction and the convergence between traditional face-to-face learning and e-learning environments. Several BELSs, such as WebCT (www.webct.com) and Cyber University of NSYSU (cu.nsysu.edu.tw) have developed systems that integrate a variety of functions to facilitate learning activities. For example, these systems can be used to integrate instructional material (via audio, video, and text), e-mail, live chat sessions, online discussions, forums, quizzes and assignments. With these kinds of systems, instructional delivery and communication between instructors and students can be performed at the same time (synchronously) or at different times (asynchronously). Such systems can provide instructors and learners with multiple, exible instructional methods, educational technologies, interaction mechanisms or learning resources and applying them in an interactive learning environment to overcome the limitations of classroom and e-learning. As a result, these online learning systems may better accommodate the needs of learners or instructors who are geographically dispersed and have conicting schedules (Pituch & Lee, 2006). As BELS emerge as perhaps the most prominent instructional delivery solution, it is vital to explore what determines learning satisfaction in a blended e-learning environment. 2.2. Social cognitive theory Social cognitive theory (Bandura, 1986) serves as an initial foundation in this study for exploring what determines student learning satisfaction in a blended e-learning environment. Social cognitive theory is a widely accepted and empirically validated model for understanding and predicting human behavior and identifying methods in which behavior can be changed. Several studies have applied it as a theoretical framework to predict and explain an individuals behavior in IS settings. The theory argues that the meta progress of a human being occurs through consecutive interactions with the outside environment and the environment must be subjected to ones cognition process before they affect ones behavior. It proposes that a triadic reciprocal causation among cognitive factors, environmental factors, and human behavior exists. Behavior is affected by both cognitive factors and environmental factors (Wood & Bandura, 1989). Cognitive factors refer to the personal cognition, affect and biological events. Environmental factors refer to the social and physical environments that can affect a persons behavior. Environments inuence an individuals behavior through his or her cognitive mechanisms. Hence, social cognitive theory posits two critical cognitive factors: performance expectations and self-efcacy that inuence individual behavior. It gives prominence to the concept of self-efcacy dened as ones judgments and beliefs of his/her condence and capability to perform a specic behavior recognizing that our performance expectations of a behavior will be meaningless if we doubt our capability to successfully execute the behavior in the rst place. It can enhance human accomplishment and well-being, help determine how much effort people will expend on a behavior, how long they will persevere when confronting obstacles and how resilient they will be in the face of adverse situations. The theory further

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

157

argues that self-efcacy inuences performance expectations and performance expectations also inuence behavior. Thus, self-efcacy and performance expectations are held to be the principal cognitive determinants of individual behavior. Regarding environmental factors, there is ample educational literature and research that shows the learning environment affects a learners behavior and performance. Traditionally, a learning environment was dened in terms of the physical and social environments in a classroom setting. Piccoli et al. (2001) expanded the traditional denition of learning environment and identied ve environmental factors that clarify how an e-learning environment differs from classroom-based education, including technology, content, interaction, learning model, and learner control. These factors can be classied into two categories that particularly are relevant to BELS-specic environments. The rst category relates to the technological environment that includes system functionality and content feature. The second category relates to social environments that include interactions (between learners and instructors or between learners and other learners) and learning climate. 3. Research model and hypotheses Based on the foregoing theoretical underpinnings, we consider that the social cognitive theory is applicable to the BELS learning context. Accordingly, three factors: learners cognitive beliefs (self-efcacy and performance expectations), technological environment (system functionality and content feature), and social environment (interaction and learning climate) are identied and elucidated as the primary dimensions of student learning satisfactions with BELS, as shown in Fig. 1. 3.1. Cognitive factors Cognitive factors refer to the learners cognitive beliefs that inuence their behaviors in using BELS. Two main cognitive variables: computer self-efcacy and performance expectations are believed to be the most relevant factors affecting human behavior in using an information system (IS) (Compeau & Higgins, 1995; Compeau, Higgins, & Huff, 1999; Venkatesh, Morris, Davis, & Davis, 2003). The social cognitive theory dened performance expectations as the perceived consequences of a behavior and further noted they are a strong force guiding individuals actions. The performance expectations are derived from individual judgments regarding valuable outcomes that can be obtained through a requisite behavior. Individuals are more likely to perform behaviors that they believe will result in positive benets than those which they do not perceive as having favorable consequences. Performance expectations are dened as the degree to which a learner believes that using BELS will help him or her to attain gains in learning performance. The denition is similar to the concepts of perceived usefulness, based on Daviss (1989) technology acceptance model (Venkatesh et al., 2003). The inuence of performance expectations on individual behavior of using computer systems has been demonstrated by Compeau and Higgins (1995), Compeau et al. (1999) and Venkatesh et al. (2003). Prior research in education or computer-mediated learning has found that performance expectations are positively related to students learning performance (Bolt & Koh, 2001) and satisfaction (Martins & Kellermanns, 2004; Shih, 2006). Individual attitudes are a function of beliefs, including the behavioral beliefs directly linked to a persons intention to perform a dened behavior (Ajzen & Fishbein, 1980). User acceptance is an important indicator that measures a users positive attitudes toward the IS and predicts their behaviors while using the system, based on theory of reasoned action (Taylor & Todd, 1995). Satisfaction is a good surrogate for user acceptance and is often used to measure learners attitude in computer-mediated learning studies (Chou & Liu, 2005; Piccoli et al., 2001). Thus, we conceptualize the students attitude toward BELS as the learning satisfaction with the BELS dened as the sum of students behavioral beliefs and attitudes that result from aggregating all the benets that a student receives from using BELS. Therefore, the following hypothesis is proposed. H1: A higher level of performance expectations for BELS use will positively associate with a higher level of learning satisfaction with BELS. The second cognitive factor to be applied in this research is self-efcacy. In general, it refers to an individuals beliefs about his or her capabilities to successfully perform a particular behavior. According to social cognitive theory, individuals form their perceptions of self-efcacy toward a task based on cue they receive from the four information sources: (1) past experience and familiarity with similar activities, (2) vicarious learning, (3) social support and encouragement, and (4) attitudes toward the task. Bandura (1986) noted self-efcacy is task-specic and its measures should be tailored to the targeted domain context. Accordingly, several studies have investigated self-efcacy beliefs towards tasks such as computers and IS-related behaviors (Compeau & Higgins, 1995; Compeau et al., 1999). Derived from the general denition of self-efcacy, computer self-efciency was dened as the individual ability to use information technology to accomplish computer-related tasks or jobs (Marakas, Yi, & Johnson, 1998). Computer self-efcacy was also validated as a determinant of IS acceptance and use. We dene computer self-efcacy as the condence in ones ability to perform certain learning tasks using BELS. Prior research has shown that increases in computer self-efcacy improve initiative and persistence, which lead to improved performance or outcome expectations (Francescato et al., 2006; Johnston, Killion, & Oomen, 2005; Piccoli et al., 2001), including attitude and behavioral intention (Venkatesh & Davis, 2000). In the context of computer-mediated learning, empirical evidence indicates that increases in computer self-efcacy improve students condence in their computer-related capabilities, which in turn leads to a perception of positive performance expectations to the learning courses (Bolt & Koh, 2001; Jawahar & Elango, 2001; Santhanam et al., 2008; Shih, 2006). That is, computer self-efcacy could reduce learning barriers in using BELS. If students have higher computer self-efcacy and can control BELS, they will perceive the systems usefulness and value, which in turn motivates their intention to use BELS. Accordingly, the following hypothesis is proposed: H2: A higher level of individuals computer self-efcacy will positively associate with a higher level of performance expectations for BELS use.

