I
Andrea Miccichè - 11375337
University of Amsterdam
Faculty of Economics and Business
in Business Administration
Specialization in
Supervisor:
II This document is written by student Andrea Miccichè who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
III Technology is changing the way people live their life. Old ways of doing things cannot keep up with the endless possibilities rising from the development of new software and devices that reach the market every day. This research aims to understand what drives people in adopting new technologies. By linking together two of the major theories on technology use and acceptance, Social Cognitive Theory (Bandura, 1989) and the Technology Acceptance Model (TAM) (Davis, 1989), the study investigates if intrinsic motivational variables, such as CA (Computer Anxiety) and MLSE (Mobile Learning Self-efficacy), and variables on how technology is perceived, such as PEU (Perceived Ease of Use) and PU (Perceived Usefulness) and BI (Behavioral Intention to use the given platform), are related. And if those variables can predict the actual use that university students make of a mobile learning platform. Data were processed with the means of multiple mediated regressions, using SPSS. Results show that there is a strong positive relationship between MLSE and all the variables of TAM, but no significant relationship with the actual use. Also, CA was found to significantly and directly relate only with PEU. Furthermore, multiple indirect effects take place between the mentioned variables. This research has multiple practical implication for professors, academic institutes, and developers of mobile learning systems. Guiding them in what to focus when building a mobile platform, and how to educate students to make the best use out of a given platform.
___________________________________________________________________________ : mobile learning, technology acceptance model, social cognitive theory, university
IV
Table of Contents
1. Introduction ... 6
2. Literature Review ... 8
2.1. Mobile Learning... 8
2.2. Technology Acceptance Model (TAM) ... 10
2.2. Self-efficacy ... 11
2.2.1. Computer Self-efficacy (CSE) ... 122
2.2.2. CSE and TAM... 133
2.3. Computer Anxiety (CA) ... 144
2.3.1. CA and TAM ... 144
3. Theoretical Framework ... 155
4. Method ... 188
4.1. Participants and Procedure ... 188
4.2. Measures ... 19
4.3. Analysis and Results ... 222
5. Discussion... 355 5.1. Findings... 355 5.2. Limitations ... 388 5.3. Practical Implication ... 39 6. Conclusions ... 40 References………..43 Appendix………51
V
List of Figures and Tables
Figure 1 – Research Design ... 18
Table 1 - Means, SD, Correlation and Cronbach's Alpha for all variables. ... ..25
Table 2 - Multiple mediated Regression MLSE - USE... ... 29
Table 3 - Indirect effects MLSE - TAM - USE... ... .30
Table 4 - Multiple mediated Regression CA – USE ... 34
Table 5 - Indirect effects CA - TAM - USE ... 35
6
1. Introduction
Technology is getting more and more important in every aspect of people’s life, from work and education to leisure and even personal health. In the last ten years, thousands of new technological devices with various features entered onto market, dramatically changing how people interact with many of the social agents in their life (Borgmann, 2009).
Education is a field that has been under much of technological pressure. Especially for Higher Education institutes, such as universities, changes have been dramatic; the way students’ approach their studies; the way they interact with each other, with faculty members, and with the administration itself. For example, massive open online courses (MOOCs) are available for almost any subject from a significant number of universities; most of the student-professor exchange takes place over emails and web-based platforms; all necessary materials are available through e-learning platforms; in some cases, even the exams are computer-based. In this changing environment, it is important to understand what drives students in the adoption of these new technologies, to make their transition to this new way of learning as smooth as possible.
Previous studies that focused on understanding what influences people in adopting and using new technologies explored how intrinsic and extrinsic motivations help to shape the attitude toward new technologies. Concepts such as general and application-specific Computer Self-efficacy (CSE), Computer Anxiety (CA) (Barbeite & Weiss, 2004; Beckers & Schmidt, 2001; Chua, Chen, & Wong, 1999; Compeau & Higgins, 1995b; Durndell & Thomson, 1997; Durndell & Haag, 2002; Glass & Knight, 1988; Green, 1999; Hackett, 1985; Heinssen, Glass, & Knight, 1987a; Loyd & Gressard, 1984; Loyd, Loyd, & Gressard, 1987; McIlroy, Bunting, Tierney, & Gordon, 2001; Powers, 1973; Rosen, Sears, & Weil, 1987; Whitley, 1997; Yau & Leung, 2016), Perceived Ease of Use (PEU), Perceived Usefulness (PU), Behavioral Intention
7 (BI), which are part of the Technology Acceptance Model (TAM), (Davis, 1989; Kim, Chun, & Song, 2009; Legris, Ingham, & Collerette, 2003; Li, Qi, & Shu, 2008; Park, 2009; Turner, Kitchenham, Brereton, Charters, & Budgen, 2010; Yi & Hwang, 2003) emerged as key variables, able to predict acceptance of new technologies and actual use from its intended users. All these variables have been mostly used isolated one another. For instance, Social Cognitive Theory used in multiple research the concept of CA and CSE to predict usage and acceptance and subsequent usage of a given technology (Bandura, 1989; Compeau, Higgins, & Huff, 1999; Compeau & Higgins, 1995a). Focusing on the intrinsic characteristic of a person to infer on its actions. At the same time, Davis (1989) developed TAM. This model focuses more on how the given technology at study is perceived by the potential user, including both intrinsic characteristic of the subject at study, BI, and characteristic of the specific technology and how it is seen by the user, PU and PEU.
This research paper aims at expanding current literature by linking together two streams of research to create a unified model that considers all the variables cited above, in a university context by answering the following questions: Do CSE and CA affect TAM in mobile learning platforms for university students? If so, what is the interaction between CSE, CA and the distinct TAM’s variables? Do CSE and CA have a direct relationship with the actual use of these platforms? Alternatively, is it fully mediated by TAM?
In the following paragraphs, the relevant literature will be reviewed to provide clear definitions of the key concepts of this research, such as mobile learning, CSE, CA, TAM and the relationship that occurs between the different constructs will be explored. The methodology will be described, and the paper will end with the results, and the conclusions and recommendation sections.
8
2. Literature Review
The pervasion of new technology in the life of students and teachers has dramatically changed the process of teaching and learning in Universities (Laurillard, 2013). Nowadays, most students extensively use computers, tablets, and smartphones for their everyday university life. From researching additional materials and papers to working on their assignments or exchanging information with faculty members to even taking entire courses online. This trend will likely spread even more as new technologies, such as wearable technologies, virtual reality, or augmented reality, will become cheaper and more widely used.
To explore this intriguing world, this study builds on the extensive literature of two major streams of research, the Social Cognitive Theory (Bandura, 1989) and the Technology Acceptance Model (Davis, 1989). In both these literature streams, researchers link people’s intrinsic motivations and perceptions to predict, respectively, human actions, and in this case the acceptance and actual use of a given technology for university students. The following paragraphs will provide an extensive overview of these theories, the relationship that occurs between the main variables that are part of these theories, linking them with mobile learning domain.
2.1. Mobile Learning
The concept of Mobile Learning has gained relevance in higher education domains. This can be seen by the growing number of conferences, seminars, and workshops all over the world, such as the MLEARN series of lectures, the International Workshop on Mobile and Wireless Technologies in Education (WTME), the International Association for Development of the Information Society (IADIS), etc.; and the increasing number of references on the topic at generalist academic conferences, such as the Association for Learning Technology Conference (ALT-C) (Traxler, 2007).
