Exploring awareness of metacognitive knowledge and acquisition of vocabulary knowledge in primary grades: a latent growth curve modelling approach

Abstract

This study explores how primary school children develop their metacognitive knowledge and vocabulary knowledge and how both types of knowledge are dynamically correlated from 1st to 4th grade. The longitudinal sample included 426 first-grade students (M = 6.6 years, SD = .51; 50.2% boys, 49.7% girls) from five public primary schools in China. A set of tests on metacognitive knowledge and vocabulary knowledge were administered four times over four years. The one-on-one-basis metacognitive knowledge test was based on students’ explanations about cognitive activities; the vocabulary knowledge test focused on students’ breadth and depth of vocabulary knowledge. Latent growth curve modelling was used to study developmental change. Results showed that participants’ metacognitive knowledge and vocabulary knowledge improved from 1st to 4th grade but not in a cumulative fashion. Participants’ level of metacognitive knowledge was strongly associated with their vocabulary knowledge throughout the selected school years. The development of vocabulary knowledge breadth lagged behind vocabulary knowledge depth during the study period. These findings shed light on primary school students’ development of metacognitive knowledge and vocabulary knowledge.

Keywords:

• Metacognition
• metacognitive knowledge
• vocabulary knowledge
• breadth of vocabulary knowledge
• depth of vocabulary knowledge

Introduction

Although a large body of literature exists on vocabulary knowledge, few studies have evaluated the dynamics between individuals’ metacognitive knowledge and vocabulary knowledge from a longitudinal developmental perspective. Metacognitive knowledge and vocabulary knowledge are commonly acknowledged as two slowly developing academic skills. Learners who can reflect on their cognitive activities and metacognitive knowledge may use self-directive capabilities to develop the need to acquire vocabulary knowledge or to manage cognition when obtaining vocabulary knowledge. As argued by Schraw (1994), metacognitive knowledge can help learners better direct, guide, and govern their learning and comprehension.

Extensive research on metacognition throughout the past decade has revealed the pivotal role of metacognitive knowledge in academic learning (e.g. Annevirta et al., 2007; Hacker, 1998; Teng, 2016a). Effective comprehension of texts and the learning of words from texts require active thinking, metacognitive processing, and strategic actions (e.g. Teng & Zhang, 2021). As indicated in early studies (e.g. Paris & Newman, 1990), a repertoire of metacognitive strategies or a high level of self-regulation can differentiate higher-achieving students from low achievers. Similarly, Annevirta and Vauras (2001) pointed out that primary school learners can enhance their metacognitive knowledge as they progress in age and learning experience. Research has also shown the indispensable role of metacognitive knowledge in primary school learning, yet exactly when and how young learners develop metacognitive knowledge—and how this development may be associated with vocabulary knowledge—remain unclear.

Indeed, young learners must learn to think and be conscious of their cognitive activity before they can become active agents of their own cognitive processing. Young learners also have to learn to use meta-knowledge and self-directive capabilities to control their feelings during learning (Annevirta & Vauras, 2001). Flavell (1979) noted that children cannot talk about and reflect on their learning as adults can. If children are given opportunities to think about learning, and if teachers facilitate these instances appropriately, then learners should gradually come to understand and develop metacognitive knowledge. In fact, reflecting on how one learns is a key aspect of metacognitive awareness, for which children must acquire the necessary skills to retrieve cues in relevant task situations that can ‘influence the course of the cognitive enterprise without itself entering into consciousness’ (Flavell, 1979, pp. 907–908).

In terms of language learning, metacognitive knowledge characterizes an approach of expert learners. As argued by Sato (2021), development of metacognition is a type of individual difference that can be trait-like and state-like. For example, individual differences in metacognitive knowledge determines whether a language learner can develop an awareness of learning, facilitate recall, accelerate their progress, and enhance the quality and speed of cognitive engagement (Wenden, 1998). The importance of metacognitive knowledge also extends to vocabulary learning, a complex process involving a dynamic interplay of metacognitive skills and components of vocabulary knowledge (Teng & Zhang, 2021). According to Ellis (2005), the secret to effective vocabulary learning/acquisition lies in the various cognitive processes and conditions that shape language learners’ progression towards better mastery of vocabulary knowledge.

Vocabulary knowledge is considered a multidimensional construct (Read, 2000). This type of knowledge is crucial in determining whether learners can gain knowledge of a word and use it appropriately during academic English learning (Schmitt, 2008). Vocabulary knowledge also facilitates the acquisition of the form and meaning of new words and helps learners achieve deeper comprehension of spoken and written language (Schmitt, 2010; Teng, 2021). As Webb and Nation (2017) proposed, the development of vocabulary knowledge represents a pressing concern in the language-learning process. Numerous factors, such as metacognitive awareness, may affect university students’ acquisition of vocabulary knowledge (Teng & Zhang, 2021). However, based on a longitudinal study (Pham & Kohnert, 2014), children’s acquisition of vocabulary knowledge differs from that of adult learners, in that children are more likely to demonstrate stable growth when learning English lexical items. Song et al. (2015) contended that the development of young learners’ vocabulary knowledge relies on language- related metacognitive skills, which can improve as students move through school or accumulate learning experience.

Thus far, few studies have concentrated on metacognitive knowledge among children in early elementary school (Haberkorn et al., 2014), let alone its dynamic interaction with vocabulary knowledge. The present study aims to bridge this gap by examining developmental dynamics between primary school young learners’ metacognitive knowledge and vocabulary knowledge along with their cumulative learning experience. In particular, the present study marks an attempt to examine the possible mutual relationship between learners’ metacognitive knowledge and development of the breadth and depth of vocabulary knowledge.

Literature review

Metacognitive knowledge

Research on metacognitive development began in the early 1970s. Flavell (1979) proposed a pioneering definition of metacognition: any knowledge or cognitive activity that takes as its cognitive object, or that regulates, any aspect of any cognitive activity. This early conceptualization focused on individuals’ knowledge of information-processing skills, cognitive tasks, and strategies when implementing tasks. Metacognition is a multifaceted concept, which includes the knowledge that human thinkers possess about their own and others’ cognitive processes and executive skills related to planning, monitoring, and evaluating one’s own cognitive activities. Flavell (1979) also described three major facets of metacognition, namely metacognitive knowledge, metacognitive experiences, and metacognitive skills.