3.2. Technological environment The quality and reliability of an e-learning system, as well as easy access to appropriate educational technologies, material content, and course-related information are important determinants of e-learning effectiveness (Piccoli et al., 2001). Thus, system functionality and

158

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

content features are identied as critical technological environment factors for BELS. They are expected to inuence the learner to use and accept BELS. Prior research has shown that system functionality signicantly affected user beliefs in various computer-related contexts (Igbaria, Gamers, & Davis, 1995; Venkatesh & Davis, 2000). For instance, research ndings showed that specic system functionality is a critical factor that inuences e-learning system usage (Hong, Thong, Wong, & Tam, 2002; Pituch & Lee, 2006). Pituch and Lee (2006) dened system functionality as the perceived ability of an e-learning system to provide exible access to instructional and assessment media. Accordingly, we dene system functionality as the perceived ability of BELS to provide exible access to instructional and assessment media. Such media, for example, allows students to access course materials and content, turn in homework assignments, complete tests and quizzes online. In general, content is used to identify various divergent formats and types of information. In this study, content refers to technologybased materials and course-related information that may provide value for learners in the context of BELS. BELS achieves its goals of sharing and delivering course content through various forms of media such as tutorials, online discussions, or web-based courses. Due to the diversity of delivery methods, it is a considerable issue that how to design and represent the hybrid content in appropriate formats or types best suited to delivery or access by BELS (So & Brush, 2008). Appropriate BELS content features, as well as effective design, representing hybrid course content and transparent content knowledge transfer, are core components of BELS design (Piccoli et al., 2001). Drawing on the previous research (Zhang, Keeling, & Pavur, 2000), we dene content feature as the characteristics and presentation of course content and information in BELS. Text, hypertext, graphics, audio and video, computer animations and simulations, embedded tests, and multimedia information are some examples of content features in BELS environment. System functionality and content feature have the potential to directly affect perceived usefulness of IS (Hong et al., 2002; Pituch & Lee, 2006) that are thought to be similar concepts in performance expectation. Several empirical evidences have argued that both content features (Zhang et al., 2000) and system functionality (Pituch & Lee, 2006) affects the effectiveness of computer-mediated learning. That is to say, learners perceiving a higher level of system functionality and content features in BELS will lead to a higher level of performance expectations for BELS use. In addition, in the BELS environment, the diverse content features can be delivered and accessed depending upon the support of appropriate system functionality BELS facilitated (Pituch & Lee, 2006; So & Brush, 2008). Thus, we consider that the content feature highly depends on the power and quality of system functionality of BELS. Therefore, the following hypotheses are proposed: H3: A higher level of system functionality of BELS will positively associate with a higher level of performance expectations for BELS use.

