TEACHER AUTONOMY SUPPORT FOR YOUNG BEGINNERS’ 
APP-BASED LANGUAGE LEARNING BEYOND THE CLASSROOM: 
A SELF-DETERMINATION THEORY PERSPECTIVE 

By 

Hyun-Bin Hwang 

A DISSERTATION 

Submitted to 
Michigan State University 
in partial fulfillment of the requirements   
for the degree of 

Second Language Studies – Doctor of Philosophy 

2025 

   
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ABSTRACT 

With the importance of second language (L2) learning beyond the classroom (e.g., Reinders et 

al., 2022), language learning apps (e.g., Mango Languages, Duolingo) have shown potential for 

promoting self-directed L2 learning. Despite research indicating a positive relationship between 

the extent of in-app activities and L2 proficiency gains (e.g., Loewen et al., 2020), high attrition 

rates pose a significant challenge to this learning method in self-study contexts (Hwang et al., 

2024). Thus, teacher support is crucial for developing learners’ agency in their independent use 

of technology out of class (Godwin-Jones, 2019). 

In self-determination theory (SDT, Ryan & Deci, 2017; Noels et al. 2019a in L2 

contexts), autonomy—a fundamental human psychological need—refers to a sense of volition 

and self-endorsement in one’s action. When this need is satisfied, learners take ownership of 

their own learning and engage in activities out of interest and enjoyment. This, in turn, leads to 

greater learning success and well-being. In this dissertation, I explored how teachers can support 

L2 learners’ autonomy need for app-based language learning out of class, thereby influencing 

their app engagement, app usage, and L2 learning. Particularly, the study examined the 

moderating role of learners’ initial motivation for English learning in this process. 

Additionally, recognizing the potential influence of socio-ecological structures, I 

investigated the implementation of app-based language learning in South Korea’s distinctive 

educational context. In South Korea, the high-stakes nature of English tests often leads learners 

to rely on hagwons—for-profit, private educational institutions. Within this landscape, language 

learning apps are one of many resources available for English learning. Building on this, I 

explored whether learners’ perceived opportunity cost of using language learning apps (i.e., the 

sense of sacrificing other valued learning activities to use apps) affects their out-of-class app 

   
usage, and how teacher autonomy support can help mitigate this perceived opportunity cost. 

Participants were seventh-grade beginner learners (N = 258) in South Korea. While 

learners independently used a commercially available language learning app over 13 weeks, their 

teachers provided Reeve and Cheon’s (2021) autonomy-supportive instructional behaviors to 

enhance out-of-class app usage. Using questionnaires, I measured learners’ (a) initial L2 

motivation, (b) perceived teacher autonomy support, (c) autonomy need satisfaction, (d) app 

engagement (behavioral, emotional, cognitive, and agentic dimensions), and (e) perceived 

opportunity cost. Additionally, in-app usage and learning gain data were collected. 

Using structural equation modeling, I conducted mediation and moderated mediation 

analyses, revealing three key findings: (1) greater app usage positively predicted vocabulary 

learning gains; (2) teacher autonomy support indirectly increased app usage by enhancing 

autonomy need satisfaction and reducing perceived opportunity cost; and (3) teacher autonomy 

support was effective even for learners with controlled L2 motivation. 

These findings suggest that classroom-based autonomy support encourages L2 learners to 

use technology beyond the classroom and builds their resilience against disengagement from it, 

enabling sustained self-directed L2 learning. Pedagogical implications are discussed regarding 

the importance of creating a structured learning climate to ensure consistent and reliable teacher 

autonomy support. 

   
 
This dissertation is dedicated to my family. 
Thank you for always believing in me. 

  iv 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ACKNOWLEDGEMENTS 

On August 16th, 2021, we arrived at a small airport in Lansing, Michigan. It is strange how I can 

vividly recall that day yet struggle to remember recent events. We were burdened with large 

suitcases, and after a 15-hour flight, my 2-year-old son was whining, exhausted, and asleep on 

his mother’s back. Jealously watching others being greeted by familiar faces, we felt utterly 

alone, as we were the only people we knew in this foreign country. We took an Uber, driving 

past the sign for 1855 family housing, and entered our new home. An unfamiliar scent greeted us 

as we opened the door. Bo-ram and I started unpacking and cleaning, while Ha-jun explored his 

new surroundings with curiosity. Lying in bed that night, we struggled to fall asleep amid the 

noisy train sounds coming from nearby and the discomfort of being in an unfamiliar place. The 

overnight change in everything felt surreal, almost dreamlike, leaving us full of nervous 

excitement and wonder about what the coming days would bring. 

Over the last four years, I’ve come to acknowledge that not every day is filled with 

happiness and joy. There have been stressful struggles, constant efforts to improve, and moments 

of wondering when I would finally reach the end of this long tunnel. However, looking back on 

our time here, my family has made wonderful friends who have always been kind to us and stood 

by us through both good times and bad. It’s clear that we have been able to live safely and thrive 

without major concerns over the past few years. The good memories we’ve made with them have 

been a true blessing in our lives and will stay with us forever. We are no longer alone. 

To celebrate the completion of this significant milestone, I want to express my sincere 

gratitude for the help, care, guidance, and encouragement I have received from those around me. 

First and foremost, I am deeply grateful to my family for their unconditional love throughout this 

journey. Bo-ram and Ha-jun moved to the U.S. with me, leaving behind their lives in Korea to 

  v 

support my dream. They have been wonderful companions along the way, and I deeply 

appreciate their sacrifices. My parents also deserve special recognition for their unwavering love 

and support, which have laid the foundation for my successes and shaped me into the person I 

am today. Their presence in my life has been a constant source of strength, inspiration, and 

perseverance. I am proud to be their son. 

I consider myself incredibly fortunate to have shared my academic journey with my Fall 

2021 cohort, the Three Idiots—Matt and Vashti. These comrades have been a tremendous 

blessing, providing not only intellectual engagement but also deep personal connections that 

have enriched both my life and my family’s. Our friendship, humor, and support, along with 

unforgettable cohort meals, made my PhD experience all the more enjoyable. 

Professionally, I have been profoundly influenced by the dedication and support of my 

dissertation committee. My advisor, Dr. Shawn Loewen, has been an incredible mentor and 

unwavering supporter in all circumstances. His commitment goes beyond academic guidance; he 

has been a steadfast source of support and advocacy during difficult times, enabling me to ‘keep 

calm and carry on.’ I am also immensely indebted to Drs. Charlene Polio, Paula Winke, Aline 

Godfroid, and Yuichi Suzuki, whose insights and encouragement have not only strengthened my 

research, including this dissertation, but have also played a pivotal role in shaping my expertise 

and perspectives as a junior researcher. 

I feel truly blessed to be part of the SLS community and to cherish the memories I’ve 

shared with Ryo, Curtis, Robert, Kiyo, Dylan, Kevin, Brittany, Amr, Yingzhao, Ayşen, Carlo, 

Steve, Adam, Philip, and Won-Young. They have generously shared their ‘secret sauce’ for 

thriving in the program, offering invaluable survival skills that have enriched my well-being, 

productivity, and sense of belonging. They are not just friends but also mentors, making my 

  vi 

academic journey both enjoyable and fulfilling. Another highlight of my time in Wells Hall has 

been my interactions with Dr. Fred Poole. He is an incredibly kind, smart, down-to-earth, and 

humorous person, and I truly love our water-cooler conversations. He has introduced me to the 

world of technology and has been an endless source of new research ideas and creativity. 

Outside the SLS program, I am especially grateful to Kaitlyn Tagarelli for her 

tremendous support and feedback on my research and for generously allowing me to use Mango 

Languages in multiple projects. Working with her has provided me with invaluable hands-on 

experience and insights into the synergy between research and industry collaboration. I am also 

deeply thankful to my friend Rob Mucklo, who has always welcomed me with open arms 

whenever I visit Pittsburgh, making the city feel like a second home in this country. 

Last but not least, I deeply appreciate the faculty at Korea National University of 

Education, where this journey first began. It was there that my foundational knowledge, skills, 

mindset, and identity as a teacher were nurtured. In particular, Dr. Sang-Ki Lee has had a 

profound impact on my growth, inspiring and encouraging me to aim higher in my academic 

pursuits. 

This dissertation would not have been possible without the generous financial support of 

the Mango Languages Dissertation Award, the Language Learning Dissertation Grant, the 

NFMLTA/MLJ Dissertation Writing Support Grant, and the MSU graduate school Dissertation 

Completion Fellowship. I am also deeply grateful to the Korean-American Educational 

Commission for the Fulbright scholarship and to the Sejong City Office of Education for 

supporting my academic pursuits. Their endorsement has enabled me to fully dedicate myself to 

research and make meaningful contributions to the field. 

 vii 

 
 
TABLE OF CONTENTS 

INTRODUCTION .......................................................................................................................... 1 

BACKGROUND LITERATURE ................................................................................................... 2 

THE PRESENT STUDY .............................................................................................................. 14 

METHOD ..................................................................................................................................... 19 

RESULTS ..................................................................................................................................... 38 

DISCUSSION ............................................................................................................................... 90 

CONCLUSION ........................................................................................................................... 103 

REFERENCES ........................................................................................................................... 104 

APPENDIX ................................................................................................................................. 111 

 viii 

INTRODUCTION 

With a growing emphasis on second language (L2) learning beyond the classroom (Reinders et 

al., 2022), mobile technology has emerged as a valuable tool in facilitating L2 learning in self-

study contexts by providing flexible and diverse learning opportunities. In particular, language 

learning apps (e.g., Mango Languages, Duolingo) are specifically designed as self-study 

resources and incorporate curricular components (e.g., lessons, feedback, assessment) to simulate 

classroom instruction in out-of-class contexts. Unlike general content platforms (e.g., YouTube) 

and auxiliary tools (e.g., online dictionaries), language learning apps are expected to offer a more 

structured approach to L2 learning beyond the classroom. 

Empirical studies demonstrate the benefits of app-based language learning on L2 

proficiency (e.g., Bang et al., 2024; García Botero et al., 2021; Jiang et al., 2021, 2024; Kessler 

et al., 2023; Loewen et al., 2019, 2020; Smith et al., 2024; Sudina & Plonsky, 2024). However, 

high attrition rates in self-study contexts pose significant challenges, as only a small subset of 

learners persist long enough to achieve meaningful gains (Hwang et al., 2024). Addressing these 

issues requires understanding how initial affective and linguistic traits (e.g., L2 motivation, L2 

proficiency) and educational contexts influence app usage as well as exploring how teachers can 

systematically support learners’ self-directed L2 learning through apps. Relatedly, learners’ 

agency is crucial for using digital resources effectively beyond the classroom (Godwin-Jones, 

2019), yet there is little consensus on how learner autonomy is conceptualized (Chong & 

Reinders, 2022). In this regard, using self-determination theory (SDT; Ryan & Deci, 2017) as a 

theoretical framework, I explored how teachers can support learners’ autonomy need for app-

based language learning out of class, thereby influencing their app engagement, app usage, and 

L2 learning. 

  1 

BACKGROUND LITERATURE 

Autonomy Support for L2 Learning Beyond the Classroom 

Self-determination theory (SDT) is an integrative approach to human motivation and personality 

(Ryan & Deci, 2017, 2020). As a macro-theory, SDT integrates its six mini-theories, each of 

which uniquely addresses specific research questions about the nature of human motivation and 

how social conditions affect it: (a) basic psychological needs theory, (b) cognitive evaluation 

theory, (c) causality orientations theory, (d) organismic integration theory, (e) goal contents 

theory, and (d) relationships motivation theory. What unites these mini-theories into a single 

macro-theory is shared assumptions for humans’ inherent nature for personal growth and 

proactive interaction with the environment to achieve their flourishing. 

This dissertation focuses on one of the six mini-theories of SDT: basic psychological 

needs theory (Ryan & Deci, 2001, 2017; Ryan et al., 2008). The main question of this theory is 

to understand how need satisfaction or frustration leads to effective functioning and well-being 

or, conversely, to maladaptive functioning and ill-being. Basic psychological needs theory 

identifies the three universal psychological needs for (a) autonomy, (b) relatedness, and (c) 

competence as inherent inner motivational resources. Autonomy concerns a sense of personal 

ownership during one’s action. Relatedness concerns a sense of belonging and emotional 

connection with others. Competence concerns a sense that one can succeed and grow in one’s 

skills and capacities. A key theoretical implication of basic psychological needs theory is that the 

satisfaction of these basic psychological needs is fundamental for optimal functioning and 

healthy development of individuals, while their frustration undermines these outcomes, 

manifesting as passivity, defiance, and ill-being (Ryan & Deci, 2017).  

  2 

 
In L2 contexts, Noels et al. (2019a) proposed a model for the L2 learning motivational 

process based on SDT. In their model, as demonstrated in Figure 1, socially important others 

(e.g., family, teachers, L2 community) support learners’ psychological needs and shape their L2 

motivational orientation, which influences the manner and intensity of L2 engagement, 

(non-)linguistic outcomes, and psychological wellness. With empirical evidence substantiating 

this motivational process (e.g., Alamer & Lee, 2019; McEown et al., 2014; Noels et al., 2019b; 

Oga-Baldwin & Nakata, 2017, 2020; Oga-Baldwin et al., 2017), basic psychological needs are 

considered as inherent inner motivational resources, whose satisfaction is the essential nutriment 

for learning and growth (see Al-Hoorie et al., 2022; McEown & Oga-Baldwin, 2019, for the 

overview of possible applications of SDT to L2 learning research and pedagogy). 

Figure 1 Noels et al.’s (2019a) language learning motivational process (p. 100). Reprinted with 
permission from the author. 

  3 

 
While each psychological need uniquely contributes to human motivation, the focus has 

primarily been on autonomy need because it is seen as an anchor construct that promotes the 

satisfaction of the other needs (Reeve, 2022a). Given the importance of autonomy need 

satisfaction, SDT researchers in educational contexts have explored diverse autonomy-supportive 

strategies such as classroom climates and educational practice (Ryan & Deci, 2020). 

Specifically, when autonomy need is satisfied, learners experience a sense of volition and self-

endorsement in initiating and regulating their learning. This sense of ownership drives them to 

get involved in activities out of interest and enjoyment (i.e., intrinsic goal pursuit), further 

creating opportunities for greater autonomy need satisfaction (Reeve, 2022b). Building on this 

theoretical assumption, Reeve and Cheon (2021) propose seven strategies for autonomy-

supportive instructional behaviors (ASIB).  

Figure 2 Reeve and Cheon’s (2021) seven autonomy-supportive instructional behaviors (p. 56). 
Reprinted with permission from the author. 

  4 

 
As demonstrated in Figure 2, the essence of autonomy-supportive teaching is adopting a learner-

centered approach (ASIB #1) with two major goals: (a) support intrinsic motivation, and (b) 

support internalization. Each goal can be achieved by relevant autonomy support behaviors. 

First, teachers may support learners’ intrinsic motivation by providing them with learning 

activities that align with their personal interests (ASIB #2) and by allowing them to decide their 

learning guided by their interests and goals (ASIB #3). Second, teachers can provide four 

autonomy supports to facilitate learners’ internalization, a process of taking in values, beliefs, or 

behaviors from external sources and incorporating them into their own way of thinking and 

acting. These strategies include explaining the value and personal relevance of learning activities 

(ASIB #4), acknowledging negative feelings (ASIB #5), using invitational language (ASIB #6), 

and displaying patience to let learners work at their own pace and in their own way (ASIB #7). 

Reeve and Cheon’s (2021) systematic review provided empirical evidence highlighting the 

benefits of teacher autonomy support for educationally important outcomes (e.g., learning 

engagement, academic achievement, prosocial behavior). Overall, the findings highlight that 

teachers can effectively satisfy learners’ autonomy need through autonomy-supportive teaching 

(see Reeve & Cheon, 2021, for a comprehensive list of empirical studies exploring the effects of 

autonomy-supportive instructional behaviors and their outcomes). 

While existing research on autonomy need support primarily focuses on classroom 

settings, it holds particular significance in L2 contexts, where much of language learning occurs 

beyond the classroom (Mynard & Shelton-Strong, 2022). In self-study contexts, external 

accountability (e.g., teacher guidance, peer pressure) is often absent, requiring learners to 

manage their learning independently (Reinders & Benson, 2017). In this regard, there is growing 

scholarly attention to exploring technology as a potential tool to facilitate autonomous L2 

  5 

learning out of class. Empirical research has shown that various types of digital resources can 

effectively support learners’ autonomy need, thereby enhancing their self-directed L2 learning 

(e.g., Alamer & Al Khateeb, 2023, for a social networking app; Dincer & Işık, 2022, for digital 

resources in general; Hsu, 2023, for language massive open online courses; Jeon, 2022, for a 

speaking-practice app; Zeng & Fisher, 2024, for Duolingo’s gamification). Overall, it is 

noteworthy that these studies primarily focused on resource-based autonomy need support where 

features within technologies are the source of need-satisfying experiences (i.e., autonomy need 

satisfaction through technology usage). 

However, there is a lack of research exploring how teachers can support autonomy need 

for technology usage in self-study contexts (i.e., teacher-provided autonomy support to enhance 

out-of-class technology use). To truly fulfill learners’ autonomy need through technology usage, 

teachers first need to create autonomy-supportive environments that encourage L2 learners to 

take ownership of using technology and managing their learning behaviors and strategies beyond 

the classroom. Applying Noels et al.’s (2019a) motivational process model to technology-

mediated L2 learning contexts, I investigate how teacher autonomy support can promote a sense 

of volition and ownership of technology use for L2 learning beyond the classroom, as well as its 

effects on technology engagement, actual usage, and L2 learning outcomes. Among diverse 

digital resources available, language learning apps are the focal technology in this study, as they 

are specifically designed for self-study purposes. 

L2 Motivation as the Precondition in App-Based Learning 

Research on app-based language learning has shown that the extent of in-app activities (e.g., 

total time spent on apps, the number of lessons completed) is a positive predictor of L2 

proficiency gains (e.g., Bang et al., 2024; García Botero et al., 2021; Jiang et al., 2021, 2024; 

  6 

Kessler et al., 2023; Loewen et al., 2019, 2020; Smith et al., 2024; Sudina & Plonsky, 2024). 

However, high dropout rates are a significant challenge of this learning method, particularly in 

self-study settings, where learners often need to find other reasons to persist (Hwang et al., 

2024). For example, García Botero et al. (2019) found that among 574 university students 

introduced to a focal app, 149 reported using it, and only 12 completed a language course on the 

app over a year. Jeon (2022) showed a similar finding with a younger population. Only nine of 

179 primary school students used a focal app for more than a month during a two-month period. 

Fifty-nine students did not use the app at all. Notably, Hwang et al. (2024) highlighted that a 

significant portion of dropouts occurred in the early stage of app-based language learning 

irrespective of individuals’ dispositions toward technology. Overall, in real-world scenarios 

where learners juggle daily responsibilities, these findings suggest that only a small group of 

learners are likely to use language learning apps long enough to achieve meaningful L2 gains in 

self-study contexts. 

Methodologically, previous studies have applied eligibility criteria to include participants 

who met minimum usage thresholds (i.e., persistence-based inclusion criteria), thereby excluding 

those with lower app usage from subsequent analysis. However, this approach may introduce 

bias by favoring persistent app users whose initial affective and linguistic traits could affect the 

extent of app usage in the first place. Consequently, the effectiveness of app-based language 

learning may have been overestimated, suggesting that it is effective only if learners can ensure a 

certain level of usage—a condition that many learners in self-study contexts find challenging to 

meet. This potential bias, where individuals predisposed to persistent behaviors are 

overrepresented in data and inflate the apparent efficacy of interventions, has been documented 

in other research fields. This phenomenon is known as the healthy initiator effect or healthy 

  7 

adherer effect, a well-recognized bias in medicine (e.g., Olawore et al., 2024; Shrank et al., 

2011). It occurs when individuals who are more likely to initiate or adhere to a treatment of 

interest are initially healthier or engage in other healthy behaviors (e.g., screening tests, 

immunizations) than their non-healthy counterparts. This predisposition to healthier behaviors, 

unrelated to the intervention itself, increases the likelihood of sustained participation in medical 

research, with their data being included on adherence-based inclusion criteria (e.g., those who 

take medicine for more than three months). As a result, individuals who are initially healthier 

may be overrepresented, leading to an overestimation of the effectiveness of a medical 

intervention. Therefore, persistence-based inclusion criteria in app-based language learning 

research may conflate app effectiveness with pre-existing individual traits (e.g., L2 motivation, 

technology acceptance) that affect persistence. In this regard, this raises the question of whether 

the potential benefits of app-based language learning beyond the classroom are conditional on 

individual differences existing before using apps. 

Despite increasing attention to understanding digital resources for L2 learning from 

technology acceptance perspectives (e.g., Hwang et al., 2024), little research systematically 

considers L2-specific variables as preconditions for the benefits of app-based language learning. 

One typical example of such a variable is L2 motivation. For example, Loewen et al. (2020) 

found that learners with greater L2 motivation were likely to use the app more, which in turn, led 

to greater growth in oral proficiency. Similarly, Sudina and Plonsky (2024) reported that diverse 

in-app usage indices were positively correlated with initial L2 motivation level and L2 

proficiency gains, respectively. However, they did not find clear associations between L2 

motivation and L2 proficiency gains. To sum up, these studies suggest that L2 motivation 

existing before learners start using apps may provide an initial advantage in app-based language 

  8 

learning. This emphasizes the importance of considering learners’ initial L2 motivation to better 

understand potential causal relationships between app usage and L2 gains. 

It is noteworthy that L2 motivation was conceptualized as a single-dimensional construct 

indicated by high or low levels (e.g., I work hard at studying French/Spanish, I am interested in 

learning Spanish) in previous studies (e.g., Loewen et al., 2020; Sudina & Plonsky, 2024). In 

contrast, as demonstrated in Figure 3, the SDT framework arranges different forms of motivation 

along an autonomous-controlled continuum, reflecting their relative satisfaction of autonomy 

need for an activity (Ryan & Deci, 2017).  

Figure 3 Taxonomy of motivation on the self-determination continuum (Reeve, 2022a, p. 21). 
Reprinted with permission from the author. 