9 Despite interest in such subject increased, there is not a clear definition of the concept of Mobile Learning. Researchers focused on different characteristics of this evolving trend, and thus, Mobile Learning has been defined in numerous ways. Some scholars focused only on the technologies and its hardware, and for them, Mobile Learning is learning that occurs with the help of mobile devices and wireless transmission (Pinkwart, Hoppe, Milrad, & Perez, 2003; Quinn, 2000). Other scholars believe that this techno-centric approach is too restrictive and that the focus should shift from devices to humans and their experiences. In this stream of research, scholars take into consideration how people learn; they learn with their peers and teachers, through examples, by doing and most importantly, people learn in a well-defined learning environment. Until a few years ago, the learning environment was limited by classroom walls and teachers as the main source of knowledge. With the rise of computers and, more generally, all the mobile devices available on the market, the boundary of the learning environment became blurred, allowing the learners to move beyond its classroom. This means that in Mobile Learning, the learner is the one who should be able to move freely. They define mobility as the increased learner’s ability to move their learning environment as they move ((Barbosa & Geyer, 2005; Laouris & Eteokleous, 2005). To sum up, Mobile Learning is a complex concept that has not been clearly defined so far. What is clear is that it is a notion that includes various parameters, such as the technology, the user’s learning experience, the mobility of the hardware, the learner and its learning environment. This research focuses on Mobile Learning Platforms, which we define as any application, website, or device that allows students to access their respective learning environment from anywhere, anytime, and on what drives users in the use of the technology. To do so, the researcher will adapt TAM for the current setting. The model has been extensively used in the context of e-learning platforms and other educational systems, both in an academic context as well as in work-related environment (Hsia, Chang & Tseng, 2014; Park, 2009). Knowing that it can be assumed it represents a good
10 fit for the study. By including MLSE and CA in TAM, the study hopes to increase the understanding in what drives students to use a mobile learning platform.
2.2. Technology Acceptance Model (TAM)
Davis (1989) first developed TAM, based on the Theory of Reasoned Action (TRA), to explain the potential users’ BI to use a new technology. The original construct was formed by four main variables; PEU and PU, as independent variables; BI, as dependent variable; and the attitude, which was a mediator between the independent and the dependent variables. Subsequent studies demonstrated how attitude is a weak mediator and thus, it was discarded from the model (Davis, Bagozzi, & Warshaw, 1992; Szajna, 1996).
Many researchers have demonstrated the internal relationship between the different variables composing TAM (Chau, 1996; Davis & Venkatesh, 1996; Li et al., 2008; Park, 2009; Szajna, 1996; Venkatesh & Davis, 2000). To better understand the relationship between the variables of the model the meta-analysis (2008) wrote by Li and colleagues, which examined thirty-four papers on TAM, was reviewed. The meta-analysis presents mixed results. As for the correlation between PEU and PU, twenty-four papers found a significant positive relationship, while in the remaining seven articles this relationship was not significant. Only eleven out of the thirty-four papers conclude that there was a significant direct relationship between PEU and BI. Also, nine articles describe a significant direct link between PEU and actual use; five present a positive relationship while four a negative relationship. However, this does not directly affect the intention to use or the actual use the technology. For PU, twenty-four papers discovered a significant direct relationship with BI and only five with the actual use. Finally, seven papers reported in their results a significant direct correlation between BI and actual use (Hung & Chang, 2005; Szajna, 1996; Taylor & Todd, 1995). In practice, this means that the easier the technology is perceived by its users, the higher their level of PEU, the greater will be the benefit
11 they perceive in term of usefulness (PU). However, this does not directly influence their intention to use, BI, or the actual use, USE, of the technology. On the other hand, PU directly influences the intention to use the technology, but it does not have a direct effect on the actual use. Finally, it is unclear what is the effect of the intention to use new technology to the real use. This is strongly related to what kind of measurements are used to gather data on usage.
The model presents strong external validity as well. Many researchers adopted it in different industries and for different kind of technologies and they all verified its consistency. Holden & Karsh (2010) in their review on the use of TAM in healthcare collected more than twenty studies. Other studies tested its validity in the context of online banking (Lai & Li, 2005; Pikkarainen, Pikkarainen, Karjaluoto, & Pahnila, 2004), and e-commerce (Pavlou, 2003; Wu & Wang, 2005). While e-learning systems have been studied both at the academic level (Park, 2009), and in companies (Hsia, Chang, & Tseng, 2014).
2.2. Self-efficacy
Researchers conceptualised three different ways in which human agency operates. Autonomous agency, mechanical agency and emergent interactive agency. Social cognitive theorist believe that people are not autonomous agents or mechanical agents of animating environmental influences. Instead, they make the causal contribution to their motivation and operations throughout a system of reciprocal causation (Bandura, 1989; Davis et al., 1992). Therefore, any model that wants to understand the determinant of human action must include self-generated influences as principal actors. One of the main determinants of human action is one’s belief of self-efficacy which operate through cognitive, motivational, and affective processes. (Bandura, 1989). First, regarding the cognitive processes, self-efficacy influences thought that could either be of self-promotion or self-impediment. Since a significant part of one’s behaviour is regulated by pre-decided goals. Those personal goals are set through
self-12 evaluation of one’s abilities, stronger levels of SE are linked to higher goals and with a greater commitment to them (Locke, Frederick, Lee, & Bobko, 1984; Taylor, Locke, Lee, & Gist, 1984). The perception people have on their level of efficacy affects the kind of anticipatory thoughts they create in their mind. Scenarios in which people see themselves executing activities successfully increase the actual performance of the task (Bandura, 1986). Second, self-efficacy regulates people’s motivation level. The more they believe in their abilities, the greater is their effort and persistence (Bandura, 1988). Everyday life is filled with difficulties, failures, inequities, and to overcome these impediments people need a solid level of self-efficacy. Successful people, innovators and more in general social reformers tend to have an optimistic view of their efficacy to influence the events in their life. If not exaggerating, these beliefs support the effort needed to obtain personal and social fulfilment (Bandura, 1989; Taylor & Brown, 1988). Lastly, self-efficacy influences the affective processes on people. It influences stress and depression levels in certain situations. This emotional response has both a direct and indirect effect on actions (Bandura, 1989). People who believe that they can manage difficult or negative situations do not get flustered by them. On the contrary, those who believe they are not capable of managing them, suffer from high levels of stress and anxiety, and in turn, this reduces their performances (Brosnan, 1998).
2.2.1. Computer Self-efficacy (CSE)
Researchers have been using the concept of self-efficacy in studies concerned with what affects the adoption process of new technologies, and subsequent use. Compeau and Higgins, (1995a) define CSE as “a judgment of one’s capability to use a computer” (p. 192). In the same paper, they adapt three relevant dimensions of self-efficacy for CSE. These dimensions are: magnitude, which refers to the judgment on one’s level of expected capability when using a computer; strength, which relates to the confidence about the given judgement of the expected capability when using a computer; and generalization, which refers to the degree to which the
13 judgment is limited to a specific situation (e.g., hardware, application system, software package).