H4: A higher level of content features in BELS will positively associate with a higher level of performance expectations for BELS use.

H5: A higher level of system functionality in BELS will positively associate with a higher level of content features in BELS. 3.3. Social environment In computer-mediated instructional design, there is an increasing focus on facilitating human interaction in the form of online collaboration, virtual communities, and instant messaging in the BELS context (Graham, 2006). From the group interactions perspective, social environment factors, such as collaborative learning (Francescato et al., 2006), learning climate (Chou & Liu, 2005) and social interaction (Johnston et al., 2005) are important antecedents of beliefs about using an e-learning system. Prior research (Pituch & Lee, 2006) shows that social interaction has a direct effect on the usage of an e-learning system. The interactions among students, between faculty and students and learning collaboration are the keys to learning process effectiveness. In addition, the emotional learning climate is an important indicator of learning effectiveness. Interaction is dened in our study as the social interactions among students themselves, the interactions between instructors and students, and collaboration in a BELS environment. Learning climate is dened as the learning atmosphere in the BELS context. Johnston et al. (2005) argued that contact and interaction with instructors and learners is a valid predictor of performance. A positive learning climate encourages and stimulates the exchange of ideas, opinion, information, and knowledge in the organization that will lead to better learning satisfaction (Prieto & Revilla, 2006). That is, when learners believe that BELS provides effective student-to-student and student-to-instructor interactions and improves learning climate, they will be more satised with BELS. Therefore, the following hypotheses are proposed: H6: A higher level of interaction will positively associate with a higher level of performance expectations for BELS use.

H7: A higher level of interaction will positively associate with a higher level of learning climate.

H8: A higher level of learning climate will positively associate with a higher level of learning satisfaction with BELS.

4. Method 4.1. Instrument development To develop the self-report instrument, a number of prior relevant studies were reviewed to ensure that a comprehensive list of measures were included. All measures for each construct were taken from previously validated instruments and modied based on the BELS context.

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

159

For instance, the measures for learning satisfaction were selected from Chiu, Hsu, and Sun (2005) and Wu and Wang (2005). Measures for computer self-efcacy and performance expectations were taken from Compeau and Higgins (1995). The measures for content feature were adapted from Zhang et al. (2000) and Molla and Licker (2001). The measures for functionality were taken from Pituch and Lee (2006). The measures for student and instructor interactions were taken from Johnston et al. (2005), Kreijns, Kirschner, and Jochems (2003), and Pituch and Lee (2006). Finally, the measures for the learning climate were selected from Chou and Liu (2005). Supplementary material lists the denition of each construct, its measures, and the references. The questionnaire consisted of two major parts including a portion for the respondents basic data and another for the responses to our research constructs. The basic data portion recorded the subjects demographic information (e.g., gender, age, highest education, computer experiences, and so forth). The second part recorded the subjects perception of each variable in the model. It includes items for each construct. All items are measured via a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree). Once the initial questionnaire was developed, an iterative personal interview process with professionals, instructors, and students from blended learning courses (including four instructors and ve students from three different universities) was conducted to verify the completeness, wording, and appropriateness of the instrument and to conrm the content validity. Feedback from the interview processes served as the basis for correcting, rening, and enhancing the experimental scales. For example, scale items were eliminated if they represented the same aspects with only slightly different wording and modied if the semantics were ambiguous in order to enhance the psychometric properties of the survey instrument. At the end of the pre-test, there were seven constructs with 21 items in total to be used for the survey. 4.2. Participants The empirical data were collected using a cross-sectional survey methodology. Participants for this study were students that had the opportunity to take courses via BELS. We distributed 518 paper-based and online questionnaires to target universities. The target universities were purposively selected for the universities or colleges actually implemented BELS courses in Taiwan. Because of the applications of BELS are still at an early stage in Taiwan, the target universities are relatively rare. Data were collected via snowball and convenient sampling. Due to the conventional expectation of low survey response rates in survey studies, we endeavored to nd a specic local contact person for each target university who was placed in charge of distributing the questionnaire. Three hundred and seven-six questionnaires were returned. Sixty-four responses were incomplete and had to be discarded. This left 212 valid responses for the statistical analysis, and a valid response rate of 40.93% of the initial sample. Among the valid responses, 84 responses were received from physical classrooms and 128 responses were gathered from online learning environments. The potential non-response bias was assessed by comparing the early versus late respondents that were weighed on several demographic characteristics. The results indicated that there were no statistically signicant differences among demographics between the early (the rst semester) and late (the second semester) respondents. These results suggest that non-response bias was not a serious concern. The respondent proles and the non-response bias analysis results are shown in Table 1. 5. Results Partial least squares (PLS) method was applied for the data analysis in this study. An analytical method is, in general, recommended for predictive research models emphasized on theory development, whereas Linear Structural Relationships (LISREL) is recommended for conrmatory analysis and requires a more stringent adherence to distributional assumptions (Jreskog & Wold, 1982). PLS performs a Conrmatory Factor Analysis (CFA). In a CFA, the pattern of loadings of the measurement items on the latent constructs was explicitly specied in the model. The t of this pre-specied model is then examined to determine its convergent and discriminant validities. This factorial validity deals with whether the loading patterns of the measurement items corresponds to the theoretically anticipated factors (Gefen & Straub, 2005). Convergent validity is shown when each measurement item correlates strongly with its assumed theoretical construct, while discriminant validity is shown when each measurement item correlates weakly with all other constructs except for the one to which it is theoretically associated. The evaluation of the model t was conducted in two stages (Chin, 1998; Gefen & Straub, 2005). First, the measurement validation was assessed, in which construct validity and reliability of the measures were assessed. The structural model with hypotheses was then tested. The statistical analysis strategy involved a two-phase approach in which the psychometric properties of all scales were rst assessed through CFA and the structural relationships were then validated using bootstrap analysis. 5.1. Measurement validation For the rst phase, the analysis was performed in relation to the attributes of individual item reliability, construct reliability, average variance extracted (AVE), and discriminant validity of the indicators as measures of latent variables. The assessment of item loadings, reliability, convergent validity, and discriminant validity was performed for the latent constructs through a CFA. Reective items should be uni-dimensional in their representation of the latent variable and therefore correlated with each other. Item loadings should be above 0.707, showing that more than half of the variance is captured by the constructs. The results indicate that all items of the instrument had signicant loadings higher than the recommended value of 0.707. As shown in Table 2, all constructs exhibit good internal consistency as evidenced by their composite reliability scores. The composite reliability coefcients of all constructs and the AVE in the proposed model (see Fig. 1) are more than adequate, ranging from 0.821 to 0.957 and from 0.605 to 0.849, respectively. To assess discriminant validity: (1) indicators should load more strongly on their corresponding construct than on other constructs in the model and (2) the AVE should be larger than the inter-construct correlations (Chin, 1998). AVE measures the variance captured by a latent construct, that is, the explained variance. For each specic construct, it shows the ratio of the sum of its measurement item variance as extracted by the construct relative to the measurement error attributed to its items. As a rule of thumb, the square root of the AVE of each construct should be larger than the correlation of the specic construct with any of the other constructs in the model (Chin, 1998) and should be at least 0.50 (Fornell & Larcker, 1981). As the results show in Table 3, all constructs meet the above mentioned requirements. The