At the one end of this continuum, intrinsic motivation involves engaging in activities ‘for their 

own sake,’ while at the other end, amotivation refers to a lack of interest and value when an 

activity is perceived as personally irrelevant. In between, extrinsic motivations exist, which 

concern reasons other than inherent interest and enjoyment, and are divided into four subtypes: 

  9 

 
integration, identification, introjection, and external regulation. Each is characterized by the 

degree to which motive is seen as coming from within an individual’s sense of value for an 

activity or from externally imposed sources (e.g., rewards). Integrated and identified regulations 

are autonomous extrinsic motivations, whereas introjected and external regulations are controlled 

forms of extrinsic motivation (Noels et al., 2019a). While most intentional behaviors involve 

both autonomous and controlled motivation simultaneously (Ryan & Deci, 2020) and the unique 

role of controlled motivation in supporting autonomous motivation is acknowledged (Al-Hoorie, 

2024), a large body of empirical research has demonstrated that more autonomous types of 

motivation predict greater learning engagement and positive outcomes (Ryan & Deci, 2020; see 

Alamer & Alrabai, 2023; Alamer & Lee, 2019; Alrabai & Alamer, 2024; McEown et al., 2014; 

Noels et al., 2019b; Oga-Baldwin et al., 2017; Shirvan & Alamer, 2024, for L2 learning). 

In this study, grounded in the SDT framework, I defined L2 motivation as the extent to 

which learners are willing to take ownership of their L2 learning (i.e., levels of autonomy need 

satisfaction for L2 learning), ranging from autonomous to controlled motivation. Given the 

critical role of autonomous L2 motivation in shaping out-of-class L2 learning experiences (e.g., 

Hoang et al., 2022), I hypothesized that pre-existing L2 motivation might influence app-based 

language learning beyond the classroom, leading to varying levels of app engagement and usage. 

Opportunity Cost of App-Based Language Learning in a Test-Driven Context 

The present study was conducted in South Korea, where English is taught as a foreign language. 

English education in South Korea is predominantly test-oriented, with school English tests 

primarily assessing reading and grammar skills. While a strong emphasis is placed on achieving 

high grades to secure better educational and career opportunities, the participants in the study 

received only two 45-minute English classes per week at school, which is insufficient for test 

 10 

preparation. This limitation makes it almost mandatory for learners to study English out of class 

to supplement in-class instruction. Yet, such independent learning can be particularly 

challenging for young learners who lack effective self-study skills and strategies. As a result, 

parents and learners often turn to hagwons—for-profit private educational institutions—for 

additional support to excel in tests. In this context, language learning apps may offer a practical 

alternative to promote self-directed English learning, potentially reducing reliance on hagwons. 

As the importance of integrating dynamic real-world contexts into language teaching and 

learning is emphasized in The Douglas Fir Group (2016) and Noels et al.’s (2019a) motivation 

process model (i.e., socio-structural and socio-cultural components in Figure 1), it is essential to 

consider a specific local context, beyond the confines of research design, that may determine the 

success of app-based language learning. In this regard, learners’ perceived opportunity cost of 

using language learning apps could influence their app usage. In economics, opportunity cost 

refers to the value of the alternative that you give up when making a choice.  

Figure 4 Constant opportunity cost between goods X and Y. 

 11 

 
Figure 4 visualizes the concept of opportunity cost with two goods: Good X (e.g., books) and 

Good Y (e.g., meals). The line demonstrates the trade-offs involved in reallocating resources 

between the production of the two goods under conditions of constant opportunity cost. For 

instance, at Point A, more resources are allocated to Good Y, resulting in the production of 75 

meals, while fewer resources are allocated to Good X, producing only 25 books. Conversely, at 

Point B, the allocation shifts to prioritize Good X, resulting in the production of 75 books but 

only 25 meals. The dashed lines connecting the points to the axes visually emphasize the 

opportunity cost. Moving from Point A to Point B involves sacrificing 50 meals to produce 50 

additional books, and vice-versa. This trade-off highlights the fundamental principle of 

opportunity cost, which underscores the efficient use of limited resources such as time, energy, 

or money, and the necessity of making choices when resources are scarce. 

In educational contexts, the concept of opportunity cost has been explored within the 

situated expectancy-value model of motivation (Eccles & Wigfield, 2020; Gladstone et al., 

2022). This model includes two main components: expectancy, which examines one’s perceived 

ability to perform an activity, and value, which evaluates whether one finds the activity 

worthwhile. Within this framework, ‘cost’ is a subcomponent of value, including components 

such as the perceived amount of effort required to complete a task, and the extent to which 

engaging in one activity limits the ability to participate in other valued tasks (i.e., loss of 

resources). Empirically, higher perceived costs associated with an activity can lower the overall 

value of the activity, leading to a decline in motivation and an increase in avoidance of the 

activity (e.g., Jiang et al., 2018; Robinson et al., 2019). Recent research has elevated cost to an 

independent component to propose the expectancy-value-cost model (Barron & Hulleman, 2015; 

Kosovich et al., 2015). While SDT does not explicitly address the concept of cost, it highlights 

 12 

the role of teacher autonomy support in developing an initial sense of value and personal 

relevance for a task to facilitate the internalization process (i.e., ASIBs #4 to #7) by explaining 

its utility value aligned with learners’ perspective, rather than teacher’s (Reeve & Cheon, 2021). 

Therefore, I relied on the expectancy-value(-cost) model to explore the perceived opportunity 

cost of app-based language learning. 

In the context of this study, Korean learners typically face tightly packed extracurricular 

schedules, including various (non-)academic subjects. With limited available time and energy, 

every additional activity must justify its value in terms of opportunity cost. As such, learners 

must make deliberate choices to prioritize activities that maximize academic performance or 

personal enjoyment, reflecting the trade-offs between competing options. In this respect, 

language learning apps represent one of many available tools for improving English skills, and 

these apps must compete with well-established non-app learning resources (e.g., hagwons), 

which are integral to learners’ routines and offer a proven return on investment. For language 

learning apps to be meaningfully used out of class, their perceived benefits must outweigh the 

opportunity cost of time and effort diverted from other alternative non-app learning methods 

through a cost-benefit analysis. Based on this understanding, I hypothesize that a high perceived 

opportunity cost associated with language learning apps will negatively influence learners’ app 

usage, whereas a lower perceived opportunity cost will lead to greater app usage. Furthermore, I 

aim to explore the extent to which teacher autonomy support could alleviate such perceived 

opportunity cost, potentially promoting greater app engagement and usage out of class. 

 13 

 
 
Mediation and Moderated Mediation 

THE PRESENT STUDY 

Figure 5(a) demonstrates a mediation relationship. With variables X, Y, and M, X has a direct 

effect on Y (i.e., X → Y). When the effect of X on Y is transmitted through an intervening 

variable M (i.e., X → M → Y), M mediates the relationship between X and Y. In this case, M 

serves as both an outcome of X and a cause of Y, with its effect on Y being conditional on X. For 

example, if M positively affects Y (i.e., M → Y), X indirectly affects Y by increasing M (i.e., X → 

M). When X has both direct and indirect effects on Y, this is known as partial mediation. If X 

affects Y only through M, it is full mediation. 

For example, in the present study, perceived teacher autonomy support (X) may directly 

influence out-of-class app usage (Y). Additionally, this relationship may involve an indirect 

effect of perceived teacher autonomy support through autonomy need satisfaction (M). While 

higher autonomy need satisfaction can directly lead to greater app usage (i.e., the direct effect of 

M on Y), it is perceived teacher autonomy support that determines the levels of autonomy need 

satisfaction from the outset. In other words, the extent to which perceived teacher autonomy 

support influences autonomy need satisfaction (i.e., X → M ) indirectly shapes the strength of the 

relationship between autonomy need satisfaction and app usage (i.e., M → Y). This reflects a 

mediation effect where the impact of X on Y is partially or fully transmitted through M. 

However, one key assumption of mediation analysis is that there are no omitted 

confounders for any pair of variables among X, Y, and M. This means that the relationship 

between X and M should not be affected by any other unobserved variables. Omitted 

confounders (cf. healthy initiator and adherer effects) can introduce bias into the estimated 

 14 

 
mediation effect, making it difficult to accurately assess the role of M as a mediator between X 

and Y. Violating this assumption could mean that the observed relationships could be spurious. 

If indirect effects are driven by unmeasured variables, the use of the term ‘mediation’ is 

not warranted. In this case, both mediation and moderation can be estimated in the same analysis. 

Figure 5(b) demonstrates moderated mediation, where X directly affects Y, and W moderates the 

indirect effect of X on Y through M. When the indirect effect of X on Y through M depends on a 

moderator W, this reflects how the strength or direction of causal mechanisms depend on 

individual differences or situation (Kline, 2023). In a moderated mediation model, X and W 

interact in their effects on M in addition to their direct effects, indicating that the direction or 

strength of the relationship between X and M may vary over the levels of W. 

In the earlier example, learners’ initial L2 motivation may serve as a moderator (W). 

Specifically, learners with more autonomous L2 motivation may respond more meaningfully to 

teacher autonomy support than those with more controlled L2 motivation. Consequently, for the 

autonomous L2 motivation group, greater autonomy need satisfaction could be achieved with the 

same level of perceived teacher autonomy support compared to the controlled L2 motivation 

group.  

M

(mediator)

W

(moderator)

M

(mediator)

X

Y

X

Y

(a) Mediation

(b) Moderated mediation

Figure 5 Skeletal diagrams for mediation and moderated mediation. 

 15 

 
 
 
Research Questions and Hypotheses 

In self-study settings, where learners’ agency in using digital resources is crucial for successful 

L2 learning (Godwin-Jones, 2019), autonomously motivated L2 learners are more likely to 

respond positively to teacher autonomy support for app-based language learning beyond the 

classroom. This support can effectively satisfy their autonomy need for app usage, leading to 

higher-quality app engagement. In turn, this engagement may proactively recruit further 

autonomy support from teachers (cf. Reeve et al., 2020), creating a positive self-reinforcing loop 

that enhances app usage and L2 learning. Under this assumption, teacher autonomy support will 

directly influence app usage, which serves as a predictor of L2 learning gains. To ensure the 

robustness of these gains, non-app English learning experiences out of class are controlled for in 

the analysis. Autonomy need satisfaction, app engagement, and perceived opportunity cost are 

included as mediators, transmitting the effect of teacher autonomy support to app usage in 

unique ways. Additionally, learners’ initial levels of self-determined (i.e., autonomous) 

motivation for English learning will moderate these relationships. Figure 6 presents a skeletal 

diagram of these hypotheses, where the + signs indicate positive relationships between variables, 

and the – signs represent negative relationships. The research questions and corresponding 

hypotheses of the present study are provided as follows: 

 16 

 
 
Autonomy
Need
Satisfaction

+

+

Teacher 
Autonomy 
Support

+

+

App 
Engagement

+

+

App Usage

+

Learning Gain

–

Perceived
Opportunity 
Cost

–

Non-App 
English
Learning
(covariate)

Figure 6 Hypothesized model for the present study. 

Research Question 1. To what extent does teacher autonomy support affect autonomy need 

satisfaction, app engagement, perceived opportunity cost, and app usage? 

Hypotheses 1a, 1b and 1c. Teacher autonomy support will positively predict autonomy need 

satisfaction, app engagement, and app usage, respectively. 

Hypothesis 1d. Teacher autonomy support will negatively predict perceived opportunity cost. 

Research Question 2. To what extent does autonomy need satisfaction affect app engagement 

and app usage? 

Hypotheses 2a and 2b. Autonomy need satisfaction will positively predict app engagement 

and app usage, respectively.  

Hypotheses 2c and 2d. Autonomy need satisfaction will mediate the relationship between 

teacher autonomy support and app engagement, and between teacher autonomy support and 

app usage, respectively. 

 17 

 
 
 
Research Question 3. To what extent does app engagement affect app usage? 

Hypothesis 3a. App engagement will positively predict app usage. 

Hypothesis 3b. App engagement will mediate the relationship between teacher autonomy 

support and app usage. 

Hypothesis 3c. App engagement will mediate the relationship between teacher autonomy 

support and app usage through autonomy need satisfaction. 

Research Question 4. To what extent does perceived opportunity cost affect app usage? 

Hypothesis 4a. Perceived opportunity cost will negatively predict app usage. 

Hypothesis 4b. Perceived opportunity cost will mediate the relationship between teacher 

autonomy support and app usage. 

Research Question 5. To what extent does app usage affect learning gains from app-based 

learning? 

Hypothesis 5. App usage will positively predict learning gains after non-app English learning 

experiences are controlled. 

Research Question 6. To what extent does initial L2 motivation influence the relationships 

among the variables? 

Hypothesis 6. Greater autonomous L2 motivation will positively moderate the relationships 

among the variables (i.e., moderated mediation). 

 18 

 
 
Participants 

METHOD 

I recruited two Korean English teachers and their combined group of 305 seventh graders (aged 

12-13) from the same public middle school in South Korea. These young adolescent learners 

have been studying English for at least five years in school as required by the national 

curriculum. Additionally, they have participated in varying levels of private English learning 

activities out of class. Such informal English learning experiences, independently from the 

school curriculum, have resulted in a diverse range of English ability and motivation profiles 

among the learners. 

To measure the breadth of receptive vocabulary knowledge as a proxy of learners’ initial 

English proficiency prior to app-based language learning, I administered the updated Vocabulary 

Levels Test A (Webb et al., 2017). This test was chosen because it is recommended as a suitable 

test for measuring the vocabulary size and vocabulary knowledge of elementary and intermediate 

learners of English as a foreign language. Additionally, due to the overall low proficiency, I 

adapted the item definitions by translating them into Korean. For example, as shown in Figure 7, 

the original item definitions—‘picture’, ‘place where things grow outside’, and ‘cost’—were 

provided with their Korean equivalents—‘사진’, ‘정원’, and ‘가격, 비용’, respectively—and 

presented in blue. In the vocabulary level test, the learners were asked to match correct English-

Korean word pairs by choosing corresponding buttons.  

 19 

Figure 7 Example of adapted Webb et al.’s (2017) updated vocabulary levels test. 

The test included five frequency levels, and each contained 30 items. As proposed by Webb et al. 

(2017), I applied a cutoff point of 29/30 for mastery of the 1k, 2k, and 3k levels, and a cutoff 

point of 24/30 for mastery of the 4k and 5k levels. The result shows that about 73.44% of the 

learners (224 of 305) had not achieved mastery of the most frequent 1,000-word families, while 

an additional 13.44% (41 of 305) reached mastery of this level. To explore the impact of app-

based language learning on beginner learners, I included only those learners whose vocabulary 

level was at or below the 1,000-word level, removing 40 participants from the data. Additionally, 

11 more participants withdrew during the project. Consequently, a total of 254 learners were 

included in the present study. 

Focal Technology: Mango Languages 

The focal app of this study is a commercially available language learning app, Mango Languages 

(https://mangolanguages.com). It provides structured, conversation-based courses that cover 

reading, listening, speaking, vocabulary, pronunciation, grammar, and culture in over 70 

languages. As of March 2024, the app had approximately 100,000 monthly active global users. 

Learners enrolled in Mango Languages’ premium course, designed specifically for native 

 20 

 
Korean speakers. The course offered 95 English lessons, along with review and assessment 

features. As illustrated in Figure 8, Mango Languages is structured to support learners’ self-

study of English beyond the classroom.  

Mango Languages provided free subscriptions and access to usage data for this study but 

did not impose any restrictions or exert influence on any aspect of the research process. None of 

the learners had prior experience with the focal app before this study. Over three months, they 

used the app independently from their regular school English curriculum. 

 21 

Figure 8 Key self-study features of Mango Languages (retrieved from 
https://mangolanguages.com/how-it-works). 

 22 

 
Teacher Autonomy Support for App-Based Language Learning Beyond the Classroom 

Based on Reeve and Cheon’s (2021) autonomy-supportive instructional behaviors (ASIB), the 

two English teachers recruited for this study provided various types of classroom-based 

autonomy support, with the goal of developing a sense of volition and ownership of app-based 

language learning beyond the classroom. Using their own digital devices, learners used the app 

in a self-directed way with full freedom to decide how and to what extent they used it (ASIB #2). 

No external accountability (e.g., rewards, punishments) was imposed for app usage (ASIB #3). 

The teachers also demonstrated the app’s key content and features and helped raise learners’ 

awareness of the value of using the app (ASIB #4), highlighting its potential benefits not 

available in non-app learning methods (e.g., algorithm-based spaced vocabulary learning, 

pronunciation practice using visual waveforms). Additionally, the teachers assisted learners by 

providing strategies for setting both short-term and long-term usage goals (e.g., He & Loewen, 

2022). They also made efforts to recognize learners’ progress and challenges, either by 

addressing them publicly during class or privately to build confidence in using the app out of 

class (ASIB #5, ASIB #6, and ASIB #7). This teacher autonomy support was consistently 

provided throughout the semester as a routine of regular English classes. This also served as an 

informal reminder, encouraging learners to stay focused on their app learning goals, being 

reinforced at least once a week. 

Materials 

I collected three types of data for this study: (a) app usage time, (b) questionnaire responses, and 

(c) vocabulary test scores. All questionnaires were rated on a 6-point Likert scale, ranging from 1 

(“strongly disagree”) to 6 (“strongly agree”), and were administered in learners’ L1, Korean. The 

Korean language versions of the questionnaires are provided in Appendix. 

 23 

In-App Usage Time 

To reliably collect and manage in-app usage data, I used the Mango administrative portal (MAP) 

system (see Figure 9). Before app-based language learning began, I set up the MAP system and 

assigned learners to a user group of each teacher, accordingly. As demonstrated in Figure 10, the 

MAP system allowed the researcher and teachers to track individual learners’ in-app activities by 

providing detailed app usage data in real time. This data included the time spent during each app 

session (recorded in seconds), organized by date and activity. 

After app-based language learning ended on June 30th, I extracted the entire usage data, 

resulting in 12,126 data points. For ease of interpretation, I transformed the raw time data from 

seconds into hours, calculating the usage time per day and the total time spent on the app. 

Figure 9 Mango administrative portal (MAP) system interface. 

 24 

 
Figure 10 Individual app usage data on the Mango administrative portal (MAP) system. 

Perceived Teacher Autonomy Support Questionnaire 

Table 1 presents questionnaire items for perceived teacher autonomy support. To measure 

learners’ perceptions of teachers’ instructional behaviors as autonomy-supportive—using Reeve 

and Cheon’s (2021) ASIB strategies—for their out-of-class app usage, I adapted items from the 

learning climate questionnaire (e.g., Jang et al., 2016; Oga-Baldwin et al., 2017). The 

questionnaire included four items that focus on how teachers created an autonomy-supportive 

climate for app-based language learning beyond the classroom. Cronbach’s alpha value of .89 

indicates a high level of reliability. 

Table 1 Perceived teacher autonomy support questionnaire 

Item 

Question 

Support1  My teacher has helped me set my own goals for using Mango. 

Support2  My teacher has supported the ways I use Mango. 

Support3  My teacher has recognized my efforts to make good use of Mango. 

Support4  My teacher has carefully responded to my questions about Mango. 

 25 

 
Autonomy Need Satisfaction Questionnaire 

Table 2 presents questionnaire items for autonomy need satisfaction for app-based language 

learning. To assess the extent to which learners’ autonomy need was satisfied for app-based 

language learning (i.e., their sense of volition and personal ownership in out-of-class app usage), 

I adapted items from Leeming and Harris’ (2022) basic psychological needs satisfaction and 

frustration scale. The questionnaire included three items. Cronbach’s alpha value of .84 indicates 

a high level of reliability. 

Table 2 Autonomy need satisfaction questionnaire 

Item 

Question 

Autonomy1 

I feel that I have the freedom to decide whether or not to use Mango. 

Autonomy2 

I feel that I have the freedom to plan my use of Mango. 

Autonomy3 

I feel that I use Mango in the way I truly want. 

App Engagement Questionnaire 

Table 3 presents questionnaire items for app engagement. In contemporary education, 

engagement is a multidimensional construct (Skinner & Raine, 2022; Hiver et al., 2021, 2024 in 

L2 contexts). In the context of app-based language learning beyond the classroom, I measured 

learners’ behavioral, emotional, cognitive, and agentic engagement by adapting the scales from 

previous studies (Oga-Baldwin, 2019; Reeve, 2013; Reeve & Lee, 2014; Zhou et al., 2021). Each 

engagement dimension included three items, resulting in a total of 12 items. Cronbach’s alpha 

value for each construct indicates a high level of reliability. 

 26 

 
 
Table 3 App engagement questionnaire 

Scale  Question 

Behavioral Engagement (BE) (α = .94) 

BE1 

I used Mango diligently. 

BE2 

I stayed focused while using Mango. 

BE3 

I did my best while using Mango. 

Emotional Engagement (EE) (α = .95) 

EE1  Mango was fun. 

EE2 

It was interesting to learn new things on Mango. 

EE3 

The learning activities on Mango were enjoyable. 

Cognitive Engagement (CE) (α = .91) 

CE1 

I knew what and how to study while using Mango. 

CE2 

I had a plan for when to use Mango. 

CE3 

I reviewed what I learned on Mango. 

Agentic Engagement (AE) (α = .92) 

AE1 

I aim to use Mango beyond the usage goal I had set. 

AE2  When I faced difficulties using Mango, I actively sought help from those around me. 

AE3  When I had questions while studying on Mango, I endeavored to resolve them. 

Perceived Opportunity Cost Questionnaire 

Table 4 presents questionnaire items for perceived opportunity cost. To evaluate learners’ 

perceived opportunity cost of app usage, I adapted the cost component of the Expectancy-Value-

Cost (EVC) Scale (e.g., Kosovich et al., 2015). Four items were used, with higher scores 

indicating that learners perceive greater app usage as involving a greater loss of valued 

alternatives in their other daily commitments. Cronbach’s alpha value of .82 indicates a high 

level of reliability. 

 27 

 
 
Table 4 Perceived opportunity cost questionnaire 

Item 

Question 

Cost1 

Mango requires too much time and effort in my current situation. 

Cost2 

Cost3 

Cost4 

Due to other academic commitments, I currently do not have time to invest in 
using Mango. 

