{"id":24175,"date":"2026-07-19T03:07:28","date_gmt":"2026-07-19T11:07:28","guid":{"rendered":"https:\/\/tesl-ej.org\/wordpress\/?page_id=24175"},"modified":"2026-07-20T17:04:21","modified_gmt":"2026-07-21T01:04:21","slug":"ej118a6","status":"publish","type":"page","link":"https:\/\/tesl-ej.org\/wordpress\/issues\/volume30\/ej118\/ej118a6\/","title":{"rendered":"The Role of Vocabulary Learning Strategies in Longitudinal Vocabulary Growth"},"content":{"rendered":"

August 2026 – Volume 30, Number 2<\/h2>\n

https:\/\/doi.org\/10.55593\/ej.30118a6<\/strong><\/p>\n

Masaki Akase<\/strong>
\nNational Institute of Technology, Nagano College, Japan
\n<m_akase\"atmark\"nagano-nct.ac.jp><\/p>\n

Abstract<\/h3>\n

This study examines the role of vocabulary learning strategies (VLS) in vocabulary size (VS) and vocabulary growth among science, technology, engineering, and mathematics (STEM)-oriented upper secondary students in Japan. Using a three-year longitudinal mixed-methods design, 189 participants completed three forms of the Vocabulary Size Test (VST), a fourth form composed of items from VSTs 1\u20133, and a VLS questionnaire administered at four time points. Total VS was estimated using a Rasch-based method to ensure comparability across test forms. Cluster analysis identified three strategy-use groups (low, medium, high), and a two-way mixed ANOVA revealed significant main effects of time and strategy cluster, as well as their interaction, indicating that more frequent strategy use was modestly but consistently associated with vocabulary growth, with small but cumulative effects over time. Open-ended responses further revealed differences in strategy depth and diversity across clusters. Integrating quantitative and qualitative findings, this study identifies patterns of vocabulary development among students at Japanese Colleges of Technology (KOSEN), underscores the importance of sustained strategy engagement, and offers implications for vocabulary assessment and longitudinal monitoring in diverse EFL contexts, including technical and higher education settings.<\/p>\n

Keywords<\/em><\/strong>: vocabulary learning strategies, vocabulary size, Japanese EFL learners, longitudinal study, Rasch measurement<\/p>\n

Background to the Study<\/h3>\n

Vocabulary is central to foreign language learning; as Wilkins (1972, p. 111) famously observed, \u201cwithout grammar very little can be conveyed, without vocabulary nothing can be conveyed.\u201d Adequate vocabulary supports both comprehension and expression, providing a foundation for all four language skills. Reflecting this importance, the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) increased the target number of English words taught from approximately 3,000 to 4,000\u20135,000 across elementary to high school levels (MEXT, 2018). Nevertheless, a nationwide survey reported that 71.8% of first-year high school students found vocabulary learning difficult (Benesse, 2019), suggesting that learners\u2019 actual acquisition may fall short of policy expectations. Moreover, Nation (2006) estimated that roughly 8,000 word families are required to reach 98% lexical coverage for adequate comprehension, indicating that Japanese EFL learners face a substantial and cumulative vocabulary burden.<\/p>\n

Given that vocabulary knowledge underpins communicative competence, an important question arises regarding which VLS are associated with vocabulary growth. Fostering the use of VLS may support learner autonomy and engagement (Nation, 2022). Understanding which types of strategies are most effective, and how their use evolves over time, is therefore critical for providing both theoretical insights and practical guidance in EFL instruction. The present study investigates these questions in the context of Japanese EFL learners enrolled in a five-year STEM-oriented educational program at KOSEN, aiming to clarify the relationship between VLS use and VS growth.<\/p>\n

Literature Review<\/h3>\n

Vocabulary knowledge is commonly conceptualized in terms of size and depth (Schmitt, 2014). VS, or breadth, refers to the quantitative dimension of lexical knowledge\u2014that is, the number of words a learner knows at a basic level, including knowledge of word forms and primary meanings (Anderson & Freebody, 1981; Nation, 2022; Qian, 1999; Read, 2000). In contrast, vocabulary depth represents the qualitative aspects of word knowledge, including how lexical items are organized and interconnected within the mental lexicon (Read, 2000; Schmitt, 2008; St\u00e6hr, 2009).<\/p>\n

Previous research has emphasized the importance of VS because of its central role in language use and overall language proficiency (Laufer et al., 2004; Meara, 2002; Nation, 2013; Schmitt, 2010; Schmitt & Schmitt, 2014). VS is especially informative for learners with limited lexical repertoires (Meara, 1996) and is often considered a more representative indicator of overall vocabulary knowledge than depth measures (Read, 2000). In addition, VS shows strong associations with general language proficiency and performance across the four skills\u2014reading, listening, writing, and speaking (Milton & Hopwood, 2023). It also contributes to reading and listening comprehension (Webb & Paribakht, 2015) and facilitates incidental vocabulary learning (Teng & Uchihara, 2024).<\/p>\n