Other studies argue that CSE is a multilevel variable that works both at the general level (GCSE) and at the application-specific level, Mobile Learning Self-efficacy (MLSE), in this case, and that they are two separate theoretical constructs. Even if strongly related one another they should not be interchanged (Marakas, Yi, & Johnson, 1998). On the one hand, there is GCSE which is the product of all the previous experiences in any computer-related activity and is defined as “an individual judgement of efficacy across multiple computer application domains” (Marakas et al., 1998, p. 129). On the other hand, there is the application-specific self-efficacy which refers to “an individual’s perception of efficacy in performing specific computer-related tasks within the domain of general computing” (Marakas et al., 1998, p. 128). The latter allows measuring the individual perceived capability in completing specific computer-related tasks, excluding the cross-domain skills fundamental in the performance of a task that requires the use of a computer.
2.2.2. CSE and TAM
CSE has been shown to affect the actual use of a given technology both directly and indirectly, through numerous significant variables and constructs. Many studies demonstrated the strong impact that CSE has on TAM (Agarwal, Sambamurthy, & Stair, 2000; Davis & Venkatesh, 1996; Yi & Hwang, 2003). Davis and Venkatesh (1996) empirically proved the causal flow from CSE to system-specific PEU. Agarwal et al. (2000) building on these findings found that application-specific CSE presents a stronger relation with PEU, compared to GCSE. For the other variables incorporated in TAM, previous studies have not found any significant relationship with CSE (Venkatesh & Davis, 1996; Venkatesh & Davis, 2000; Venkatesh, 2000).
14
2.3. Computer Anxiety (CA)
From the second half of the ’80 and throughout the ’90, the concept of CA gained relevance among researchers. The rise of computers to mainstream tools came together with an increase in computer-related discomforts and phobias (Hackbarth, Grover, & Mun, 2003, Glass & Knight, 1988; Torkzadeh & Angulo, 1992). Such concept has been defined in many ways. Some researchers refer to it as a change in physiological measures, such as blood pressure, heart rate, sweaty palms, dizziness, and so on, when dealing with computers (Hemby, 1998; LaLomia & Sidowski, 1993; Powers, 1973). Others focused on the attitude that computers generate in people (Reece & Gable, 1982). Lastly, affective factors, such as fear, have been identified in subjects who suffer from CA (Chua et al., 1999; Heinssen, Glass, & Knight, 1987b). Beckers and Schmidt (2001) identified four elements that recur in many studies, “low confidence in one’s own ability to use computers; negative affective responses to them; becoming aroused while using a computer or thinking about it, and negative beliefs about the role of the computer in our lives.” (Beckers & Schmidt, 2001, p. 36). As stated, there is a close relationship between CSE and CA. In their paper, Beckers and Schmidr (2001) showed that this link is mediated by computer literacy, which is one’s experience with computers.
2.3.1. CA and TAM
In many studies in the information system (IS) domains, researchers have analysed various intrinsic motivational variables together with TAM to understand if those variables influence PEU, PU, BI or the actual use of a new technology or system and to understand the strength and direction of this influence. What they found is that both positive and negative intrinsic motivation affect TAM and in turn actual usage (Yi & Hwang, 2003). As for CA, Hackbarth and colleagues (2003) discovered a significant direct negative relationship with PEU, but it appears that it does not produce any significant effect on the other variables in the model.
15
3. Theoretical Framework
Following the findings of the Social Cognitive Theory. CSE play a major role in the creation of behaviour and actions toward computers in people. CA has been identified as an important construct that influences many variables that then predict actual use of a technology (Bandura, 1986; Igbaria & Iivari, 1995). Venkatesh and Davis (1996) found experimental evidence for the causal flow from CSE to PEU. This finding is supported by later research; Agarwal et al. (2000), while analysing this phenomenon discovered that application-specific self-efficacy, from now on MLSE, has a stronger effect on PEU than GCSE. This means that students with a higher level of MLSE will perceive the given mobile learning platform easier to use. This finding is supported by Igbaria and Iivary (1995), who also found a direct relationship between CA and PEU. On the other hand, they have not found any direct effect between CSE, CA on PU. Instead, they found that their effect on PU is fully mediated by PEU. In practice, this means that people with a higher level of MLSE will perceive a mobile learning platform more useful because of the increased ease of use they perceive, while using it. At the same time, people with a higher level of CA will perceive the platform less useful given their perceived difficulty when using the platform. Finally, many studies have found a direct relationship between CSE and the actual use of technology (Compeau et al., 1999; Compeau & Higgins, 1995a). The study will validate once again these relationships in the context of a mobile learning platform for university students through the following hypothesis:
H1a: MLSE has a direct positive relationship with PEU of a given mobile learning system. H1b: PEU mediates the indirect positive relationship between MLSE and PU of a given mobile learning system.
16 H2b: PEU mediates the indirect negative relationship between CA and PU of a given mobile learning system.
H3: MLSE has a direct positive effect on the actual use of a given mobile learning system. The internal relationship between the variables incorporated in TAM has been tested and validated many times and in different settings (Davis, 1989; Igbaria & Iivari, 1995; Venkatesh & Davis, 1996). The current study hopes to replicate once again these relationships in the context of a mobile learning platform for university students through the following hypothesis:
H4a: PEU has a direct positive relationship with the PU of a given mobile learning system. H4b: PU mediates the indirect positive relationship between PEU and BI of a given mobile learning system.
H5: PU has a direct positive relationship with the BI of a given mobile learning system. H6: PU has a direct positive relationship with the actual use of a given mobile learning system. H7: BI has a direct positive relationship with the actual use of a given mobile learning system.
17
18
4. Method
4.1. Participants and Procedure
This research aims to explore the relationships occurring between two motivational variables (MLSE and CA) with the well know and adopted TAM, composed by PEU, PU and BI, and use of a mobile learning platform in a university environment. To achieve the given goal, a quantitative study which used multiple mediated regressions was conducted.
Data were collected through a questionnaire created with Qualtrics. Participants freely agreed to participate in this research study beforehand. Moreover, were assured of the confidentiality of the responses, and a follow up of the study was promised to whoever was interested. The population for this study is composed by students currently enrolled in either a Bachelor’s or a Master’s degree at the University of Amsterdam. The research used a convenience sampling technique. The survey was distributed through different digital means such as University group emails, University’s Facebook groups, and private messages, but only UvA students had the possibility to complete the questionnaire. Because of different means through which the questionnaire has been sent, the final sample is diverse.
After cleaning the data from missing values, the number of valid responses amounts to 98. The sample’s average age is of 23.84 years (SD = 2.3). Moreover, 52% of the respondents were male, and the remaining 48% were female. Most of the participants were enrolled in Business Administration courses, but also Sociology, Physics, Computer Science, Psychology, Neuroscience, Molecular Sciences, and Art students took part in the questionnaire.
4.2. Measures
The complete questionnaire consists of thirty-four items, clustered to measure the different variables. The first block of questions aims at collecting demographic information about the participants, such as age, gender, and field of study. The remaining four parts of the
19 questionnaire measured: general CSE, MLSE, CA, PEU, PU, BI, attitude and actual use of mobile learning systems for each participant. Not all the variables were needed for the analysis but since it did not impact questionnaire’s length, general CSE and attitude items were added as back-up data.