In my current situation, I cannot afford the extra time and effort needed to use 
Mango consistently. 

To use Mango diligently, I would have to give up too much of my currently 
available time. 

English Learning Orientation Scale 

Table 5 presents the English learning orientation scale. To understand the learners’ general 

English learning motivation within SDT, I adapted Leeming and Harris’ (2022) language 

learning orientation scale. They improved upon the original scale developed by Noels et al. 

(2000) and further validated it within an East Asian context (i.e., Japan) similar to the present 

study. The questionnaire included six SDT-based motivational constructs: Intrinsic, integrated, 

identified, introjected, external motivation, and amotivation (cf. Figure 3). Each construct 

represents the extent to which their motivation for English learning is driven by themselves (e.g., 

self-satisfaction, enjoyment) or by outside influences (e.g., external incentives, pressure). The 

questionnaire consisted of 21 items. Cronbach’s alphas for each construct indicate a high level of 

reliability, except for external motivation whose alpha value is slightly below the .70 benchmark. 

Table 5 English learning orientation scale 

Scale  Question 

Intrinsic Motivation (α = .92) 

Item1  Because I like English. 

Item2  Because learning English is fun. 

Item3  Because I like using English. 

 28 

Table 5 (cont’d) 

Integrated Motivation (α = .83) 

Item1  Because English is an important component in understanding who I am. 

Item2  Because English represents the things I consider important in life (values, beliefs). 

Item3  Because English is a natural part of my life. 

Item4  Because my ability to speak English reflects who I am as a person. 

Identified Motivation (α = .81) 

Item1  Because English helps my personal growth and development. 

Item2  Because English helps me achieve the goals I aspire to. 

Item3  Because English provides me with more opportunities for the future success. 

Introjected Motivation (α = .82) 

Item1 

Item2 

Item3 

Item4 

Because those around me (family, friends, teachers) have high expectations for my 
English skills. 

Because the people around me (family, friends, teachers) view me positively when I 
do well in English. 

Because the people around me (family, friends, teachers) expect me to study English 
diligently. 

Because I want to be seen as good at English by those around me (family, friends, 
teachers). 

External Motivation (α = .65) 

Item1  Because I want to achieve a good score on my English exams. 

Item2  Because English is a required subject that I must take, not a choice I can make. 

Item3  Because studying English is mandatory, regardless of how much I like it. 

Item4 

Because being good at English can help me get into a better upper school (high school, 
university). 

Amotivation (α = .83) 

Item1  Studying English is a waste of time for me. 

Item2  English is meaningless to me. 

Item3 

If possible, I would choose not to learn English. 

 29 

Non-App English Learning Experiences Beyond the Classroom Questionnaire 

Figure 11 shows an example of the questionnaire to measure non-app English learning 

experiences beyond the classroom. I collected data on learners’ non-app English learning 

activities during the study period. The questionnaire asked learners to report, on average, how 

many days per week they were involved in each of the following out-of-class activities where 

English served as the primary medium: reading, listening, watching, and communication (Figure 

11a). For each activity type, learners also rated the intensity of specific English skills involved, 

using a scale from 1 (very rarely: less than 10% of an activity) to 7 (almost always: more than 

90% of an activity) (Figure 11b). These skills included reading, grammar, listening, vocabulary, 

writing, pronunciation, and speaking. To create a single index, I multiplied the number of days 

for each activity by the intensity of each skill (i.e., frequency × intensity). For example, if a 

learner reported that listening skill “almost always” occurred during a watching activity, which 

happened 3 days a week, the listening score for watching was 21 (i.e., 7 × 3). The resulting 

scores were summed across all activity types. A higher value indicates a greater level of 

experience of non-app English learning activities out of class. 

 30 

a

b

Figure 11 Example of non-app English learning experiences questionnaire. 

Meaning Recall Test 

I measured the impact of app-based language learning with the number of English words learned 

after using the app. The target words were selected through a multi-step process. Initially, 417 

words were extracted from all 95 lessons of Mango Languages. Of these, 349 words and their 

related word families were removed as they either appeared in learners’ English textbooks at 

school used during the study or were basic words likely learned in earlier school stages. From the 

remaining words, those from the first 18 lessons were excluded because the learners completed 

 31 

 
these lessons for in-class app training sessions with the teachers. To maximize the benefits of 

learners’ participation, words less relevant to their school curriculum (e.g., busser, locksmith) 

were excluded. This process resulted in a final set of 40 target words, as listed in Table 6. 

Table 6 Final list of the target words 

Item  Target Word 

Item  Target Word 

Item  Target Word 

Item  Target Word 

1 

2 

3 

grocery 

avenue 

direction 

4  medicine 

5 

6 

7 

8 

9 

downtown 

freeway 

account 

exchange 

form 

10 

deposit 

11  withdraw 

12 

13 

14 

15 

16 

17 

18 

19 

20 

dizzy 

repair 

trouble 

tow 

hire 

resume 

reference 

similar 

application 

21 

22 

23 

24 

25 

26 

27 

28 

29 

30 

shift 

address 

rent 

inexpensive 

available 

compact 

unlock 

repeat 

bill 

stuff 

31 

day off 

32  midnight 

33 

34 

35 

36 

37 

38 

drugstore 

license 

emergency 

fix 

straight 

fever 

39  metal 

40 

sign 

In a meaning recall test, the learners were asked to provide Korean translations of target English 

stimuli. Responses were scored on a binary correct/incorrect scale for the test. The test was 

administered twice as a pretest (March) and a posttest (July). The order of the items was 

randomized across learners and testing times. Cronbach’s alphas indicated a high level of 

reliability, with .88 for the pretest and .92 for the posttest. The difference between pretest and 

posttest scores was used as an indicator of learning gains in the analysis. 

Procedure 

Figure 12 shows an overview of the study procedure. This research was approved by the 

institutional review board at Michigan State University (STUDY00010031) in December 2023. 

During December 2023 and January 2024, I recruited in-service Korean EFL teachers in public 

 32 

schools by advertising nationwide in online English teacher communities. Over two weeks, an 

initial pool of 47 teacher candidates was created. The selection criteria included teachers who 

met all of the following: those who (a) were currently teaching more than 100 learners across 

multiple classes and (b) were willing to promote out-of-class English learning using technology. 

After reviewing the willing participant pool, two local English teachers from the same public 

middle school were selected. Both teachers were female; one had eight years of teaching 

experience, while the other had two years of teaching experience. In Spring 2024, each teacher 

taught about 150 seventh graders (aged 12-13) across six classes, totaling 305 learners. In 

February, the two participating teachers and I prepared for the study via online meetings or text 

chats. During this preparation stage, I shared the purpose of this project and provided a 

comprehensive overview of Mango Languages and its key features. Additionally, I introduced 

the teachers to Reeve and Cheon’s (2021) seven ASIBs and discussed how these autonomy-

supportive strategies could be effectively implemented in their classrooms to enhance learners’ 

self-directed app-based language learning. 

Before the study began in March, learners and their parents agreed to participate in this 

study and completed an English learning orientation questionnaire. In March, the teachers 

introduced the app to learners and provided in-class training sessions to familiarize participants 

with its functions and usage. During this initial phase, the learners completed a pretest of the 

meaning recall test. From April through June, they engaged in self-directed app usage for 13 

weeks out of class, with consistent classroom-based teacher autonomy support provided 

throughout the usage period. At the end of May, learners completed a perceived teacher 

autonomy support questionnaire, and in mid-June, they were asked to fill out an autonomy need 

satisfaction questionnaire. At the end of June, they completed a perceived opportunity cost 

 33 

questionnaire. The app usage ended as of June 30th, followed by a posttest, along with 

questionnaires on app engagement and non-app English learning experiences beyond the 

classroom. 

March
(Weeks 1 to 4)

April
(Weeks 5 to 9)

May
(Weeks 10 to 13)

June
(Weeks 14 to 17)

July
(Week 18)

• App introduction
•
In-class app 
trainings

• App usage over 13 weeks (out of class, self-directed)
• App usage data collected in real time.

Teacher autonomy support was provided throughout the app training and usage period.

• Consent form
• Meaning recall 
test (pretest)
• English learning
orientation scale

• Perceived teacher autonomy support questionnaire (End of 

May)

• Autonomy need satisfaction questionnaire (Mid June)
• Perceived opportunity cost questionnaire (End of June)

• App usage 
ended.

• App usage data 

extracted.

• Meaning recall 
test (posttest)
• App engagement 
questionnaire
• Non-app English 

learning 
experiences 
questionnaire

Figure 12 Overview of the study procedure. 

Data Analysis 

Analysis 1: Mediation Model 

To answer research questions from 1 to 5, I built a mediation model within the framework of 

structural equation modeling (SEM), using the lavaan R package. Although the dataset had no 

missing values, the results of the multivariate non-normality test indicate that the assumption of 

multivariate normality was not satisfied (see Table 7). To address the nonnormal distributions 

and accommodate both continuous and ordinal indicators, I applied the robust Maximum 

Likelihood method with the mean-adjusted Satorra-Bentler scaled chi-square estimator (MLM) 

(Kline, 2023). 

 34 

 
 
 
Table 7 Results of multivariate non-normality test 

Test 

Mardia Skewness 

Mardia Kurtosis 

Statistic 

6805.73 

43.14 

p value 

< .001 

< .001 

Normality 

No 

No 

Next, I used confirmatory factor analysis (CFA) to build and assess the measurement model, 

establishing construct validity—the degree to which each scale accurately measures its intended 

construct. This was achieved by ensuring unidimensionality, whereby the indicators of each 

latent variable represent only that specific variable. I evaluated this through multiple forms of 

evidence. First, I checked whether the standardized factor loadings of individual indicators were 

greater than .50. Second, I examined construct reliability (i.e., how consistent the indicators are 

with each other in measuring the same construct) to confirm that omega (ω) values exceeded the 

recommended threshold of .70. Additionally, convergent validity (i.e., the extent to which the 

latent variable explains the variance of its indicators) was supported when the average variance 

extracted (AVE) for each latent variable was greater than .50. Additionally, discriminant validity, 

which ensures that a construct is distinct from others, was examined to ensure that each construct 

is distinct from others by verifying that the square root of each latent variable’s AVE was greater 

than its correlations with other variables. Based on this information, I determined the number of 

latent variables of the measurement model. 

Finally, I compared the relative fit of measurement invariance models to ensure that the 

questionnaire items were interpreted equivalently across class groups taught by two teachers. 

The stepwise procedure involved testing for several levels of invariance: configural invariance, 

which assesses whether the same factor structure holds across groups; weak invariance (i.e., 

equality of unstandardized factor loadings) to determine if the relationships between items and 

latent variables were consistent across groups; strong invariance (i.e., equality of unstandardized 

 35 

intercepts) to establish that item response levels were comparable; and strict invariance (i.e., 

equality of error variances and covariances) to confirm that residual variances were equivalent 

across groups. 

For both the measurement model and the structural model, I evaluated model fit at both 

the local and global levels. For local fit, I examined normalized residuals, considering values 

within the range of ±2.0 as acceptable. For global fit, I reported the scaled χ² statistic alongside 

robust versions of three fit indices: Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), and 

Root Mean Square Error of Approximation (RMSEA). In addition, I also reported Standardized 

Root Mean Square Residual (SRMR). The thresholds for acceptable model fit were CFI and TLI 

values ≥ .90, and RMSEA and SRMR values ≤ .08, with values up to .10 considered marginally 

acceptable. Lastly, I compared the hypothesized model in Figure 6 against an alternative model 

to reduce the risk of confirmation bias. 

Analysis 2: Cluster Analysis and Moderated Mediation Model 

To answer research question 6, I first identified distinct subgroups among young beginner 

Korean EFL learners based on their motivation for English learning. I used a hybrid (hierarchical 

k-means) approach to cluster analysis (e.g., Crowther et al., 2021; Hwang et al., 2024). Using the 

NbClust and factoextra R packages, I determined the number of clusters with an agglomerative 

hierarchical clustering algorithm with Ward’s linkage and squared Euclidean distance methods. 

Subsequently, I used the resulting clusters as a moderator of the mediation model built in 

Analysis 1 (i.e., moderated mediation model). To ensure that questionnaire items were 

interpreted equally across different L2 motivation groups, I checked measurement invariance in a 

stepwise fashion. Next, I compared a constrained model in which parameters (e.g., path 

coefficients, intercepts) are constrained to be equal across motivational groups and an 

 36 

unconstrained model in which they are allowed to vary freely. If the chi-square difference 

between the fit of the constrained and the unconstrained model is significant, it indicates that the 

structural relationships differ across groups. 

 37 

 
 
Descriptive Statistics 

RESULTS 

Perceived Teacher Autonomy Support and Autonomy Need Satisfaction 

Tables 8 and 9 present the descriptive statistics for perceived teacher autonomy support and 

autonomy need satisfaction for app-based language learning beyond the classroom, respectively. 

On the six-point Likert scale used in the survey, mean scores above 4 indicate positive responses, 

whereas mean scores below 3 are negative responses to corresponding items. As summarized in 

Figure 13, the overall results show neutral-positive trends for the measured constructs, with 

mean scores close to 4, suggesting that teachers’ instructional behaviors were generally 

perceived as autonomy-supportive and that learners who felt them as such experienced a sense of 

volition and personal ownership. 

Table 8 Descriptive statistics of perceived teacher autonomy support 

Item 

Mean 

Support1 

Support2 

Support3 

Support4 

3.96 

4.11 

3.35 

3.91 

95% CI 

LL 

3.77 

3.93 

3.14 

3.70 

UL 

4.15 

4.29 

3.56 

4.12 

SD 

SE 

Median  Min. 

Max. 

1.55 

1.47 

1.71 

1.70 

0.10 

0.09 

0.11 

0.11 

4 

5 

3 

4 

1 

1 

1 

1 

6 

6 

6 

6 

Table 9 Descriptive statistics of autonomy need satisfaction 

Item 

Mean 

Autonomy1 

3.83 

Autonomy2 

4.28 

Autonomy3 

3.40 

95% CI 

LL 

3.63 

4.09 

3.19 

UL 

4.04 

4.46 

3.61 

SD 

SE 

Median  Min. 

Max. 

1.66 

1.47 

1.69 

0.10 

0.09 

0.11 

4 

5 

4 

1 

1 

1 

6 

6 

6 

 38 

 
Figure 13 Perceived teacher autonomy support and autonomy need satisfaction: means and 95% 
confidence intervals. 

App Engagement 

Table 10 presents the descriptive statistics for app engagement beyond the classroom. Overall, 

the results indicate low to moderate engagement levels across behavioral, emotional, cognitive, 

and agentic types, with most mean scores falling below the positive benchmark of 4. Figure 14 

summarizes these findings. 

 39 

 
 
 
Table 10 Descriptive statistics of app engagement 

Item 

Mean 

BE1 

BE2 

BE3 

EE1 

EE2 

EE3 

CE1 

CE2 

CE3 

AE1 

AE2 

AE3 

2.89 

3.27 

3.38 

3.06 

3.24 

3.23 

3.47 

3.08 

2.83 

2.93 

3.05 

3.22 

95% CI 

LL 

2.68 

3.05 

3.17 

2.87 

3.05 

3.03 

3.27 

2.88 

2.64 

2.73 

2.85 

3.01 

UL 

3.09 

3.48 

3.59 

3.26 

3.44 

3.42 

3.67 

3.27 

3.02 

3.13 

3.24 

3.42 

SD 

SE 

Median  Min. 

Max. 

1.66 

1.72 

1.71 

1.59 

1.61 

1.59 

1.61 

1.57 

1.56 

1.60 

1.59 

1.66 

0.10 

0.11 

0.11 

0.10 

0.10 

0.10 

0.10 

0.10 

0.10 

0.10 

0.10 

0.10 

2 

3 

4 

3 

3.5 

3 

4 

3 

2 

3 

3 

3 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

Figure 14 App engagement: means and 95% confidence intervals. The acronyms BE, EE, CE, 
and AE represent behavioral, emotional, cognitive, and agentic engagement, respectively. 

 40 

 
 
 
Non-App English Learning Experiences and Perceived Opportunity Cost 

Table 11 presents the descriptive statistics for non-app English learning experiences across 

language skills. The results indicate that listening was the most frequently practiced activity 

among young beginners beyond the classroom (e.g., listening to English songs) while speaking, 

pronunciation, and writing skills were the least practiced. However, the wide 95% confidence 

intervals in Figure 15a suggest substantial variation among the learners. 

Table 12 presents the descriptive statistics for perceived opportunity cost of app usage, 

indicating overall neutral-negative responses, indicating learners did not evaluate using the app 

as overly time- and effort-intensive. Figure 15b summarizes these trends, showing the mean 

scores ranged between 3 and 4. 

Table 11 Descriptive statistics of non-app English learning experiences 

Scale 

Mean 

95% CI 

LL 

UL 

SD 

SE 

Median  Min. 

Max. 

Reading 

16.04 

12.89 

19.19 

25.49 

Listening 

21.65 

17.85 

25.44 

30.73 

Writing 

11.59 

Speaking 

10.27 

Grammar 

13.14 

8.79 

7.54 

9.99 

14.40 

22.70 

13.00 

22.10 

16.29 

25.48 

Vocabulary 

16.63 

13.29 

19.97 

27.03 

Pronunciation 

10.36 

7.44 

13.28 

23.61 

1.60 

1.93 

1.42 

1.39 

1.60 

1.70 

1.48 

6 

10 

11.59 

0 

0 

3 

0 

0 

0 

0 

0 

0 

0 

0 

189 

189 

189 

189 

189 

189 

189 

 41 

 
 
 
Table 12 Descriptive statistics of perceived opportunity cost 

Item 

Mean 

Cost1 

Cost2 

Cost3 

Cost4 

3.12 

3.75 

3.37 

3.17 

95% CI 

LL 

2.94 

3.55 

3.19 

2.98 

UL 

3.29 

3.94 

3.55 

3.37 

SD 

SE 

Median  Min. 

Max. 

1.43 

1.58 

1.45 

1.59 

0.09 

0.10 

0.09 

0.10 

3 

4 

3 

3 

1 

1 

1 

1 

6 

6 

6 

6 

Figure 15 Non-app English learning experiences and perceived opportunity cost: means and 
95% confidence intervals. R = Reading; L = Listening; W = Writing; S = Speaking; G = 
Grammar; V = Vocabulary; P = Pronunciation. 

App Usage 

Table 13 presents the total amount of time spent on the app. Overall, the learners used the app for 

1.65 hours (95% CI [1.36, 1.93]) out of class over 13 weeks. Notably, the learners in the top 10% 

of app usage (n = 26) showed about 7.02 hours (95% CI [5.36, 8.68]) of app usage on average. 

As shown in Figure 16, the red trend line indicating the average cumulative app usage over time 

reveals that the mean values were skewed toward the lower end due to a subset of learners with 

minimal app usage.  

 42 

 
 
Table 13 Descriptive statistics of app usage 

Group 

Mean 

All (n = 254) 

1.65 

Top 10% (n = 26) 

7.02 

95% CI 

LL 

1.36 

5.36 

UL 

1.93 

8.68 

SD 

SE  Median  Min.  Max. 

2.33 

4.11 

0.15 

0.81 

1.01 

5.28 

0 

19.70 

3.67 

19.70 

Figure 16 Individual cumulative app usage over time. 

Panel A in Figure 17 represents the daily total app usage time at a group level. Overall, the app 

usage displayed repetitive fluctuations. A nearly two-week period of national holidays in May 

led to decreased usage as learners did not attend school, suggesting a diminished effect of teacher 

autonomy support during this time. Similarly, the performance assessment season in June for 

most school subjects shifted learners’ focus away from the app. However, it is noteworthy that 

the learners repeatedly returned to the app after pauses in their usage, rather than failing to 

resume app usage once it was disrupted. Panel B in Figure 17 shows the distribution of the total 

app usage time of individual learners, suggesting great variations among the learners. 

 43 

 
 
Figure 17 Distribution of app usage data. 

Meaning Recall Test 

Table 14 and Figure 18 present the results of the meaning recall test. After using the app over 13 

weeks, the learners demonstrated an overall improvement on the posttest, with the top 10% app 

usage group (n = 26) showing greater gains. 

Table 14 Descriptive statistics of meaning recall test 

Group 

Time  Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

All 

Pretest 

5.61 

4.97 

6.26 

5.19 

0.33 

(n = 254)  Posttest 

9.48 

8.57 

10.40 

7.39 

0.46 

4 

8 

Top 10%  Pretest 

6.23 

4.11 

8.35 

5.26 

1.03 

4.50 

(n = 26) 

Posttest 

13.04 

9.58 

16.50 

8.57 

1.68 

13 

0 

0 

0 

0 

23 

28 

16 

28 

 44 

 
Figure 18 Meaning recall test scores: means and 95% confidence intervals. 

Correlational Relationships Among the Variables 

Figure 19 displays a correlation plot illustrating the strength of correlations between items. The 

colors represent the magnitude and direction of the relationships, with stronger positive 

correlations depicted in blue and stronger negative correlations in red shades. The results reveal 

three key patterns. First, the items from perceived teacher autonomy support, autonomy need 

satisfaction, app engagement, app usage time, and learning gains positively correlated with each 

other. As outlined in Noels et al.’s (2019) motivation process model, these findings indicate a 

potential positive chain reaction: learners who perceived teachers’ instructional behaviors as 

autonomy-supportive were more likely to experience greater volition and personal ownership in 

app usage. This, in turn, enhanced their app engagement and increased their actual app usage out 

of class, ultimately leading to improved learning gains. Second, the items from perceived 

opportunity cost demonstrated negative correlations with the positively correlated group, 

indicating that learners who perceived app-based language learning as time- and effort-intensive 

were more likely to have negative experiences with app usage. Conversely, those who viewed it 

 45 

 
as requiring less time and effort were more likely to have positive app usage experiences. 

Finally, non-app English learning activities beyond the classroom showed no clear positive or 

negative correlations with the other variables. 

Figure 19 Correlation plot of individual variables. 