Considering its theoretical and pedagogical importance, VS is commonly operationalized and measured through simple recognition tests of word forms and meanings. Such assessments are widely used as indicators of learners\u2019 overall vocabulary proficiency across the four language skills. For beginning EFL learners, including junior and senior high school students, vocabulary breadth\u2014that is, basic knowledge of many words\u2014may be more immediately relevant than deeper lexical knowledge. Although these measures capture only surface-level aspects of vocabulary knowledge, they offer a widely accepted and practical approach to evaluating vocabulary breadth and tracking developmental change over time, particularly in classroom-based longitudinal research.<\/p>\n

Measuring Vocabulary Size<\/h4>\n

VS is a robust predictor of reading, writing, general language proficiency, and academic achievement in EFL contexts (Laufer & Goldstein, 2004). Several instruments, including the VST (Nation & Beglar, 2007) and the Updated Vocabulary Levels Test (Webb et al., 2017), assess learners\u2019 knowledge of word forms and their associated meanings, often in bilingual formats. While test formats are similar, they differ in corpus selection and sampling, typically using stratified-random sampling by frequency level (Schmitt, 2010). VS is estimated from the proportion of known words per level, multiplied by 1000, following the conventional raw-score approach, although Rasch-based scaling has increasingly been adopted for improved measurement precision.<\/p>\n

Recent research has emphasized rigorous validity assessment for VST development (Schmitt et al., 2019). Rasch analysis (Rasch, 1960), a one-parameter item response theory model, has been widely applied to evaluate unidimensionality\u2014that is, whether responses can be explained by a single underlying latent trait\u2014as well as item fit, reliability, and linking of different test versions (Henning et al., 1985). Rasch difficulty estimates and fit statistics can be used to evaluate whether individual items contribute appropriately to the reliability of the test as a whole (Bond & Fox, 2001). Rasch analysis can also be used to link different versions of a test to a single scale, so long as the versions have some items in common (Bond et al., 2020).<\/p>\n

Several studies have validated the VST for Japanese EFL learners (Beglar, 2010; Hamada et al., 2021; Kasahara, 2006; Koizumi & Mochizuki, 2011). Kasahara (2006) revised Mochizuki\u2019s (1998) VST using FACETS Rasch software, replacing misfitting items with items from the JACET 8000 Basic Words list (JACET, 2003) to enhance unidimensionality, reliability, and validity. Beglar (2010) validated the VST created by Nation and Beglar (2007) with a large group of learners, confirming good Rasch fit, unidimensionality, reliability, and invariance. Hamada et al. (2021) developed a VST based on the New JACET List, validated it with a three-parameter logistic IRT model, and demonstrated that it predicted reading and listening proficiency. Koizumi and Mochizuki (2011) also applied the Rasch model to Mochizuki\u2019s VST and found evidence of measurement consistency, including few misfitting items, increasing item difficulty across frequency levels, and a strong correlation with TOEIC\u00ae<\/sup> L&R scores.<\/p>\n

The VST can also track changes in learners\u2019 VS over time using conventional raw-score methods (e.g., Akase, 2021; Katagiri, 2009). Building on this line of research, the present study adopts the equated VST forms developed by Akase (2022), which linked three original versions created by Aizawa and Mochizuki (2010) and a newly created fourth form through Rasch analysis, thereby providing reliable longitudinal measures of VS. Based on Rasch person ability estimates, the estimated number of words known in each frequency band was calculated as the mean probability of success for that level multiplied by 1000. Total VS was then obtained by summing the estimated numbers of words known across all frequency levels, yielding Rasch-based VS estimates for subsequent analyses.<\/p>\n

Vocabulary Learning Strategies<\/h4>\n

Since the mid-1980s, second language acquisition research has highlighted the pivotal role of VLS in vocabulary acquisition and overall language proficiency. VLS are broadly defined as \u201cactions that learners take to help themselves understand and remember vocabulary\u201d (Cameron, 2001, p. 92). These include memorization, repetition, association, use of the keyword method\u2014which links the meaning of an L2 word with an L1 word that has a similar spoken form and has been shown to support durable long-term vocabulary learning (Webb & Nation, 2017)\u2014as well as contextualization and dictionary use (Nyikos & Fan, 2007). Systematic investigation of VLS effects on VS has been enabled by widespread VST implementation (Gu & Johnson, 1996; Kojic-Sabo & Lightbown, 1999). While some strategies, such as visual repetition and rote memorization, have been found to correlate negatively with VS, most show positive associations.<\/p>\n

Research examining the relationship between VLS and VS has identified several consistent patterns. First, learners with higher proficiency and larger VS tend to use a wider variety of strategies more frequently than less successful learners (e.g., Fan, 2003; Gu, 2010; Lawson & Hogben, 1996). Second, as learners gain proficiency, their strategy use gradually shifts from simple strategies, such as rote repetition, to more complex strategies, including imagery, inferencing, association, and organization (e.g., Horino & Ichikawa, 1997; Lai, 2016; Schmitt, 1997; Zhang & Lu, 2015). Third, the appropriate, structured, and balanced use of strategies is associated with more successful vocabulary learning and higher proficiency (e.g., Akase, 2021; Mizumoto & Takeuchi, 2009a; Sanaoui, 1995). Fourth, metacognitive strategies and cognitive strategies\u2014particularly memory strategies\u2014play a central role in effective vocabulary learning, and explicit strategy instruction has been demonstrated to enhance learners\u2019 VS (e.g., Atay & Ozbulgan, 2007; Mizumoto & Takeuchi, 2009b; Rasekh & Ranjbary, 2003). Finally, individual difference variables, including gender, age, academic discipline, and learning context, influence both the choice and frequency of VLS use (e.g., Catal\u00e1n, 2003; Nakamura, 2002; Tseng & Schmitt, 2008).<\/p>\n