The present study revolves around metacognitive knowledge, which, within the broader construct of metacognition, has been conceptualized as the declarative component besides children’s procedural activities in regulating and monitoring memory performance during a task (Flavell, 1979). According to Schraw (1994), metacognitive knowledge includes person knowledge (i.e. knowledge of individual strength or weakness that influence one’s cognitive activity), task knowledge (i.e. knowledge of processing demands required to complete a task), and strategy knowledge (i.e. knowledge of different strategies required to successfully accomplish a task). The three dimensions of metacognitive knowledge function as a prerequisite for learners to internalize metacognitive knowledge and adopt it as a capacity to regulate their cognitive activity. Based on Flavell’s (1979) conceptualization, Brown (1987) depicted metacognitive knowledge as including declarative knowledge (i.e. knowledge of factors that influence one’s performance and role as a learner), procedural knowledge (i.e. knowledge of how to execute procedural skills), and conditional knowledge (i.e. knowledge of how to determine the time and conditions in which to apply various cognitive actions).

Metacognitive knowledge can be acquired formally or informally, deliberately or incidentally. Metacognitive knowledge therefore refers to what learners know about themselves. Learners can become conscious of and articulate what they know. Based on Teng’s (2016a) work, access to metacognitive knowledge facilitates learners’ thinking and self-regulation. However, the participants were university students. In a recent study (Sato & Dussuel Lam, 2021), the findings suggested that although development of metacognition for young learners can be obscure or inaccurate, metacognitive instruction can improve L2 young learners’ metacognition as well as classroom participation patterns.

Overall, whereas Flavell and Wellmann (1977) characterized metacognitive knowledge as late-developing, other researchers have framed metacognitive knowledge as the beliefs or thinking abilities possessed by learners of all ages, from primary school students (Annevirta et al., 2007) to secondary school learners (e.g. Schneider et al., 2017) and university students (Cotterall & Murray, 2009). One’s understanding of metacognitive knowledge (e.g. person, task, and strategy knowledge) develops significantly beginning in the early primary school period and does not peak until young adulthood (Schneider, 2008).

Longitudinal studies on metacognitive knowledge

Although Flavell (1979) called for more research to evaluate young learners’ development of metacognitive knowledge, only a few longitudinal studies have delineated this process in primary school learners. In Schneider and Sodian (1991) study, two groups of primary school students (ages 4 and 6, respectively) were involved in a sort-recall task. Learners were required to choose the best strategy among given alternatives and provide explanations for their choices. Results showed that learners built an awareness of their effectiveness and increased metacognitive knowledge as they grew older. However, the learners varied in their development of metacognitive knowledge. A longitudinal study on the development of metacognitive knowledge from pre-school to 3rd grade involving 196 primary school students in Finnish schools (Annevirta & Vauras, 2001) revealed that, although many learners developed metacognitive knowledge during the first three years, some 3rd-grade learners still had low levels of metacognitive knowledge. In other words, many learners still did not understand the role of the learner in facilitating cognitive ability. In another longitudinal study focusing on primary school students in Finland (Annevirta et al., 2007), results showed that opportunities to enhance metacognitive knowledge varied individually. Learners followed different developmental paths in acquiring certain types of metacognitive knowledge. For instance, some learners entering 2nd grade acquired more cognitive processing than some learners entering 3rd grade.

Recent studies (e.g. Marulis et al., 2016) have evaluated the development of metacognitive knowledge in 3- to 5-year-old preschool children. Findings suggest that older children achieve higher scores in metacognitive knowledge over the length of a school year. However, variations across children have also been detected. As Irak and Çapan (2018) contended, the ability to articulate metacognitive knowledge and apply this knowledge to a learning task can predict subsequent metacognitive development and academic success. Essentially, metacognitive knowledge is already present in young learners and develops steadily over the elementary school years. Yet Schneider (2008) argued that differences exist in learners’ developmental trends related to declarative and procedural metacognitive knowledge. For example, despite slow and steady improvement in declarative metacognitive knowledge through childhood, trends are not as clear-cut for procedural metacognitive knowledge.

Hence, learners tend to display a growing proportion of metacognitive knowledge grade by grade (Teng & Zhang, 2021). Enhanced acquisition of metacognitive knowledge acts as the foundation for understanding metacognitive strategies and consequently using those strategies to identify, evaluate, and enhance learning. When progressing to higher school grades, students’ opportunities to enhance metacognitive knowledge are assumed to increase along with their accumulated learning experience. However, primary school students’ understanding of metacognitive knowledge appears rudimentary or inconsistent to some extent, highlighting the complexity of developing this type of knowledge. Young learners have been shown to improve their metacognitive knowledge at different rates, which may explain why their academic success varies in kind (Smortchkova & Shea, 2020). In addition, disagreement persists regarding whether metacognitive knowledge can be developed in specific domains (e.g. vocabulary) before broad skills or whether broad skills are acquired before knowledge in specific areas (for a review, see Schneider & Löffler, 2016). More longitudinal research is therefore needed to explore the development of metacognitive knowledge and to test the assumption that metacognitive knowledge can be developed in particular domains (e.g. vocabulary knowledge) with increasing age, ideally leading to ‘metamnemonically sophisticated persons’ in vocabulary learning (Flavell, 1979, p. 908).

Vocabulary knowledge (VK)

Vocabulary knowledge is an essential construct in language acquisition (Read, 2000). The breadth of vocabulary knowledge (BVK; i.e. the number of words for which a learner has at least some superficial knowledge of their meaning) and depth of vocabulary knowledge (DVK; i.e. how well learners understand a lexical item) have been widely discussed. Having strong mastery of both dimensions facilitates unassisted comprehension of written (Qian, 2002) and spoken texts (Teng, 2016b). In particular, learners’ receptive L2 vocabulary size had an influence on the number of form-meaning elaborations they could make (Candry et al., 2017). Per Read (2000), having a superficial understanding of the meaning of a word is not sufficient; learners should develop rich and detailed knowledge of each individual lexical item that can help them function within a language.