160

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

Table 1 Respondents prole and the results of non-response bias analysis (N = 212). Variables Classication Total (%) Early respondents (%) 0.500 0.500 0.476 0.387 0.108 0.019 0.009 0.038 0.142 0.269 0.047 0.278 0.226 0.000 0.047 0.547 0.406 0.071 0.198 0.495 0.236 0.165 0.448 0.226 0.052 0.019 0.090 0.208 0.203 0.142 0.061 0.047 0.340 0.292 0.354 0.203 0.094 0.028 0.028 10.7 (years) 73 72 48 41 14 2 1 3 12 27 5 36 23 0 1 60 45 7 20 53 26 18 50 25 6 2 5 24 22 15 9 6 30 33 33 22 10 4 4 27.076 (0.133) 0.344 0.340 0.453 0.387 0.132 0.019 0.009 0.014 0.057 0.127 0.024 0.170 0.108 0.000 0.005 0.283 0.212 0.033 0.094 0.250 0.123 0.085 0.236 0.118 0.028 0.009 0.024 0.113 0.104 0.071 0.042 0.028 0.142 0.156 0.156 0.104 0.047 0.019 0.019 Late respondents (%) 33 34 53 41 9 2 1 5 18 30 5 23 25 0 9 56 41 8 22 52 24 17 45 23 5 2 14 20 21 15 4 4 42 29 42 21 10 2 2 0.156 0.160 0.500 0.387 0.085 0.019 0.009 0.024 0.085 0.142 0.024 0.108 0.118 0.000 0.042 0.264 0.193 0.038 0.104 0.245 0.113 0.080 0.212 0.108 0.024 0.009 0.066 0.094 0.099 0.071 0.019 0.019 0.198 0.137 0.198 0.099 0.047 0.009 0.009 2.695 (0.747)

v2 (Sig.)

Gender

Male Female 1830 3140 4150 5160 >61 Student Industry Manufacturing Service Finance Others Senior high school College (2 years) University (4 years) Graduate school Pure physical classroom experience Pure virtual classroom experience Physical experience more than virtual experience Virtual experience more than physical experience <0.5 years 0.51 years 2 years 3 years 4 years >4 years 1 times 2 times 3 times 4 times 5 times P6 times <1 h 13 h 35 h 57 h 79 h >9 h 11.79 (years)

106 106 101 82 23 4 2 8 30 57 10 59 48 0 10 116 86 15 42 105 50 35 95 48 11 4 19 44 43 30 13 10 72 62 75 43 20 6 6 13.7 (years)

0.022 (0.50) 1.344 (0.855)

Age

Types of Jobs

4.806 (0.440)