Despite a substantial body of research on VLS, most studies rely on cross-sectional designs. This leaves unanswered questions about how VLS use develops over time and how such development influences VS growth. As Boers (2024) argues, adopting a longitudinal perspective is essential for understanding not only how learning strategies are acquired but also how they can be effectively evaluated. Considering the crucial role of VS in communicative competence, examining the dynamic interplay between VS and VLS longitudinally can provide important theoretical and pedagogical insights (Gu, 2019).<\/p>\n

To address this gap, the present study investigates patterns of VLS use and their relationship with VS development through both cross-sectional and longitudinal analyses. Given that VS provides a strong indication of learners\u2019 general linguistic progress toward communicative goals (Milton & Hopwood, 2023), the investigation is guided by four research questions (RQs):<\/p>\n

RQ1:<\/strong> What types of VLS do learners use, and to what extent are these strategies employed at Time 4 (the final measurement point)?<\/p>\n

RQ2:<\/strong> What patterns of VS and VS growth emerge in relation to changes in learners\u2019 VLS use over time?<\/p>\n

RQ3:<\/strong> To what extent can four parallel forms of the VST, together with a VLS questionnaire, be used to capture relationships between patterns of VLS use and VS growth?<\/p>\n

RQ4:<\/strong> Based on text mining of learners\u2019 open-ended responses, which specific VLS can be identified?<\/p>\n

Method<\/h3>\n

This study employed a quantitative longitudinal design to examine changes in VS and VLS over time. Data were collected using four parallel forms of the VST and a VLS questionnaire administered at multiple measurement points.<\/p>\n

Participants<\/h4>\n

The participants were 189 students (36 females and 153 males; aged 16\u201318 years) enrolled in the first three years of a five-year STEM program at a Japanese college of technology (KOSEN), a tertiary institution that admits students after junior high school and provides specialized engineering education. This period corresponds to the senior high school stage, during which learners typically experience substantial vocabulary growth within a consistent English curriculum. The researcher taught these students throughout this period, ensuring stable instructional conditions and minimizing external variation. Although the researcher also served as the instructor, testing followed standardized procedures. Prior to data collection, students were informed of the purpose of the study, that participation was voluntary, and that they could withdraw from the study at any time without penalty.<\/p>\n

Students\u2019 English proficiency ranged from A2 to B1 on the CEFR, with a few students reaching B2. Across five engineering-related departments\u2014mechanical engineering, electrical and electronic engineering, electronics and control engineering, computer science, and civil engineering\u2014English was regarded as an important academic and professional skill. About half of the students advanced to higher education, while the remainder entered the workforce, making English an important component of their academic and professional development.<\/p>\n

Instruments<\/h4>\n

Vocabulary Size Test. Data were collected at four time points: Time 1 (April 2018, first year), Time 2 (April 2019, second year), Time 3 (April 2020, third year), and Time 4 (December 2020, third year). Three forms of Aizawa and Mochizuki\u2019s (2010) VST, along with an equated version, VST 4, developed by Akase (2022), were administered. These tests use Japanese definitions and are suitable for lower-intermediate EFL learners. To balance test length, difficulty, and reliability, six frequency bands (1,000\u20136,000 words) were included; higher levels were excluded to reduce guessing, which could otherwise overestimate VS and compromise measurement accuracy (Stewart, 2014). Each band contained 26 items, resulting in 156 items per form.<\/p>\n

VST 4 employed a Rasch equating procedure to align the original three forms, thereby enabling reliable longitudinal measurement while minimizing practice effects. Items were carefully selected from VSTs 1\u20133 to cover the full difficulty range and ensure representation across each frequency band (see Appendix A<\/a> in the Supplementary File). Rasch equating also allows estimation of the expected number of correct responses per frequency band by averaging item difficulties, from which total VS can be inferred. This approach ensures valid longitudinal comparisons and supports robust analyses of VS development.<\/p>\n

Vocabulary Learning Strategies Questionnaire. An inventory of VLS was compiled from the literature to develop the Vocabulary Learning Strategies Questionnaire (VLSQ), drawing on instruments by Fan (2003), Gu and Johnson (1996), Horino and Ichikawa (1997), Mizumoto and Takeuchi (2009a), and Schmitt (1997). Following D\u00f6rnyei\u2019s (2005) observation that \u201cit is difficult to distinguish the difference between engaging in an ordinary learning activity and a strategic learning activity\u201d (p. 143), VLS in this study are defined as deliberate or goal-directed actions, techniques, or behaviors that learners employ to acquire and practice vocabulary (cognitive strategies) or to plan, monitor, and regulate their vocabulary learning (metacognitive strategies) (Fan, 2003; Oxford, 1990; Schmitt, 1997, 2000). While this conceptualization emphasizes intentional strategy use, some strategies may become automatic or habitual with repeated practice, reflecting the dynamic nature of strategy development over time (Cohen, 2011; D\u00f6rnyei, 2005; Griffiths, 2018).<\/p>\n

