Exploring the Drivers Predicting Behavioral Intention to Use m-Learning Management System: Partial Least Square Structural Equation Model

This study aimed at reporting the drivers predicting Pre-Service English Teachers’ (PSETs) behavioral intention to use m-learning Management Systems (m-LMS) in their learning activities. To achieve the purpose of the study, an extended Technology Acceptance Model (TAM) was administered. Seven variables were included to examine 11 paths of the framework. The instrument was adapted from previous studies and validated through content validity. Further, it was piloted to 76 PSETs for reliability. Two hundred and ten responses were analyzed for the main study. The results of the study were obtained by implementing the procedures of Partial Least Squares Structural Equation Modeling (PLS-SEM). The extended TAM-based drivers are reported to be valid and reliable in the measurement model stage. In the assessment of the structural model, nine out of eleven hypotheses are significant. The strongest significant relationship is reported to emerge between perceived usefulness and attitude toward m-LMS, while the weakest significant relationship is reported between self-efficacy and perceived usefulness. On the other hand, the two paths that are not positively correlated are between; 1) supporting condition and perceived usefulness; 2) subjective norm and behavioral intention to use m-LMS. Recommendations for future research regarding the use of m-LMS in developing countries are provided.


I. INTRODUCTION
Mobile learning or m-learning is a common thing in technology-based education. m-learning is gradually developed due to the rapid innovation in mobile technologies. Any learning activity that is encouraged by the use of mobile devices can be categorized as m-learning based activity [1], [2]. This m-learning use supports people to have a personal-based and customized learning within mobile technologies [2]. The integration of m-learning into teaching and learning activities is always facilitated by the rapid innovation of new mobile applications [3].
The innovation includes the m-Learning Management System (m-LMS) [4]. The m-LMS in education is defined as The associate editor coordinating the review of this manuscript and approving it for publication was Debashis De . an approach in the forms of an online platform with mobile devices provided by educational institutions for faculties and students without being limited by time and space [5]. A number of studies have been conducted regarding the use of m-LMS in education [5]- [7]. However, few studies have been conducted in the context of South-East Asia [8]. Therefore, this study was done to enrich the use of m-LMS in various contexts and settings. Through the Technology Acceptance Model (TAM), six drivers were used to predict Indonesian Pre-Service English Teachers' (PSETs) behavioral intention to use m-LMS in their learning activities.

A. RESEARCH SIGNIFICANCE AND PROBLEMS
Studies have addressed the adoption of m-learning in educationally advanced countries [9], including factors affecting the acceptance of m-learning for the teaching and learning process, such as in the USA [10], South Korea [7], and the UK [11]. However, few studies have been conducted to examine m-learning adoption in developing nations like Indonesia. Therefore, propound investigation of less advanced countries in education should also be promoted. In this study, m-LMS is the focus on the m-learning approach, which leads to two research problems; is the model proposed in this study to measure Behavioral Intention (BI) to use m-LMS in Indonesia valid and reliable? And how are the relationships among drivers in the proposed model?
Technology adoption in education is affected by nationallevel initiatives [9]. Thus, this study can be a guideline for national-level stakeholders in issuing policies regarding the use of m-learning in higher education institutions. The findings are also hoped to help future researchers regarding technology integration, especially in developing countries. For practitioners, the study addresses its contribution to improving the practical activities for teacher educators to more engage m-LMS technology to teach pre-service teachers.
The paper is organized as follows. The introduction presents the background of the paper. The literature review informs the theoretical background of the study. In the method, we explored the approaches to answer the research problems from instrumentation to main data analysis. Following this, the results of the study are presented. The discussion presents the findings and implications of the study. The final section presents the conclusion, limitations, recommendations, and future work.