General CSE. To measure general CSE, the current study adopted the new computer self-efficacy scale for an internet sample (Barbeite & Weiss, 2004). It consists of a four- item scales for general computer activities (“I feel confident making selections from an on-screen menu”; “I feel confident using the computer to write a letter or essay”; “I feel confident escaping or
exiting from a program or software”; “I feel confident calling up a data file to view on the monitor screen”) rated on a seven-point Likert scale (1 = Strongly disagree; 7 = Strongly agree). The computer self-efficacy scale for an internet sample for general computer activities
showed good internal reliability (Cronbach’s Alpha = .830). The corrected item-total correlations indicate that all items have a good correlation with the total score of the scale (> .50). None of the items would substantially affect reliability if deleted.
MLSE. To measure MLSE, the study adopted the 2-items scale based on the Technology Acceptance Model (TAM), rated on a Seven-point Likert scale (1 = Strongly disagree; 7 =
Strongly agree) (Park, 2009). The items were adjusted to measure mobile learning systems
attitude by changing the world “e-learning” with “Mobile Learning Systems” (“I feel confident
finding information in the mobile learning system”; “I have the necessary skills for using a mobile learning system”). A definition of the concept of mobile learning was given before
participants answered the questions. The MLSE scale showed good internal reliability (Cronbach’s Alpha = .868). The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .70). None of the items would substantially affect reliability if deleted.
20 CA. To measure CA, the study adopted the new computer anxiety scale for an internet sample (Barbeite & Weiss, 2004). It consists of an eight-items scale, which measure one’s CA when using the computer (“Working with a computer would make me very nervous”; “I get a sinking
feeling when I think of trying to use a computer”; “Computers make me feel uncomfortable”; “Computers make me feel uneasy and confused”) and one’s CA in computer-related activities
(I feel anxious when… “Learning computer terminology”, “Thinking about prepackaged
(software packages) programs for a computer”; “Visiting a computer store”, “Taking a class about the uses of computers”), rated on a seven point Likert scale (1 = Strongly disagree; 7 = Strongly agree). The computer anxiety scale for an internet sample when using the computer
has high reliability (Cronbach’s Alpha = .945). The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .70). None of the items, in either one of the scales, would substantially affect reliability if deleted.
PEU. To measure mobile learning PEU, the study adopted a 3-items scale based on the Technology Acceptance Model (TAM), rated on a seven-point Likert scale (1 = Strongly
disagree; 7 = Strongly agree) (Park, 2009). The items were adjusted to measure mobile
learning systems attitude by changing the world “e-learning” with “Mobile Learning Systems” (“I find mobile learning systems easy to use”; “Learning how to use a mobile learning system
is easy for me”; “It is easy to become skillful at using a mobile learning system”). The PEU
scale adjusted for mobile learning has high reliability, with Cronbach’s Alpha = .926. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (< .80). None of the items would considerably affect reliability if deleted.
PU. To measure mobile learning PU, the study adopted a 3-items scale based on the Technology Acceptance Model (TAM), rated on a seven-point Likert scale (1 = Strongly
disagree; 7 = Strongly agree) (Park, 2009). The items were adjusted to measure mobile
21 (“mobile learning would improve my learning performance”; “mobile learning would
increase academic productivity”; “mobile learning could make it easier to study course content”). The PU scale adjusted for mobile learning has high reliability, with Cronbach’s
Alpha = .879. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .70). None of the items would considerably affect reliability if deleted.
Attitude. To measure the attitude toward mobile learning, the study adopted a 3-items scale based on TAM, rated on a seven-point Likert scale (1 = Strongly disagree; 7 = Strongly agree) (Park, 2009). The items were adjusted to measure mobile learning systems attitude by changing the world “e-learning” with “Mobile Learning Systems” (“Studying through mobile learning
is a good idea”; “Studying through mobile learning is a wise idea”; “I am positive toward mobile learning”). The attitude scale adjusted for mobile learning has high reliability, with
Cronbach’s Alpha = .951. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .80). None of the items would considerably affect reliability if deleted.
BI. To measure mobile learning BI, the study adopted a 2-items scale based on the Technology Acceptance Model (TAM), rated on a seven-point Likert scale (1 = Strongly disagree; 7 =
Strongly agree) (Park, 2009). The items were adjusted to measure mobile learning systems
attitude by changing the world “e-learning” with “Mobile Learning Systems” (“I intend to
check announcements from mobile learning systems frequently”; “I intend to be a heavy user of mobile learning system”). The BI scale adjusted for mobile learning has good internal
reliability, with Cronbach’s Alpha = .837. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (< .70). None of the items would considerably affect reliability if deleted.
22 Actual USE. To measure this last variable, the researcher created a block of four questions that aim at analyzing the effective use of mobile learning systems. Of these four questions, only two were used in the analysis. One of them aimed at measuring the frequency of use of a mobile learning platform (“How often do you use on AVERAGE a mobile learning system in a week?”) with a 7-points response scale (1 = Once; 2 = Twice, 3 = Three times; 4 = Four times; 5 =
Five times; 6 = More than five times; 7 = None). The other one aimed at measuring the intensity
of the usage (“How much time do you spend on AVERAGE on mobile learning systems in a
day?”) with a 4-points response scale (1 = 5-10; 2 = 11-30; 3 = 31-60; 4 = More than 60).
The other two questions (“What do you mostly use mobile learning systems for?”; “What
device do you mostly use when using mobile learning platforms?”) were not included in the
analysis.
Appendix A includes the complete questionnaire that participants filled out.
4.3. Analysis and Results
Once the data collection process was completed, data were downloaded and exported in SPSS to proceed with the statistical analysis. Frequencies were checked, and missing values were excluded from the analysis. After the data polishing procedure, reliability tests were run for every variable, to check whether the items in the questionnaire were measuring the desired variable or not.
A correlation analysis between age, gender (control variables), and the main variables was run. The results present interesting findings that seem to provide initial support for some of the hypothesis. MLSE (M = 5.78, SD = 1.11) is strongly positively correlated with PEU (r = 0.731,
p <.01), PU (r = 0.633, p <.01), BI (r = 0.572p <.01), and actual USE (r = 0.237, p <.05),
while it presents a strong negative correlation with CA (r = -0.489, p <.01). These findings support the initial hypotheses H1a and H3. CA (M = 2.12, SD = 1.16) is strongly negatively
23
p <.01), BI (r = -0.393, p <.01), and actual USE (r = -0.184), but this last correlation is not
statistically significant. This means that students that differ by one unit on CA are estimated to differ by -0.613 units on PEU, 0.392 units on PU, -0.393 units on BI. These findings support the initial hypotheses H2a. PEU (M = 5.53, SD = 1.26) is strongly positively correlated with
PU (r = 0.651, p <.01), BI (r = 0.526, p <.01), and actual USE (r = 0.350, p <.01). This means that students who differ by one unit on PEU are estimated to differ by 0.651 units in PU, 0.526 units on BI, 0.373 units on actual use. These findings support initial hypotheses H4a. PU (M =
5.16, SD = 1.22) is strongly positively correlated with BI (r = 0.602, p <.01), and still
positively correlated with the actual USE (r = 0.272). This means that students who differ by one unit on PU are estimated to differ by 0.602 units on BI and, 0.272 units in actual use. These findings support the initial hypotheses H5 and H6. Lastly, BI (M = 5.84, SD = 1.44) is positively correlated with the actual USE of a mobile learning system (r = 0.185), but the relationship is not statistically significant. The correlation analysis also showed that the two control variables added do not present any significant correlation with the main variables of the research model.