In particular, BVK is relatively amenable to intentional learning, whereas DVK is much more challenging to acquire and may need to be learned through extensive exposure to the target language (Schmitt, 2010). Accordingly, DVK may be learned or mastered sooner than BVK (Schmitt, 2008). This pattern provides implications for the current research, as students in primary grades may assume different stages of vocabulary knowledge acquisition. Given the incremental nature of acquiring vocabulary (Schmitt, 2010), examining developmental trends in vocabulary knowledge is worthwhile.

Longitudinal studies on vocabulary knowledge

The ability to understand and use vocabulary knowledge is an important stimulus for language comprehension. Acquiring vocabulary knowledge is often regarded as a difficult task, as it requires personal interpretation by attending to contextual clues (Teng, 2019, 2020a), mapping lexical form to meaning (Chilton & Ehri, 2015), and applying metacognitive strategies (Yamada, 2018). The task of orchestrating a multiplicity of cognitive processes to acquire vocabulary knowledge is essential, and understanding how various dimensions of vocabulary knowledge develop highlights the value of longitudinal research. In an early longitudinal study focusing on a young child (Clark, 1973), vocabulary knowledge acquisition appeared less differentiated for a young child than for an older person: the child could make more progress and pick up missing features of vocabulary knowledge as their age increased.

Zhang and Lu (2014) conducted a longitudinal study measuring growth in BVK and vocabulary fluency as well as how these two variables were correlated. Participants consisted of 300 students at a Chinese university who received a vocabulary levels test at three points in time over the course of 22 months. Results revealed the effect of frequency level on the acquisition of BVK and vocabulary fluency. Findings also showed that the acquisition of vocabulary fluency lagged behind that of BVK. In a later study by Zhang and Lu (2015), compared with BVK, DVK was a more significant predictor of language learning performance. Relatedly, compared with BVK, DVK was a more significant predicter in reading test outcomes (Qian, 2002) and lexical inferencing (Nassaji, 2006). In an earlier study (Gu & Johnson, 1996), metacognitive strategies such as self-initiation and selective attention were closely associated with vocabulary knowledge.

Although the above studies shed light on the development of vocabulary knowledge, their main focus was on tertiary-level students. One study (Mcbride-Chang et al., 2008) involved 211 kindergartners from Hong Kong, 288 kindergarteners from Beijing, and 164 kindergartners from Korea. The authors intended to evaluate morphological awareness and vocabulary knowledge acquisition concurrently and longitudinally. Results indicated that vocabulary knowledge explained a unique proportion of variance in young learners’ morphological awareness, demonstrating the bidirectional influences of vocabulary development and morphological awareness. In a longitudinal study using growth curve analysis to explore 149 primary school students’ vocabulary knowledge and reading comprehension (Cheng et al., 2017), findings indicated that learners’ initial status and growth rates of vocabulary knowledge (i.e. compounding awareness) made a significant direct contribution to reading comprehension. Results also showed that compounding awareness developed quickly during the early primary school years. In a study focusing on the Spanish EFL context (Gallego & Llach, 2009), 224 learners in their 4th, 5th and 6th grades of primary education completed the Vocabulary Levels Test. The findings documented the incremental nature in developing receptive English vocabulary size for the learners. Learners significantly increased their receptive vocabulary knowledge as they moved up a grade. In another study (Kieffer & Lesaux, 2012), the focus was on young learners’ English vocabulary development. A total of 90 Spanish-speaking learners was followed from fourth through seventh grade. Latent growth modelling results suggested a relationship between derivational morphological awareness and vocabulary. In addition, the learners demonstrated rapid, linear growth in morphological awareness and vocabulary knowledge across grades. The above studies underline the possibility of young learners’ development in vocabulary knowledge in the early school years.

Research gap and research questions

Judging by the literature review, primary school students’ metacognitive knowledge and vocabulary knowledge can be expected to deepen throughout primary grades. Previous studies provided some details about the relationship between metacognitive strategies and vocabulary knowledge; even so, relatively little is known about the longitudinal developmental dynamics between metacognitive knowledge and vocabulary knowledge, particularly in primary grades. Specifically, empirical data is needed to support the developmental dynamics between metacognitive knowledge and vocabulary knowledge across primary school grades. The developmental patterns of vocabulary knowledge and metacognitive knowledge may follow the Matthew Effect (Perc, 2014), in that learners with better levels of vocabulary knowledge and metacognitive knowledge tend to exhibit strong vocabulary knowledge and metacognitive knowledge in their later schooling. In addition, learners possessing greater metacognitive knowledge may perform better in terms of their long-term development of vocabulary knowledge. This hypothesis highlights the importance of longitudinal research in delineating developmental trends in metacognitive knowledge and vocabulary knowledge during primary school. Three research questions framed this study:

1. What developmental trends related to metacognitive knowledge emerge during primary grades?

2. What developmental trends related to vocabulary knowledge (i.e. BVK & DVK) emerge during the primary school years?

3. What developmental interaction exists between metacognitive knowledge and vocabulary knowledge (i.e. BVK & DVK) during primary grades?

Method

Research design

This study was designed to examine the development of young learners’ metacognitive knowledge and vocabulary knowledge along with the dynamics of developing these types of knowledge across four time periods. The adopted tests were identical across all periods (Years 1–4). Repeated tests may lead students to deliberately memorize some items, representing one limitation of longitudinal studies. However, the test results in the present study did not exhibit floor or ceiling effects on any measure (e.g. per the skewness and kurtosis values in Table 1).

Table 1. Descriptive statistics for 4 time-point measurements.