II. LITERATURE REVIEW
A. MOBILE LEARNING m-learning is defined as learning activities that are conducted through mobile tools. However, it has different considerations if seen based on research interests and perspectives. It was defined as a new kind of learning supported by mobile devices, including ubiquitous communications technology and intelligent user interfaces [12]. m-learning is defined by focusing on mobile computational tools that reflect the significance of various functional parts of mobile tools in learning. Meanwhile, it is also described as e-learning that utilizes mobile tools in the application.
There is a shifting focus on m-learning from students' active participation to meaningful shared. Mobile tools utilized in situational learning have strong power in guaranteeing a good impact on the behavior of student learning. So, educators should have good consideration for all m-learning dimensions in education, such as people, design, and institutions. The impact of the theory of activity in learning should also be addressed in reintroducing ''m-learning'' in education. Some experts [4], [5] reported that active participation and engagement of students could ensure the success of the m-learning application in educational contexts. In this way, m-learning should be deeply discussed for its suitableness with the students' characteristics that are needed to be applied in learning courses.

B. MOBILE DEVICES AND m-LEARNING IN INDONESIA
The development of mobile technology has facilitated opportunities for every user to communicate more quickly and get information more easily. The number of smartphone users in Asia has been 49% of mobile device users across the world. Based on the latest report [13], there are 175.4 million internet users in Indonesia in 2020. Compared to the previous year, there is an increase of 17% or 25 million internet users in this country. Based on Indonesia's total population of 272.1 million, 64% of the population has access to the Internet. Indonesia is a very potential market for the use of mobile devices and the Internet, including for m-learning. The vast number of mobile device users in Indonesia has given a significant opportunity for its use as a medium of learning. Baggaley [14] informed the possibility that Asia countries like Indonesia will be the pioneer in the adoption of mobile learning. Therefore, studies about m-learning adoption are important, including in higher education. m-learning has helped Indonesian students to improve the accessibility, flexibility, and interactivity in learning [15]. For Indonesian students, m-learning provides opportunities for personalized, customized, and context-aware learning support services [15].

C. TECHNOLOGY ADOPTION AND ACCEPTANCE THEORIES
Successful technology integration, especially information and communication technology, is suggested to be very important. Theories and models investigating successful technology integration are widely defined. It could be categorized in some ways. Even though the success of an in-depth understanding of various areas in technology acceptance has been dominating current research in technology integration, the hope for more predictability of the successful technology integration in different cultures and contexts should be promoted. In promoting technology acceptance, some models were established to examine users' attitude and intention to integrate new technologies. These models are, among others; the Technology Acceptance Model (TAM) by [16], Theory of Planned Behavior (TPB) by [17], and Unified Theory of Acceptance and Use of Technology (UTAUT) by [18]. From those models, TAM, which is the extension of the Theory of Reasoned Action (TRA) [19], has been a major model in comprehending the predictors of behavior towards technology and its integration acceptance.

D. TECHNOLOGY ACCEPTANCE MODEL (TAM)
TAM has been widely used as a framework to investigate technology integration in education, even though it was originally used to explain technology integration in economics [20]. Even though TAM had limited predictive power and practical value [21], it has been reported to become a beneficial framework in elaborating the behavioral intention to use a management system that involves the Internet as educational media [22]. TAM has obtained an important eminence because of its transferability to different settings and samples. It has also explained various uses of digital technology in education [23], [24]. The simplicity of structural equation modeling is also one of the TAM benefits [25]. Several studies have utilized TAM to explore technology integration in education [23], [24], [26]- [29]. In addition, some studies have facilitated the understanding that TAM can be the framework that may predict behavioral intention to use m-learning [5], [30]- [34]. The success of TAM in supporting studies to predict behavioral intention to use m-LMS in learning among PSETs encourages the establishment of this study's proposed model ( Figure 1). Three main drivers of TAM used in this study were extended by three external drivers. The three TAM drivers (Davis, 1989) used in this study are (1) perceived ease of use, (2) perceived usefulness, and (3) attitudes that could be significant in predicting behavioral intention to use m-LMS (BI). In addition to the TAM's main drivers, three additional drivers were included (1) subjective norms, (2) self-efficacy, and (3) supporting condition.