24 Table 2 - Means, Standard Deviations, Correlation and Cronbach's Alpha for all study variables
Variables M SD 1 2 3 4 5 6 7 8 Age 23.84 2.31 - Gender 1.47 0.50 -0.054 - CA 2.12 1.16 0.026 -0.024 (.945) MLSE 5.78 1.11 -0.015 -0.006 -0.489** (.868) PEU 5.53 1.26 0.050 0.084 -0.613** 0.731** (.926) PU 5.16 1.22 -0.048 -0.128 -0.392** 0.633** 0.651** (.879) BI 5.84 1.44 -0.060 0.097 -0.393** 0.572** 0.526** 0.602** (.837) USE 3.53 1.32 0.005 0.077 -0.184 0.237* 0.350** 0.272** 0.185 - N = 95
Note: Gender and Age are control variables; Cronbach’s Alpha values in brackets on diagonal Note: **. p < .01; *. p < .05
25 After the correlation analysis, multiple regressions were run to test the different hypothesis of the model. For the first regression, PROCESS model 6 by Hayes was used to run the multiple mediation analysis having MLSE as independent variable, USE as dependent variable and the three variables part of TAM, namely PEU, PU and BI as mediators. The direct effect of MLSE on USE of a given mobile learning platform is not statistically significant (c’1 = -0.323, t(90) =
-0.1729, ns). Even so, among two students with same levels of PEU, PU and BI, the one who
experience one unit’s higher MLSE is estimated to score 0.317 units higher in their use of a mobile learning platform (c1 = 0.3172, SE = 0.119, p < .01, CI = .080 to .554). This is because
of the effect MLSE has on the other variables of the model. There are seven indirect effects that illuminate the underlying process.
The first indirect effect is the specific indirect effect of MLSE on actual use of a mobile learning platform through PEU. This effect showed that students who experience a higher level of MLSE, experience significant increase in their PEU (a1 = 0.838, p < .001), which is further associated
with increased usage of mobile learning platforms (d1 = 0.349, p < .05), independently of their
PU and BI. This indirect effect can be interpreted as non-significant because zero is part of the bootstrap confidence interval (indirect effect1 = 0.292, SE = 0.1482, CI = -.016 to .572). The
second indirect effect is the effect of MLSE on actual use of a mobile learning platform through PEU and PU in serial. Students who experience a higher level of MLSE experience a higher level of PEU, which was also related to a higher level of PU (a3 = 0.429, p < .001). Meanwhile,
the relationship between PU and actual use was not-significant (d2 = 0.139, ns). The overall
indirect effect is not statistically significant given the bootstrap confidence interval contains zero (indirect effect2 = 0.050, SE = 0.051, CI = -.047 to .161). The third is the effect of MLSE on
USE, through PEU and BI in serial. As mentioned the higher MLSE’s level, the greater is student’s PEU but there is no significant relationship between the latest and BI (b2 = 0.039, ns)
26 this indirect effect was found not-significant (indirect effect3 = -0.02, SE = 0.015, CI = -.053 to
.017). The fourth effect is the one of MLSE on the actual use, through PEU, PU, and BI. As
previously stated, students who experience a higher level of MLSE are related to a higher level of PEU which in turn increase their PU level. In the current sample, students who score one unit higher in their PU are expected to score 0.496 units higher in their BI (b3 = 0.496, p < .001) but
this, as showed for the third indirect effect, is not significantly related to an increase in the actual USE. The overall indirect effect was not-significant since the bootstrap confidence interval included zero in its values (indirect effect4 = -0.010, SE = 0.020, CI = -.063 to .023). The fifth
effect is the effect of MLSE on the actual USE of a mobile learning platform, through PU. Students with a higher MLSE are expected to score higher on their PU (a2 = 0.331, p < .01) but
there is not a significant relationship between higher levels of PU and greater usage of a mobile learning platform. The underlying indirect effect was statistically not-significant as well (indirect effect5 = 0.046, SE = 0.047, CI = -.041 to .147). The sixth effect is the effect of MLSE
on the USE, through PU and BI. Students with a higher level of MLSE are expected to score higher in their level of PU, which in turn translates to a higher level of BI. While, BI and actual USE are not significantly related to each other. This indirect effect was found not-significant (indirect effect6 = -0.009, SE = 0.019, CI = -.060 to .020). The seventh and last effect of the
first multiple mediated regression is the effect of MLSE on USE, through BI. Students with higher level of MLSE are expected to score higher in their BI (b1 = 0.346, p < .05), but there is
not a significant direct relationship between a higher level of BI and greater usage of a given mobile learning platform. Also, this last indirect effect was found not-significant (indirect effect7
= -0.019, SE = 0.043, CI = -.148 to .036). To test the hypothesis H1b a simple mediated
regression, through PROCESS model 4 by Hayes was run, using MLSE as the dependent, PU as independent and PEU as mediator. The analysis showed that the MLSE has both a direct (a2
27 effect on PU, which means that the relationship between MLSE and PU is partially mediated by PEU. To test the internal relationship between MLSE and the TAM variables other multiple regressions were run. It was found that MLSE has a positive indirect effect on BI, both through PEU and PU and through PU alone (indirect effect11 = 0.164, SE = 0.068, CI = .056 to .325)
(indirect effect10 = 0.179, SE = 0.071, CI = .009 to .357). Meanwhile, there is not a significant
indirect effect of MLSE on BI through PEU (indirect effect9 = 0.033, SE 0.122, CI = -.194 to
.288).
The findings from this first sets of regression, using MLSE as the dependent variable, shed light on the underlying relationship between MLSE and the variables incorporated in TAM, and they allow some consideration regarding some of the initial hypothesis of the current study. It was found that MLSE has a positive direct relationship with PEU, PU, and BI which support the hypothesis H1a. Contrary to H1b the relationship between MLSE and PU is only partially
mediated by PEU. Also, the results showed that there is not any direct positive relationship between MLSE and actual use of a mobile learning platform, contrary to what was speculated in H3.