Four-Factor Measurement Model 

Evaluating the measurement model confirms that the observed variables accurately reflect the 

underlying constructs, establishing a solid foundation for any subsequent structural analysis of 

relationships among constructs. First, I built a seven-factor measurement model including 

perceived teacher autonomy support, autonomy need satisfaction, four types of engagement, and 

 46 

 
perceived opportunity cost. However, this initial attempt revealed methodological issues that 

required adjustments to the model. Although engagement is a theoretically multidimensional 

construct (e.g., Hiver et al., 2021, 2024), the factor correlations among the four engagement 

latent variables were very high, reaching above .90. Such high interdependence indicates that the 

latent variables measure almost the same underlying construct. Indeed, an exploratory factor 

analysis with the oblimin rotation and the minres estimation method in Figure 20 confirmed a 

single-factor structure of app engagement, with all 12 items strongly loading on a single factor, 

and parallel analysis scree plots in Figure 21 showed the eigenvalues of any additional factors 

being negligible. Furthermore, I encountered a convergence issue in a subsequent structural 

model, which failed to find a proper solution due to possible extreme multicollinearity. 

To address this problem, I conceptualized engagement as a single construct, consistent 

with previous studies (e.g., Oga-Baldwin et al., 2017). Furthermore, I applied a parceling 

technique by grouping individual items into multiple parcels and used these parcel values as 

indicators of a latent construct. Parceling, by specifying a target construct with fewer indicators, 

can help address measurement error associated with individual item-level data, leading to more 

stable parameter estimates and better model fit (see Little et al., 2013, for a pro-parcel argument 

for multidimensional constructs). Following Matsunaga’s (2008) recommendation, I grouped 12 

engagement items into three parcels—Engagement 1, Engagement 2, and Engagement 3—

considering an even distribution of item-specific factor loadings across parcels (i.e., factorial 

algorithm). As summarized in Table 15, these parcels, each comprising four items, served as 

indicators of a single latent variable, app engagement.  

 47 

Figure 20 Exploratory factor analysis of app engagement items. 

Figure 21 Parallel analysis scree plots. 

 48 

 
 
 
Table 15 Results of factorial-algorithm-based parceling 

Item 

CE2 

EE3 

AE1 

BE2 

BE3 

AE3 

EE2 

EE1 

CE1 

BE1 

Standardized Factor Loading 

Parcel 

0.90 

0.90 

0.89 

0.88 

0.88 

0.88 

0.88 

0.87 

0.87 

0.86 

Engagement1 

Engagement2 

Engagement3 

Engagement3 

Engagement2 

Engagement1 

Engagement1 

Engagement2 

Engagement3 

Engagement3 

To further check the unidimensionality of items within each parcel, I conducted a series of 

single-factor CFA models for each parcel. The results in Table 16 show that the standardized 

factor loadings for all indicators were greater than the recommended threshold of 0.7, providing 

empirical support for aggregating the items into parcels. 

Table 16 Results of confirmatory factor analysis by parcel 

Parcel 

Indicators 

Engagement1 

Engagement2 

CE2 

AE3 

EE2 

AE2 

EE3 

BE3 

EE1 

CE3 

b* 

.86 

.94 

.81 

.90 

.95 

.83 

.92 

.79 

 49 

p-value 

Error variances 

< .001 

< .001 

< .001 

< .001 

< .001 

< .001 

< .001 

< .001 

.26 

.12 

.35 

.18 

.11 

.32 

.15 

.38 

 
 
Table 16 (cont’d) 

Engagement3 

AE1 

BE2 

CE1 

BE1 

.85 

.90 

.83 

.91 

< .001 

< .001 

< .001 

< .001 

.27 

.19 

.31 

.17 

After parceling, I built a four-factor measurement model, consisting of (a) perceived teacher 

autonomy support, (b) autonomy need satisfaction, (c) app engagement, and (d) perceived 

opportunity cost. The inspection of normalized residuals of the four-factor measurement model 

revealed all residuals, except for three cases, remained within the acceptable range of ±2.0, 

supporting the model’s local fit (see Table 17). As shown in Table 18, the standardized factor 

loadings were statistically significant at p < .001 and consistently high across items, with the 

lowest loading observed at .65 for Cost 1. Figure 22 summarizes the finalized four-factor 

measurement model. Global fit indices indicated strong model fit (scaled χ² (71) = 131.36, p 

< .001; CFI = . 97; TLI = . 96; RMSEA = .07, 90% CI [.05, .08]; SRMR = .04). 

 50 

 
Table 17 Normalized residuals for a four-factor measurement model (local fit) 

Variable 

1 

2 

3 

4 

5 

6 

7 

8 

9 

10 

11 

12 

13 

14 

1. Support1 

0.00 

2. Support2 

0.67 

0.00 

3. Support3 

–0.32  –1.06 

0.00 

4. Support4 

–0.40  –0.14 

0.73 

0.00 

5. Autonomy1 

–1.04  –0.36 

0.78 

0.50 

0.00 

6. Autonomy2 

–0.37 

0.81 

–0.65 

1.25 

0.29 

0.00 

7. Autonomy3 

–0.24  –0.23 

1.55 

0.75 

0.00 

–0.46 

0.00 

8. Engagement1 

–0.54  –0.95 

2.39 

–0.81  –0.72  –1.44  –0.07 

0.00 

9. Engagement2 

0.00 

–0.23 

2.46 

0.25 

0.67 

–0.46 

0.27 

0.01 

0.00 

10. Engagement3 

–0.17  –0.62 

2.22 

–0.03 

0.29 

–0.38 

0.07 

0.08 

–0.07 

0.00 

11. Cost1 

12. Cost2 

13. Cost3 

14. Cost4 

–1.08  –0.01  –1.22  –0.71  –0.38 

0.29 

–0.18  –1.07  –0.96  –0.97 

0.00 

–0.56 

0.20 

–1.42 

0.68 

1.04 

0.53 

–0.04 

0.02 

–0.14 

0.69 

–0.06 

0.00 

0.94 

0.76 

–0.82 

0.09 

0.31 

0.51 

–0.54  –0.06 

0.54 

0.30 

–0.18 

0.26 

0.00 

–0.01 

0.11 

–0.89 

0.44 

–0.45 

0.01 

–0.30 

0.07 

–0.04 

0.22 

0.10 

–0.06  –0.06 

0.00 

Note. Residual values within the range of ±2.0 are acceptable.

 51 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 18 Robust maximum likelihood estimates for a four-factor measurement model 

Parameter 

b 

SE 

b* 

SE 

z 

Perceived Teacher Autonomy Support 

Factor Loadings 

Support1 

Support2 

Support3 

Support4 

Autonomy Need Satisfaction 

Autonomy1 

Autonomy2 

Autonomy3 

App Engagement 

Engagement1 

Engagement2 

Engagement3 

Perceived Opportunity Cost 

Cost1 

Cost2 

Cost3 

Cost4 

Support1 

Support2 

Support3 

Support4 

Autonomy1 

Autonomy2 

Autonomy3 

1.00 

0.96 

0.93 

1.05 

1.00 

0.68 

0.96 

1.00 

1.02 

1.03 

1.00 

1.18 

1.20 

1.35 

0.05 

0.05 

0.04 

0.07 

0.04 

0.03 

0.03 

0.14 

0.13 

0.14 

Error (Unique) Variances 

0.10 

0.10 

0.15 

0.12 

0.11 

0.15 

0.13 

0.56 

0.47 

1.33 

0.85 

0.60 

1.16 

0.88 

 52 

0.87 

0.88 

0.73 

0.84 

0.89 

0.68 

0.83 

0.92 

0.93 

0.95 

0.65 

0.70 

0.77 

0.79 

0.24 

0.22 

0.46 

0.30 

0.22 

0.54 

0.31 

0.03 

0.03 

0.03 

0.03 

0.02 

0.05 

0.03 

0.02 

0.02 

0.01 

0.05 

0.05 

0.05 

0.04 

0.05 

0.05 

0.05 

0.04 

0.04 

0.07 

0.05 

34.08 

32.50 

21.72 

32.24 

37.33 

12.98 

30.73 

57.53 

57.90 

69.61 

13.68 

12.94 

14.70 

17.99 

5.25 

4.58 

9.27 

6.84 

5.17 

7.57 

6.87 

 
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
  
Table 18 (cont’d) 

Engagement1 

Engagement2 

Engagement3 

Cost1 

Cost2 

Cost3 

Cost4 

0.40 

0.33 

0.24 

1.17 

1.27 

0.85 

0.94 

Perceived Teacher Autonomy Support 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

Perceived Teacher Autonomy Support 

Factor Variances 

1.82 

2.16 

2.13 

0.86 

Covariances 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

App Engagement 

1.60 

1.43 

–0.28 

1.56 

–0.48 

0.07 

0.07 

0.07 

0.13 

0.20 

0.17 

0.17 

0.19 

0.20 

0.16 

0.15 

0.17 

0.14 

0.11 

0.16 

0.12 

0.16 

0.13 

0.10 

0.58 

0.52 

0.41 

0.38 

1.00 

1.00 

1.00 

1.00 

0.81 

0.72 

–0.22 

0.73 

–0.35 

0.03 

0.03 

0.03 

0.06 

0.08 

0.08 

0.07 

— 

— 

— 

— 

0.04 

0.04 

0.08 

0.05 

0.08 

5.39 

4.30 

3.69 

9.31 

6.87 

5.03 

5.39 

— 

— 

— 

— 

19.02 

20.02 

–2.75 

14.97 

–4.66 

Perceived Opportunity Cost 

–0.44 

0.11 

–0.33 

0.07 

–4.65 

 53 

 
 
 
 
 
 
 
 
 
 
 
 
  
  
  
  
  
Perceived
Teacher
Autonomy 
Support

Autonomy
Need 
Satisfaction

App 
Engagement

Perceived
Opportunity 
Cost

.874

.883

.734

.838

.885

.679

.831

.918

.933

.951

.651

.697

.770

.791

Teacher
Support 1

Teacher
Support 2

Teacher
Support 3

Teacher
Support 4

Autonomy 1

Autonomy 2

Autonomy 3

Engagement 1
(CE2, AE3, EE2, AE2)

Engagement 2
(EE3, BE3, EE1, CE3)

Engagement 3
(AE1, BE2, CE1, BE1)

Cost 1

Cost 2

Cost 3

Cost4

.236

.220

.461

.298

.217

.538

.309

.158

.129

.096

.576

.515

.406

.375

Figure 22 Four-factor measurement model. Covariances are not shown for simplicity. 

 54 

 
The findings in Table 19 also confirmed the strong reliability and validity of the four-factor 

measurement model. The ω values exceeded the recommended threshold of .70 for all latent 

variables, indicating good construct reliability. Convergent validity was supported by AVE, 

which surpassed the .50 threshold for all constructs. Additionally, perceived teacher autonomy 

support, autonomy need satisfaction, and app engagement were positively correlated with one 

another, while all four latent variables showed negative correlations with perceived opportunity 

cost. The square roots of AVE were consistently higher than the correlations with other 

constructs, confirming discriminant validity. These results collectively provide robust evidence 

for the hypothesized factor structure and unidimensionality of the four latent variables. 

Table 19 Reliability and validity of the measurement model 

1. Perceived Teacher Autonomy Support 

2. Autonomy Need Satisfaction 

3. App Engagement 

4. Perceived Opportunity Cost 

ω 

.89 

.85 

.95 

.82 

AVE 

1 

2 

3 

4 

.69 

.66 

.87 

.54 

.83 

.80 

.72 

.81 

.73 

.93 

–.22 

–.35 

–.33 

.73 

Note. The shaded values represent the square roots of the AVE for each latent variable. The 
correlations between latent variables are listed below the shaded diagonal. 

As indicated in Table 20, measurement invariance was also established across class groups 

taught by two teachers. All tested invariance models demonstrated good global fit, and the non-

significant chi-square difference test results between models (ps > .05) indicated that the 

configural, weak, strong, and strict invariance models were retained. These findings confirm that 

the constructs were interpreted and perceived equally across the two teacher class groups.

 55 

 
 
 
 
 
 
 
Table 20 Invariance (multi-group) models across class groups taught by two teachers 

Model 

χ² 

df 

p 

Δχ² 

Δdf 

p 

CFI 

TLI 

RMSEA 

Configural  292.17 

Weak 

304.63 

Strong 

310.55 

Strict 

327.37 

142 

152 

162 

176 

< .001 

< .001 

12.72 

< .001 

< .001 

6.28 

9.85 

10 

10 

14 

0.24 

0.79 

0.77 

.96 

.96 

.96 

.96 

.95 

.95 

.96 

.96 

.08 

.07 

.07 

.07 

RMSEA 90% CI 

LB 

0.05 

0.05 

0.05 

0.04 

UB 

0.09 

0.09 

0.09 

0.08 

SRMR 

0.05 

0.06 

0.06 

0.06 

 56 

 
 
 
 
 
Structural Model of Mediation for Research Questions 1 to 5 

Proposed vs. Alternative Models 

As illustrated in Figure 23, I compared the two competing models to reduce confirmation bias. In 

the originally proposed model, app engagement precedes app usage, which predicts learning 

gains (i.e., app engagement → app usage → learning gains). In the alternative model, app usage 

precedes app engagement, which predicts learning gains (i.e., app usage → app engagement → 

learning gains).  

A. Proposed Model

Autonomy
Need
Satisfaction

+

+

Teacher 
Autonomy 
Support

+

+

App 
Engagement

+

+

App Usage

+

Learning Gain

–

Perceived
Opportunity 
Cost

–

Non-App 
English
Learning
(covariate)

B. Alternative Model

Autonomy
Need
Satisfaction

+

+

Teacher 
Autonomy 
Support

+

+

App Usage

+

+

App 
Engagement

+

Learning Gain

–

Perceived
Opportunity 
Cost

–

Non-App 
English
Learning
(covariate)

Figure 23 Proposed and alternative models. 

 57 

 
The results of the model comparison in Table 21 reveal that the proposed model, with app 

engagement predicting app usage, explained more variance in app usage (i.e., 18.8%) and 

learning gains (i.e., 8.1%), with lower information criterion values (i.e., AIC, BIC, SABIC) 

compared to the alternative model. Therefore, the proposed model was retained for further 

analysis. 

The proposed structural model demonstrated good global fit to the data, as evidenced by 

scaled χ²(112) = 230.44, p < .001, CFI = .95, TLI = .95, RMSEA = .07 (90% CI [.06, .08]), and 

SRMR = .07. As shown in Table 22, the model’s local fit was supported by the minimal 

proportion of significant normalized residuals exceeding ±2.0: eight out of 153 total cases.

 58 

Table 21 Comparison between the proposed and the alternative models 

Model 

χ² 

df 

p 

AIC 

BIC 

SABIC  CFI  TLI  RMSEA 

Proposed 

230.44 

Alternative  239.76 

112 

112 

< .001  15998.36  16139.85  16013.04 

.96 

.95 

< .001  16008.28  16149.78  16022.97 

.95 

. 94 

.07 

.07 

Variance Explained (%) 

RMSEA 90% CI 

LB 

.06 

.06 

UB 

.08 

.09 

SRMR 

.07 

.06 

Autonomy Need 
Satisfaction 

App Engagement 

Perceived 
Opportunity Cost 

App Usage 

Learning Gains 

Proposed 

Alternative 

66.5% 

66.4% 

66.5% 

68.3% 

7.1% 

6.5% 

18.8% 

14.1% 

8.1% 

4.7% 

Note. AIC = Akaike information criterion; BIC = Bayesian information criterion; SABIC = Sample-size adjusted Bayesian criterion.

 59 

 
 
Table 22 Normalized residuals for a structural model: local fit 

1 

2 

3 

4 

5 

6 

7 

8 

9 

10 

11 

12 

13 

14 

15 

16 

17 

1 

2 

0.00 

0.94 

0.00 

3  –0.49  –1.12  0.00 

4  –0.38  –0.01  0.42 

0.00 

5  –1.07  –0.31  0.46 

0.34 

0.00 

6  –0.32  0.92  –0.80  1.21 

0.37 

0.00 

7  –0.29  –0.19  1.24 

0.56  –0.06  –0.40  0.00 

8 

0.08  –0.54  2.48 

0.50 

0.24  –0.67  0.75 

0.00 

9  –0.32  –1.04  2.53  –0.32  –0.09  –1.02  –0.04  0.00 

0.00 

10  –0.66  –1.26  2.74 

0.08 

0.12  –0.92  0.37  –0.05  0.03 

0.00 

11  –0.76  0.29  –0.85  –0.35  –1.49  –0.54  –1.24  –1.53  –2.30  –2.37  0.00 

12  –0.15  0.59  –1.00  1.13  –0.14  –0.44  –1.14  –1.13  –1.10  –1.77  –0.13  0.00 

13  1.32 

1.11  –0.41  0.51  –1.03  –0.54  –1.79  –1.17  –1.30  –1.87  –0.10  0.21 

0.00 

14  0.42 

0.51  –0.44  0.89  –1.87  –1.08  –1.65  –1.51  –1.69  –2.41  0.14  –0.19  0.02 

0.00 

15  –0.42  –1.15  1.53 

0.37 

0.47  –0.25  0.13  –0.04  0.01 

0.81  –0.18  –1.05  0.13  –0.90  0.04 

16  0.57 

0.43 

2.52 

1.87 

1.84 

1.89 

2.40 

1.55 

1.41 

1.52  –0.66  1.27 

0.20 

0.78 

0.07 

0.02 

17  0.45  –0.33  0.67 

0.95 

1.12 

0.10 

1.47 

1.58 

1.72 

1.66  –1.72  –0.15  –0.59  0.03 

0.85 

0.29 

0.00 

Note. Residual values within the range of ±2.0 are acceptable; Items 1 to 4 = Support 1 to 4;  Items 5 to 7 = Autonomy 1 to 3; Items 8 
to 10 = Engagement 1 to 3; Items 11 to 14 = Cost 1 to 4; Item 15 = App usage time; Item 16 = Learning gains; Item 17 = Non-app 
English learning experiences.

 60 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Tables 23 and 24 present direct and indirect path coefficients from robust maximum likelihood 

estimation for the proposed structural model, respectively. Figure 24 demonstrates the proposed 

mediation model with standardized path coefficients, where solid and dashed lines indicate 

significant and non-significant paths at the p < .05 level, respectively. Each path is discussed in 

relation to its corresponding latent variables. 

Table 23 Robust maximum likelihood estimates for a structural model: direct effects 

Direct Effect Path 

b* 

95% CI 

UL 

LL 

SE 

z 

p 

® 

Autonomy Need  
Satisfaction 

.82 

.74 

.90 

.04 

19.92  < .001 

Perceived  
Teacher 
Autonomy 
Support 

®  App Engagement 

.42 

.18 

.66 

.12 

3.46  < .001 

® 

Perceived 
Opportunity Cost 

–.27 

–.42 

–.11 

.08 

–3.38  < .001 

®  App Usage 

.07 

–.08 

.21 

.08 

0.87 

.38 

®  App Engagement 

.44 

.18 

.69 

.13 

3.35  < .001 

®  App Usage 

.05 

–.09 

.19 

.07 

0.72 

.47 

®  App Usage 

.26 

.13 

.39 

.07 

3.92  < .001 

®  App Usage 

–.18 

–.27 

–.09 

.05 

–3.83  < .001 

Autonomy Need 
Satisfaction 

App  
Engagement 

Perceived  
Opportunity Cost 

App Usage 

®  Learning Gains 

.28 

.13 

.43 

.08 

3.68  < .001 

Non–App 
English Learning 

®  Learning Gains 

.06 

–.02 

.14 

.04 

1.41 

.16 

 61 

 
 
Table 24 Robust maximum likelihood estimates for a structural model: indirect effects 

Mediation Path 

T ® A ® E 

T ® A ® U 

T ® E ® U 

T ® A ® E ® U 

b* 

.36 

.04 

.11 

.09 

95% CI 

UL 

.15 

–.07 

.03 

.02 

LL 

.56 

.15 

.19 

.17 

SE 

z 

0.11 

0.06 

0.04 

0.04 

3.32 

0.72 

2.65 

2.50 

p 

.00 

.47 

.01 

.01 

T ® C ® U 
Note. T = Perceived teacher autonomy support; A = Autonomy need satisfaction; E = App 
engagement; C = Perceived cost opportunity; U = App usage; G = Learning gains. 

2.57 

0.02 

.08 

.01 

.05 

.01 

Autonomy
Need
Satisfaction

.435

App 
Engagement

The proportion of variance explained (R2)
• Autonomy Need Satisfaction: .665
• App Engagement: .665
• Perceived Opportunity Cost: .071
• App Usage: .188
• Learning Gain: .081

.815 

.421

.050

.262

Teacher 
Autonomy 
Support

.066

App Usage

.278

Learning Gain

–.267

Perceived
Opportunity 
Cost

–.177

.056

Non-App 
English
Learning
(covariate)

Figure 24 Moderation model. Error terms are not shown for simplicity. 

 62 

 
 
Effects of Perceived Teacher Autonomy Support 

Perceived teacher autonomy support positively affected autonomy need satisfaction (b* = .815, 

95% CI [.735, .896], p < .001) and app engagement (b* = .421, 95% CI [.182, .659], p = .001), 

supporting Hypotheses 1a and 1b, respectively. However, there was no direct effect of teacher 

autonomy support on app usage (b* = .066, 95% CI [–.082, .214], p = .383), rejecting 

Hypothesis 1c. These findings indicate that learners who perceive teachers’ instructional 

behaviors as autonomy-supportive tend to exhibit higher levels of personal ownership in app 

usage and greater app engagement. However, this perception of teacher autonomy support does 

not directly result in increased app usage. 

Additionally, teacher autonomy support negatively affected perceived opportunity cost of 

app usage (b* = –.267, 95% CI [–.421, –.112], p = .001), supporting Hypothesis 1d. This finding 

indicates that learners who perceive greater teacher autonomy support are less likely to view app 

usage as a trade-off with other valuable learning activities. 

Effects of Autonomy Need Satisfaction 

Autonomy need satisfaction had a positive effect on app engagement (b* = .435, 95% CI 

[.181, .690], p = .001), but its effect on app usage was not significant (b* = .050, 95% CI 

[–.086, .185], p = .472). These findings support Hypothesis 2a but reject Hypothesis 2b, 

respectively. These findings indicate that while learners who experience higher volition in app-

based language learning are more likely to engage with the app, they do not necessarily 

demonstrate greater app usage. 