E. SUPPORTING CONDITION (SC)
SC was used as a degree to which PSETs believe that their institution would support the use of m-LMS [18]. SC can be infrastructure availability, technical support, and professional development opportunities, as well as policies to promote the use of m-LMS. In research focusing on students' use of technology, a positive correlation between facilitating condition or supporting condition and PEU was reported [24].
Similarly, in predicting PU, SC was also reported to be significant [35].
H1: SC is significantly predicting PU H2: SC is significantly predicting PEU

F. SELF-EFFICACY (SE)
Adapting Bandura's [36] self-efficacy theory, SE in this study is defined as PSETs belief in their ability to use m-LMS. A study by [37] found that SE was a significant antecedent predicting PU and PEU regarding behavioral intention to use the Internet in higher education in India. H3: SC is significantly predicting PU H4: SC is significantly predicting PEU

G. PERCEIVED EASE OF USE (PEU)
In the context of the current study, PEU indicates the degree to which PSETs believe the use of m-LMS would be effort-free. Joo [4] reported that when m-LMS was perceived ease to use while [5] informed that no correlation emerged between PEU and PU regarding the use of m-LMS in their study. Besides, PEU was reported to be significant in predicting AT [5]. H5: PEU is significantly predicting PU H6: PEU is significantly predicting AT

H. SUBJECTIVE NORMS (SN)
SN, which is one of the theories of planned behavior variables developed by [17], is defined as PSETs beliefs that their important people would support the use of m-LMS. Findings from previous studies informed that SN has a significant relationship with PU of Web 2.0 use [24] and behavioral intention to use m-learning [38]. H7: SN is significantly predicting PU H8: SN is significantly predicting BI.

I. PERCEIVED USEFULNESS (PU)
Within this study context, PU is defined as the level to which PSETs believe the use of m-LMS would improve their learning. Some studies with TAM as their framework reported that PU was one of the key drivers predicting AT and behavioral intention to use m-learning [4], [5], [39]. H9: PU is significantly predicting AT H10: PU is significantly predicting BI.

J. ATTITUDES (AT)
AT in this study refers to the PSETs positive or negative feelings towards the use of m-LMS in learning. AT to the use of m-LMS has been significant [5], [30], [39]. Therefore, AT is also considered as a significant predicting driver for behavioral intention to use m-learning. H11. AT is significantly predicting BI.

III. METHOD
This study was a survey conducted after a-semester use of m-LMS in two Indonesian universities. We took the survey as the evaluation process of the m-LMS use and the intention to use m-LMS. The instrument was adapted based on previous related studies [16], [30], [39]. Besides, the validation was done through content validity. We piloted the instrument for its reliability. The main data were computed through PLS-SEM; measurement and structural model.

A. m-LMS IN THIS STUDY
In this study, Moodle and Edmodo were the two applications of m-LMS used by Pre-Service English Teachers from two Indonesian universities. The teacher educators were asked to use these two applications during their teaching activities. They assigned the PSETs to learn with the applications throughout one semester in 2019. Besides, the submission of assignment and tasks were also conducted through these tools. At the end of the semester, the PSETs were asked to fill in the validated survey instrument. An extended TAM was attributed to the instrument. In this study context, the implementation of m-LMS was addressed for English teaching and learning.

B. INSTRUMENTATION
The questionnaire of the current study includes two parts; 1) demographic information items and 2) extended TAMbased items. The extended TAM-based items refer to the original TAM scales, which first established by [16] as well as other related studies [30], [39]. Twenty-five items were adapted from this process. Three educational technology experts were invited to get involved in discussions to assess the instrument to fit the Indonesian context and setting. It was part of the content validity process [40]. Four items were dropped based on the advice of the experts; the items are redundant or not in line with the Indonesian context. Afterward, we translated the instrument from English to Indonesian, involving two translation experts as part of back translation strategies [41]. We sent the instrument to experts for Content Validity Index (CVI) [42]. The experts' majors are educational technology and policy. Fifteen experts were contacted; five of them rejected our request. The experts were requested to rate the items' relevance and clarity. The attributes of the items were rated on a 4-point scale (1 = not relevant/ not clear/ to 4 = very relevant/ very clear [43]. In analyzing the output, we measured the item level-CVI and scale level-CVI. The item-CVI was evaluated by facilitating a score of 3 or 4 (positive responses) divided by the total number of experts. With ten experts, the item-CVI must not be lower than 0.780 [44]. In scoring the scale-CVI, the average portion of the items on one scale rated 3 or 4 were calculated. The threshold value of scale -CVI is 0.800. The items-CVI values and scale-CVI were above the threshold values.
We piloted the instrument to seventy-six respondents and reported the reliability through Cronbach's alpha; all values exceeded .700 as the threshold for alpha [45]. As a result, twenty-one indicators were included in the main data collection. A 7-point Likert scale from 1 (strongly disagree) to 7 (strongly agree) was applied.