Based on this framework, 33 items were developed to assess seven strategy categories: metacognitive (6), input-seeking (4), written rehearsal (5), oral rehearsal (3), social (4), organization (6), and elaboration (5). Participants rated their strategy use on a 6-point Likert scale (1 = strongly disagree, 6 = strongly agree). The questionnaire items are provided in Appendix B<\/a> in the Supplementary File. Table 1 presents an overview of the quantitative instruments used in the study.<\/p>\n

Table 1. Summary of Quantitative Measures Used in the Study<\/strong><\/p>\n\n\n\n\n\n\n\n\n\n\n\n
Measure<\/td>\nItems<\/td>\nPossible range<\/td>\n<\/tr>\n
Vocabulary size test (VST 1, 2, 3, 4)<\/td>\n156 \/ VST<\/td>\n0\u20136000<\/td>\n<\/tr>\n
Metacognitive strategies<\/td>\n6<\/td>\n1\u20136<\/td>\n<\/tr>\n
Input-seeking strategies<\/td>\n4<\/td>\n<\/tr>\n
Written rehearsal strategies<\/td>\n5<\/td>\n<\/tr>\n
Oral rehearsal strategies<\/td>\n3<\/td>\n<\/tr>\n
Social strategies<\/td>\n4<\/td>\n<\/tr>\n
Organization strategies<\/td>\n6<\/td>\n<\/tr>\n
Elaboration strategies<\/td>\n5<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

The VLSQ was administered alongside the four VST forms at each measurement point. For the present analysis, only Time 4 VLS data were used to represent learners\u2019 habitual strategy use, as strategy use tends to stabilize with maturation (Akase, 2021) and become more automatic with experience (Griffiths, 2018). By the third year, learners had accumulated substantial instructional exposure, making Time 4 a reasonable proxy for relatively stable strategy repertoires and typical strategy profiles. This approach was intended to provide a relatively stable index of learners\u2019 strategy use and to align the VLS data with the longitudinal VS analyses.<\/p>\n

Because the analysis relies on end-point strategy data, no causal claims are made regarding earlier vocabulary growth, and associations between strategy use and vocabulary outcomes are interpreted cautiously.<\/p>\n

Open-Ended Survey on Vocabulary Learning. At the end of their third year (March 2021), participants completed a voluntary open-ended online survey administered via Google Forms. Of the 189 students invited, 178 responded (36 females, 142 males) to the prompt: \u201cWhat techniques or strategies do you use specifically when learning vocabulary?\u201d The online format allowed learners to respond at their convenience, reduced transcription burden, minimized researcher influence, and facilitated anonymous data collection.<\/p>\n

In contrast to the structured VLSQ, these open-ended responses provided self-generated descriptions of strategy use, including strategies not fully captured by the predefined questionnaire categories. The qualitative data played a complementary role in explaining the quantitative clusters and interpreting differences among strategy groups. Specifically, themes derived from the open-ended responses were used to elaborate, confirm, and nuance patterns identified statistically.<\/p>\n

Following an explanatory sequential mixed-methods design (see Figure 1), the study first collected and analyzed longitudinal quantitative data on VLS use and VS, then used qualitative data to explain and contextualize the results (Creswell, 2015). Although collected at a single time point, the qualitative data were not intended to establish causality but to illuminate patterns observed in the longitudinal analyses. By integrating both strands at the interpretation stage, the study provides a more comprehensive and theoretically grounded understanding of learners\u2019 VLS and their association with VS development.<\/p>\n

\"Figure<\/p>\n

Figure 1. Explanatory Sequential Design<\/strong><\/p>\n

Note. adapted from Creswell (2015, p. 39).<\/p>\n

Procedures<\/h4>\n

Data collection followed institutional approval and cooperation from course instructors across four measurement points (Time 1\u20134). At each point, students completed the VST and VLSQ under standardized classroom conditions. An open-ended online survey was administered at the end of the third year (March 2021) to capture learners\u2019 self-generated descriptions of vocabulary learning practices beyond the structured questionnaire.<\/p>\n

All data were de-identified and compiled in Excel. Quantitative analyses were conducted using Winsteps 4.7 (Linacre, 2019) for Rasch calibration of the VST forms (Akase, 2022). Using the dichotomous Rasch model, item difficulty and learner ability were estimated on the same latent scale. Descriptive statistics, principal component analysis (PCA), hierarchical cluster analysis, and repeated-measures mixed ANOVA were conducted using JASP 0.12.2 (JASP Team, 2020). These procedures yielded relatively stable estimates of VS and strategy-use profiles, which informed the qualitative phase. Cluster analysis enabled pedagogically interpretable group comparisons and examined whether different levels of strategic engagement were associated with distinct longitudinal vocabulary growth trajectories.<\/p>\n

Following an explanatory sequential mixed-methods design (Creswell, 2015), qualitative analyses were guided by the quantitative results. Open-ended responses were linked to cluster membership and analyzed using KH Coder 3.0 (Higuchi, 2014) to identify salient keywords and recurrent themes for each strategy group. This integration allowed qualitative findings to elaborate on, qualify, and interpret patterns observed in the quantitative analyses, thereby enhancing the explanatory depth and validity of the study.<\/p>\n