Autonomy need satisfaction mediated the effect of teacher autonomy support to app 

engagement (i.e., teacher autonomy support → autonomy need satisfaction → app engagement) 

(b* = .355, 95% CI [.146, .564], p = .001), supporting Hypothesis 2c. However, autonomy need 

 63 

 
satisfaction did not mediate the effect of teacher autonomy support to app usage (i.e., teacher 

autonomy support → autonomy need satisfaction → app usage) (b* = .041, 95% CI 

[–.070, .151], p = .473), rejecting Hypothesis 2d. These findings indicate that learners who 

perceive greater teacher autonomy support are more likely to experience higher levels of 

autonomy need satisfaction, which, in turn, enhances their app engagement. However, the same 

effect was not observed for app usage. 

Effects of App Engagement  

App engagement positively affected app usage (b* = .262, 95% CI [.131, .393], p < .001). This 

finding supported Hypothesis 3a, indicating that learners with higher levels of app engagement 

tend to display greater app usage. 

App engagement mediated the effect of teacher autonomy support to app usage (i.e., 

teacher autonomy support → app engagement → app usage) (b* = .110, 95% CI [.029, .191], p 

= .008). In a serial mediation (i.e., teacher autonomy support → autonomy need satisfaction → 

app engagement → app usage), app engagement mediated the indirect effect of autonomy need 

support transmitted through autonomy need satisfaction on app usage (b* = .093, 95% CI 

[.020, .166], p = .012). These findings support Hypotheses 3b and 3c, respectively, indicating 

that learners who perceive teachers’ instructional behaviors as autonomy-supportive are more 

likely to engage with the app, which subsequently contributes to higher app usage, both directly 

and through enhanced autonomy need satisfaction (i.e., a sense of volition and self-endorsement 

in initiating and regulating their app usage beyond the classroom). 

 64 

Effects of Perceived Opportunity Cost 

Perceived opportunity cost negatively affected app usage (b* = –.177, 95% CI [–.267, –.086], p 

< .001), supporting Hypotheses 4a. This finding indicates that learners who perceive a higher 

opportunity cost—greater time and effort involved in app usage—are less likely to use it. 

However, perceived opportunity cost mediated the effect of teacher autonomy support to 

app usage (i.e., teacher autonomy support → perceived opportunity cost → app usage) (b* 

= .047, 95% CI [.011, .083], p = .010), supporting Hypothesis 4b. This finding indicates that 

learners who perceive greater teacher autonomy support are likely to view app usage as less of 

an opportunity cost, which, in turn, increases app usage. 

Effect of App Usage 

App usage had a positive effect on L2 learning gains (b* = .278, 95% CI [.130, .427], p < .001) 

after controlling for non-app English learning experiences as a covariate (b* = .056, 95% CI 

[–.022, .135], p = .158). This finding supported Hypothesis 5, indicating that learners who use 

the app more tend to achieve higher L2 learning gains, regardless of their non-app English 

learning experiences. 

Summary of the Findings 

Table 25 presents the summary of the findings. Instead of having a direct effect on app usage, 

perceived teacher autonomy support indirectly impacted it by enhancing autonomy need 

satisfaction and app engagement (i.e., full mediation). Additionally, higher perceived 

opportunity cost decreased app usage, but teacher autonomy support mitigated this negative 

effect. The four variables collectively explained about 18.8% of the variance in app usage, 

indicating a medium-large effect size. In turn, app usage was a significant predictor of 

 65 

 
vocabulary learning gains, explaining about 8.1% of its variance—a medium effect size—even 

after controlling for non-app English learning experiences beyond the classroom. 

Table 25 Summary of the findings 

Hypothesis 

Finding 

Meaning 

1a 

1b 

1c 

1d 

2a 

2b 

2c 

2d 

3a 

3b 

3c 

4a 

4b 

5 

Supported 

Positive effect of teacher support on autonomy need satisfaction 

Supported 

Positive effect of teacher support on app engagement 

Rejected 

Effect of teacher support on app usage not found 

Supported  Negative effect of teacher support on opportunity cost 

Supported 

Positive effect of autonomy need satisfaction on app engagement 

Rejected 

Effect of autonomy need satisfaction on app usage not found 

Supported  Mediation of autonomy need satisfaction to app engagement 

Rejected 

Mediation of autonomy need satisfaction to app usage not found 

Supported 

Positive effect of app engagement on app usage 

Supported  Mediation of app engagement to app usage 

Supported 

Serial mediation of app engagement to app usage 

Supported  Negative effect of opportunity cost on app usage 

Supported  Mediation of opportunity cost to app usage 

Supported 

Positive effect of app usage on L2 learning gains 

Cluster Analysis: Two Types of Initial English Learning Motivation 

As shown in Figure 25, nine of 24 clustering validity indices suggested that the optimal number 

of clusters was two, representing distinct initial L2 motivational orientations among young 

beginners in South Korea. Following the majority rule, I applied the two-cluster solution to 

hierarchical k-means clustering. The clusters in Figure 26 contain 117 and 137 learners, 

respectively. 

 66 

Figure 25 Optimal number of clusters for each validation index.

 67 

 
Figure 26 Results of hierarchical k-means clustering with two-cluster solution. 

Table 26 and Figure 27 summarize the means and 95% confidence intervals for the 

corresponding motivational constructs. Cluster 1 displayed positive responses in intrinsic and 

integrated motivation (i.e., +Autonomous), whereas Cluster 2 showed lower responses to these 

variables (i.e., –Autonomous). It is noteworthy that both clusters showed high levels of identified 

and external motivation (i.e., +Controlled), with relatively weaker introjected motivation. 

Additionally, Cluster 2 showed relatively higher amotivation for English learning. Taken 

together, Cluster 1 displays a balance of autonomous and controlled motivation for English 

learning, while Cluster 2 is predominantly driven by controlled motivation, with less reliance on 

autonomous reasons. I therefore labeled Cluster 1 as the +Autonomous/+Controlled motivation 

group, and Cluster 2 as the –Autonomous/+Controlled motivation group. 

 68 

 
 
 
Table 26 Descriptive statistics of English learning orientation scale by L2 motivation group 

Group 

Scale 

Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

Intrinsic 

4.28 

4.08 

4.49 

1.10 

0.10 

4 

Integrated 

4.26 

4.09 

4.44 

0.96 

0.09 

4.25 

1 

1 

+Autonomous/ 
+Controlled 
(n = 117) 

Identified 

5.16 

5.05 

5.28 

0.62 

0.06 

5 

3.67 

Introjected 

3.72 

3.50 

3.94 

1.19 

0.11 

3.75 

1 

External 

4.46 

4.29 

4.63 

0.94 

0.09 

4.50 

1.25 

–Autonomous/ 
+Controlled 
(n = 137) 

Amotivation  1.72 

1.60 

1.83 

0.62 

0.06 

1.67 

Intrinsic 

2.23 

2.10 

2.37 

0.80 

0.07 

2.33 

Integrated 

2.92 

2.76 

3.08 

0.93 

0.08 

Identified 

4.03 

3.86 

4.21 

1.03 

0.09 

3 

4 

Introjected 

3.27 

3.07 

3.47 

1.16 

0.10 

3.25 

External 

4.23 

4.07 

4.39 

0.97 

0.08 

4.25 

Amotivation  3.08 

2.92 

3.25 

0.99 

0.08 

3 

1 

1 

1 

1 

1 

1 

1 

6 

6 

6 

6 

6 

3.3 

4 

5 

6 

6 

6 

5.67 

Figure 27 Two groups of English learning orientation: means and 95% confidence intervals. 

 69 

 
 
 
 
Descriptive Statistics by L2 Motivation Group 

Perceived Teacher Autonomy Support and Autonomy Need Satisfaction 

Tables 27 and 28 present the mean scores for perceived teacher autonomy support and autonomy 

need satisfaction for app-based language learning out of class. As illustrated in Figure 28a, the 

+Autonomous/+Controlled motivation group reported higher levels of perceived teacher support 

across all scale items. However, overlapping 95% confidence intervals in autonomy need 

satisfaction in Figure 28b indicate less pronounced group differences. 

Table 27 Descriptive statistics of perceived teacher autonomy support by L2 motivation group 

Group 

Item 

Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

+Autonomous/ 
+Controlled 
(n = 117) 

–Autonomous/ 
+Controlled 
(n = 137) 

Support1 

4.26 

3.99 

4.54 

1.51 

0.14 

Support2 

4.41 

4.16 

4.66 

1.36 

0.13 

Support3 

3.50 

3.18 

3.83 

1.79 

0.17 

Support4 

4.21 

3.91 

4.52 

1.65 

0.15 

Support1 

3.69 

3.43 

3.95 

1.53 

0.13 

Support2 

3.85 

3.59 

4.10 

1.51 

0.13 

Support3 

3.21 

2.94 

3.49 

1.62 

0.14 

Support4 

3.65 

3.36 

3.94 

1.69 

0.14 

5 

5 

3 

5 

4 

4 

3 

4 

1 

1 

1 

1 

1 

1 

1 

1 

6 

6 

6 

6 

6 

6 

6 

6 

 70 

 
 
 
Table 28 Descriptive statistics of autonomy need satisfaction by L2 motivation group 

Group 

Item 

Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

+Autonomous/ 
+Controlled 
(n = 117) 

–Autonomous/ 
+Controlled 
(n = 137) 

Autonomy1 

3.97 

3.65 

4.28 

1.71 

0.16 

Autonomy2 

4.56 

4.30 

4.81 

1.41 

0.13 

Autonomy3 

3.58 

3.26 

3.90 

1.73 

0.16 

Autonomy1 

3.72 

3.45 

4.00 

1.62 

0.14 

Autonomy2 

4.04 

3.79 

4.29 

1.49 

0.13 

Autonomy3 

3.25 

2.97 

3.53 

1.64 

0.14 

5 

5 

4 

4 

5 

3 

1 

1 

1 

1 

1 

1 

6 

6 

6 

6 

6 

6 

Figure 28 Perceived teacher autonomy support and autonomy need satisfaction by L2 
motivation group: means and 95% confidence intervals. 

App Engagement 

Table 29 and Figure 29 display the mean scores for app engagement beyond the classroom 

across four engagement constructs. Overall, the +Autonomous/+Controlled motivation group 

reported higher levels of app engagement across all scale items, with only a small number of 

95% confidence intervals overlapping. 

 71 

 
 
Table 29 Descriptive statistics of app engagement by L2 motivation group 

Group 

Item 

Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

+Autonomous/ 
+Controlled 
(n = 117) 

–Autonomous/ 
+Controlled 
(n = 137) 

BE1 

BE2 

BE3 

EE1 

EE2 

EE3 

CE1 

CE2 

CE3 

AE1 

AE2 

AE3 

BE1 

BE2 

BE3 

EE1 

EE2 

EE3 

CE1 

CE2 

CE3 

AE1 

AE2 

AE3 

3.15 

2.83 

3.46 

1.71 

0.16 

3.61 

3.29 

3.93 

1.75 

0.16 

3.67 

3.34 

3.99 

1.76 

0.16 

3.50 

3.19 

3.80 

1.67 

0.15 

3.66 

3.36 

3.96 

1.65 

0.15 

3.64 

3.33 

3.95 

1.67 

0.15 

3.79 

3.51 

4.08 

1.58 

0.15 

3.37 

3.07 

3.66 

1.61 

0.15 

3.09 

2.80 

3.39 

1.61 

0.15 

3.26 

2.95 

3.56 

1.66 

0.15 

3.30 

3.00 

3.60 

1.63 

0.15 

3.57 

3.26 

3.88 

1.69 

0.16 

2.67 

2.40 

2.94 

1.59 

0.14 

2.98 

2.70 

3.26 

1.66 

0.14 

3.13 

2.86 

3.41 

1.63 

0.14 

2.69 

2.45 

2.94 

1.43 

0.12 

2.89 

2.64 

3.14 

1.48 

0.13 

2.88 

2.63 

3.12 

1.43 

0.12 

3.19 

2.92 

3.46 

1.58 

0.14 

2.83 

2.58 

3.09 

1.50 

0.13 

2.61 

2.35 

2.86 

1.49 

0.13 

2.65 

2.39 

2.90 

1.51 

0.13 

2.83 

2.57 

3.09 

1.53 

0.13 

2.91 

2.65 

3.18 

1.57 

0.13 

3 

4 

4 

4 

4 

4 

4 

3 

3 

3 

3 

4 

2 

3 

3 

2 

3 

3 

4 

3 

2 

2 

3 

3 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

1 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

6 

 72 

 
 
 
Figure 29 App engagement by L2 motivation group: means and 95% confidence intervals. The 
acronyms BE, EE, CE, and AE represent behavioral, emotional, cognitive, and agentic 
engagement, respectively. 

Non-App English Learning Experiences and Perceived Opportunity Cost 

Table 30 presents the mean scores for non-app English learning experiences out of class. Higher 

scores indicate greater and more intensive involvement in English skill sets. Among young 

beginners, listening emerged as the most frequently practiced activity beyond the classroom, 

while speaking, pronunciation and writing skills were the least practiced. Although the 

+Autonomous/+Controlled motivation group showed overall higher levels of learning activities, 

Panel A in Figure 30 reveals substantial overlap in 95% confidence intervals, suggesting no 

significant differences in non-app English learning patterns between the two L2 motivation 

groups.  

Panel B in Figure 30 and Table 31 show the mean scores for perceived opportunity cost 

of app usage. Overall, both groups perceived a similar level of opportunity cost across all scale 

items, with mean values ranging between 3 and 4, further indicating a comparable proportion of 

 73 

 
learners who viewed app usage as involving a significant loss of valued alternatives and those 

who did not. 

Table 30 Descriptive statistics of non-app English learning experiences by L2 motivation group 

Group 

Skill  Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

+Autonomous/ 
+Controlled 
(n = 117) 

–Autonomous/ 
+Controlled 
(n = 137) 

R 

L 

19.68 

14.16 

25.19 

30.10 

2.78 

23.64 

17.42 

29.86 

33.96 

3.14 

W 

13.93 

8.88 

18.99 

27.61 

2.55 

S 

G 

V 

P 

R 

L 

13.15 

8.19 

18.10 

27.07 

2.50 

13.79 

8.81 

18.78 

27.21 

2.52 

19.26 

13.47 

25.07 

31.67 

2.93 

11.74 

6.62 

16.85 

27.95 

2.58 

12.93 

9.49 

16.37 

20.36 

1.74 

19.94 

15.26 

24.62 

27.70 

2.37 

W 

9.60 

6.67 

12.52 

17.31 

1.48 

S 

G 

V 

P 

7.82 

5.04 

10.60 

16.44 

1.40 

12.58 

8.53 

16.64 

24.00 

2.05 

14.38 

10.63 

18.13 

22.18 

1.89 

9.18 

5.94 

12.42 

19.17 

1.64 

9 

12 

0 

0 

0 

6 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

0 

189 

189 

189 

189 

189 

189 

189 

127 

146 

114 

120 

142 

114 

114 

Note. R = Reading; L = Listening; W = Writing; S = Speaking; G = Grammar; V = Vocabulary; 
P = Pronunciation. 

 74 

Figure 30 Non-app English learning experiences and perceived opportunity cost by L2 
motivation group: means and 95% confidence intervals. R = Reading; L = Listening; W = 
Writing; S = Speaking; G = Grammar; V = Vocabulary; P = Pronunciation. 

Table 31 Descriptive statistics of perceived opportunity cost by L2 motivation group 

Group 

Item 

Mean 

95% CI 

LL 

UL 

SD 

SE  Median  Min.  Max. 

Cost1 

Cost2 

Cost3 

Cost4 

Cost1 

Cost2 

Cost3 

Cost4 

2.85 

2.62 

3.09 

1.31 

0.12 

3.74 

3.46 

4.01 

1.53 

0.14 

3.22 

2.98 

3.47 

1.34 

0.12 

3.00 

2.72 

3.28 

1.55 

0.14 

3.34 

3.09 

3.59 

1.49 

0.13 

3.76 

3.49 

4.03 

1.62 

0.14 

3.50 

3.24 

3.75 

1.52 

0.13 

3.32 

3.05 

3.59 

1.61 

0.14 

3 

4 

3 

3 

3 

4 

3 

3 

1 

1 

1 

1 

1 

1 

1 

1 

6 

6 

6 

6 

6 

6 

6 

6 

+Autonomous/ 
+Controlled 
(n = 117) 

–Autonomous/ 
+Controlled 
(n = 137) 

App Usage 

As demonstrated in Figure 31, two L2 motivation groups showed similar repetitive fluctuations 

overall, with no significant differences in the patterns of app usage between them. The red trend 

 75 

 
line indicating the average cumulative app usage over time in Figure 32 also reveals similar 

patterns across the two L2 motivation groups. 

Figure 31 Daily total app usage by L2 motivation group. 

Figure 32 Individual cumulative app usage over time by L2 motivation group. 

Table 32 presents the total app usage time by L2 motivation group. On average, the 

+Autonomous/+Controlled motivation group used the app for 1.76 hours (95% CI [1.32, 2.21]), 

while the –Autonomous/+Controlled motivation group used it for 1.55 hours (95% CI [1.16, 

 76 

 
 
1.93]). However, the mean values of both groups were skewed toward the lower end due to a 

subset of learners with minimal app usage, as shown in Figure 33. Notably, on average, the top 

10% of learners of each group used the app for 7.36 hours (95% CI [4.72, 10.00]) and 6.72 hours 

(95% CI [4.29, 9.15]) in the +Autonomous/+Controlled and –Autonomous/+Controlled 

motivation groups, respectively. 

Table 32 Descriptive statistics of app usage by L2 motivation group 

Group 

Mean 

95% CI 

LL 

UL 

A. +Autonomous/+Controlled motivation group 

SD 

SE  Median  Min.  Max. 

All (n = 117) 

1.76 

1.32 

2.21 

2.41 

Top 10% (n = 12) 

7.36 

4.72 

10.00 

4.16 

0.22 

1.20 

B. –Autonomous/+Controlled motivation group 

All (n = 135) 

1.55 

Top 10% (n = 14) 

6.72 

1.16 

4.29 

1.93 

9.15 

2.27 

4.21 

0.19 

1.12 

1.01 

5.94 

0.98 

4.65 

0 

19.70 

4.51 

19.70 

0 

16.74 

3.63 

16.74 

Figure 33 Distribution of app usage data by L2 motivation group. 

 77 

 
 
 
Meaning Recall Test 

Table 33 and Figure 34 present the results of the meaning recall test. While the 

+Autonomous/+Controlled motivation group showed a tendency to know more target words than 

the –Autonomous/+Controlled motivation group on the pretest, both groups demonstrated 

improvement on the posttest. The mean gain score for the +Autonomous/+Controlled motivation 

group was 4.90 (95% CI [4.05, 5.75]), which was higher than that of 2.99 (95% CI [2.30, 3.69]) 

for the –Autonomous/+Controlled motivation group. 

Table 33 Descriptive statistics of meaning recall test by L2 motivation group 

Group 

Time  Mean 

95% CI 

LL 

UL 

A. +Autonomous/+Controlled motivation group 

SD 

SE  Median  Min.  Max. 

All 

Pretest 

6.88 

5.91 

7.85 

5.31 

0.49 

(n = 117)  Posttest 

11.78 

10.40 

13.20 

7.54 

0.70 

7 

11 

Top 10%  Pretest 

9.25 

6.14 

12.36 

4.9 

1.41 

8.50 

(n = 12) 

Posttest 

18.58 

14.37 

22.80 

6.64 

1.92 

19 

B. –Autonomous/+Controlled motivation group 

All 

Pretest 

4.53 

(n = 137)  Posttest 

7.53 

Top 10%  Pretest 

3.5 

3.71 

6.40 

1.07 

5.35 

8.65 

5.93 

4.85 

0.41 

6.68 

0.57 

3 

6 

4.2 

1.12 

2.50 

(n = 14) 

Posttest 

8.36 

4.23 

12.48 

7.14 

1.91 

7 

0 

0 

2 

6 

0 

0 

0 

0 

23 

28 

16 

28 

19 

25 

14 

21 

 78 

Figure 34 Meaning recall test scores by L2 motivation group: means and 95% confidence 
intervals. 

Multi-Group Measurement Model 

The four-factor multi-group measurement model, incorporating the two L2 motivation groups, 

demonstrated robust global fit (scaled χ²(160) = 249.55, p < .001; CFI = .96; TLI = .95; RMSEA 

= .07, 90% CI [.05, .09]; SRMR = .05). As for each L2 motivation group, local fit was supported 

with only four cases showing residuals exceeding ±2.0 (see Tables 34 and 35, respectively for 

each L2 motivation group). Additionally, all standardized factor loadings were significant (p 

< .001) and consistently high across items, with the lowest loading being .56 for Cost 1 in the 

+Autonomous/+Controlled motivation group (see Tables 36 and 37, respectively for each L2 

motivation group).

 79 

 
Table 34 Normalized residuals for a measurement model: local fit of the +Autonomous/+Controlled motivation group 

Variable 

1 

2 

3 

4 

5 

6 

7 

8 

9 

10 

11 

12 

13 

14 

1. Support1 

0.00 

2. Support2 

0.35 

0.00 

3. Support3 

–0.28  –0.72 

0.00 

4. Support4 

–0.13 

0.01 

0.22 

0.00 

5. Autonomy1 

–0.73  –0.27 

0.82 

0.27 

0.00 

6. Autonomy2 

–0.01 

0.93 

0.75 

0.68 

0.10 

0.00 

7. Autonomy3 

–0.33  –0.39 

1.62 

0.20 

0.11 

–0.53 

0.00 

8. Engagement1 

–0.14  –0.50 

2.04 

–0.53  –0.67  –0.52 

0.06 

0.00 

9. Engagement2 

–0.07  –0.65 

1.44 

–0.32  –0.01  –0.10 

0.05 

0.17 

0.00 

10. Engagement3 

0.20 

–0.13 

1.65 

–0.23 

0.04 

0.28 

0.25 

–0.04  –0.04 

0.00 

11. Cost1 

12. Cost2 

13. Cost3 

14. Cost4 

–0.99  –0.61  –1.48  –1.27  –1.18  –0.89  –0.81  –1.05  –0.83  –1.28 

0.00 

–1.17  –0.10  –0.77 

0.03 

0.72 

–0.52  –0.22  –0.56  –0.26  –0.15 

0.13 

0.00 

0.11 

1.20 

–1.09 

0.35 

0.53 

0.63 

0.36 

0.03 

0.50 

0.22 

–0.43 

0.40 

0.00 

0.26 

0.54 

–0.05 

0.05 

–0.52 

0.06 

0.11 

0.58 

0.40 

0.19 

0.18 

–0.41 

0.09 

0.00 

Note. Residual values within the range of ±2.0 are acceptable. 