C. DATA COLLECTION AND PREPARATION
We distributed the instrument after we implemented the use of m-LMS in English teaching for PSETs. The evaluation aimed at predicting their intention to use m-LMS in their future learning. In determining the sample, we attributed * G power to test multiple linear regression [46]. * G power is free statistical software to explore and address issues relating to sample size [46]. For six predicting factors involved in this study, the computing process in the * G power informed 146 minimal numbers of samples. However, we managed to process 210 responses out of 250 distributed questionnaires giving a response rate of 84% (Table 1). Forty responses were eliminated because they were partially completed.
In preparing the data, we converted them to Microsoft excel. Outliers and missing data were revised [47]. The data preparation was conducted for the completeness, accuracy of the data, as well as the assurance that the data had no issues with outliers, missing values, non-normal distributions, and/or errors inputting the data [48]. Outliers were identified by a box plot for all items. The amount of missing data in the current research ranged from 0 to 0.5% each item. Missing data were completely random [47]. For the univariate normality of a variable in a measurement model for a latent variable, the skewness and kurtosis values for each item should be in the range of −1.96 to +1.96 at 0.05 significance level. The data were normally distributed ( Table 2) and included in the analysis tool, SmartPLS 3.2.8.

A. MEASUREMENT MODEL
We assessed the measurement model of the proposed model to examine the reliability and validity of the variable measures in the main data collection stage. We reported the reflective indicator loadings, internal consistency reliability, convergent validity, and discriminant validity [45].

B. REFLECTIVE INDICATOR LOADINGS
Using PLS-SEM, the reflective indicator loadings were informed for all seven variables. From the result of the process, most items in all variables meet the threshold value of loading (>0.708) [45]. However, one value (SC1) was dropped due to its low loading (0.680).

C. INTERNAL CONSISTENCY RELIABILITY
Internal consistency reliability was used to evaluate the result consistency among the instrument items [45]. Cronbach's alpha and Composite Reliability (CR) are suggested to be reported [45]. Cronbach's alpha and CR values should be between 0.700 and 0.950 [45]. Table 2 informs the detail of Cronbach's alpha and CR values. The Cronbach's alpha and the CR for most variables have appropriate internal consistency reliability values. However, one variable (SE) had a value of above 0.950. We deleted one item (SE2) to solve the issue because it has a similar pattern of a statement or semantic redundancy with SE1.

D. CONVERGENT AND DISCRIMINANT VALIDITY
We reported the Average Variance Extracted (AVE) for the evaluation of the model's convergent validity [43], [45]. AVE values are recommended to be higher than 0.500, which informs 50% or more of the variable item variance. In calculating the AVE values, we utilized PLS-SEM algorithm in the SmartPLS. The AVE values of all variables exceed 0.500 (Table 3).
According to [45], discriminant validity is ''the extent to which a variable is empirically distinct from other variables in the structural model.'' To evaluate the discriminant validity, we first reported the Fornell-Larcker criterion; the shared variance for variables should not be higher than their AVEs (Fornell & Larcker, 1981). Table 4 informs that the AVE of all variables in this study is higher than their shared variance. Discriminant validity can also be appropriate when a loading value on a variable is greater than that of all loadings of its other variables' cross-loading value. Table 5 performs that the outer loadings (in bold) for every variable was higher than its other variables' cross-loadings. From the criterion of Fornell-Larcker and cross-loading values, it is summarized that the discriminant validity was established.

E. STRUCTURAL MODEL
The structural model stage was begun with the examination of collinearity issues. Afterward, the structural model relationship was reported through path coefficients (β), t values, and p values. The coefficient of determination (R 2 ), the effect size (f 2 ), predictive relevance (Q 2 ) were also assessed [45].