28 Table 2 - Multiple mediated Regression MLSE - USE
DV
PEU (M) PU(M) BI (M) USE (Y)
IV Coeff. SE P Coeff. SE P Coeff. SE p Coeff. SE p
MLSE (X) a1 0.838 0.078 < .001 a2 0.331 0.119 < .01 b1 0.346 0.155 < .05 c'1 -0.032 0.187 ns PEU (M) a3 0.429 0.105 < .001 b2 0.039 0.144 ns d1 0.349 0.168 < .05 PU (M) b3 0.496 0.131 < .001 d2 0.139 0.165 ns BI (M) d3 -0.054 0.186 Ns Age (Control) v1 0.037 0.038 Ns v3 -0.400 0.039 Ns V5 -0.018 0.049 ns V7 -0.005 0.058 ns Gender (Control) V2 0.217 0.175 Ns V4 -0.404 0.179 < .05 V6 0.426 0.231 ns V8 0.191 0.275 ns Constant i1 -0.525 1.075 Ns i2 2.432 1.091 < .05 i3 0.872 1.409 ns I4 1.180 1.654 ns Rsq = 0,556 Rsq = 0,512 Rsq = 0,442 Rsq = 0,1442 F(3,93) = 38.802; p < .001 F(4,92) = 24.121; p < .001 F(5,91) = 14.436; p < .001 F(6,90) = 2.528; p < .05
29 Table 3 - Indirect effects MLSE - TAM - USE
Effect SE p LLCI ULCI
Direct effect on use c1' -0.032 0.187 ns -0.403 0.339
Total effect on use c1 0.317 0.119 < .01 0.040 0.554
Boot SE Boot LLCI Boot ULCI
Indirect effect1 a1d1 0.292 0.148 -0.016 0.573 Indirect effect2 a1a3d2 0.050 0.051 -0.047 0.161 Indirect effect3 a1b2d3 -0.002 0.015 -0.053 0.016 Indirect effect4 a1a3b3d3 -0.010 0.020 -0.063 0.023 Indirect effect5 a2d2 0.046 0.047 -0.041 0.147 Indirect effect6 a2b3d3 -0.009 0.019 -0.060 0.020 Indirect7 b1d3 -0.019 0.043 -0.148 0.036 Indirect effect8 a1a3 0.360 0.086 0.209 0.549 Indirect effect9 a1b2 0.033 0.122 -0.194 0.288 Indirect effect10 a1a3b3 0.179 0.071 0.069 0.357 Indirect effect11 a2b3 0.164 0.068 0.056 0.325
Note: Indirect effect1 = MLSE→ PEU → USE Indirect effect2 = MLSE → PEU → PU → USE Indirect effect3 = MLSE → PEU → BI → USE Indirect effect4 = MLSE → PEU → PU → BI → USE Indirect effect5 = MLSE → PU → USE
Indirect effect6 = MLSE → PU → BI → USE Indirect effect7 = MLSE → BI → USE Indirect effect8 = MLSE → PEU → PU Indirect effect9 = MLSE → PEU → BI Indirect effect10 = MLSE → PEU → PU → BI Indirect effect11 = MLSE → PU → BI
30 After that, a new multiple mediated regression, using again PROCESS model 6 by Hayes was run to investigate how CA interact with TAM’s variables and actual use of a mobile learning platform. Mediator and dependent variables were the same as in the first regression but, of course, the independent variable used was CA. The direct effect of CA on the actual use of a mobile learning platform was found statistically not-significant (c’1 = 0.047, t(88) = 0.3227,
ns). Even so, among two students with same levels of PEU, PU and BI, the one who experience
a higher level of CA is estimated to use less a given mobile learning platform (total indirect
effect = -0.255, SE = 0.136, CI = -.576 to -.046). This is because CA affects the other variables
present in the model. Also in this regression, there are seven indirect effects that explain the process in which CA influences the actual use of a mobile learning platform.
The first indirect effect is the specific indirect effect of CA on the real use of a given mobile learning platform through PEU. The analysis showed that students who experience a higher level of CA are expected to score lower in their PEU (a1 = -0.668, p < .001). As consequence,
these students experience lower usage, given the positive relationship between PEU and use (d1 = 0.330, p < .05). The indirect effect was found not-significant, the bootstrap confidence
interval contained zero within its values (indirect effect1 = -0.221, SE = 0.130, CI = -.502 to
.0166). The second effect is the effect of CA on MLSE, through PEU and PU. Students with
greater CA experience lower PEU which, given the positive relationship between the latest and PU (a3 = 0.668, p < .001), translate in lower PU. There is not significant direct relationship
between PU and use (d2 = 0.126, ns) and the underlying indirect effect is not-significant as
well (indirect effect2 = 0.056, SE = 0.061, CI = -.186 to .060). The third effect is the effect of
CA on USE, through PEU and BI. In our sample, the last two variables do not have a significant direct relationship (b2 = 0.135, ns), nor BI and actual use (d3 = -0.039, ns). Therefore, also the
third indirect effect is not statistically significant (indirect effect3 = 0.004, SE = 0.160, CI =
31 PU, and BI. As mentioned, CA has a direct negative correlation with PEU, which has a direct positive relationship with PU. The analysis shows that PU has a direct positive correlation with BI (b3 = 0.581, p < .001), but there is not a direct relation between BI and USE. This means
that students with higher level of CA are expected to score lower in their PEU which in turn will lower their PU’s scores, which will again translate into a lower level of BI; but this decrease will not affect the use of a given mobile learning platform. The fifth effect is the one of CA on the actual use, through PU. It was found that there is not a significant direct relationship between CA and PU (a2 = 0.030, ns), as it was for PU and use. Also, the indirect
effect was found statistically not-significant (indirect effect5 = 0.004, SE = 0.028, CI = -.029
to .073). the sixth indirect effect, CA on the use through PU and BI, was statistically
non-significant (indirect effect6 = -0.001, SE = 0.008, CI = -.025 to .011). The seventh indirect
effect is the one of CA on use, through BI. In the analysis, there was not a significant direct relationship for either CA and BI (b1 = -0.151, ns) or BI and the actual use. The indirect effect
was not-significant as well (indirect effect7 = 0.006, SE = 0.023, CI = -.022 to .076). To verify
the internal relationship of the TAM variables with CA other multiple regressions were run, using CA as the independent variable and each of the TAM variables as either mediator or dependent. To test H2b a regression using PROCESS model 4 by Hayes was run, having PU as
dependent and PEU as mediator. It showed that CA does not have a significant direct effect on PU (a2). Even so, greater level of CA negatively influences PU of a given mobile learning
platform through its effect on PEU (indirect effect8 = -0.446, SE = 0.117, CI = -.734 to -.255).
Statistically speaking the relationship between CA and PU is fully mediated by PEU. Other regressions were run to have a clear picture of the indirect effect existing within CA and the TAM model. It was found that CA has a statistically significant indirectly affect BI through PEU and PU (indirecteffect10 = -0.259, SE = 0.088, CI = -.505 to -.129), while the other
32 indirect effect was not-significant (indirecteffect9 = -0.090, SE = 0.107, CI = -.326 to .101)
(indirecteffect11 = 0.018, SE = 0.077, CI = -.155 to -.137)
The findings of these regressions allow to give some considerations regarding the initial hypotheses of the study. It was found that CA has a direct negative relationship only with PEU, which confirms H2a. The effect of CA on the remaining variables of the model is fully mediated
by PEU, which supports H2b. Regarding the internal relationship between the variables that
form TAM, the study found that, as expected in H4a and H4b, PEU has a direct correlation with
PU but its effect on BI is fully mediated by PU. Confirming the initial hypothesis H5, the results showed the existence of a direct positive relationship between PU and BI. Finally, contrary to the hypotheses H6 and H7, there was no direct relationship between either PU or BI with the actual use of a mobile learning platform. Meanwhile, PEU showed a significant relationship with the real use in both cases, CA and MLSE.