Data analysis<\/h4>\n

The data were analyzed using quantitative statistical procedures. Descriptive statistics (means and standard deviations) were calculated, and the reliability of the questionnaire was assessed using Cronbach\u2019s alpha. Cluster analysis was conducted to identify profiles of VLS use, followed by a two-way mixed ANOVA to examine differences in VS development across the identified groups over time.<\/p>\n

Results<\/h3>\n

Descriptive Statistics for the Vocabulary Size Tests<\/h4>\n

Table 2 presents descriptive statistics for the 189 participants across VSTs 1\u20134. Rasch-estimated VS, equated across test forms, increased steadily over time. The distributions of VS scores were approximately normal at each wave (skewness and kurtosis between \u22121.5 and +1.5; Tabachnick & Fidell, 2013), supporting the use of parametric analyses in subsequent longitudinal comparisons.<\/p>\n

Table 2. Descriptive Statistics for the VST Scores<\/strong><\/p>\n\n\n\n\n\n\n\n
VST<\/td>\nTime<\/td>\nM<\/em><\/td>\nSD<\/em><\/td>\nSkewness<\/td>\nKurtosis<\/td>\nk<\/em><\/td>\n\u03b1<\/em><\/td>\n<\/tr>\n
VST 1<\/td>\n1<\/td>\n2734.31<\/td>\n486.63<\/td>\n\u00a00.00<\/td>\n\u00a00.05<\/td>\n156<\/td>\n.87<\/td>\n<\/tr>\n
VST 2<\/td>\n2<\/td>\n3035.44<\/td>\n652.33<\/td>\n-0.44<\/td>\n\u00a00.29<\/td>\n156<\/td>\n.93<\/td>\n<\/tr>\n
VST 3<\/td>\n3<\/td>\n3401.77<\/td>\n661.30<\/td>\n\u00a00.01<\/td>\n-0.24<\/td>\n156<\/td>\n.93<\/td>\n<\/tr>\n
VST 4<\/td>\n4<\/td>\n3734.34<\/td>\n745.43<\/td>\n-0.24<\/td>\n\u00a00.51<\/td>\n156<\/td>\n.94<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Note. N<\/em> = <\/strong>18<\/strong>9<\/strong>; <\/strong>k<\/em><\/strong> = number of items; VST = Vocabulary Size Test.<\/strong><\/p>\n

Validation of a Vocabulary Learning Strategy Questionnaire<\/h4>\n

PCA with Oblimin rotation was conducted to examine the underlying structure of the 33-item VLSQ and to identify components for subsequent analyses. This approach was selected as an appropriate data-reduction technique for identifying coherent item groupings while allowing correlations among components, which are theoretically expected in strategy use.<\/p>\n

An initial seven-component solution (eigenvalues > 1.0; scree plot inspection) largely yielded loadings above .40. However, one item cross-loaded, two items failed to load clearly, and the seventh component had a marginal eigenvalue (1.15) with limited interpretability. Considering these results alongside theoretical frameworks (D\u00f6rnyei & Dewaele, 2022; Gu & Johnson, 1996; Schmitt, 1997), a six-component solution was adopted. Three items (VLS-14, VLS-15, VLS-26) were removed, resulting in a final 30-item instrument. This refined structure balances empirical adequacy with theoretical coherence, providing support for the construct validity of the instrument.<\/p>\n

The six components were interpreted and labeled as follows:<\/p>\n

    \n
  1. Metacognitive strategies: input-seeking and regulation of learning (Fan, 2003; Mizumoto & Takeuchi, 2009a; Schmitt, 1997).<\/li>\n
  2. Grouping strategies: organizational and memory-based word association strategies (Horino & Ichikawa, 1997; Schmitt, 1997).<\/li>\n
  3. Contextualizing strategies: imagery and contextual use of target vocabulary (Gu & Johnson, 1996).<\/li>\n
  4. Traditional strategies: focused practices such as flashcards, vocabulary notebooks, and written rehearsal (Mizumoto & Takeuchi, 2009a; Schmitt, 1997).<\/li>\n
  5. Social strategies: collaborative learning and peer interaction (Schmitt, 1997).<\/li>\n
  6. Oral rehearsal strategies: repeating and practicing vocabulary aloud (Schmitt, 1997).<\/li>\n<\/ol>\n

    Table 3 presents descriptive statistics and reliability indices for the six VLSQ components at Time 4. Mean scores for metacognitive and grouping strategies were around 3, indicating moderate-to-low levels of use, while traditional and social strategies were used even less frequently. In contrast, contextualizing and oral rehearsal strategies were employed more often, with mean scores of approximately 3.8. Cronbach\u2019s alpha coefficients exceeded .70 for all components except oral rehearsal strategies, which demonstrated marginal but acceptable reliability given its limited number of items.<\/p>\n