Dimensionality	Descriptive statistics		Normality index
	M	SD	Skewness	Kurtosis
MCK (Year 1)	29.143	6.811	−0.058	−1.147
MCK (Year 2)	36.76	12.580	0.030	−1.123
MCK (Year 3)	45.75	13.297	0.080	−1.002
MCK (Year 4)	53.997	13.362	0.013	−0.848
DVK (Year 1)	14.089	6.759	0.231	−0.631
DVK (Year 2)	25.593	8.873	0.212	−0.851
DVK (Year 3)	30.692	9.808	0.238	−0.825
DVK (Year 4)	36.084	10.625	0.312	−0.467
BVK (Year 1)	10.6257	5.850	−0.028	−0.248
BVK (Year 2)	34.340	10.681	0.737	0.368
BVK (Year 3)	44.166	12.787	0.656	−0.364
BVK (Year 4)	53.108	12.859	0.621	−0.318

Participants

The original sample consisted of 456 primary school first-grade students. Participants were from five public primary schools in a large city in southern China. Their English instruction began in primary school Grade 1. However, many participants reported that they had started to learn basic English during kindergarten. The participants were monolingual Chinese students learning English as a foreign language (EFL). Participation in this study was supported by the dean of the teaching department at each participating school. Participants attended this study on a voluntary basis, and their parents signed consent forms. Some families moved to other places during the study. The final dataset included 426 young learners (M = 6.6 years, SD=.51), including 214 boys (50.2%) and 211 girls (49.7%). The 426 learners were followed from first grade to fourth grade. Thirty students who could not attend the entire study were excluded from data analysis. However, Singer and Willett (2003) stated that longitudinal methods do not require full participants across waves, suggesting that students who did not attend the entire study could be included in analysis. We chose not to include them in our analysis because participant retention is vital to ensuring the power and internal validity of longitudinal research. Including those participants could increase the risk of bias, as they were lost to follow-up.

Measures

Measures in the present study included metacognitive knowledge and vocabulary knowledge (BVK and DVK).

Metacognitive knowledge (MCK)

Young learners’ metacognitive knowledge (MCK) was assessed using two tests. The first MCK test included a series of verbally- and pictorially-presented tasks. It measured young learners’ knowledge of cognitive processes, specifically their abilities to remember, understand, and learn something. Each cognitive process entailed eight tasks. In each task, participants were presented with two or three pictures. Each picture depicted a situation in which a young boy or girl was expected to remember, understand, or learn something. A detailed description of the 24 tasks can be found in Annevirta and Vauras (2001). This test was also applied in Teng and Zhang (2021). Figure 1 shows a sample picture task that measured participants’ knowledge about understanding the rules of a game and the strategy in seeking explanation.

Figure 1. Example of task in MCK test.

The experimenter verbally described each situation in the task. The young learners, while listening to the experimenter’s explanation, independently looked at the drawings. The learners then selected the best response choice. The chosen option reflected the learners’ reported way of remembering, understanding, or learning something during a cognitive task (e.g. answering the question ‘What was the best way for the boy to remember the details of a story?’). The learners made their own choices and were then asked to verbally explain why they selected this option (e.g. ‘Why did the boy you chose can remember the details of a story?’) (Figure 1). This recall interview served as the second MCK test. If some young learners found the questions difficult to answer, the experimenter could guide the learners in answering by instructing them to imagine themselves as the young child in the given situation. A concrete prompt was also provided to help participants imagine what they might do to remember, understand, or learn something. The prompts were as follows: ‘How would you try to remember as much information as possible from this activity?’ (memory), ‘How would you try to understand some information and teach it to other students?’ (comprehension), and ‘How would you learn to help your classmates solve problems?’ (learning). Through these prompts, learners were guided to reflect on and explain the factors that would influence their cognitive activities. The young learners were expected to provide explanations and form answers based on their own thoughts and experiences.

As noted, each of the 24 tasks included two or three options. The plan was to score the selected options. Although most participants chose the best option in many tasks, random guesses could not be ruled out. Only participants’ responses when explaining their reasons for choosing related options were included in data analysis.

Participants’ explanations were assessed on a 4-point scale. Zero points were given if the participants did not explain or their explanations were irrelevant or naïve (e.g. ‘I just think it’s good’ or ‘I am not sure, but I just like it’). One point was awarded for implicit, indirect references to the interviewee’s cognitive processing (e.g. ‘Reading a book can help me know more’, ‘I can find something new in the book’, or ‘The teacher said reading books is good’). Two points were awarded for fairly adequate cognitive mental processing (e.g. ‘You need to show how the game was played in order to explain the rule’ or ‘You can know more while doing’). Three points were awarded for more explicit explanations related to cognitive processing (e.g. ‘Rules can be better understood through combining practice and knowledge’).

Scores for explanations of each cognitive task were then summed. The resulting MCK score (min = 0, max = 72) was applied in statistical analyses. Participants’ scores reflected the quality of oral explanations related to metacognitive knowledge. Young learners’ metacognitive knowledge was based on their reflection and explanations of their academic routines and cognitive mental processing, which, according to Annevirta et al. (2007), accorded well with Flavell’s (1976) theoretical conceptualization that metacognitive knowledge reflects learners’ comprehension of their cognitive processes.

Coefficient alpha values for the MCK tests were .77–.80 (1st grade), .78–.81 (2nd grade), .81-.83 (3rd grade), and .78–.83 (4th grade). These values ensured a sound degree of internal consistency for each primary school grade. Three independent judges who did not teach the participants were recruited for scoring. They rated participants’ verbal explanations twice. Interrater reliability between the three judges reached a unanimous level of 86–95% for answers in the second round of rating. Disagreements were resolved through discussion.

Vocabulary knowledge (VK)

The test measuring participants’ breadth of vocabulary knowledge (BVK) was adapted from Nation and Beglar (2007) Vocabulary Size Test, a reliable test of learners’ receptive vocabulary size. The test originally had 140 items. We included the first 100 items, as others were too challenging for the participants to answer. The difficulty level for the 100 items was 70%, equal to ideal difficulty levels for 5-response multiple-choice items in terms of discrimination potential (Feldt, 1993). The difficulty levels of excluded items were lower than 50%, indicating they would be too difficult for participants to answer (Feldt, 1993). Each multiple-choice item had five options. Participants were encouraged to choose the meaning that corresponded to the given word. If they were unsure, they could choose the option ‘I don’t know’. In contrast to Nation and Beglar (2007) study, a native-language (L1) explanation was provided rather than a second-language (L2) explanation; the 10 teachers in the study agreed on the appropriateness of using L1 for Chinese primary school students and that L1 explanations may better reflect learners’ incremental receptive knowledge. An example item is provided below:

Solider: He is a solider.