Education level

8.824 (0.32)

BELS experience

0.371 (0.946)

BELS experience: participating in BELS (years)

BELS experience: participating in BELS (times)

4.710 (0.452)

BELS experience:spending time in the BELS (1 week)

4.729 (0.450)

Average years of computer usage experience

Table 2 Results of conrmatory factor analysis. Construct Computer self-efcacy (CSE) System functionality (SF) Content feature (CF) Interaction (I) Performance expectations (PE) Learning climate (LC) Learning satisfaction (LS) Items 3 3 2 3 3 3 4 Composite reliability 0.821 0.905 0.890 0.915 0.940 0.926 0.957 AVE 0.605 0.761 0.802 0.782 0.838 0.807 0.849

values for reliability are all above the suggested minimum of 0.7 (Hair, Anderson, Tatham, & Black, 1998). Thus, all constructs display adequate reliability and discriminant validity. All constructs share more variance with their indicators than with other constructs. Thus, the convergent and discriminant validity of all constructs in the proposed research model can be assured.

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

161

Computer Self -efficacy

H2

System Functionality H5 Content Feature

H3

Performance Expectations

H1 Learning Satisfaction

H4 H6 Learning Climate H7 H8

Interaction

Fig. 1. The research model for BELS learning satisfaction.

Table 3 Correlation between constructs. CSE CSE SF CF PE I LC LS

a

SF 0.872 0.609 0.534 0.507 0.513 0.534

CF

PE

LC

LS

0.778a 0.539 0.492 0.527 0.389 0.425 0.44

0.896 0.596 0.608 0.593 0.601

0.916 0.662 0.761 0.798

0.884 0.727 0.614

0.898 0.74

0.921

The shaded numbers in the diagonal row are square roots of the average variance extracted.

5.2. Hypotheses testing In the second phase of the statistical analysis, the structural model was assessed to conrm to what extent the relationships specied by the proposed model were consistent with the available data. The PLS method does not directly provide signicance tests and path coefcient condence interval estimates in the proposed model. A bootstrapping technique was used to estimate the signicance of the path coefcients. Bootstrap analysis was performed with 200 subsamples and the path coefcients were re-estimated using each of these samples. The parameter vector estimates was used to compute parameter means, standard errors, signicance of path coefcients, indicator loadings, and indicator weights. This approach is consistent with recommended practices for estimating signicance of path coefcients and indicator loadings (Lhmoeller, 1984) and has been used in prior information systems studies (Chin & Gopal, 1995; Hulland, 1999). Hypotheses and corollaries testing were performed by examining the size, the sign, and the signicance of the path coefcients and the weights of the dimensions of the constructs, respectively. Results of the analysis for the structural model are presented in Fig. 2. The estimated path coefcient (standardized) and its associated signicance level are specied next to each link. The R2 statistic is indicated next to the dependent construct. The statistical signicance of weights can be used to determine the relative importance of the indicators in forming a latent construct. We found that all specied paths between constructs in our research model had signicant path coefcients. The results provide support for our model. One indicator of the predictive power of path models is to examine the explained variance or R2 values (Barclay, Higgins, & Thomson, 1995; Chin & Gopal, 1995). R2 values are interpreted in the same manner as those obtained from multiple regression analysis. They indicate the amount of variance in the construct that is explained by the path model (Barclay et al., 1995). The results indicate that the model explained 67.8% of the variance in learning satisfaction. Similarly, 37.1% of the variance in content feature, 55.1% of the variance in performance expectations and 52.9% of the variance in learning climate were explained by the related antecedent constructs. The path coefcient from computer self-efcacy to performance expectations is .229 and from interaction to learning climate is 0.727. The magnitude and signicance of these path coefcients provides further evidence in support of the nomological validity of the research model. Table 4 summarizes the direct, indirect, and total effects for the PLS analysis. As for the cognitive factors, Hypotheses H1 and H2, effectively drawn from computer self-efcacy to performance expectations and performance expectations to learning satisfaction are supported by the signicant path coefcients, respectively. That is, students who had higher computer self-efcacy will have higher performance expectations, which in turn will lead to higher learning satisfaction. As for the technological environment factors, with the signicant path coefcients, the analysis results also provide support for the hypotheses H3 and H4, effectively drawn from system functionality and content feature to performance expectations. In addition, Hypothesis H5, effectively drawn from system functionality to content feature is also supported by the signicant path coefcients. However, it is interesting to note that the indirect effect of system functionality on performance expectations was stronger than its direct effect (see Table 4). This seems to indicate that system functionality alone may not be sufcient for improving performance expectations when the BELS content features are not well-matched or designed.

162

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

Computer Self-efficacy 0.229*** 0.092* 0.171**

System Functionality 0.609*** Content Feature (R2 =37.1%)

Performance Expectations (R 2=55.1%) Learning Climate (R =52.9%)

2

0.557*** Learning Satisfaction 0.315*** (R 2=67.8%)

0.422***

0.727*** Interaction * P<0.05 **P<0.01 ***P<0.001

Fig. 2. PLS analysis results.