    Table 3. Descriptive Statistics for the Six Components of VLS at Time 4<\/strong><\/p>\n\n\n\n\n\n\n\n\n\n
    VLS<\/td>\nk<\/em><\/td>\nItems<\/td>\nM<\/em><\/td>\nSD<\/em><\/td>\nSkewness<\/td>\nKurtosis<\/td>\n\u03b1<\/em><\/td>\n<\/tr>\n
    MET<\/td>\n8<\/td>\n2, 10, 12, 20, 22, 24, 29, 33<\/td>\n3.12<\/td>\n1.00<\/td>\n0.02<\/td>\n-0.43<\/td>\n.83<\/td>\n<\/tr>\n
    GRP<\/td>\n6<\/td>\n1, 8, 11, 23, 30, 31<\/td>\n3.21<\/td>\n0.98<\/td>\n-0.09<\/td>\n-0.23<\/td>\n.74<\/td>\n<\/tr>\n
    CON<\/td>\n7<\/td>\n3, 6, 7, 17, 19, 21, 27<\/td>\n3.87<\/td>\n0.86<\/td>\n-0.62<\/td>\n0.88<\/td>\n.75<\/td>\n<\/tr>\n
    TRD<\/td>\n4<\/td>\n4, 25, 28, 32<\/td>\n2.71<\/td>\n1.13<\/td>\n0.51<\/td>\n-0.22<\/td>\n.71<\/td>\n<\/tr>\n
    SOC<\/td>\n2<\/td>\n5, 16<\/td>\n2.74<\/td>\n1.35<\/td>\n0.36<\/td>\n-0.77<\/td>\n.78<\/td>\n<\/tr>\n
    ORL<\/td>\n3<\/td>\n9, 13, 18<\/td>\n3.83<\/td>\n1.13<\/td>\n-0.37<\/td>\n-0.25<\/td>\n.65<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    Note<\/strong>.<\/em><\/strong> N<\/em> = <\/strong>189; <\/strong>k <\/em><\/strong>= number of items; VLS = vocabulary learning strategies; MET = metacognitive strategies; GRP = grouping strategies; CON = contextualizing strategies; TRD = traditional strategies; SOC = social strategies; ORL = oral rehearsal strategies<\/strong>.<\/p>\n

    <\/strong><\/p>\n

    Cluster Analysis<\/h4>\n

    A cluster analysis (Ward\u2019s method, squared Euclidean distance) of the Time 4 VLSQ data yielded three clusters that reflected overall differences in strategy-use frequency: high (Cluster 1, n<\/em> = 53), medium (Cluster 2, n<\/em> = 101), and low (Cluster 3, n <\/em>= 35), rather than qualitatively distinct strategy profiles. To facilitate interpretation, the clusters were labeled high-, medium-, and low-strategy-use groups.<\/p>\n

    As shown in Figure 2, between-cluster differences were most pronounced for metacognitive, grouping, and contextualizing strategies, with Cluster 1 scoring above and Cluster 3 below the overall sample mean, whereas differences were smaller for traditional, social, and oral rehearsal strategies\u2014particularly for traditional strategies, which showed similar levels of use in Clusters 1 and 2.<\/p>\n

    \"Figure<\/p>\n

    Figure 2. Vocabulary Learning Strategy Use Cluster at Time 4<\/strong><\/p>\n

    Note. Cluster 1 (<\/strong>n<\/em><\/strong> = 53), Cluster 2 (<\/strong>n<\/em><\/strong> = 101), and Cluster 3 (<\/strong>n<\/em><\/strong> = 35). <\/strong><\/p>\n

    In general, the clusters reflect differences in the overall intensity of strategy engagement, although these differences were relatively small, particularly for traditional, social, and oral rehearsal strategies. Accordingly, the cluster analysis is interpreted as providing a descriptive account of individual differences in strategy engagement rather than evidence for categorical learner types.<\/p>\n

    Two-Way Mixed ANOVA<\/h4>\n

    A two-way mixed repeated-measures ANOVA was conducted to examine differences in Rasch-estimated VS across time, with strategy clusters as a between-subjects factor and time as a within-subjects factor. Strategy clusters were treated as an index of overall strategy engagement rather than categorical learner types. Preliminary analyses indicated that the assumptions of homogeneity of variance and sphericity were met (Levene\u2019s test; Mauchly\u2019s test).<\/p>\n

    Table 4 shows significant main effects for both strategy clusters and time, suggesting that learners differed in overall VS according to their level of strategy engagement and that VS increased across the four measurement points. The significant cluster \u00d7 time interaction further reflects modest differences in developmental trajectories across strategy groups.<\/p>\n