1. A. 商人 (Person in a business)

2. B. 学生 (Student)

3. C. 金属工艺制造者 (Person who uses metal)

4. D. 士兵 (Person in the army)

5. E. 我不知道 (I don’t know)

Participants were expected to select one option. The correct answer was awarded 1 point, and the maximum score on this test was 100 points. Cronbach’s alpha values for the instrument were .72 (1st grade), .75 (2nd grade), .71 (3rd grade), and .78 (4th grade), indicating internal consistency. Item response theory (IRT) analysis revealed that discrimination indices ranged between .43 and .84, and slope parameter values ranged from .51 to 1.35; the BVK test was therefore valid. This test was also discriminant, enabling classification of high and low performers. The three judges were unanimous in their scoring and thus affirmed interrater reliability.

The test of participants’ depth of vocabulary knowledge (DVK) was adapted from the Word Associates Test (Read, 2004). In the present study, the DVK measure included 100 keywords learned throughout all primary school grades. These 100 words were determined through a discussion among 10 experienced primary school English teachers. The test measured depth of vocabulary knowledge based on three aspects: paradigmatic, syntagmatic, and lexical progression. The goal was to determine whether participants could identify either collocational, synonymous, part–whole, or whole–part relationships between a test item and the given words. Each test item included a stimulus word and two boxes. Each box included four given words. Learners were asked to locate the correct associate with the stimulus word from eight options; see below for an example.

Delighted

In the above example, adjectives appear in the left box with four nouns in the right box. The stimulus word (‘delighted’) has no synonymous or part–whole relationships with the given words (‘ugly’, ‘smart’, ‘lucky’, ‘angry’) in the left box. The stimulus word (‘delighted’) has a collocational relationship with one word (‘party’) in the right box. Each test item included one correct word option in either the left or right box. Each correct response was awarded 1 point; the maximum possible test score was 100.

Cronbach’s alpha values for this measure were .72 (1st grade), .71 (2nd grade), .75 (3rd grade), and .77 (4th grade), indicating sound reliability. Interrater reliability among the three judges reached a unanimous level of 100%. Again, IRT model analysis suggested that based on discrimination indices (from .42 to .95) and slope parameter values (from .52 to 1.45), the DVK test was valid in assessing learners with different proficiency levels.

Procedure

Data collection began in the second semester of 1st grade because learners needed time to become familiarized with the learning situation in their respective schools. The tests were repeated annually in the second semesters of 2nd, 3rd, and 4th grade. MCK assessments were administered to participants on a one-by-one basis by the experimenter. All participants’ verbal explanations were recorded and transcribed. VK tests were administered individually in paper-and-pencil format. The two VK tests were administered in each school by the experimenter. All tests were presented in Mandarin Chinese. One may argue that there could be a testing effect problem because the same tests were used over the four years of the study. To hopefully prevent learners from deliberately memorizing items, new test items were added each year. All new items were high-frequency words that participants would easily understand. These additions were developed based on the format of VK tests. Assuming that the tests were administered once per year and that the same number of new test items would be added to each test yearly, testing effects were likely minimized. New items added to each test were not included in data analysis but were used to divert participants’ sole attention on target test items.

Statistical analysis

To address the need to investigate participants’ developmental changes, latent growth curve (LGC) analysis under the linear structural equation modelling framework was employed. The Mplus program was used to evaluate longitudinal data based on identical multiple indicators at each time point (Geiser, 2013). The measurement of each variable (i.e. MCK, DVK, and BVK) over time (from Year 1 to Year 4 in this study) was divided into several latent components to measure linear or quadratic trends. Other means, including the chi-square test and various goodness-of-fit statistics, were employed to evaluate whether the data fit the models. Index values to evaluate model fitness included the root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), goodness-of-fit index (GFI), and comparative fit index (CFI). The chi-square evaluates the discrepancy between an observed covariance matrix and the covariance matrix implied by a given model. The RMSEA explains the goodness of fit and per degree of freedom of the model compared to the population covariance matrix. A cut-off value for indicating a good fit is close to 0.06 for RMSEA (Hu & Bentler, 1999). The SRMR is the average of the standardized residual of the predicted covariance matrix from the observed covariance matrix, with a good-fit threshold of 0.09 (Hu & Bentler, 1999). The Tucker–Lewis index (TLI) assesses a model’s improvement relative to the null model, with a value close to 0.96 considered adequate (Hu & Bentler, 1999). The CFI is an index of the extent to which a model’s fit improves compared with the independent model; a value close to 0.96 is the agreed-upon critical value for fit indices such as GFI and CFI (Hu & Bentler, 1999).

Results

Data analysis consisted of two steps. The first step involved using LGC models to perform a uni-construct analysis, focusing on associations between the initial level of a test skill (MCK, BVK, and DVK) and its growth over time. The second step was to build the uni-construct models of MCK, BVK, and DVK through multivariate LGC analyses. Table 1 presents the normality index, means, and standard deviations of observed variables. Figure 2 depicts individual trajectories of MCK, BVK, and DVK.

Figure 2. Mean vectors of MCK and VK (DVK & BVK) skills.

Separate analyses based on LGC models measured the degree to which MCK and VK would be associated with the same variables of MCK and VK at the uni-construct level. The extent to which MCK and VK would be associated with MCK and VK refers to two growth factor components: an intercept growth factor (level) and the linear growth rate estimated for each variable. The first step was to evaluate the linear growth model. If the model did not fit with linear growth, then the second step was to determine whether the quadratic growth rate (quadratic trend) fit the data well. The model was built by fixing the loadings of observed variables across primary school years 1–4. Tables 2–4 present the model results; Figures 3–5 illustrate the constructed model based on model results.

Figure 3. Growth rate model for MCK.