Table 4 Standardized causal effects of PLS analysis. Dependent latent variables Independent latent variables Standardized causal effects Direct Content feature Performance expectations System functionality Computer self-efcacy System functionality Content feature Interaction Interaction Computer self-efcacy System functionality Content feature Interaction Performance expectations Learning climate 0.609 0.229 0.092 0.171 0.422 0.727 Indirect Total 0.609 0.229 0.196 0.171 0.422 0.727 0.128 0.109 0.095 0.465 0.557 0.315 11.849*** 3.717*** 1.358* 2.011** 5.203*** 18.849*** 3.693*** 2.307** 1.802* 7.175*** 7.006*** 3.804*** T-statistics

0.104

Learning climate Learning satisfaction

0.128 0.109 0.095 0.465 0.557 0.315

* ** ***

P < 0.05. P < 0.01. P < 0.001.

As for the social environment factors, hypotheses H6 and H7, the paths from interaction to performance expectations and learning climate are supported. That is, interaction apparently inuences the performance expectations and learning climate, respectively. Hypothesis H8, effectively drawn from learning climate to learning satisfaction is also supported by the signicant path coefcients. That is, learning climate inuences learning satisfaction. Overall, both performance expectations and positive learning climate have a direct effect on learning satisfaction; performance expectations provide the greatest contribution (total effect) to learning satisfaction. 6. Conclusion BELS environments have become the most prominent instructional delivery alternative when employed in e-learning systems. This study presents a theoretical model that was based on social cognitive theory for investigating the key determinants of student learning satisfaction in a BELS environment. The results provide strong evidence for the nomological validity of each construct and the effects on learning satisfaction, as shown in Fig. 2. The estimate of 0.551 for the performance expectations construct (R2 = 55.1%) for these paths provides good support for the hypothesized impact of computer self-efcacy, system functionality, content feature, and interaction on the dependent variable, performance expectations. In addition, the estimate of 0.371 for the content feature construct (R2 = 37.1%) for the path provides support for the hypothesized impact of system functionality on the content feature. The 0.529 estimate for the leaning climate construct (R2 = 52.9%) for these paths provides support for the hypothesized impact of interaction on the dependent variable, learning climate. In addition, the 0.678 estimate for the learning satisfaction construct (R2 = 67.8%) denotes that the learning satisfaction as perceived by learners is directly and indirectly mediated by the performance expectations and learning climate. Therefore, as a whole, the model has strong explanatory power for the student learning satisfaction with BELS. The signicant path coefcients, effect size and the value of the R2 reinforce our condence in the hypotheses testing results and provide support for the association with learning satisfaction in the BELS setting. The results demonstrated that the BELS learning satisfaction is affected by the interaction among cognitive, technological environment, and social environment factors. We conrmed that technology alone does not cause learning to occur. It is consistent with the theoretical perspective of social cognitive theory: human behavior as a reciprocal interplay of cognitive factors, environment, and behavior (Bandura, 1986).

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

163

The empirical results indicate that performance expectations and learning climate are two strong determinants of learning satisfaction with BELS. The computer self-efcacy, system functionality, content feature, and interaction provided an indirect contribution to learning satisfaction via the above determinants. Thus, as students become more condent and capable of learning with BELS and more accustomed to the BELS learning environments, they will likely expect more benets from the use of BELS, foster positive learning climate, and, overall, be more satised with the BELS learning. These ndings provide initial insights into those factors that are likely signicant antecedents for planning and implementing BELS to enhance student learning satisfaction. The contributions and implications of this study include the following: A BELS environment should enhance students performance expectations and foster positive learning climate. Our ndings indicate that performance expectations provide the most contribution to learning satisfaction. This suggests that instructors should take advantage of BELS effectiveness in designing and teaching courses to enhance students beliefs that they would be able to achieve improved outcomes with BELS. A positive learning climate signicantly affects students learning satisfaction. This suggests that both instructors and learners should foster and motivate the positive learning atmosphere within the BELS learning context. Consequently, if students believe that using BELS is worthwhile, valuable and simple, they will be more likely to accept it resulting in greater satisfaction. Education institutions should provide incentives and supports to enhance students computer self-efcacy. The empirical results demonstrate that computer self-efcacy have a signicant positive inuence on performance expectations. This implies that learners should have the computer competence necessary to exploit BELS and control over his/her learning activities. Therefore, educational institution administrators and instructors should provide sufcient incentives and administrative supports to encourage students to actively participate in BELS courses and to enhance their computer self-efcacy. BELS should provide built-in help to t various learners needs in different learning circumstances. BELS should offer appropriate system functionality and content features with multimedia presentation and exibility. The results show that system functionality and content features have a positive inuence on perceived expectations. These ndings suggest that: (1) BELS should offer useful information with synchronous and asynchronous learning and content-rich design that satisfy students needs; (2) BELS should provide various types of content presentation (e.g., multimedia), customized functions to allow learners control over the system, and exible access to t various students learning requirements. It seems reasonable to note that education institutions may offer BELS-related technical training, awareness programs to the students to enhance students comprehension of BELS. BELS should provide effective interaction tools and instructors should motivate interaction publicly. The results demonstrate that participant interaction had a signicant positive inuence on both performance expectations and learning climate. In addition, interaction has the most contribution (total effect) to the performance expectations. These ndings suggest that when implementing BELS courses, the instructors should motivate the positive interaction publicly to increase participant communication and collaborative learning via the system. In general, learning climate is a function and positive feedback of participant interaction in a BELS environment. A positive learning climate can make learning easy and natural. Thus, if BELS could support a good social environment to facilitate the students-to-student and student-to-instructor connectivity interaction (e.g., interactive communication and collaborative learning), learners will be more likely to actively participate in interaction, so as to foster better learning climate and to perceive greater BELS performance expectations and learning satisfactions. Although our study provides insights into what determines student learning satisfaction in a BELS environment, it has several limitations that also represent opportunities for future research. First, the model was validated using sample data gathered from the target universities in Taiwan. The fact that the participants come from one country limits the generalizability of the results. Other samples from different nations, cultures, and contexts should be gathered to conrm and rene the ndings of this study. Second, given the self-report instrument used (e.g. for measuring computer self-efcacy, system functionality, and content feature), therefore, the typical shortcomings associated with self-report measures must be recognized when interpreting the results. Third, this research sets a timely stage for future research in understanding the determinants of learning satisfaction in a BELS environment. It would be interesting to use a longitudinal design to examine the relationships among the identied research variables might be a useful extension to the current study. Finally, the results cannot be exhaustive and future works should endeavor to uncover additional determinants of student learning satisfaction with BELS.

Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.compedu.2009.12.012. References
Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice-Hall. Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood Cliffs, NJ: Prentice Hall. Barclay, D., Higgins, C., & Thomson, R. (1995). The partial least squares approach to causal modeling, personal computer adoption and use as an illustration. Technology Studies, 2, 285309. Bolt, M. A., & Koh, H. C. (2001). Testing the interaction effects of task complexity in computer training using the social cogitative model. Decision Sciences, 32, 119. Bonk, C. J., Olson, T. M., Wisher, R. A., & Orvis, K. L. (2002). Learning from focus group: An examination of blended learning. Journal of Distance Education, 17(3), 97118. Chin, W. W. (1998). The partial least squares approach to structural equation modeling. In G. A. Marcoulides (Ed.), Modern methods for business research (pp. 298336). Mahwah, NJ: Erlbaum. Chin, W. W., & Gopal, A. (1995). Adoption intention in GSS: Relative importance of beliefs. The Data Base for Advances in Information Systems, 26, 4263. Chiu, C. M., Hsu, M. H., & Sun, S. Y. (2005). Usability, quality, value and e-learning continuance decisions. Computers & Education, 45, 399416. Chou, S. W., & Liu, C. H. (2005). Learning effectiveness in a web-based virtual learning environment: A learner control perspective. Journal of Computer Assisted Learning, 21, 6576. Compeau, D. R., & Higgins, C. A. (1995). Computer self-efcacy: Development of a measure and initial test. MIS Quarterly, 19, 189211. Compeau, D. R., Higgins, C. A., & Huff, S. (1999). Social cognitive theory and individual reactions to computing technology: A longitudinal study. MIS Quarterly, 23, 145158. Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13, 318339. EL-Deghaidy, H., & Nouby, A. (2008). Effectiveness of a blended e-learning cooperative approach in an Egyptian teacher education programme. Computers & Education, 51, 9881006. Fornell, C. D., & Larcker, F. (1981). Evaluating structural equation models with unobservable variables and measurement errors. Journal of Marketing Research, 18(2), 3950.