F. COLLINEARITY
The Variance Inflation Factor (VIF) is utilized to assess the collinearity. When the values of VIF are greater, the collinearity level will be higher. The values (≥5) have an indication that collinearity issues emerged among the variables. All VIF values are below 5. Therefore, collinearity has not been an issue within this study (Table 6).

G. STRUCTURAL MODEL RELATIONSHIP
For the relationship significance, a bootstrapping process was done with 5,000 sub-samples. With 5% significance level, most variable are significantly related. The strongest significant relationship emerges between PU and AT (β = 0.515; t = 5.787) while the weakest significant relationship is between SE and PU (β = 0.199; t = 3.119). Two paths are not significantly related; SC −> PU (β = 0.096; t = 1.156) and SN −> USE (β = 0.079; t = 1.065). Table 7 and Figure 2 show the complete result of the bootstrapping.

H. COEFFICIENT OF DETERMINATION (R 2 )
R 2 is the value measuring the model's accuracy. A variance measured by R 2 is explained in each dependent variable, the model's explanatory power measures [45], [49]. R 2 values are in between 0 and 1, a higher value have an indication for a greater level of predictive accuracy (0.750 = substantial, 0.500 = moderate, and 0.250 = weak) [45]. Table 8 presents a good level of R 2 . All values are above 0.500 (moderate);  the strongest R 2 is BI (0.722), while the least strong is AT (0.635).

I. f 2 EFFECT SIZE
The effect sizes (f 2 ) measures a variable driving impact on a dependent variable [49]. The f 2 assesses the change in the values of R 2 when an independent variable is removed from the model [50]. This is to measure the real impact of an independent variable on the dependent variable with the guideline values (0.020 = small, 0.150 = medium, 0.350 = large) [45]. Table 8 exhibits f 2 effect size. All drivers have effects on endogenous variables except for SC− > PU and SN− > PU (Table 9). VOLUME 8, 2020

J. ASSESSING PREDICTIVE RELEVANCE, Q 2
We involved the assessment of predictive relevance (Q 2 ) of the model [51]. When the model informs appropriate Q 2 values, the prediction for the indicators' points would be accurate [45]. Q2 value (>.0) indicates that the model's predictive relevance for the variable is achieved (0.020 = small; 0.150 = medium 0.350 = large) [45]. We did a blindfolding process in the SmartPLS tool to understand the Q 2 values of  the dependent variables. Table 10 performs Q 2 values for the variables; they are above 0. The results provide an insight that the model's predictive relevance of all dependent variables is supported.