33 Table 4 - Multiple mediated Regression CA - USE
DV
PEU (M) PU(M) BI (M) USE (Y)
IV Coeff. SE P Coeff. SE p Coeff. SE p Coeff. SE p
CA (X) a1 -0,668 1,150 < .001 a2 0,030 0,103 ns b1 -0,151 0,127 ns c'1 0,047 0,145 ns PEU (M) a3 0,668 0,095 < .001 b2 0,135 0,145 ns d1 0,330 0,166 < .05 PU (M) b3 0,581 0,130 < .001 d2 0,126 0,163 ns BI (M) d3 -0,039 0,120 ns Age (Control) V1 0,038 0,045 ns v3 -0,050 ..0411 ns v5 -0,020 0,051 ns v7 -0,003 0,058 ns Gender (Control) V2 0,183 0,207 ns v4 -0,465 0,189 < .05 v6 0,419 0,240 ns v8 0,186 0,276 ns constant I1 5,769 1,150 < .001 i2 3,279 1,180 < .05 i3 2,265 1,510 ns i4 0,967 1,732 ns Rsq = 0,386 Rsq = 0,466 Rsq = 0,424 Rsq = 0,133 F(3,91) = 19.041; p < .001 F(4,90) = 19.666; p < .001 F(5,89) = 13.084; p < .001 F(6,88) = 2.242; p < .05
34 Table 5 - Indirect effects CA - TAM - USE
Effect SE p LLCI ULCI
Direct effect on use c1' -0,032 0,187 ns -0,403 0,339
Total effect on use c1 0,317 0,119 < .01 0,040 0,554
Boot SE Boot LLCI Boot ULCI
Indirect effect1 a1d1 -0,221 0,130 -0,502 0,017 Indirect effect2 a1a3d2 -0,056 0,061 -0,186 0,061 Indirect effect3 a1b2d3 0,004 0,016 -0,014 0,060 Indirect effect4 a1a3b3d3 0,010 0,029 -0,039 0,083 Indirect effect5 a2d2 0,004 0,028 -0,029 0,073 Indirect effect6 a2b3d3 -0,001 0,008 -0,025 0,011 Indirect effect7 b1d3 0,006 0,023 -0,022 0,076 Indirect effect8 a1a3 -0,446 0,117 -0,734 -0,255 Indirect effect9 a1b2 -0,090 0,107 -0,326 0.101 Indirect effect10 a1a3b3 -0.259 0.088 -0.505 -0.129 Indirect effect11 a2b3 0.018 0.077 -0.155 0.137
Note: Indirect effect1 = CA → PEU → USE Indirect effect2 = CA → PEU → PU → USE Indirect effect3 = CA → PEU → BI → USE Indirect effect4 = CA → PEU → PU → BI → USE Indirect effect5 = CA → PU → USE
Indirect effect6 = CA → PU → BI → USE Indirect effect7 = CA → BI → USE Indirect effect8 = CA → PEU → PU Indirect effect9 = CA → PEU → BI Indirect effect10 = CA → PEU → PU → BI Indirect effect11 = CA → PU → BI
35
5. Discussion
The following section will discuss research’s results, considering research questions and all other studies on the topic, to connect the findings of the current study in the context of previous studies. Limitations and threat to the validity of the research will be discussed as well.
5.1. Findings
Table 6 – Table of findings
Hypotheses Results
H1a MLSE direct relationship PEU Supported H1b MLSE indirect relationship PU, through PEU Partially supported H2a CA direct relationship PEU Supported H2b CA indirect relationship PU, through PEU Supported
H3 MLSE direct relationship USE Not supported
H4a PEU direct relationship PU Supported H4b PEU indirect relationship BI, through PU Supported H5 PU direct relationship BI Supported
H6 PU direct relationship USE Not supported
H7 BI direct relationship USE Not supported
The research model presented explored and explain the elemental relationship between two motivational variables, namely MLSE and CA, the variables part of TAM, namely PEU, PU and BI, and the actual use of a given mobile learning platform. From the analysis, it was found that six of the initial hypotheses were supported, one was only partially supported, and finally, three hypotheses were rejected. These findings expand prior research on the technology’s acceptance model for new technologies. Linking the classic TAM variables with major motivational variables and the self-reported actual usage of the mobile learning platform, we increase the understanding of what kind of motives or heuristics drive students in adopting and using new technologies.
MLSE showed to significantly affect all the variables of TAM positively. However, no direct relationship with use was found. In particular, the relationship between MLSE and PEU has
36 been theorized and empirically demonstrated by various research, but they did not inquire about the direct effect of self-efficacy on the actual system usage (Agarwal et al., 2000; Venkatesh & Davis, 1996; Venkatesh, 2000). The findings of this study confirm the previous research regarding the relationship between self-efficacy and PEU. In the same stream of research, previous studies demonstrated that PU is one of the principal determinants of actual use of an IS, both directly and indirectly through BI (Chau, 1996; Davis, 1989; Mathieson, 1991; Taylor & Todd, 1995). The current study partially contrasts these previous studies given that it was indeed found a direct relationship between PU and BI, but neither BI or PU had a significant relationship with actual use. Earlier social psychology research empirically demonstrated that CSE influences the actual use of an IS, but they did not include BI as a determinant of the actual use (Compeau et al., 1999; Compeau & Higgins, 1995a). Assimilating the two concepts in one single research model, this study wanted to confirm that both mentioned variables are determinants of the actual system use. Unfortunately, findings of the present study discard these hypotheses. Neither BI, or MLSE, or PU resulted significant predictors of the actual usage of a mobile learning platform for students. On the other hand, only PEU resulted in a determinant of the real use of a mobile learning system. These findings are in contrast with the many studies that confirm the link between BI and self-efficacy, both general and application-specific, to actual use of a system (Compeau et al., 1999; Compeau & Higgins, 1995a; Compeau & Higgins, 1995b; Hung & Chang, 2005; Koeszegi, Vetschera, & Kersten, 2004; Venkatesh & Davis, 1996). In practice, this means that students with higher level of MLSE are expected to perceive greater usefulness when using a mobile learning platform. It will be easier for them to use it, and their intention to use the platform will be higher compared to students with lower MLSE. However, this will not affect the actual use of the platform, at least in our study.
As of CA, in the current study, it was found that it strongly affects PEU in the expected direction. This finding is in line with previous research on social cognitive theory, which
37 advocates that CA be one of the main determinants of one’s attitude (Bandura, 1986); and in the original TAM, attitude was closely related with PEU and PU (Davis, 1989). So thus, it enriches the previous research by adding new variables as important determinant of one of the main variables of TAM.
Past research on TAM has had mixed findings regarding the effect of PEU on PU and BI. Some studies found that PEU had a significant effect on both BI and PU (Venkatesh & Davis, 2000). Other studies found that PEU does not have any effect on PU (Agarwal & Karahanna, 2000). Other studies found that right after the introduction of the system PEU had a significant effect on BI and no effect on PU; but after subjects get familiar with the system, PEU had a significant effect on PU and no significant effect on BI (Davis, 1989). In the current study, PEU had a significant indirect effect on BI, fully mediated by PU. Since the study was conducted at the end of the school year, students were already familiar with the mobile learning system in place at the university and thus this study support Davis (1989) findings.