Figure 4. Growth rate model for DVK.

Figure 5. Growth rate model for BVK.

Table 2. Results for MCK model.

	Estimate	S.E.	t	sig.
Means
I	28.837	0.588	49.039	0.000
S	7.731	0.188	41.124	0.000
Q	0.194	0.060	3.203	0.001
Variance
I	129.268	10.072	12.835	0.000
S	−0.389	5.027	−0.077	0.938
Q	0.854	0.161	5.316	0.000
Covariance
I–S	19.678	5.994	3.283	0.001
S–Q	−4.808	1.504	−3.197	0.001
Q–I	−0.759	0.889	−0.854	0.393
n	426
LL	−4965.675
AIC	9957.351
BIC	10010.058
Linear model	χ^2 = 294.827, df = 5, CFI = 0.767, TLI = 0.720, RMSEA = 0.522, SRMR = 0.815
Quadratic model	χ^2 = 28.367, df = 1, CFI = 0.992, TLI = 0.954, RMSEA = 0.253, SRMR = 0.017

Table 3. Results for DVK model.

	Estimate	S.E.	t	sig.
Means
I	13.440	0.429	31.332	0.000
S	11.032	0.314	35.174	0.000
Q	−1.298	0.081	−15.976	0.000
Variance
I	56.217	6.144	9.150	0.000
S	28.514	4.792	5.950	0.000
Q	0.501	0.263	1.905	0.057
Covariance
I–S	−3.225	4.497	−0.717	0.473
I–Q	1.351	1.043	1.296	0.195
S–Q	−4.881	0.843	−5.788	0.000
n	426
LL	−4910.066
AIC	9846.132
BIC	9898.840
Linear model	χ^2 = 483.584, df = 5, CFI = 0.433, TLI = 0.320, RMSEA = 0.670, SRMR = 0.931
Quadratic model	χ^2 = 236.285, df = 1, CFI = 0.912, TLI = 0.475, RMSEA = 0.743, SRMR = 0.011

Table 4. Results for BVK model.

	Estimate	S.E.	t	sig.
Means
I	18.849	0.306	61.598	0.000
S	14.985	0.486	30.811	0.000
Q	−1.249	0.107	−11.670	0.000
Variance
I	37.572	4.567	8.226	0.000
S	88.999	7.528	11.823	0.000
Q	3.251	0.389	8.352	0.000
Covariance
I–S	2.478	4.046	0.613	0.540
S–Q	−0.569	0.868	−0.656	0.512
Q–I	−17.753	1.494	−11.879	0.000
n	426
LL	−5167.251
AIC	10360.503
BIC	10413.210
Linear model	χ^2 = 4395.492, df = 5, CFI = 0.325, TLI = 0.190, RMSEA = 0.605, SRMR = 0.925
Quadratic model	χ^2 = 67.056, df = 1, CFI = 0.973, TLI = 0.837, RMSEA= 0.394, SRMR = 0.057

In terms of the univariate model for MCK (Table 2 and Figure 3), the level and linear trend components were stochastic, indicating that variance in the component was significant and reflected individual variability in participants’ initial levels and rates of change over time. In addition, the variance of the quadratic term was statistically significant, suggesting that this quadratic trend was also stochastic. The covariance between the latent level and linear term was statistically non-significant (p = .393). Thus, no significant association existed between the level and growth components of MCK. The positive means of the level and the stochastic trend and deterministic quadratic term were α¹(SE) = 10.072, α²(SE) = 5.027, and α³(SE) = .161, respectively. These values indicated that the MCK improvement from 2nd to 3rd grade was greater than that observed from 1st to 2nd grade. However, the stochastic trend and deterministic quadratic term [α¹(SE) = 12.835, α²(SE) = −.077, and α³(SE) = 5.316] revealed that MCK improvement from 3rd to 4th grade was less than that observed from 2nd to 3rd grade. Overall, CFI, TLI, RMSEA, and SRMR values showed that MCK development across the four measurement periods displayed a quadratic slope; however, developmental growth did not follow a cumulative trend.

In terms of the univariate model for DVK (Table 3 and Figure 4), the variance in the linear and quadratic terms was α¹(SE) = 6.144, α²(SE) = 4.792, and α³(SE) = .0.263. These values implied that both developmental trends were stochastic. The mean parameters of latent components were statistically significant (p<.05). The positive deterministic linear trend suggested that DVK increased each year. However, the improvement in DVK from 2nd to 3rd grade was not as pronounced as that from 1st to 2nd grade. No significant difference emerged in the improvement from 3rd to 4th grade or from 2nd to 3rd grade. The CFI, TLI, RMSEA, and SRMR values did not indicate that DVK development across the four time periods followed a fully linear or quadratic slope. In addition, DVK development did not follow a cumulative trend.

The univariate model for BVK (Table 4 and Figure 5) revealed a quadratic growth model which fit the data. The variance in the latent level and quadratic term was statistically significant (p<.05), implying that the latent-level component and quadratic trend were each stochastic. The positive means of the level and the stochastic trend and deterministic quadratic term were α¹(SE) = 4.567, α²(SE) = 5.528, and α³(SE) =.389, respectively. These values revealed that MCK improvement from 3rd to 4th grade was less than that observed from 2nd to 3rd grade. Similarly, the stochastic trend and deterministic quadratic term [α¹(SE) = 8.226, α²(SE) = 11.823, and α³(SE) = 8.352] showed that the enhancement in MCK from 2nd to 3rd grade was less than that from 1st to 2nd grade. The mean parameters were statistically significant (p<.05), suggesting that growth in young learners’ BVK was quadratic throughout the four measurement periods. Similarly, BVK development did not follow a cumulative pattern.

Development trends for the MCK, DVK, and BVK models are depicted in Figure 6. In particular, the MCK and BVK models had a better fit to the quadratic model requirements. Comparatively, the DVK model data did not fit the requirements of the linear or quadratic model.

Figure 6. Developmental trend of MCK, DVK, and BVK across four measurement times.