V. DISCUSSION
Studies have investigated the use of m-learning in developed countries [7], [10], [11]. However, few studies were done in developing countries. To fill the gap, more reports regarding m-learning adoption in developing countries like Indonesia should always be promoted. Thus, this study was conducted in an attempt to fill the gap, identifying factors affecting Indonesian PSETs behavioral intention to use m-Learning Management System (m-LMS) in their learning activities.
Through the assessment of the measurement model, we assessed the validity and reliability of the proposed model. Through this process, it was reported that the model is valid and reliable. Further, the data were computed in the Smart-PLS to examine the structural model, coefficient determination, and effect size, as well as predictive relevance. These four examination processes were previously suggested for studies that utilized PLS-SEM as a statistical approach [45].
According to the results of the computation, nine out of eleven hypotheses were supported. PEU was significantly predicted by SC (β = 0.502; p < .001). Similarly, it was informed that SE influence PEU (β = 0.362; p < .001) [16], [24]. In this study context, supporting infrastructures, training, and technical support can improve PSETs perceived ease of use. Besides, PEU was also significantly influenced by SE; the more PSETs believe in their ability to use m-LMS during their learning process in Indonesian universities, the better their PEU. A similar result was informed that revealed that SE is a strong antecedent for PEU [37].
On the other hand, PU was not significantly correlated with SC (β = 0.096; p = 0.248). This result might refer that the infrastructure is not related to PU regarding the increased performance offered by infrastructure availability. The result is not in line with a previous finding [35] that informed SC as a strong driver for PU among pre-service teacher's acceptance use of technology. However, SE was reported to be one of the significant drivers for PU (β = 0.199; p < .005). A similar result regarding behavioral intention to use the Internet in Indian universities was also reported [37]. When PSETs has high SE, the PU will be more increased. The personal belief of the ability to use the tool triggers the increase of benefits of the use of the m-LMS. PU also has a significant relationship with SN (β = 0.289; p < .001). As a country that supports the culture of respecting other people, this result might represent other peoples' influences that have a significant role in influencing the PSETs' PU. Further, PEU was also reported significant predicting PU [24], which is similar to the result of this study. A plausible reason might be that PSETs' PEU is a significant factor in improving their academic performance in the process of using m-LMS. They are also advanced in using mobile devices. The part of these findings can be a guideline for the government to improve the perceived benefits of the use of m-LMS.
The structural model reported PEU (β = 0.326; p < .005) and PU (β = 0.515; p < .001) have a significant correlation with AT; both variables have medium f 2 effects on AT. The results are consistent with prior researches [5], [30], [39]. The fact that attitudes towards the BI to use m-LMS were driven by PU and PEU among PSETs indicates that the easiness of use and the benefits of m-LMS support the level of the respondents AT.
Finally, the result revealed that PU has a significant positive influence on BI to use m-LMS (β = 0.515; p < .001). It is consistent with the previous studies on technology adoption [24] and in the context of m-LMS [5]. The correlation between AT and BI is weaker (β = 0.381; p < .001). This might happened because PSETs were more dependent on their laptop-LMS version rather than that of m-LMS. However, SN is reported to be insignificant impact on BI (β = 0.079; p = 0.287). Community-based cultures of Indonesian people seem to become an insignificant major effect on BI. The finding contrasts previous related studies reporting that SN has a significant role in predicting BI [24], [38].
From the academic perspective, this study can be beneficial because it extends the statistical understanding of the behavioral intention to use m-learning in a developing country. The study has a specific contribution to the elaboration of the adoption of m-LMS, which enhances current literature reviews on technology integration and acceptance, especially in the context of developing countries. The expansion of technology acceptance model elaboration that is reported within this study should be beneficial for future researchers with similar interests. It reports the validity and reliability of the survey instrument as well as the robustness and explanatory power of the hypothesized extended TAM model in predicting behavioral intention to use m-LMS in the context of Indonesian higher education.
The findings of the study would help people understand how Indonesian student teachers have interaction and value regarding the behavioral intention to use m-LMS during their learning that will affect their ways of teaching in the future. The findings also offer information about the design and implementation of m-LMS for educational activities. The relationships between variables proposed by this study could be a guideline and very beneficial for student teachers to use m-LMS in their learning. The higher institution supports, such as the availability of infrastructure, tools, and human resources, are significant to support the m-LMS. Therefore, the implications offered by this study are expected to go beyond the report of the validation of the structural model.

VI. CONCLUSION, LIMITATIONS, RECOMMENDATIONS, AND FUTURE WORK
This study was done to understand Indonesian pre-service English teachers' intention to use m-LMS in learning activities. A proposed model based on extended TAM with eleven research hypotheses was validated, tested, and reported with a sample of 210 respondents from two universities. PEU, PU, and AT as three core variables of TAM, the proposed model contains SE, SC, and SN as external variables. We adapted variables from previous related studies; the final findings support nine out of eleven hypotheses.
All relationships between TAM core variables are significant. The results were consistent with prior related studies. From the external variables, most relationships are also significant, but between SC and PU, as well as SN and BI. As far as we know, this empirical study is among the pilot studies on the adoption of m-LMS in the Indonesian higher education context, especially in teacher education programs. Therefore, the findings could help future researchers and practitioners from developing countries as a guideline to conduct their future project.
For future research projects, some limitations of this study should also be considered. We are aware that the study only tested the proposed model and hypotheses with Indonesian pre-service English teachers as respondents. Thus, more diverse respondents are needed for a similar study with different contexts and settings. This study is also limited because the sample is relatively small. The research model constructed for this study included only the acceptance model of technology. Some demographic information could be included in future research. Different study approaches, e.g., observation and interview, are also encouraged.