 80 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 35 Normalized residuals for a measurement model: local fit of the –Autonomous/+Controlled motivation group 

Variable 

1 

2 

3 

4 

5 

6 

7 

8 

9 

10 

11 

12 

13 

14 

1. Support1 

0.00 

2. Support2 

0.56 

0.00 

3. Support3 

–0.14  –0.75 

0.00 

4. Support4 

–0.44  –0.22 

0.85 

0.00 

5. Autonomy1 

–0.69  –0.16 

0.24 

0.50 

0.00 

6. Autonomy2 

–0.61 

0.12 

–1.57 

1.04 

0.32 

0.00 

7. Autonomy3 

0.11 

0.22 

0.68 

0.97 

–0.21  –0.01 

0.00 

8. Engagement1 

–0.64  –0.86 

1.81 

–0.58  –0.20  –1.87  –0.08 

0.00 

9. Engagement2 

0.17 

0.53 

2.50 

0.89 

1.20 

–0.73 

0.46 

–0.05 

0.00 

10. Engagement3 

–0.52  –0.80 

1.77 

0.21 

0.41 

–1.18  –0.20 

0.09 

–0.07 

0.00 

11. Cost1 

12. Cost2 

13. Cost3 

14. Cost4 

–0.34 

0.74 

–0.41 

0.38 

0.43 

1.41 

0.32 

–0.13  –0.32 

0.18 

0.00 

–0.13 

0.04 

–1.46 

0.56 

0.50 

0.85 

–0.15 

0.16 

–0.39 

0.75 

–0.12 

0.00 

1.13 

0.01 

–0.25  –0.20  –0.22 

0.10 

–1.17  –0.16 

0.23 

0.17 

0.22 

0.02 

0.00 

–0.22  –0.31  –1.27 

0.56 

–0.26  –0.02  –0.64  –0.52  –0.49 

0.15 

–0.02 

0.14 

–0.20 

0.00 

Note. Residual values within the range of ±2.0 are acceptable.

 81 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 36 Robust maximum likelihood estimates for a four-factor measurement model of the 
+Autonomous/+Controlled motivation group 

Parameter 

b 

SE 

b* 

SE 

z 

Perceived Teacher Autonomy Support 

Factor Loadings 

Support1 

Support2 

Support3 

Support4 

Autonomy Need Satisfaction 

Autonomy1 

Autonomy2 

Autonomy3 

App Engagement 

Engagement1 

Engagement2 

Engagement3 

Perceived Opportunity Cost 

Cost1 

Cost2 

Cost3 

Cost4 

Support1 

Support2 

Support3 

Support4 

Autonomy1 

Autonomy2 

Autonomy3 

1.00 

0.91 

1.01 

1.07 

1.00 

0.60 

0.98 

1.00 

1.04 

1.08 

1.00 

1.30 

1.46 

1.65 

0.08 

0.09 

0.06 

0.10 

0.06 

0.05 

0.05 

0.32 

0.30 

0.29 

Error (Unique) Variances 

0.14 

0.13 

0.17 

0.15 

0.16 

0.13 

0.16 

4.27 

4.41 

3.50 

4.21 

3.97 

4.56 

3.58 

 82 

0.88 

0.89 

0.75 

0.86 

0.90 

0.66 

0.87 

0.89 

0.93 

0.96 

0.56 

0.62 

0.80 

0.77 

2.84 

3.26 

1.97 

2.56 

2.33 

3.24 

2.08 

0.04 

0.05 

0.04 

0.03 

0.03 

0.08 

0.04 

0.03 

0.02 

0.02 

0.07 

0.10 

0.09 

0.06 

0.23 

0.30 

0.11 

0.20 

0.18 

0.34 

0.14 

22.03 

17.32 

18.10 

26.05 

30.26 

8.33 

24.00 

31.96 

39.78 

57.10 

7.51 

6.27 

8.68 

13.54 

12.15 

10.82 

17.35 

12.64 

12.65 

9.67 

15.35 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 36 (cont’d) 

Engagement1 

Engagement2 

Engagement3 

Cost1 

Cost2 

Cost3 

Cost4 

3.50 

3.66 

3.64 

2.86 

3.74 

3.22 

3.00 

Perceived Teacher Autonomy Support 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

Perceived Teacher Autonomy Support 

Factor Variances 

1.75 

2.34 

2.18 

0.53 

Covariances 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

App Engagement 

1.67 

1.48 

–0.24 

1.79 

–0.43 

0.15 

0.15 

0.15 

0.12 

0.14 

0.12 

0.14 

0.30 

0.31 

0.25 

0.18 

0.26 

0.22 

0.14 

0.25 

0.17 

2.10 

2.23 

2.19 

2.19 

2.46 

2.42 

1.94 

1.00 

1.00 

1.00 

1.00 

0.83 

0.76 

–0.25 

0.79 

–0.38 

0.13 

0.15 

0.15 

0.13 

0.14 

0.13 

0.10 

— 

— 

— 

— 

0.06 

0.05 

0.12 

0.06 

0.12 

16.13 

14.95 

14.97 

17.08 

17.13 

18.06 

18.66 

— 

— 

— 

— 

14.95 

15.75 

–2.07 

12.31 

–3.34 

Perceived Opportunity Cost 

–0.34 

0.14 

–0.32 

0.11 

–2.89 

 83 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 37 Robust maximum likelihood estimates for a four-factor measurement model of the –
Autonomous/+Controlled motivation group 

Parameter 

b 

SE 

b* 

SE 

z 

Perceived Teacher Autonomy Support 

Factor Loadings 

Support1 

Support2 

Support3 

Support4 

Autonomy Need Satisfaction 

Autonomy1 

Autonomy2 

Autonomy3 

App Engagement 

Engagement1 

Engagement2 

Engagement3 

Perceived Opportunity Cost 

Cost1 

Cost2 

Cost3 

Cost4 

Support1 

Support2 

Support3 

Support4 

Autonomy1 

Autonomy2 

Autonomy3 

1.00 

1.00 

0.88 

1.03 

1.00 

0.71 

0.91 

1.00 

1.01 

1.01 

1.00 

1.20 

1.10 

1.26 

0.05 

0.07 

0.06 

0.10 

0.06 

0.05 

0.04 

0.15 

0.15 

0.16 

Error (Unique) Variances 

0.13 

0.13 

0.14 

0.14 

0.14 

0.13 

0.14 

3.69 

3.85 

3.21 

3.65 

3.72 

4.04 

3.25 

 84 

0.86 

0.88 

0.72 

0.81 

0.88 

0.69 

0.79 

0.94 

0.92 

0.95 

0.70 

0.77 

0.74 

0.81 

2.42 

2.56 

1.98 

2.17 

2.31 

2.72 

1.98 

0.04 

0.03 

0.05 

0.04 

0.04 

0.07 

0.04 

0.02 

0.03 

0.02 

0.06 

0.06 

0.07 

0.06 

0.16 

0.18 

0.11 

0.13 

0.15 

0.20 

0.11 

24.47 

31.61 

13.59 

19.77 

24.07 

9.54 

19.38 

54.73 

35.51 

43.33 

11.25 

12.60 

11.19 

12.97 

15.13 

14.00 

18.62 

16.42 

15.53 

13.91 

18.85 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 37 (cont’d) 

Engagement1 

Engagement2 

Engagement3 

Cost1 

Cost2 

Cost3 

Cost4 

2.69 

2.89 

2.88 

3.34 

3.76 

3.50 

3.32 

Perceived Teacher Autonomy Support 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

Perceived Teacher Autonomy Support 

Factor Variances 

1.74 

2.04 

1.79 

1.06 

Covariances 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

Autonomy Need Satisfaction 

App Engagement 

Perceived Opportunity Cost 

App Engagement 

1.46 

1.16 

–0.23 

1.24 

–0.42 

0.12 

0.13 

0.12 

0.13 

0.14 

0.13 

0.14 

0.23 

0.26 

0.21 

0.22 

0.22 

0.16 

0.16 

0.19 

0.16 

1.89 

1.96 

2.02 

2.26 

2.33 

2.30 

2.07 

1.00 

1.00 

1.00 

1.00 

0.78 

0.66 

–0.17 

0.65 

–0.29 

0.10 

0.10 

0.11 

0.12 

0.14 

0.13 

0.11 

— 

— 

— 

— 

0.07 

0.06 

0.11 

0.07 

0.10 

19.83 

19.01 

19.12 

19.28 

16.82 

18.26 

19.17 

— 

— 

— 

— 

11.90 

11.55 

–1.49 

9.35 

–2.86 

Perceived Opportunity Cost 

–0.41 

0.14 

–0.30 

0.09 

–3.19 

 85 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
The findings in Table 38 also confirmed the strong reliability and validity of the four-factor 

measurement model for each L2 motivation group. The ω values exceeded the recommended 

threshold of .70 for all latent variables, indicating good construct reliability. Convergent validity 

was supported by AVE, which surpassed the .50 threshold for all constructs. Additionally, 

perceived teacher autonomy support, autonomy need satisfaction, and app engagement were 

positively correlated with one another, while all four latent variables showed negative 

correlations with perceived opportunity cost. The square roots of AVE were consistently higher 

than the correlations with other constructs, confirming discriminant validity. These results 

collectively provide robust evidence for the hypothesized factor structure and unidimensionality 

of the four latent variables for both L2 motivation groups. 

Table 38 Reliability and validity of the measurement model by L2 motivation group 

ω 

AVE 

1 

2 

3 

4 

A. +Autonomous/–Controlled Motivation Group 

1. Perceived Teacher Autonomy Support 

2. Autonomy Need Satisfaction 

3. App Engagement 

4. Perceived Opportunity Cost 

.90 

.87 

.95 

.79 

B. –Autonomous/–Controlled Motivation Group 

1. Perceived Teacher Autonomy Support 

2. Autonomy Need Satisfaction 

3. App Engagement 

.89 

.84 

.95 

.70 

.69 

.86 

.49 

.67 

.63 

.87 

.84 

.83 

.76 

.83 

.79 

.93 

–.25 

–.38 

–.32 

.70 

.82 

.77 

.66 

.80 

.65 

.93 

4. Perceived Opportunity Cost 
Note. The shaded values represent the square roots of the AVE for each latent variable. The 
correlations between latent variables are listed below the shaded diagonal. 

–.16 

–.30 

–.28 

.84 

.58 

.76 

 86 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Table 39 shows the results of measurement invariance tests across the two L2 motivation groups. 

It was found that only configural and weak invariance models were retained (i.e., p > .05), 

meaning that the relationships between the items and the latent construct were equivalent across 

two L2 motivation groups. However, strong invariance was not supported (i.e., p = .03), 

suggesting that item intercepts differed between the L2 motivation groups. 

Consequently, while the +Autonomous/+Controlled motivation group descriptively 

showed higher mean scores for perceived teacher support and app engagement than the –

Autonomous/+Controlled motivation group as shown Figures 25 and 26, such mean comparisons 

cannot be reliably interpreted as true differences due to the presence of potential group-specific 

response bias, which systematically affects observed scores in a particular group apart from 

respondents’ true levels on the factor (Kline, 2023). 

Therefore, I only compared path coefficients and the structural relationships between the 

two L2 motivation groups in the subsequent multi-group structural model.

 87 

Table 39 Invariance (multi–group) models across the L2 motivation groups 

Model 

χ² 

df 

p 

Δχ² 

Δdf 

p 

CFI 

TLI 

RMSEA 

Configural  275.41 

Weak 

283.87 

Strong 

299.81 

142 

152 

162 

< .001 

< .001 

9.09 

< .001 

19.64 

10 

10 

.52 

.03 

0.97 

0.97 

0.96 

0.96 

0.96 

0.96 

0.07 

0.07 

0.07 

RMSEA 90% CI 

LB 

0.05 

0.05 

0.05 

UB 

0.09 

0.09 

0.09 

SRMR 

0.06 

0.06 

0.06 

 88 

 
 
 
 
Multi-Group Structural Model of Moderated Mediation for Research Question 6 

In this multi-group structural model, the group variable, L2 motivation, was included as a 

moderator in the proposed mediation model shown in Figure 6. To examine differences in path 

coefficients between the +Autonomous/+Controlled and the –Autonomous/+Controlled 

motivation groups, I first built a constrained model in which the path coefficients were fixed to 

be equal across the two L2 motivation groups. Next, I also built an unconstrained model in 

which the path coefficients were allowed to vary freely between the groups. 

Tables 40 and 41 present the result of the model comparison and its meaning. A chi-

square difference test comparing these two models yielded non-significant results (Δχ² = 5.41, p 

= .862), suggesting that constraining the path coefficients to be equal did not significantly 

worsen model fit (i.e., the strengths and directions of the effects were comparable between the 

L2 motivation groups). Therefore, Hypothesis 6 is rejected, indicating that teacher autonomy 

support functioned similarly regardless of learners’ L2 motivation types existing prior to app-

based language learning. 

Table 40 Result of the constrained and unconstrained model comparison 

Model 

χ² 

Constrained 

392.94 

Unconstrained 

387.55 

df 

234 

224 

AIC 

BIC 

Δχ² 

Δdf 

p 

16050 

16410 

16064 

16460 

5.41 

10 

0.86 

Table 41 Summary of the finding 

Hypothesis 

Finding 

Meaning 

6 

Rejected 

Effect of initial L2 motivation on the structural relationships not 
found 

 89 

 
 
 
 
 
 
DISCUSSION 

In this study, I explored the effects of teacher autonomy support on young beginners’ app-based 

language learning in a self-study context. Despite overall low app usage, I found that teacher 

autonomy support influenced out-of-class app usage in two main ways: (a) by enhancing 

autonomy need satisfaction for app usage and (b) by reducing perceived opportunity cost of app 

usage. Notably, the study found that these pathways were equally effective for learners with 

controlled L2 motivation. These findings are discussed through the lens of self-determination 

theory. 

Enhancing a Positive Self-Reinforcing Loop by Supporting Intrinsic Motivation 

As outlined in SDT (Ryan & Deci, 2017, 2020) and Noels et al.’ (2019a) L2 motivation process 

model, I found a positive self-reinforcing loop, where teacher autonomy support initiates a series 

of interlinked instructional pathways for facilitating app-based language learning beyond the 

classroom: teacher autonomy support → autonomy need satisfaction → app engagement → app 

usage → L2 learning. That is, learners who perceived teachers’ instructional behaviors as 

autonomy-supportive were more likely to experience greater volition and personal ownership in 

app usage. This, in turn, enhanced their app engagement and increased their actual app usage out 

of class, ultimately leading to improved learning gains. This finding provides empirical support 

for Noel et al.’s (2019) framework, highlighting its relevance and applicability to technology-

mediated L2 learning beyond the classroom. 

The effects of teacher autonomy support can be attributed to its emphasis on supporting 

intrinsic motivation for app-based language learning beyond the classroom (ASIB #2 Inviting 

students to pursue their interests, and ASIB #3 Presenting learning activities in need-satisfying 

ways, in Reeve & Cheon, 2021). In this study, the teachers supported learners using the app in 

 90 

ways that aligned with their personal interests in app-based language learning. Specifically, 

learners were encouraged to freely explore app features and activities that piqued their curiosity 

and interest (e.g., flashcards, automatic speech recognition, lesson topics) rather than strictly 

following predetermined learning paths and goals (e.g., completing all activities in one lesson 

before moving to the next). Through this process, learners could experiment with diverse ways 

of using the app (e.g., focusing on pronunciation practice with sound wave visualization, reading 

grammar notes) and might have felt that their behaviors and decisions were guided by their own 

interests, goals, and preferences, fostering a sense of volition and ownership in app-based 

language learning beyond the classroom. This sense of choice fuels intrinsic motivation for self-

directed app usage, as learners can engage in in-app activities they find enjoyable and 

meaningful. Consequently, these types of teacher autonomy support, which provided learners 

with maximal flexibility to engage with the app based on their interest in it, enabled them to 

integrate app-based language learning into their intrinsic goal pursuit. This, in turn, may have 

created more frequent opportunities for sustained effort and meaningful progress in their learning 

through app usage (Reeve, 2022b). To summarize, teacher autonomy support in this study might 

have promoted learners’ intrinsic motivation for app-based language learning by fostering a 

sense of volition and personal ownership, which in turn empowered them to personalize their 

app usage in ways that aligned with their interests in app-based language learning. 

These findings may extend beyond the particular app used in this study, highlighting the 

critical role of classroom-based teacher autonomy support in facilitating the integration of a 

broader range of digital resources into L2 learning beyond the classroom. This is because L2 

learning beyond the classroom “does not exclude the classroom but rather connects with it” (p. 

563, Reinders & Benson, 2017). Indeed, while many self-study L2 learning tools are readily 

 91 

available, learners still expect teacher support for their out-of-class learning (Lai et al., 2016). 

Furthermore, to fully benefit from these resources, learners are required to have necessary skills 

and experiences to use them effectively (e.g., self-regulated learning in García Botero et al., 

2021). However, since teachers do not directly guide out-of-class learning as they do in the 

classroom, this support and relevant training should begin in the classroom. Particularly, for 

most young beginner learners, classrooms may serve as the primary and, sometimes, the only 

place where they can receive proper need support (Oga-Baldwin, 2022). In the classroom, 

teachers can systematically implement diverse instructional strategies that promote students’ 

intrinsic motivation for technology use—voluntarily engaging with technologies for L2 learning 

in ways that bring them inherent satisfactions and joys, rather than using it due to external 

pressure. This foundational classroom-based autonomy support may first empower learners to 

take personal ownership of their L2 learning with digital tools, further enabling them to benefit 

from autonomy need satisfaction through technology use (i.e., resource-based autonomy). 

Reducing Perceived Opportunity Cost by Supporting Internalization 

Aligned with the expectancy-value(-cost) model (e.g., Eccles & Wigfield, 2020), learners were 

less inclined to use the app if they felt that it was not worth the value lost from other important 

activities (Hypothesis 4a supported). In South Korea’s high-pressure academic context, this 

might be particularly relevant to the heavy reliance on hagwons, which are seen as reliable 

investments for test success due to their emphasis on test-prep English learning activities. In 

contrast, language learning apps that focus on broader language skills may be perceived as less 

relevant for immediate test performance and, therefore, a less effective use of students’ limited 

resources (e.g., time, energy). As emphasized in Noels et al.’s (2019a) L2 motivation model, 

which situates motivational dynamics in particular socio-structural and socio-cultural contexts 

 92 

(as in Figure 1), these findings highlight the importance of considering unique educational 

contexts when implementing app-based language learning, where apps must compete with well-

established learning methods. 

More importantly, higher levels of perceived teacher autonomy support were associated 

with reduced perceived trade-off of using the app (Hypothesis 1d supported). Moreover, the 

positive mediation effect of perceived opportunity cost (i.e., the product of two negative 

coefficients) suggests that teacher autonomy support may help learners evaluate app usage as 

less demanding, thereby making them more likely to use it (Hypothesis 4b supported). 

According to SDT, this effect may be due to the role of teacher autonomy support in facilitating 

internalization—helping learners discover the value and personal relevance of an activity (ASIB 

#4 Providing explanatory rationales, ASIB #5 Acknowledging negative feelings, ASIB #6 

Relying on invitational language, and ASIB #7 Displaying patience, in Reeve & Cheon, 2021). 

Initially, app-based language learning may have perceived a high opportunity cost since the app 

was externally introduced by teachers, rather than emerging from learners’ intrinsic interests. 

However, teachers mitigated this sense of opportunity cost by framing the app as a valuable self-

study tool relevant to their English learning goals. Teachers also responded to learners’ 

challenges during app usage (e.g., technical glitches) and praised their efforts to alleviate 

negative feelings (e.g., anxiety, frustration, confusion) that could potentially overwhelm their 

volitional motivation for app-based language learning. By lowering perceived opportunity cost, 

teachers might create internalization-enabling opportunities that prepare learners motivationally 

ready to accept app-based language learning as a self-endorsed practice, further enabling them to 

engage with and benefit from it. 

 93 

 
These findings suggest a meaningful intersection between the expectancy-value(-cost) 

model and SDT, offering insights into why and how learners choose which technologies for their 

L2 learning in self-study settings. The findings also highlight the critical role teachers play in 

supporting this decision-making process in relation to the value (and cost) of technology use 

within the constraints of local educational contexts. 

Controlled Type of L2 Motivation as a Primary Driver in a Test-Driven Context 

Diverging from previous studies (e.g., Loewen et al., 2020; Sudina & Plonsky, 2024), this study 

did not find clear effects of initial L2 motivation on app-based language learning. This result 

might be attributed to different methodological approaches to conceptualizing motivation and 

engagement across studies. Specifically, this study adopted the view that motivation represents 

the initial desire or intention that energizes learning, whereas engagement is the goal-directed 

and purpose-driven enactment of this energy (i.e., the action component of motivation) (Hiver et 

al., 2021, 2024; Skinner & Raine, 2022). Engagement, in this framework, refers to the intensity 

and quality of what actively involved learners would do (behavioral), feel (affective), think 

(cognitive), and how they take an active role (agentic) during the learning process. In this regard, 

as posited in Noels et al.’s (2019a) L2 motivational process model, motivation serves as an 

antecedent to engagement (see Reschly & Christenson, 2022, for a comprehensive discussion on 

the relationship between motivation and engagement). In this context, the questionnaires used to 

measure L2 motivation in previous research focused on learning behavior (e.g., I work hard at 

studying French/Spanish in Sudina & Plonsky, 2024) or interest (e.g., I am interested in learning 

Spanish in Loewen et al., 2020), aligning more closely with the concept of engagement 

(behavior and affective engagement, respectively, in this study) rather than motivation itself. In 

contrast, the present study conceptualized L2 motivation within the framework of SDT as 

 94 

learners’ willingness to take ownership—the initial desire or intention—distinct from learning in 

action. This approach conceptually distinguished motivation from engagement. Consequently, 

the different methodological approaches made it challenging to compare the possible effects of 

initial L2 motivation on app-based language learning in self-study contexts. 