164

J.-H. Wu et al. / Computers & Education 55 (2010) 155164

Francescato, D., Porcelli, R., Mebane, M., Cuddetta, M., Klobas, J., & Renzi, P. (2006). Evaluation of the efcacy of collaborative learning in face-to-face and computer-supported university contexts. Computers in Human Behavior, 22, 163176. Gefen, D., & Straub, D. W. (2005). A practical guide to factorial validity using PLS-graph: Tutorial and annotated example. Communications of the Association for Information Systems, 16(5), 91109. Graham, C. R. (2006). Chapter 1: Blended learning system: Denition, current trends, future directions. In C. J. Bonk & C. R. Graham (Eds.), Handbook of blended learning. San Francisco, CA: Pfeiffer. Hair, J. F., Anderson, R. E., Jr., Tatham, R. L., & Black, W. C. (1998). Multivariate data analysis with readings (5th ed.). NJ: Prentice Hall. Hara, N., & Kling, R. (2000). Students distress with a web-based distance education course: An ethnographic study of participants experiences. Information, Communication and Society, 3, 557579. Hong, W., Thong, J. Y. L., Wong, W. M., & Tam, K. Y. (2002). Determinants of user acceptance of digital libraries: An empirical examination of individual differences and system characteristics. Journal of Management Information Systems, 18(3), 97124. Hulland, J. (1999). Use of partial least squares (PLS) in strategic management research: A review of four recent studies. Strategic Management Journal, 20, 195204. Igbaria, M., Gamers, T., & Davis, G. B. (1995). Testing the determinants of microcomputer usage via a structural equation model. Journal of Management Information Systems, 11(4), 87114. Jawahar, I. M., & Elango, B. (2001). The effect of attitudes, goal setting and self-efcacy on ender user performance. Journal of End User Computing Practice, 13(2), 4045. Johnston, J., Killion, J., & Oomen, J. (2005). Student satisfaction in the virtual classroom. The Internet Journal of Allied Health Sciences and Practice, 3(2). Jreskog, K. G., & Wold, H. (1982). The ML and PLS techniques for modeling with latent variables: Historical and comparative aspects. In K. G. Jreskog & H. Wold (Eds.), Systems under indirect observation: Causality structure and prediction (pp. 219243). Amsterdam: North Holland. Kerres, M., & De Witt, C. (2003). A didactical framework for the design of blended learning arrangements. Journal of Educational Media, 28(2/3), 101113. Kinshuk, D., & Yang, A. (2003). Web-based asynchronous synchronous environment for online learning. United States Distance Education Association Journal, 17(2), 517. Kreijns, K., Kirschner, P., & Jochems, W. (2003). Identifying the pitfalls for social interaction in computer-supported collaborative learning environments: A review of the research. Computers in Human Behavior, 19, 335353. Lhmoeller, J. B. (1984). LVPS 1.6 program manual: Latent variable path analysis with partial least squares estimation. Universitaet zu Koehn, Zentralarchiv fuer Empirische Sozialforschung. Maki, R. H., Maki, W. S., Patterson, M., & Whittaker, P. D. (2000). Evaluation of a web-based introductory psychology course: Learning and satisfaction in on-line versus lecture courses. Behavior Research Models, Instruments, and Computers, 32, 230239. Marakas, G. M., Yi, M. Y., & Johnson, R. D. (1998). The multilevel and multifaceted character of computer self-efcacy: Toward clarication of the construct and an integrative framework for research. Information Systems Research, 9, 126162. Martins, L. L., & Kellermanns, F. W. (2004). A model of business school students acceptance of a web-based course management system. Academy of Management Learning and Education, 3, 726. Molla, A., & Licker, P. S. (2001). E-commerce systems success: An attempt to extend and respecify the DeLone and McLean model of IS success. Journal of Electronic Commerce Research, 2(4), 111. Osguthorpe, R. T., & Graham, C. R. (2003). Blending learning environments: Denitions and directions. The Quarterly Review of Distance Education, 4(3), 227233. Piccoli, G., Ahmad, R., & Ives, B. (2001). Web-based virtual learning environments: A research framework and a preliminary assessment of effectiveness in basic it skills training. MIS Quarterly, 25, 401426. Pituch, K. A., & Lee, Y. (2006). The inuence of system characteristics on e-learning use. Computers & Education, 47, 222244. Prieto, I. M., & Revilla, E. (2006). Formal and informal facilitators of learning capability: The moderating effect of learning climate. IE working paper. WP06-09. 21-02-2006. Rovai, A., & Jordan, H. (2004). Blended learning and sense of community: A comparative analysis with traditional and fully online graduate courses. The International Review of Research in Open and Distance Learning, 5(2), 112. Santhanam, R., Sasidharan, S., & Webster, J. (2008). Using self-regulatory learning to enhance e-learning-based information technology training. Information Systems Research, 19, 2647. Shih, H. P. (2006). Assessing the effects of self-efcacy and competence on individual satisfaction with computer use: An IT student perspective. Computers in Human Behavior, 22, 10121026. Singh, H. (2003). Building effective blended learning programs. Educational Technology, 44(1), 527. So, H. J., & Brush, T. A. (2008). Student perceptions of collaborative learning, social presence and satisfaction in a blended learning environment: Relationships and critical factors. Computers & Education, 51, 318336. Taylor, S., & Todd, P. A. (1995). Decomposition and crossover effects in the theory of planned behavior: A study of consumer adoption intentions. International Journal of Research in Marketing, 12, 137156. Venkatesh, V., & Davis, F. D. (2000). A theoretical extension of the technology acceptance model: Four longitudinal eld studies. Management Science, 46, 186204. Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unied view. MIS Quarterly, 27, 425478. Wood, R., & Bandura, A. (1989). Social cognitive theory of organization management. Academy of Management Review, 14, 361384. Wu, J. H., Tennyson, R. D., Hsia, T. L., & Liao, Y. W. (2008). Analysis of e-learning innovation and core capability using a hypercube model. Computers in Human Behavior, 24, 18511866. Wu, J. H., & Wang, S. C. (2005). What drives mobile commerce? An empirical evaluation of the revised technology acceptance model. Information & Management, 42, 719729. Yang, Z., & Liu, Q. (2007). Research and development of web-based virtual on-line classroom. Computers & Education, 48, 171184. Zhang, X., Keeling, K. B., & Pavur, R. J. (2000). Information quality of commercial website home pages: An explorative analysis. In Proceeding of the 21th international conference on information systems (pp. 164175). Brisbane, Australia.