    Table 4. ANOVA Results for Strategy Clusters and Time (Growth in VS)<\/strong><\/p>\n\n\n\n\n\n\n\n\n\n\n
    Source<\/td>\nSS<\/em><\/td>\ndf<\/em><\/td>\nMS<\/em><\/td>\nF<\/em><\/td>\np<\/em><\/td>\n\u03b72<\/sup><\/td>\n<\/tr>\n
    Between Subjects<\/em><\/td>\n<\/tr>\n
    Strategy Clusters<\/td>\n18294033.40<\/td>\n2<\/td>\n9147016.70<\/td>\n8.45<\/td>\n<. 001<\/td>\n.05<\/td>\n<\/tr>\n
    Residuals<\/td>\n20145854.50<\/td>\n186<\/td>\n1082880.94<\/td>\n<\/td>\n<\/td>\n<\/td>\n<\/tr>\n
    Within Subjects<\/em><\/td>\n<\/tr>\n
    Time<\/td>\n79637893.41<\/td>\n3<\/td>\n26545964.74<\/td>\n173.40<\/td>\n<<\/em>. 001<\/td>\n.21<\/td>\n<\/tr>\n
    Strategy Clusters \u00d7 <\/strong>Time<\/td>\n4440449.65<\/td>\n6<\/td>\n740074.94<\/td>\n4.83<\/td>\n<<\/em>. 001<\/td>\n.01<\/td>\n<\/tr>\n
    Residuals<\/td>\n85425801.24<\/td>\n558<\/td>\n153092.83<\/td>\n<\/td>\n<\/td>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    Note. N<\/em> = 189; VS = <\/strong>V<\/strong>ocabulary <\/strong>S<\/strong>ize; \u03b72 <\/sup>= .01 (small), .06 (medium), .14 (large).<\/strong><\/p>\n

    Table 5 presents the mean and standard deviation of estimated VS for each strategy cluster. Figure 3 shows the longitudinal trajectories of estimated VS across strategy clusters. Follow-up analyses revealed that although VS levels were comparable at Time 1 (p<\/em> = .06, \u03b72<\/sup> = .03), learners in the high-strategy-use group showed greater cumulative gains at later time points. In particular, Cluster 1 showed significantly larger increases than Cluster 3 by Time 4 (p<\/em> < .001, \u03b72<\/sup> = .13).<\/p>\n

    Table 5. Mean Estimated VS on Time (VST) of the Three Strategy Clusters<\/strong><\/p>\n\n\n\n\n\n\n\n
    <\/td>\nCluster 1
    \n(n<\/em> = 53)<\/td>\n    		Cluster 2
    \n(n<\/em> = 101)<\/td>\n    		Cluster 3
    \n(n <\/em>= 35)<\/td>\n<\/tr>\n
    Time 1 (VST 1)<\/td>\n2861.65 (512.00)<\/td>\n2673.64 (434.43)<\/td>\n2716.56 (471.71)<\/td>\n<\/tr>\n
    Time 2 (VST 2)<\/td>\n3251.45 (689.49)<\/td>\n3016.24 (628.75)<\/td>\n2763.74 (637.56)<\/td>\n<\/tr>\n
    Time 3 (VST 3)<\/td>\n3583.74 (811.34)<\/td>\n3394.65 (621.80)<\/td>\n3146.78 (612.26)<\/td>\n<\/tr>\n
    Time 4 (VST 4)<\/td>\n4056.02 (748.90)<\/td>\n3718.90 (606.15)<\/td>\n3291.81 (710.29)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    Note. Mean (Standard Deviation)<\/strong>.<\/strong><\/p>\n

    \"Figure<\/p>\n

    Figure 3. Estimated Vocabulary Size of Three Strategy Clusters at Four Time Points<\/strong><\/p>\n

    Note. Cluster 1 (<\/strong>n<\/em><\/strong> = 53), Cluster 2 (<\/strong>n<\/em><\/strong> = 101), and Cluster 3 (<\/strong>n<\/em><\/strong> = 35). <\/strong><\/p>\n

    Overall, these findings suggest that higher levels of strategy engagement are associated with greater longitudinal vocabulary growth. Importantly, the results reflect differences in the degree of strategy engagement rather than qualitatively distinct learning pathways, and no causal claims are implied.<\/p>\n

    Analysis of Open-Ended Survey Responses<\/h4>\n

    After data cleaning procedures\u2014including the removal of blank responses, correction of typographical errors and coding inconsistencies\u2014the open-ended responses were analyzed using KH Coder 3.0 (Higuchi, 2014). A total of 3,436 word tokens were identified across 293 sentences in response to the open-ended question on VLS. All analyses were conducted on the original Japanese responses to preserve semantic accuracy, after which representative words and excerpts were translated into English for reporting purposes.<\/p>\n

    Following the exclusion of duplicated entries and function words, 1,565 content words were retained for analysis. Consistent with the explanatory sequential mixed-methods design, the qualitative data were analyzed in relation to the quantitative strategy clusters, allowing characteristic lexical patterns to be identified for each group.<\/p>\n

    Table 6 presents the top 10 distinctive words for each VLS cluster. These keywords highlight differences in learners\u2019 self-reported strategy behaviors and provide interpretive insight into the quantitative cluster profiles. In this way, the qualitative findings serve not merely as illustrative examples but as an interpretive layer that contextualizes and elaborates the statistical patterns observed in both the cluster and longitudinal analyses.<\/p>\n