Finally, in the research, the actual use of the mobile learning platform was found to have a direct relationship with PEU and nothing else. These findings are surprising given that previous research found that BI and PU should be two variables that directly affect the actual use of a new technology, and the effect of PEU should be fully mediated by PU (Davis, 1989; Li et al., 2008; Turner et al., 2010). The future researcher should investigate these relationships in other circumstances to check the validity of those findings in comparison with the previous research stream. They should also improve the way in which data regarding the actual use was collected, switching from a subjective and self-reported measurement to an objective one. By doing so, the research model will increase both its external and internal validity.
38
5.2. Limitations
The current study presents many limitations that need to be remarked. First, every model for complex behavioral manifestation is almost surely incomplete, and this is not an exception. In details, in the present study, the researcher includes only two motivational variables, CA and MLSE, as a determinant of TAM and usage. Other researchers have considered different motivational variables, such as enjoyment, learning goal orientation, computer playfulness, and so forth. (Venkatesh & Davis, 2000; Yi & Hwang, 2003). For example, Yi & Hwang (2003) demonstrated how the variables of enjoyment, learning goal orientation have a positive effect on the actual use of systems, through their effect on PEU, PU, and BI. Also, the study only focuses on intrinsic motivation, not allowing for empirical testing of the usability of the platform. Future research could include some design factors in the model to check whether they influence TAM variables or not.
Second, the non-probability convenience sampling technique used in this research, while it allows a significant number of respondents in a short time cycle, it reduces the generalizability of the findings. Third, the cross-sectional design of the research is another limitation of the study. The variables are measured only at one point in time, which is in line with previous studies on user acceptance of technology (Agarwal & Karahanna, 2000; Compeau & Higgins, 1995b; Davis & Venkatesh, 1996). However, it does not collect changes that may happen after a continued use of the system. Also, it is not possible to infer on the causal relationship between two constructs. The fourth limitation of the current research depends on the way data was collected. Since a self-reported questionnaire was used errors and personal biases can easily occur. Also, since it was distributed by online means, the researcher had very little control on the setting in which the questionnaire was conducted, and thus, it is unknown the attention level of the respondents (Fan et al., 2006). To address those issues, participants were assured before the beginning of the questionnaire that their answers were anonymous, that there were no right
39 and wrong answer, and that they should respond with honesty. Fifth, given the complexity of the model a structural equation model would have been the best fit for the analysis. This specific model is commonly used in the social science for his ability to describe the relationship between unobserved constructs. Future researchers should use SEM in in the current model to check whether there are differences in the results.
Finally, another limitation of the present study is its external validity. The present study was designed for university students with extensive computer experience. Since TAM was validated across multiple technologies and settings and almost all the relation between TAM variables were reproduced in the current study, the significant findings of this study are expected to be confirmed. Nonetheless, these findings should be validated in different settings by future research
5.3. Practical Implication
The findings reported in this research paper have several practical implications for different actors. First, the study showed the crucial role of MLSE, and CA in determining the acceptance of a new mobile learning system. Universities can organize training sessions or workshops that aim at increasing MLSE or at reducing CA should and speed-up the acceptance process, and increase the usage of the similar type of platforms for students. Second, PEU has been showed to positively affect PU, directly, and BI of students, indirectly. This means that user-friendly and intuitive systems are used more by students. This is an important finding for both the universities’ administration and systems designers. When setting up a new mobile learning platform, any given university should be accurate in their specification, with emphases on the clearness and ease of use of the platform. Also, when designing mobile learning platforms for universities, systems designers should focus much of their effort in designing user-friendly systems, realizing interfaces that give the same look and feel of a game, to engage users and
40 increase their intention to use the systems. If those suggestions are followed, universities will boost the benefits attained from their investment in mobile learning systems. Third, PU demonstrated to plays an important part in students’ behavioural intention toward the use of new technology. This finding is useful for both professors and the universities’ administration. By empathizing the usefulness of mobile systems during classes or workshops, professors have the power to increase students’ intention to use new mobile learning platforms and have a powerful tool to guide them through their learning. This can be achieved by uploading interesting extra material, videos, websites, tips on exams and so on. At the same time, the administration should set up workshops and seminars to align professors with these new systems and to instruct them on the potential that these platforms have in enhancing the study experience of students. How to enhance the PU of the named platforms for students.
6. Conclusions
The present study adds to the previous body of literature in multiple ways. First, it expands the current knowledge on TAM by including two motivational variables in the model to better understand what drive students when using mobile learning platforms, in higher education institutes. Second, it validates the findings of previous research to the internal relationship of the variables incorporated in TAM. The research model builds on top of two main streams of research, Social Cognitive theory, and TAM. Social Cognitive theory advocates that CSE and CA, two critical motivational variables, are important determinants of the attitude toward, and actual use of new technologies (Brosnan, 1998; Compeau & Higgins, 1995a; Durndell & Haag, 2002). Previous TAM research demonstrated that the variables incorporated in the model, namely PEU, PU, and BI, are good predictors of the actual use of a given system, in numerous environments (Davis, 1989; Venkatesh, 2000). By linking the two research streams, the current study wants to understand if there is a relationship between these motivational variables with the variables incorporated in TAM and the extent of these relationships.
41 Third, the findings demonstrate that the two motivational variables are strongly related, in the hypothesized direction, with the variables in TAM. In particular, both MLSE and CA, are strongly related with PEU, which confirm the findings of previous research on the topic (Agarwal et al., 2000; Davis & Venkatesh, 1996; Venkatesh & Davis, 2000). Furthermore, the study validates once again the internal relationship between the variables incorporated in TAM. What is interesting to point out is the relation between PEU and the other variables. Previous studies have found mixed results regarding this relationship, and thus, there is still much discussion on how PEU interact with the TAM and more in general with the use. In the present analysis, PEU does not have a direct effect on BI. Instead, its effect is fully mediated by PU, which confirms Davis (1989)’s findings. What differs from all the previous research is the direct relationship found between PEU and use, which has mostly been considered indirect (Hu, Chau, Sheng, & Tam, 1999; Li et al., 2008; Turner et al., 2010). On the other hand, it was not found any significant relation between MLSE, BI, and PU with the actual use of the platform. Findings inconsistent with the previous literature on the topic (Bandura, 1986; Brosnan, 1998; Chau, 1996; Davis, 1989; Holden & Karsh, 2010; Igbaria & Iivari, 1995; Lai & Li, 2005). This could be due to the way the use of the platform was measured. Future researchers should test the current model, using objective measurements of usage, such as log-in data, actual time spent log-in the platform, the number of log-in-app downloads, and so forth.
The findings of this research have important implications for universities, professors, and systems designers. It gives them tools to increase the effectiveness of their systems, which in turn increase the learning potential of the average student. Future research should focus on including other motivational variables in the model, not only intrinsic but extrinsic as well, such as A/B testing different designs elements to see which one is used more, or even designing multiple journeys for the users to see which paths is more effective. This will help the creation of guidelines and requirements for universities when setting up their mobile learning system.