In this study’s context, I identified two types of initial L2 motivation that learners carried 

into app-based language learning (i.e., +Autonomous/+Controlled and –

Autonomous/+Controlled), both of which were driven by controlled types of motivation. 

Notably, it was initially hypothesized that learners with more autonomous L2 motivation would 

benefit more from teacher autonomy support due to its potential advantages in leading to the 

enactment of better learning behaviors and greater effort when learners are autonomously 

motivated (cf. Reeve, 2022b; Ryan & Deci, 2020). However, I found that teacher autonomy 

support was consistently effective regardless of L2 motivation type (Hypothesis 6 rejected). 

These findings may reflect unique socio-ecological contexts like South Korea, where English 

learning is a high-stakes activity tied to exams and future academic or career opportunities. In 

such contexts, controlled motivation often plays a prominent role in driving learning behaviors, 

not necessarily undermining the potential for autonomous motivation to co-exist or develop (e.g., 

Al-Hoorie, 2024). Especially for after-early-childhood learners, extrinsic motivation gradually 

becomes more prevalent as they progress through school grades, while their intrinsic 

motivation—the inherent desire to learn for interest and enjoyment—declines, potentially 

limiting the impact of autonomous types of motivation (Lepper et al., 2005; Ryan & Deci, 2017). 

Evidence of this observation can be seen in the strong influence of identified motivation—an 

autonomous extrinsic form of motivation—on English learning of both L2 motivation groups 

(Figure 27). Identified motivation, where learners engage in activities because they recognize 

 95 

and value their importance (e.g., Because English helps my personal growth and development), 

represents a transitional stage where externally imposed goals (e.g., social expectations, test 

scores) begin to align with a learner’s internal values and sense of self. This form of motivation 

primarily drives English learning in the Korean context, attenuating the possible advantages of 

more autonomous types of motivation. 

Nonetheless, the findings highlight the universal applicability and benefits of teacher 

autonomy support, even for young learners predominantly driven by controlled motivation—a 

group likely to represent a significant portion of learners in typical English classrooms. The 

positive impact of teacher autonomy support on students driven by controlled L2 motivation is 

particularly critical in facilitating technology-mediated L2 learning beyond the classroom. That 

is, teacher autonomy support can help learners consider digital tools as personally meaningful 

and relevant resources for effective self-directed L2 learning, fostering a sense of ownership and 

volition to use them beyond the classroom. Following that, learners with controlled motivation 

can self-create opportunities through technology use to support their autonomy need for L2 

learning in general (i.e., resource-based autonomy). Over time, such experiences of autonomy 

need satisfaction can gradually shift their L2 motivational orientation, transforming externally 

regulated forms of motivation into more autonomous and self-determined types, facilitating a 

positive cycle of the L2 motivational process. 

Pedagogical Implications 

Persistence poses a considerable obstacle to successful app-based language learning in self-study 

contexts (e.g., García Botero et al., 2019; Hwang et al., 2024; Jeon, 2022). In this study, only a 

small group of learners meaningfully used the app beyond the classroom. This highlights the 

importance of understanding how real-world situations (e.g., holidays, school events, 

 96 

extracurricular activities) can affect app usage. In this regard, teachers are advised to set realistic 

expectations for app-based language learning and determine which learners are most likely to 

benefit from these tools. 

However, as shown in Panel A in Figure 17, it is promising that the app usage pattern did 

not display a steep drop-out phase followed by a consistent decline over time. Instead, app usage 

among young beginners displayed more dynamic patterns, with peaks and valleys, indicating that 

they repetitively returned to the app after pauses, rather than failing to resume app usage once it 

was disrupted. The cyclical phases observed in this study align with the concept of persistence 

proposed by Hwang et al. (2024), who defined persistence as a multidimensional process 

involving engagement, disengagement, dormancy, and reengagement, not merely maintaining 

uninterrupted app usage sessions or days. In this regard, teacher autonomy support may have 

played a crucial role in developing learners’ resilience against dropout, thereby enabling them to 

make consistent efforts to resume and sustain app usage. 

For young learners who may not yet be ready for self-directed learning beyond the 

classroom, classroom-based teacher support is essential because simply providing access to apps 

does not ensure effective use. When providing such support, it is important that teacher support 

avoids taking the form of controlling behaviors that push learners toward specific ways or 

outcomes. Instead, teacher support should help learners take ownership of their learning and 

must be genuinely perceived and accepted by them. One effective approach is through 

autonomy-supportive instructional behaviors, which involve providing structured support (Ryan 

& Deci, 2020). Good structure can scaffold learning by setting clear expectations and goals, 

maintaining consistent rules and guidelines, and providing positive and efficacy-oriented 

feedback. With teacher autonomy support in structure, teachers can help learners gradually 

 97 

 
develop volition and ownership for effective app-based language learning beyond the classroom, 

leading to an enhancement of higher engagement (behavioral, affective, cognitive, and 

agentic)—both in quantity and quality. For instance, as in He and Loewen (2022), teachers can 

encourage learners to make key decisions about app usage, aligning in-app activities with their 

personal goals and interests in L2 learning. Engaging learners in meaningful decision-making 

processes (e.g., setting realistic app usage goals and schedules), combined with providing 

feedback on their goal progress, can lead to greater goal achievement. This process fosters self-

endorsed accountability driven by intrinsic motivation for app-based language learning rather 

than externally imposed motives. 

Additionally, reducing perceived opportunity cost is also vital for promoting app-based 

language learning beyond the classroom. Teachers can present language apps as tools that 

complement, rather than compete with or replace, existing learning methods to which learners 

are already committed. Instead of expecting learners to use every feature within an app, teachers 

can help them focus on specific functions that effectively address gaps in their L2 learning. For 

example, learners might use a sound recognition and pattern comparison feature to improve 

speaking and pronunciation skills, as these areas are often less practiced with non-app learning 

resources (Panel A in Figure 15). By demonstrating how targeted app use contributes to 

achieving broader L2 learning goals, teachers can help learners perceive apps as a valuable and 

worthwhile investment of their time. 

To summarize, effective integration of language learning apps beyond the classroom 

requires structured classroom-based teacher autonomy support, including thoughtful planning 

and consideration of unique local contexts. More importantly, a single training session is 

inadequate and insufficient; instead, it is essential to create an autonomy-supportive learning 

 98 

climate where autonomy support can be delivered consistently and reliably throughout the 

semester. 

Limitations and Future Directions 

The findings of this study come with several limitations. First, while the study explored the role 

of teacher autonomy support, it did not directly examine its effect through a structured 

intervention design. In this study, two teachers provided the same types of autonomy support in a 

similar manner to minimize teacher-related differences for research purposes. However, in real-

world settings, teachers might adopt varying approaches and attitudes toward autonomy support. 

Future research could compare the effects of autonomy-supportive teaching styles with 

controlling teaching styles. Such a comparison would provide a clearer understanding of the 

impact of autonomy-supportive practices on learner engagement and learning outcomes.  

Second, despite the positive effects of teacher autonomy support observed in this study, 

app usage among learners remained limited overall. One possible explanation for this is that the 

participating teachers only received brief training sessions with the author prior to the study. As 

shown by Reeve and Cheon (2021), autonomy-supportive teaching is a malleable practice that 

can be enhanced through professional development. More intensive training could potentially 

increase the teachers’ ability to create an autonomy-supportive climate, thereby boosting 

learners’ app usage beyond the classroom to a greater extent. Future research should consider 

incorporating such intensive teacher training to explore whether it leads to greater and more 

sustained app usage. 

Third, this study would benefit from the adoption of a longitudinal design in future 

research. While the current study provided valuable insights into the role of teacher autonomy 

support, it relied on a single-time measurement of variables, which limits the ability to observe 

 99 

changes over time. A longitudinal design, where variables are measured at multiple time points, 

could offer a deeper understanding of how the interactions among variables evolve and influence 

one another over time. This approach would allow researchers to capture the dynamic 

development of these relationships, providing a more comprehensive picture of the long-term 

impact of autonomy-supportive teaching on (technology-mediated) L2 learning beyond the 

classroom. 

Fourth, I applied the Vocabulary Levels Test to include learners at or below the 1,000-

word mastery level. While it is conceivable that including learners with higher proficiency or 

applying different inclusion criteria (e.g., using pretest scores as an alternative cutoff) might 

offer additional insights, my decision was primarily driven by the goal of maintaining a 

homogeneous sample of beginners to ensure that the study focused on true novice populations. 

Future research could further examine the relationship between broader ranges of initial 

proficiency and learning gains from app-based language learning (e.g., Loewen et al., 2020). 

Fifth, future research should employ a broader range of measures to capture the 

multifaceted nature of language learning. In this study, learning gains were assessed solely 

through meaning-recall tests—a deliberate choice that reflected learners’ agency in choosing 

how to engage with the app. Consequently, some participants may have focused solely on 

flashcard features while others emphasized pronunciation or listening tasks. To accommodate 

these diverse usage patterns, vocabulary knowledge was measured as a key outcome, given its 

capacity to cover a broad range of learning activities on the app and its role as a fundamental 

component of second language acquisition (Jeon & In’nami, 2022). However, since the focal app 

is designed for more than just vocabulary practice, evaluating its impact on comprehensive 

language skills—such as pronunciation, orthography, and speaking fluency—beyond mere 

 100 

vocabulary growth could provide deeper insights into its overall effectiveness. With additional 

resources, future studies could further investigate the app’s influence on other language domains 

and compare its potential to enhance diverse competencies with that of more targeted 

applications. 

Relatedly, in the present study, target words were selected based on the key concepts 

associated with each lesson’s topic, ensuring a balanced representation of an average of 5 to 6 

words per lesson. This decision was made on the assumption that lessons on the app are designed 

to repeatedly expose learners to target words through various examples and activities, as is 

typical in English classroom textbooks. Consequently, the lesson content on the app was 

expected to provide consistent exposure to and across target words, thereby reducing variability 

among learners regardless of individual differences in how they used the app. 

However, while this selection method aligns with typical instructional practices and is 

ecologically justified, a more systematic approach could offer additional insights into the app’s 

effectiveness for vocabulary learning. For instance, incorporating vocabulary-related factors—

such as the frequency of target words in natural language and on the app, as well as their spatial 

distribution within the app—might better capture the influence of distinct characteristics of 

individual vocabulary items on learning outcomes. Furthermore, examining learning-related 

variables—such as the actual number of encounters for each target word and the patterns of user 

interaction with the app features and activities—could provide a more nuanced understanding of 

vocabulary learning processes and types (e.g., intentional versus incidental vocabulary learning) 

in the context of app-based language learning. Although the current study did not directly 

address these possibilities, future research incorporating detailed app usage log data could clarify 

how specific aspects of app curriculum design interact with learners’ usage experiences to 

 101 

influence learning outcomes, thereby guiding the development of more adaptive and effective 

language learning apps. 

Finally, due to logistical constraints, this study focused solely on learners’ perceptions 

rather than directly examining how teacher autonomy support was implemented in the classroom 

or how learners responded to it. More direct in-class observations (e.g., Oga-Baldwin et al., 

2017) and qualitative data—such as teacher logs, follow-up interviews, or focus groups with 

learners segmented by app usage and motivational profile—could provide a more nuanced 

understanding of these dynamics. While the current study focused on learners’ perceived 

support, future research should also examine teachers’ perceptions of autonomy support and how 

it can be more effectively implemented. For example, it would be useful to investigate whether 

incorporating external accountability (e.g., requiring, grading, or rewarding app usage)—which 

are common practices in typical classroom settings to ensure minimal engagement in learning 

activities—might undermine learners’ autonomy need for technology use (e.g., Al-Hoorie, 

2024). Insights from such studies could deepen our understanding of technology integration 

beyond the classroom, particularly in educational contexts where controlled motivation often 

drives L2 learning.  

 102 

 
 
 
CONCLUSION 

While language learning apps can facilitate self-directed L2 learning, high attrition rates 

challenge their effectiveness in self-study contexts. Grounded in self-determination theory, I 

investigated how teacher autonomy support influences app-based language learning beyond the 

classroom within South Korea’s test-driven educational context. 

Seventh-grade beginner learners used a language learning app for 13 weeks while their 

teachers provided various types of autonomy-supportive instructional behaviors to enhance out-

of-class app usage. Mediation and moderated mediation analyses revealed three key findings: (1) 

greater app usage predicted vocabulary learning gains; (2) teacher autonomy support indirectly 

increased app usage by enhancing learners’ autonomy need satisfaction and reducing perceived 

opportunity cost; and (3) teacher support benefited even learners with controlled L2 motivation. 

These findings highlight the importance of establishing an autonomy-supportive learning climate 

that promotes L2 learners’ self-directed use of technology out of class. 

 103 

 
 
REFERENCES 

Alamer, A., & Al Khateeb, A. (2023). Effects of using the WhatsApp application on language 

learners motivation: A controlled investigation using structural equation modelling. Computer 
Assisted Language Learning, 36(1-2), 149–175. 
https://doi.org/10.1080/09588221.2021.1903042 

Alamer, A., & Alrabai, F. (2023). The causal relationship between learner motivation and 
language achievement: New dynamic perspective. Applied Linguistics, 44(1), 148–168. 
https://doi.org/10.1093/applin/amac035 

Alamer, A., & Lee, J. (2019). A motivational process model explaining L2 Saudi learners’ 

achievement of English. System, 87, 102133. https://doi.org/10.1016/j.system.2019.102133 

Al-Hoorie, A. H. (2024). Metamotivational beliefs about extrinsic incentives. System, 103360. 

https://doi.org/10.1016/j.system.2024.103360 

Al-Hoorie, A. H., Oga-Baldwin, W. L. Q., Hiver, P., & Vitta, J. P. (2022). Self-determination 

mini-theories in second language learning: A systematic review of three decades of research. 
Language Teaching Research. Advance online publication. 
https://doi.org/10.1177/13621688221102686 

Alrabai, F., & Alamer, A. (2024). The effects of teacher motivational practice on learner L2 
achievement: A self-determination theory perspective using structural equation modeling. 
Language Teaching Research. Advance online publication. 
https://doi.org/10.1177/13621688241278625 

Bang, H. J., Setoguchi, E., Mackey, A., & Fujii, A. (2024). L2 learning outcomes of a research-
based digital app for Japanese children. Studies in Second Language Acquisition 46(2), 504–
534. https://doi.org/10.1017/S0272263124000147 

Barron, K. E., & Hulleman, C. S. (2015). Expectancy-value-cost model of motivation. In J. D. 
Wright (Ed.), International encyclopedia of the social & behavioral sciences (2nd ed., pp. 
503–509). Elsevier. https://doi.org/10.1016/B978-0-08-097086-8.26099-6 

Chong, S. W., & Reinders, H. (2022). Autonomy of English language learners: A scoping review 

of research and practice. Language Teaching Research. Advance online publication. 
https://doi.org/10.1177/13621688221075812 

Crowther, D., Kim, S., Lee, J., Lim, J., & Loewen, S. (2021). Methodological synthesis of 

cluster analysis in second language research. Language Learning, 71(1), 99–130. 
https://doi.org/10.1111/lang.12428 

Dincer, A., & Işık, T. (2022). Understanding the inner motivational resources of language 

learners’ out-of-class technology use of language learning. In J. Mynard & J. Shelton-Strong 
(Eds.), Autonomy support beyond the language learning classroom: A self-determination 
theory perspective (pp. 47–68). Multilingual Matters. 

 104 

Eccles, J. S., & Wigfield, A. (2020). From expectancy-value theory to situated expectancy-value 

theory: A developmental, social cognitive, and sociocultural perspective on motivation. 
Contemporary Educational Psychology, 61, 101859. 
https://doi.org/10.1016/j.cedpsych.2020.101859 

García Botero, G., Botero Restrepo, M. A., Zhu, C., & Questier, F. (2021). Complementing in-
class language learning with voluntary out-of-class MALL. Does training in self-regulation 
and scaffolding make a difference? Computer Assisted Language Learning, 34(8), 1013–1039. 
https://doi.org/10.1080/09588221.2019.1650780 

García Botero, G., Questier, F., & Zhu, C. (2019). Self-directed language learning in a mobile-

assisted, out-of-class context: Do learners walk the talk? Computer Assisted Language 
Learning, 32(1–2), 71–97. https://doi.org/10.1080/09588221.2018.1485707 

Gladstone, J. R., Wigfield, A., & Eccles, J. S. (2022). Situated expectancy-value theory, 

dimensions of engagement, and academic outcomes. In A. L. Reschly, & S. L. Christenson 
(Eds.), Handbook of research on student engagement (pp. 57–76). Springer. 

Godwin-Jones, R. (2019). Riding the digital wilds: Learner autonomy and informal language 
learning. Language Learning & Technology, 23(1), 8–25. https://doi.org/10125/44667 

He, X., & Loewen, S. (2022). Stimulating learner engagement in app-based L2 vocabulary self-

study: Goals and feedback for effective L2 pedagogy. System, 102719. 
https://doi.org/10.1016/j.system.2021.102719 

Hiver, P., Al-Hoorie, A. H., & Mercer, S. (2021). Student engagement in the language 

classroom. Multilingual Matters. 

Hiver, P., Al-Hoorie, A. H., Vitta, J. P., & Wu, J. (2024). Engagement in language learning: A 
systematic review of 20 years of research methods and definitions. Language Teaching 
Research, 28(1), 201–230. https://doi.org/10.1177/13621688211001289 

Hoang, X., Chik, A., French, R., & Ollerhead, S. (2022). Vietnamese EFL learners’ out-of-

classroom motivation viewed through self-determination theory. In J. Mynard & J. Shelton-
Strong (Eds.), Autonomy support beyond the language learning classroom: A self-
determination theory perspective (pp. 69–90). Multilingual Matters. 

Hsu, L. (2023). EFL learners’ self-determination and acceptance of LMOOCs: The UTAUT 

model. Computer Assisted Language Learning, 36(7), 1177–1205. 
https://doi.org/10.1080/09588221.2021.1976210 

Hwang, H.-B., Coss, M. D., Loewen, S., & Tagarelli, K. M. (2024). Acceptance and engagement 
patterns of mobile-assisted language learning among non-conventional adult L2 learners: A 
survival analysis. Studies in Second Language Acquisition, 46(4), 969–995. 
https://doi.org/10.1017/S0272263124000354 

 105 

Jang, H., Reeve, J., & Halusic, M. (2016). A new autonomy-supportive way of teaching that 

increases conceptual learning: Teaching in learners’ preferred ways. The Journal of 
Experimental Education, 84(4), 686–701. https://doi.org/10.1080/00220973.2015.1083522 

Jeon, E. H., & In’nami, Y. (2022). (Eds.). Understanding L2 proficiency: Theoretical and meta-

analytic investigations. John Benjamins Publishing Company. 

Jeon, J. (2022). Exploring a self-directed interactive app for informal EFL learning: A self-

determination theory perspective. Education and Information Technologies, 27, 5767–5787. 
https://doi.org/10.1007/s10639-021-10839-y 

Jiang, X., Peters, R., Plonsky, L., & Pajak, B. (2024). The effectiveness of Duolingo English 
courses in developing reading and listening proficiency. CALICO Journal, 41(3), 249–272. 
https://doi.org/10.1558/cj.26704 

Jiang, X., Rollinson, J., Plonsky, L., Gustafson, E., & Pajak, B. (2021). Evaluating the reading 
and listening outcomes of beginning-level Duolingo courses. Foreign Language Annals, 
54(4), 974–1002. https://doi.org/10.1111/flan.12600 

Jiang, Y., Rosenzweig, E. Q., & Gaspard, H. (2018). An expectancy-value-cost approach in 
predicting adolescent learners’ academic motivation and achievement. Contemporary 
Educational Psychology, 54, 139–152. https://doi.org/10.1016/j.cedpsych.2018.06.005 

Kessler, M., Loewen, S., & Gönülal, T. (2023). Mobile-assisted language learning with Babbel 

and Duolingo: Comparing L2 learning gains and user experience. Computer Assisted 
Language Learning. Advance online publication. 
https://doi.org/10.1080/09588221.2023.2215294 

Kline, R. (2023). Principles and practice of structural equation modeling (5th ed.). The Guilford 

Press. 

Kosovich, J. J., Hulleman, C. S., Barron, K. E., & Getty, S. (2015). A practical measure of 
student motivation: Establishing validity evidence for the expectancy-value-cost scale in 
middle school. The Journal of Early Adolescence, 35(5–6), 790–
816. https://doi.org/10.1177/0272431614556890 

Lai, C., Yeung, Y., & Hu, J. (2016). University student and teacher perceptions of teacher roles 

in promoting autonomous language learning with technology outside the classroom. Computer 
Assisted Language Learning, 29(4), 703–723. 
https://doi.org/10.1080/09588221.2015.1016441 

Leeming, P., & Harris, J. (2022). Measuring foreign language learners’ self-determination: A 

Rasch validation study. Language Learning, 72(3), 646–694. 
https://doi.org/10.1111/lang.12496 

Lepper, M. R., Corpus, J. H., & Iyengar, S. (2005). Intrinsic and extrinsic motivational 
orientations in the classroom: Age differences and academic correlates. Journal of 
Educational Psychology, 97(2), 184–196. https://doi.org/10.1037/0022-0663.97.2.184 

 106 

Little, T. D., Rhemtulla, M., Gibson, K., & Schoemann, A. (2013). Why the items versus parcels 

controversy needn’t be one. Psychol Methods, 18(3), 285–300. 
https://doi.org/10.1037/a0033266 

Loewen, S., Crowther, D., Isbell, D. R., Kim, K. M., Maloney, J., Miller, Z. F., & Rawal, H. 