The above findings did not show a cumulative development cycle at the uni-construct level for each variable. As such, the next step was to examine whether a possible multi-construct association existed between the level and trend components of variables. The MCK, DVK, and BVK variables were thus estimated together in the final LGC models. The first step involved estimating significant relationships between latent components; the second was to estimate whether the models demonstrated mutual relationships. In each multivariate model, the residual variance in the MCK, BVK, and DVK variables was first estimated freely and then estimated as being equal. Results are listed in Table 5; a constructed model is shown in Figure 7.

Figure 7. Multivariate LGC model for MCK, DVK, and BVK.

Table 5. Results for multivariate latent growth models.

	Estimate	S.E.	Est./S.E.	p-value
I-MCK
WITH S-MCK	0.255	0.122	2.090	0.037
Q-MCK
WITH I-MCK	−0.219	0.095	−2.301	0.021
WITH S-MCK	−0.845	0.044	−19.425	0.000
I-DVK
WITH I-MCK	0.794	0.029	27.021	0.000
WITH S-MCK	0.396	0.075	5.244	0.000
WITH Q-MCK	−0.364	0.068	−5.320	0.000
S-DVK
WITH I-MCK	0.509	0.052	9.724	0.000
WITH S-MCK	0.419	0.085	4.927	0.000
WITH Q-MCK	−0.365	0.079	−4.597	0.000
WITH I-DVK	0.177	0.110	1.611	0.107
I-BVK
WITH I-MCK	0.711	0.045	15.936	0.000
WITH S-MCK	0.070	0.066	1.052	0.293
WITH-Q-MCK	−0.027	0.073	−0.376	0.707
WITH I-DVK	0.599	0.046	13.079	0.000
WITH S-DVK	0.364	0.059	6.149	0.000
WITH Q-DVK	−0.268	0.077	−3.497	0.000
S-BVK
WITH I-MCK	0.658	0.032	20.576	0.000
S-MCK	0.558	0.090	6.218	0.000
Q-MCK	−0.474	0.073	−6.471	0.000
I-DVK	0.611	0.038	16.212	0.000
S-DVK	0.562	0.055	10.225	0.000
Q-DVK	−0.425	0.079	−5.390	0.000
I-BVK	0.201	0.083	2.433	0.015
Model fit	χ^2=390.456, df = 24, CFI = 0.965, TLI = 0.903, RMSEA = 0.189, SRMR = 0.022

The multivariate LGC model for MCK, DVK, and BVK in Figure 7 fit the data reasonably well (χ² = 390.456, df = 24, CFI = 0.965, TLI = 0.903, RMSEA = 0.189, SRMR = 0.022), and nearly all estimated parameters were statistically significant (p<.05). The correlation between latent levels of MCK, DVK, and BVK was positive. These results can be taken as evidence of a multi-construct cumulative cycle. Data analysis indicated that the more positive the change in young learners’ MCK from 1st to 4th grade, the higher DVK and BVK they achieved during this period. Similarly, young learners who showed pronounced BVK also showed pronounced DVK across the four measurement periods and vice versa.

Discussion

In the education field, metacognitive knowledge plays a crucial role in facilitating competent cognitive activity (Schneider, 2008; van der Stel & Veenman, 2010). In an attempt to think metacognitively, young learners need to acquire reflective experiences and develop an inner mental function for situations in which cognitive processes must be employed (Teng, 2020b). Metacognitive processes and performance can then help learners recognize and build vocabulary knowledge. The present study sought to refine the understanding of primary school learners’ metacognitive knowledge and vocabulary knowledge development during primary school.

Development of metacognitive knowledge (MCK)

The first research question was intended to investigate learners’ development of metacognitive knowledge. Our general findings convey an upward slope in terms of growth in this form of knowledge from primary school grades 1–4. The growth model indicates that young learners’ metacognitive knowledge increased steadily from 1st to 4th grade, supporting previous studies delineating a noticeable increase in young learners’ metacognitive knowledge with age (Annevirta & Vauras, 2001; Flavell, 1979; Roebers & Spiess, 2017). However, findings did not show a cumulative pattern in metacognitive knowledge development; that is, earlier levels did not appear to affect subsequent metacognitive knowledge development. These findings run contrary to prior studies indicating that only young learners with initially more stable memory behaviour and performance could achieve better cognitive skills in later learning (e.g. Schneider & Sodian, 1991). The results also contradict those of Roeschl-Heils et al. (2003) who found that young learners with initially low metacognitive knowledge did not improve any more than those who showed initially higher knowledge levels. Overall, young learners seemed to develop their strategic skills in a stable manner. Learners in 1st grade appeared to have mastered some basic skills in cognitive processes. As their age increased, they were better able to reflect on their cognition, asking questions such as ‘How should I learn?’, ‘In what way can I improve in my study?’, and ‘What factors are crucial in learning?’

However, most primary school children in this study were still at the beginning of their metacognitive thinking. In particular, 1st-grade learners possessed little metacognitive knowledge although they may have started to understand their roles as learners in relation to cognitive activity. Upon entering the 2nd, 3rd, and 4th grades, learners began to concentrate on more mature metacognitive thinking to some extent. One finding not identified in previous studies was that metacognitive knowledge seemingly increased more from 2nd to 3rd grade than from 1st to 2nd or from 3rd to 4th grade. There may be individual differences in primary school children’s development of metacognitive knowledge (Baker, 2016), and such development mechanisms may not align well with grade-level requirements, a proxy for cumulating ability, experience, and knowledge (Fox, 2009). This result suggests that knowing a child’s age is not sufficient to provide a full picture of their level of metacognitive knowledge development. Individual differences must therefore be considered when delineating metacognitive knowledge. As shown by LGC modelling, the developmental trend of metacognitive knowledge followed a quadratic slope. Some learners at each grade level appeared to barely develop their metacognitive thinking, whereas others’ metacognitive scores remained stable. One may argue that primary school learners began school with varied levels of metacognitive knowledge, but individual differences in such knowledge persisted even after a long period of accumulated learning experience.