    Table 6. List of Distinctive Words Extracted from Each Cluster<\/strong><\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
    No.<\/td>\nCluster 1 (n <\/em>= 51)<\/td>\n<\/td>\nCluster 2 (n <\/em>= 97)<\/td>\n<\/td>\nCluster 3 (n <\/em>= 30)<\/td>\n<\/tr>\n
    Japanese<\/td>\nEnglish<\/td>\nJaccard<\/td>\n<\/td>\nJapanese<\/td>\nEnglish<\/td>\nJaccard<\/td>\n<\/td>\nJapanese<\/td>\nEnglish<\/td>\nJaccard<\/td>\n<\/tr>\n
    1<\/td>\n\u899a\u3048\u308b<\/td>\nlearn<\/td>\n.219<\/td>\n<\/td>\n\u5358\u8a9e<\/td>\nword<\/td>\n.345<\/td>\n<\/td>\n\u66f8\u304f<\/td>\nwrite<\/td>\n.210<\/td>\n<\/tr>\n
    2<\/td>\n\u767a\u97f3<\/td>\npronunciation<\/td>\n.117<\/td>\n<\/td>\n\u899a\u3048\u308b<\/td>\nlearn<\/td>\n.263<\/td>\n<\/td>\n\u898b\u308b<\/td>\nlook<\/td>\n.103<\/td>\n<\/tr>\n
    3<\/td>\n\u805e\u304f<\/td>\nlisten<\/td>\n.096<\/td>\n<\/td>\n\u898b\u308b<\/td>\nlook<\/td>\n.141<\/td>\n<\/td>\n\u610f\u5473<\/td>\nmeaning<\/td>\n.083<\/td>\n<\/tr>\n
    4<\/td>\n\u4f7f\u3046<\/td>\nuse<\/td>\n.086<\/td>\n<\/td>\n\u8aad\u3080<\/td>\nread<\/td>\n.110<\/td>\n<\/td>\n\u767a\u97f3<\/td>\npronunciation<\/td>\n.079<\/td>\n<\/tr>\n
    5<\/td>\n\u30c6\u30b9\u30c8<\/td>\ntest<\/td>\n.082<\/td>\n<\/td>\n\u97f3\u8aad<\/td>\nreading aloud<\/td>\n.110<\/td>\n<\/td>\n\u7e70\u308a\u8fd4\u3059<\/td>\nrepeat<\/td>\n.074<\/td>\n<\/tr>\n
    6<\/td>\n\u8cea\u554f<\/td>\nquestion<\/td>\n.065<\/td>\n<\/td>\n\u7e70\u308a\u8fd4\u3059<\/td>\nrepeat<\/td>\n.102<\/td>\n<\/td>\n\u7db4\u308a<\/td>\nspelling<\/td>\n.060<\/td>\n<\/tr>\n
    7<\/td>\nTOEIC<\/td>\nTOEIC<\/td>\n.064<\/td>\n<\/td>\n\u30c6\u30b9\u30c8<\/td>\ntest<\/td>\n.089<\/td>\n<\/td>\n\u82f1\u8a9e<\/td>\nEnglish<\/td>\n.050<\/td>\n<\/tr>\n
    8<\/td>\n\u82f1\u8a9e<\/td>\nEnglish<\/td>\n.060<\/td>\n<\/td>\n\u30ce\u30fc\u30c8<\/td>\nnotebook<\/td>\n.071<\/td>\n<\/td>\n\u601d\u3044\u6d6e\u304b\u3079\u308b<\/td>\nrecall<\/td>\n.044<\/td>\n<\/tr>\n
    9<\/td>\n\u610f\u8b58<\/td>\nawareness<\/td>\n.048<\/td>\n<\/td>\n\u4f7f\u3046<\/td>\nuse<\/td>\n.071<\/td>\n<\/td>\n\u7d19<\/td>\npaper<\/td>\n.044<\/td>\n<\/tr>\n
    10<\/td>\n\u8003\u3048\u308b<\/td>\nthink<\/td>\n.047<\/td>\n<\/td>\n\u52c9\u5f37<\/td>\nstudy<\/td>\n.054<\/td>\n<\/td>\n\u6388\u696d<\/td>\nclass<\/td>\n.044<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    Note. N<\/em> = 178;<\/strong> Higher values indicate stronger associations.<\/strong><\/p>\n

    Figure 4 shows a co-occurrence network illustrating the relationships between frequently occurring words and the strategy clusters. In the network, line thickness represents the strength of co-occurrence, and circle size indicates word frequency. Larger nodes located near the center of the network represent words commonly used across all clusters, suggesting that activities such as looking up words, writing, reading, and self-testing were widely reported by learners. The network also reveals lexical tendencies associated with each strategy cluster, providing a basis for the qualitative interpretations presented below.<\/p>\n

    \"Figure<\/p>\n

    Figure 4. Co-occurrence Network in Responses to Open-ended Question<\/strong><\/p>\n

    Note. Node size represents word frequency. Edge thickness reflects co-occurrence coefficients. Japanese labels are followed by English translations.<\/strong><\/p>\n

    It should be noted that individual responses often contained multiple VLS (e.g., writing, repeating, pronouncing). In the examples below, italicized expressions highlight the lexical items most strongly associated with each cluster in the co-occurrence analysis. Cluster membership was determined based on overall co-occurrence patterns rather than the presence of a single word. Consequently, some responses may include strategies that also appear in other clusters.<\/p>\n

    Words frequently associated with Cluster 1 included listen<\/em>, use<\/em>, and question<\/em>, suggesting more active and self-regulated engagement with VLS. Representative participant responses illustrating these tendencies are presented below (translated from the original Japanese, with cluster-related expressions italicized). These examples help contextualize the quantitative cluster distinctions.<\/p>\n