(2019). Mobile-assisted language learning: A Duolingo case study. ReCALL, 31(3), 293–311. 
https://doi.org/10.1017/S0958344019000065 

Loewen, S., Isbell, D. R., & Sporn, Z. (2020). The effectiveness of app-based language 

instruction for developing receptive linguistic knowledge and oral communicative ability. 
Foreign Language Annals, 53(2), 209–233. https://doi.org/10.1111/flan.12454 

Matsunaga, M. (2008). Item parceling in structural equation modeling: A primer. 

Communication Methods and Measures, 2(4), 260–293. 
https://doi.org/10.1080/19312450802458935 

McEown, M. S., Noels, K. A., & Saumure, K. (2014). Learners’ self-determined and integrative 
orientations and teachers’ motivational support in a Japanese as a foreign language context. 
System, 45, 227–241. https://doi.org/10.1016/j.system.2014.06.001 

McEown, M. S., & Oga-Baldwin, W. L. Q. (2019). Self-determination for all language learners: 

New applications for formal language education. System, 86, 102124. 
https://doi.org/10.1016/j.system.2019.102124 

Mynard, J., & Shelton-Strong, J. (2022). (Eds.) Autonomy support beyond the language learning 

classroom: A self-determination theory perspective. Multilingual Matters. 

Noels, K. A., Lou, N. M., Vargas Lascano, D. I., Chaffee, K. E., Dincer, A., Zhang, Y. S. D., & 
Zhang, X. (2019a). Self-determination and motivated engagement in language learning. In M. 
Lamb, K. Csizér, A. Henry, & S. Ryan (Eds.), The Palgrave handbook of motivation for 
language learning (pp. 95–116). Palgrave Macmillan. 

Noels, K. A., Pelletier, L. G., Clément, R., & Vallerand, R. J. (2000). Why are you learning a 

second language? Motivational orientations and self-determination theory. Language 
Learning, 50(1), 57–85. https://doi.org/10.1111/0023-8333.00111 

Noels, K. A., Vargas Lascano, D. I., & Saumure, K. (2019b). The development of self-

determination across the language course: Trajectories of motivational change and the 
dynamic interplay of psychological needs, orientations, and engagement. Studies in Second 
Language Acquisition, 41(4), 821–851. https://doi.org/10.1017/S0272263118000189 

Oga-Baldwin, W. L. Q. (2019). Acting, thinking, feeling, making, collaborating: The 

engagement process in foreign language learning. System, 86, 102128. 
https://doi.org/10.1016/j.system.2019.102128 

Oga-Baldwin, W. L. Q. (2022). The quality of our connections matters: Relationships motivation 
theory in independent language learning. In J. Mynard & J. Shelton-Strong (Eds.), Autonomy 

 107 

support beyond the language learning classroom: A self-determination theory perspective (pp. 
149–163). Multilingual Matters. 

Oga-Baldwin, W. L. Q., & Nakata, Y. (2017). Engagement, gender, and motivation: A predictive 

model for Japanese young language learners. System, 65, 151–163. 
http://dx.doi.org/10.1016/j.system.2017.01.011 

Oga-Baldwin, W. L. Q., & Nakata, Y. (2020). How teachers promote young language learners’ 

engagement. Language Teaching for Young Learners, 2(1), 101–130. 
https://doi.org/10.1075/ltyl.19009.oga 

Oga-Baldwin, W. L. Q., Nakata, Y., Parker, P., & Ryan, R. M. (2017). Motivating young 

language learners: A longitudinal model of self-determined motivation in elementary school 
foreign language classes. Contemporary Educational Psychology, 49, 140–150. 
http://doi.org/10.1016/j.cedpsych.2017.01.010 

Olawore, O., Stϋrmer, T., Glynn, R. J., & Lund, J. L. (2024). The healthy user effect in 

pharmacoepidemiology. American Journal of Epidemiology. Advance online publication. 
https://doi.org/10.1093/aje/kwae358 

Reeve, J. (2013). How learners create motivationally supportive learning environments for 

themselves: The concept of agentic engagement. Journal of Educational Psychology, 105(3), 
579–595. https://doi.org/10.1037/a0032690 

Reeve, J. (2022a). A brief but comprehensive overview of self-determination theory. In J. 
Mynard & J. Shelton-Strong (Eds.), Autonomy support beyond the language learning 
classroom: A self-determination theory perspective (pp. 13–30). Multilingual Matters. 

Reeve, J. (2022b). What it means to ‘take ownership over one’s own learning’ in a self-

determination theory analysis. In J. Mynard & J. Shelton-Strong (Eds.), Autonomy support 
beyond the language learning classroom: A self-determination theory perspective (pp. 31–43). 
Multilingual Matters. 

Reeve, J., & Cheon, S. H. (2021). Autonomy-supportive teaching: Its malleability, benefits, and 

potential to improve educational practice. Educational Psychologist, 56(1), 54–77. 
https://doi.org/10.1080/00461520.2020.1862657 

Reeve, J., Cheon, S. H., & Yu, T. H. (2020). An autonomy-supportive intervention to develop 
learners’ resilience by boosting agentic engagement. International Journal of Behavioral 
Development, 44(4), 325–338. https://doi.org/10.1177/0165025420911103 

Reeve, J., & Lee, W. (2014). Learners’ classroom engagement produces longitudinal changes in 

classroom motivation. Journal of Educational Psychology, 106(2), 527–540. 
https://doi.org/10.1037/a0034934 

Reinders, H., & Benson, P. (2017). Research agenda: Language learning beyond the classroom. 

Language Teaching, 50(4), 561–578. https://doi.org/10.1017/S0261444817000192 

 108 

Reinders, H., Lai, C., & Sundqvist, P. (Eds.). (2022). The Routledge handbook of language 

learning and teaching beyond the classroom. Routledge. 

Reschly, A. L., & Christenson, S. L. (2022). Handbook of research on student engagement (2nd 

ed.). Springer. 

Robinson, K. A., Lee, Y.-k., Bovee, E. A., Perez, T., Walton, S. P., Briedis, D., & Linnenbrink-

Garcia, L. (2019). Motivation in transition: Development and roles of expectancy, task values, 
and costs in early college engineering. Journal of Educational Psychology, 111(6), 1081–
1102. https://doi.org/10.1037/edu0000331 

Ryan, R. M., & Deci, E. L. (2001). On happiness and human potentials: A review of research on 

hedonic and eudaimonic well-being. Annual Review of Psychology, 52, 141–166. 
https://doi.org/10.1146/annurev.psych.52.1.141 

Ryan, R. M., & Deci, E. L. (2017). Self-determination theory: Basic psychological needs in 

motivation, development, and wellness. Guilford Publishing. 

Ryan, R. M., & Deci, E. L. (2020). Intrinsic and extrinsic motivation from a self-determination 

theory perspective: Definitions, theory, practices, and future directions. Contemporary 
Educational Psychology, 61, 101860. https://doi.org/10.1016/j.cedpsych.2020.101860 

Ryan, R. M., Huta, V., & Deci, E. L. (2008). Living well: A self-determination theory 

perspective on eudaimonia. Journal of Happiness Studies, 9(1), 139–170. 
https://doi.org/10.1007/s10902-006-9023-4 

Shirvan, M. E., & Alamer, A. (2024). Modeling the interplay of EFL learners’ basic 

psychological needs, grit and L2 achievement. Journal of Multilingual and Multicultural 
Development, 45(7), 2831–2847. https://doi.org/10.1080/01434632.2022.2075002 

Shrank, W. H., Patrick, A. R., & Alan Brookhart, M. (2011). Healthy user and related biases in 

observational studies of preventive interventions: A primer for physicians. Journal of General 
Internal Medicine, 26, 546–550. https://doi.org/10.1007/s11606-010-1609-1 

Skinner, E. A., & Raine, K. E. (2022). Unlocking the positive synergy between engagement and 
motivation. In A. L. Reschly & S. L. Christenson (Eds.), Handbook of research on student 
engagement (pp. 25–56). Springer. 

Smith, B., Jiang, X., & Peters, R. (2024). The effectiveness of Duolingo in developing receptive 
and productive language knowledge and proficiency. Language Learning & Technology, 
28(1), 1–26. https://hdl.handle.net/10125/73595 

Sudina, E., & Plonsky, L. (2024). The effects of frequency, duration, and intensity on L2 

learning through Duolingo: A natural experiment. Journal of Second Language Studies, 7(1), 
1–43. https://doi.org/10.1075/jsls.00021.plo 

The Douglas Fir Group. (2016). A transdisciplinary framework for SLA in a multilingual world. 

Modern Language Journal, 100(S1), 19–47. https://doi.org/10.1111/modl.12301 

 109 

Webb, S., Sasao, Y., & Ballance, O. (2017). The updated vocabulary levels test: Developing and 
validating two new forms of the VLT. International Journal of Applied Linguistics, 168(1), 
34–70. https://doi.org/10.1075/itl.168.1.02web 

Zeng, C., & Fisher, L. (2024). Opening the “black box”: How out-of-class use of Duolingo 

impacts Chinese junior high school learners’ intrinsic motivation for English. ECNU Review 
of Education, 7(2), 283–307. https://doi.org/10.1177/20965311231171606 

Zhou, S., Hiver, P., & Al-Hoorie, A. (2021). Measuring L2 engagement: A review of issues and 
applications. In P. Hiver, A. H. Al-Hoorie, & S. Mercer (Eds.). Student engagement in the 
language classroom (pp. 75–100). Multilingual Matters. 

 110 

 
 
APPENDIX 

Table A1 Perceived teacher autonomy support questionnaire (Korean translation) 

Item 

Question 

Support1  선생님은 내가 스스로 망고 사용계획을 정할 수 있도록 도와 주신다. 

Support2  선생님은 나의 망고 사용방식을 지지해 주신다. 

Support3  선생님은 나에게 망고를 잘 사용하고 있다고 칭찬해 주신다. 

Support4  선생님은 망고 사용 관련 나의 고민을 잘 들어 주신다. 

Table A2 Autonomy need satisfaction questionnaire (Korean translation) 

Item 

Question 

Autonomy1  나는 망고를 사용할지 여부를 자유롭게 결정할 수 있다고 생각한다. 

Autonomy2  나는 망고 사용 계획을 자유롭게 결정할 수 있다고 생각한다. 

Autonomy3  나는 내가 진정으로 원하는 방식으로 망고를 사용한다고 생각한다. 

Table A3 App engagement questionnaire (Korean translation) 

Item  Question 

BE1  나는 망고를 부지런히 사용하였다. 

BE2  나는 망고를 사용할 때 집중하였다. 

BE3  나는 최선을 다해 망고를 하였다. 

EE1  나는 망고가 재밌었다. 

EE2  나는 망고에서 새로운 것을 배울 때 흥미로웠다. 

EE3  나는 망고에서 하는 학습 활동들은 즐거웠다. 

CE1  나는 망고에서 무엇을 어떻게 공부할지 잘 알고 있었다. 

CE2  나는 망고를 언제 사용할지에 대한 계획이 있었다. 

CE3  나는 망고에서 배운 내용을 복습하였다. 

AE1  나는 정해진 목표 이상으로 망고를 사용하려고 노력하였다. 

AE2  나는 망고를 사용하다가 어려움이 생기면, 주위 사람들에게 도움을 구하였다. 

AE3  나는 망고로 공부를 하다가 궁금한 점이 생기면 해결하려고 노력하였다. 

 111 

 
Table A4 Perceived opportunity cost questionnaire (Korean translation) 

Item  Question 

Cost1  망고는 현재 나의 상황에서 너무 많은 시간과 노력을 요구한다. 

Cost2  현재 하고 있는 다른 영어 공부 때문에, 나는 망고에 투자할 시간이 없다. 

Cost3  현재 망고를 꾸준히 사용하기 위한 별도의 시간과 노력을 들일 여유가 없다. 

Cost4  망고를 열심히 하려면, 내가 사용할 수 있는 시간에서 너무 많은 것을 포기해야 한다. 

Table A5 English learning orientation scale (Korean translation) 

Scale  Question 

Intrinsic Motivation 

Item1  영어를 좋아하기 때문이다. 

Item2  영어 공부가 재밌기 때문이다. 

Item3  영어를 사용하면 즐겁기 때문이다. 

Integrated Motivation 

Item1  영어는 내가 어떤 사람인지 이해하는데 중요한 요소이기 때문이다. 

Item2  영어는 내가 인생에서 중요하다고 여기는 것들(가치, 믿음)을 나타내기 때문이다. 

Item3  영어는 자연스러운 나의 일부이기 때문이다. 

Item4  영어를 할 수 있는 능력은 내가 어떤 사람인지 보여주기 때문이다. 

Identified Motivation 

Item1  영어가 나의 성장과 발전에 도움이 되기 때문이다. 

Item2  내가 이루고자 하는 목표를 달성하는데 영어가 도움이 되기 때문이다. 

Item3  영어는 나에게 더 많은 미래의 가능성과 기회를 주기 때문이다. 

Introjected Motivation 

Item1  주위 사람들이 내가 영어를 잘하기를 기대하고 있기 때문이다. 

Item2  주위 사람들이 내가 영어를 잘하면 나를 긍정적으로 평가하기 때문이다. 

Item3  주위 사람들이 내가 영어를 열심히 하기를 바라기 때문이다. 

Item4  주위 사람들에게 내가 영어를 잘하는 사람으로 보이길 원하기 때문이다. 

 112 

 
 
 
Table A5 (cont’d) 

External Motivation 

Item1  영어 시험에서 좋은 점수를 받고 싶기 때문이다. 

Item2  영어는 내가 선택할 수 없는 필수 과목이기 때문이다. 

Item3  영어는 나의 선호와 상관 없이 의무적으로 공부해야 하기 때문이다. 

ID

Item4  영어를 잘하면 상위 학교에 진학하는데 도움이 되기 때문이다. 

 1

(3 ~6 ) 

 (

Amotivation 

, 
Item1  나에게 영어 공부는 시간 낭비이다. 

', '

'

 '
.

Item2  영어는 나에게 의미가 없다. 

)

. 

, 

, 

.

Item3  가능하다면 나는 영어를 배우고 싶지 않다. 

Figure A1 Non-app English learning experiences beyond the classroom questionnaire. 

Activities

 1

 (

(3 ~6 )
, 

? 

).

 (

: 

/

, 

, 

, 

, 

, SNS, 

)

 (

: 

, 

, 

, SNS, 

)

 (

: 

, 

, 

, 

, SNS, 

)

 (

)

) 

 (

: 

, 

, 

,

Reading

 1

(3 ~6 )

SNS, 

)

.

(

) 

? 

 (

: 

/
, 

, 

, 

, 

, 

,

 1
, 

(3 ~6 )
, SNS, 

)

? 

 (

: 

/

, 

, 

,

 113 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ID

', '

'

, 

 '

.

, 

, 

.

 1

(3 ~6 ) 

 (

)

. 

Activities

 1

(3 ~6 )

 (

, 

? 

).

 (

: 

/

, 

, 

, 

, 

, SNS, 

)

 (

: 

, 

, 

, SNS, 

)

 (

: 

, 

, 

, 

, SNS, 

)

 (

)

) 

 (

: 

, 

, 

,

Reading

 1

(3 ~6 )

SNS, 

)

.

(

) 

? 

 (

: 

/

, 

, 

, 

, 

, 

,

Figure A1 (cont’d) 

 1

, 

(3 ~6 )

, SNS, 

(
(

 1
 1

, 
, 

(3 ~6 )
(3 ~6 )
)
)

, SNS, 
, SNS, 

)

/
/

) 
) 

? 

 (

: 

/

, 

, 

,

/
/

 (
 (

/
/

/
/

/
/

)
)

 (
 (

/
/

)
)

? 
? 

 (
 (

: 
: 

/
/
.
.

, 
, 

, 
, 

, 
, 

 1
 1
SNS, 
SNS, 

(3 ~6 )
(3 ~6 )
)
)

/
/

 (
 (

: 
: 
? 
? 

, 
, 

, 
, 

, 
, 
, 
, 

, 
, 

,
,

.
.

 1
 1
SNS, 
SNS, 

(3 ~6 )
(3 ~6 )
)
)

 (
 (

: 
: 

/
/
. 
. 

, 
, 

, 
, 

, 
, 

, 
, 

,
,

 (10%
 (10%
)
)

(10%-20%)
(10%-20%)

(20%-40%)
(20%-40%)

(40%-60%)
(40%-60%)

(60%-80%)
(60%-80%)

(80%-90%)
(80%-90%)

 (90%
 (90%
)
)

Watching
Watching

 114 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
/

/

 (

/

/

/

)

 (

/

)

(

) 

 1

(3 ~6 )

, 

, SNS, 

)

? 

 (

: 

/

.

, 

, 

, 

 1

(3 ~6 )

SNS, 

)

 (

: 

/

, 

, 

, 

, 

,

? 

, 

.

 1

(3 ~6 )

SNS, 

)

 (

: 

, 

, 

, 

, 

,

/

. 

 (10%

)

(10%-20%)

(20%-40%)

(40%-60%)

(60%-80%)

(80%-90%)

 (90%

)

Figure A1 (cont’d) 

Watching

 1
 1

(3 ~6 )
(3 ~6 )

(
(

? 
? 

) 
) 

 (
 (

: 
: 

, 
, 

, 
, 

, SNS, 
, SNS, 

)
)

.
.

, 
, 

 1
 1
SNS, 
SNS, 

(3 ~6 )
(3 ~6 )
)
)

? 
? 

/
/

/
/

) 
) 

? 
? 

(
(

 1
 1
)
)

(3 ~6 )
(3 ~6 )

 (
 (

: 
: 

, 
, 

, 
, 

,
,

 (
 (

/
/

/
/

/
/

)
)

 (
 (

/
/

)
)

 (
 (

: 
: 

, 
, 

, 
, 

, SNS,
, SNS,

.
.

 1
 1

(3 ~6 )
(3 ~6 )

.
.

? 
? 

 (
 (

: 
: 

, 
, 

, 
, 

, 
, 

, SNS, 
, SNS, 

)
)

 1
 1

(3 ~6 )
(3 ~6 )

 (
 (

: 
: 

, 
, 

, 
, 

, SNS, 
, SNS, 

)
)

. 
. 

 115 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 1

(3 ~6 )

 (

: 

, 

, 

, SNS, 

)

? 

, 

.

(

) 

 1

(3 ~6 )

SNS, 

)

? 

 (

: 

, 

, 

,

/

/

 (

/

/

/

)

 (

/

)

(

 1

)

) 

? 

(3 ~6 )

 (

: 

, 

, 

, SNS,

.

 1

(3 ~6 )

 (

: 

, 

, 

, SNS, 

)

? 

, 

.

Figure A1 (cont’d) 

 1

(3 ~6 )

 (

: 

, 

, 

, SNS, 

)

. 

 (10%
 (10%
)
)

(10%-20%)
(10%-20%)

(20%-40%)
(20%-40%)

(40%-60%)
(40%-60%)

(60%-80%)
(60%-80%)

(80%-90%)
(80%-90%)

 
 
 (90%
 (90%
)
)

Listening
Listening

 1
 1
, SNS, 
, SNS, 

(3 ~6 )
(3 ~6 )
)
)

.
.

(
(

) 
) 

 (
 (

: 
: 

, 
, 

, 
, 

, 
, 

? 
? 

, 
, 

 1
 1

(3 ~6 )
(3 ~6 )

, 
, 

, SNS, 
, SNS, 

)
)

? 
? 

 (
 (

: 
: 

, 
, 

,
,

/
/

/
/

 (
 (

/
/

/
/

/
/

)
)

 (
 (

/
/

) 
) 

(3 ~6 )
(3 ~6 )

, SNS, 
, SNS, 

)
)

? 
? 

 (
 (

: 
: 

(
(

 1
 1
, 
, 

)
)

.
.

, 
, 

, 
, 

 1
 1
, SNS, 
, SNS, 

(3 ~6 )
(3 ~6 )
)
)

 116 

 (
 (

: 
: 

? 
? 

, 
, 

, 
, 

, 
, 

, 
, 

 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 (10%

)

(10%-20%)

(20%-40%)

(40%-60%)

(60%-80%)

(80%-90%)

 

 (90%

)

Listening

 1

(3 ~6 )

, SNS, 

)

.

(

) 

 (

: 

, 

, 

, 

? 

, 

 1

(3 ~6 )

, 

, SNS, 

)

? 

 (

: 

, 

,

/

/

 (

/

/

/

)

 (

/

) 

(

 1

, 

(3 ~6 )

, SNS, 

)

? 

)

.

 (

: 

, 

, 

Figure A1 (cont’d) 

 1
, SNS, 

(3 ~6 )
)

 (

: 

? 

, 

, 

, 

, 

.

.

 1
, SNS, 
 1
, SNS, 

(3 ~6 )
)
(3 ~6 )
)

 (

 (

: 

: 

. 

. 

, 

, 

, 

, 

, 

, 

 (10%
)
 (10%
)

(10%-20%)

(20%-40%)

(40%-60%)

(60%-80%)

(80%-90%)

(10%-20%)

(20%-40%)

(40%-60%)

(60%-80%)

(80%-90%)

 (90%
)
 (90%
)

Communication

Communication

 1

(3 ~6 )

 1

(3 ~6 )

.

.

(

(

? 

? 

) 

) 

 1

 1

, 

, 

(3 ~6 )

)

(3 ~6 )

)

(

(

? 

? 

/

/

/

/

) 

) 

 (

 (

: 

: 

, 

, 

, 

, 

, 

, 

)

)

, 

, 

 (

 (

: 

: 

, 

, 

, 

, 

 (

 (

/

/

/

/

/

/

)

)

 (

 (

/

/

)

)

 117 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure A1 (cont’d) 

 1

(3 ~6 )

)

? 

 (

: 

, 

, 

,

.

 1

(3 ~6 )

 (

: 

, 

, 

, 

)

? 

, 

.

 1

(3 ~6 )

 (

: 

, 

, 

, 

)

. 

 (10%
)

(10%-20%)

(20%-40%)

(40%-60%)

(60%-80%)

(80%-90%)

 (90%
)

Powered by Qualtrics

 118