Development of vocabulary knowledge (VK)

The second research question related to vocabulary knowledge development. In this study, understanding developmental trends in vocabulary knowledge included learners’ breadth and depth of vocabulary knowledge (BVK and DVK). The development of DVK and BVK also did not follow a cumulative pattern: even when learners acquired certain levels of DVK and BVK, they did not necessarily acquire DVK and BVK corresponding to grade-level requirements. Thus, in contrast to Sparks and Deacon (2015), learners with initially poor vocabulary knowledge in the 1st grade might not improve any more than those who initially demonstrated better vocabulary knowledge. Likewise, learners with initially better vocabulary knowledge might not necessarily develop more than those with initially poor vocabulary knowledge. Hence, the argument about unchangeable individual differences in young learners’ development of vocabulary knowledge during primary grades was supported (Kidd et al., 2018).

Differences were observed in both univariate models of BVK and DVK. These discrepancies may be related to large residual variances in the LGC models. Although BVK and DVK each improved each year (Figure 6), LGC modelling showed that DVK followed neither a linear nor a quadratic slope whereas BVK followed a quadratic slope. Accordingly, the development of DVK neither followed a straight line nor produced a parabola. In other words, learners’ DVK might not always increase during school years but could change constantly instead. This trend may partially explain the challenges in acquiring DVK. In terms of BVK, some learners at each grade level did not develop their BVK whereas others’ BVK scores remained stable. Complementing previous studies (Kieffer & Lesaux, 2012; Zhang & Lu, 2015) are the variations in BVK and DVK development during primary school. For example, learners might not always acquire more pronounced levels of BVK and DVK from 2nd to 3rd grade than from 1st to 2nd grade. One possible explanation is the complexity of individual differences in learners’ acquisition of vocabulary knowledge.

Developmental dynamics between MCK, DVK, and BVK

The third research question attempted to examine developmental dynamics between MCK, DVK, and BVK. Results revealed an interaction between learners’ MCK, DVK, and BVK as well as some other interesting patterns.

First, findings reflected a multi-construct relationship between young learners’ BVK and DVK. The levels of both components appeared to be highly correlated with each other: learners with better BVK had better DVK per measurement and vice versa. The developmental dynamics between BVK and DVK were cumulative, suggesting that the initial level of BVK influenced subsequent development of BVK, with a similar trend for DVK. These findings partially support earlier conclusions about the role of BVK as a key antecedent of DVK (e.g. Qian, 1999). The present study extends this research stream by demonstrating the influence of learners’ earlier development of BVK/DVK on subsequent BVK/DVK development. In addition, learners showed a more pronounced level of BVK than DVK in each grade level during the first four years of primary school. Complementing previous studies (Teng, 2016b; Zhang & Lu, 2015), DVK development lags behind that of BVK and remains a challenging dimension of vocabulary knowledge acquisition.

Second, learners’ metacognitive knowledge levels were strongly associated with their DVK and BVK across school years. Consistent with Bowey (2001), vocabulary and metacognitive skills are closely related. Corresponding to early studies (Schraw, 1994), the more learners understood their mental and cognitive processing, the better DVK and BVK they could acquire. In particular, learners’ cognitive achievement (i.e. the ability to explain strategy use based on their learning experiences) appeared to predict vocabulary learning. In addition, the growth in learners’ metacognitive knowledge was associated with enhancements in DVK and BVK.

Overall, the findings highlight the Matthew effect (Perc, 2014). As discussed by Otto and Saskia (2017), a Matthew effect might occur in the learning process. Such Matthew effect might explain the findings, for which subsequent metacognitive knowledge development and vocabulary acquisition are triggered by the initial success or failure of a student. Individual differences in the initial level of metacognitive knowledge and vocabulary knowledge lead to a widening of differences in the subsequent learning outcomes. Such phenomenon was also explained in a longitudinal study of metacognitive knowledge and reading and writing (Teng & Zhang, 2021). Those findings highlight the need for teachers to focus on developing young learners’ metacognitive knowledge at an early age, which is closely related to vocabulary knowledge acquisition.

Concluding remarks

Tracing young learners’ vocabulary knowledge development over time and unravelling its longitudinal relations with metacognitive knowledge contributes to EFL vocabulary instruction. These findings shed light on practices relevant to developing metacognitive knowledge and vocabulary knowledge. Most learners entering primary school progressively improve their metacognitive knowledge and vocabulary knowledge after receiving English instruction. Metacognitive knowledge is of great value for enhancing learners’ vocabulary knowledge. However, BVK development lags behind that of DVK in early school years. Adult learners could therefore encounter similar problems in BVK and DVK acquisition. As none of the univariate models for MCK, BVK, or DVK were cumulative, future studies should address the incremental nature of these knowledge forms.

Despite intriguing findings, this study has some limitations. First, the sample may not be sufficient in terms of generalizability. The time-consuming collection of data for metacognitive knowledge did not allow for more participants. About 600 hours per year were devoted to individual sessions and data analysis. Second, our sample of only Chinese learners could be a limitation. Finally, the nature of a longitudinal study involves multiple and repeated testing of participants. The same tests were used over time; the question of whether learners had been familiarized with test items may also be pertinent. Although skewness and kurtosis values did not show floor or ceiling effects, learners might have recalled some items due to taking the test once a year. Despite its limitations, this study revealed a clearer conceptualization of early metacognitive knowledge—particularly evidence of the dynamic connections between early emerging MCK, BVK, and DVK. This work thus contributes to an in-depth understanding of metacognitive knowledge and vocabulary knowledge during primary school grades.

Disclosure statement

No potential conflict of interest was reported by the author.

Additional information

Notes on contributors

Mark Feng Teng

Mark Feng Teng obtained his PhD in Applied Linguistics from Hong Kong Baptist University. He is now an Associate Professor at Beijing Normal University. His main research interests include L2 vocabulary acquisition, and metacognition in L2 writing development. His latest publications appeared in TESOL Quarterly, Applied Linguistics, Language Teaching Research, Computer Assisted Language Learning, and other international journals. His recent monographs were published by Springer, Bloomsbury, and Routledge. He has served as a guest editor for several international peer-reviewed journals, including SSLLT, TESOL Journal, Asian EFL Journal, and Journal of Writing Research.