How Do Individual-Difference Variables Affect Adolescent Learners’ L2 English Speaking Development? A Microgenetic Study

Abstract

Researchers have found that learners’ second language development is influenced by internal and external individual differences but only few studies have adopted a longitudinal approach. In the present study, I aimed to investigate how several internal and external individual differences were interrelated and whether and how these variables predicted L2 English speaking development in adolescent learners. I conducted a dense longitudinal study with frequent measurements of L2 speaking skills. Learners in the first year of secondary school (11 to 13 years old, n = 48) did a weekly speaking task from September to May. At the start of the study, the participants also did multiple tasks, which measured various individual differences. Spearman correlations were calculated to shed light on the relationships between individual-difference variables, and generalized additive mixed models were used to model learning trajectories over time and to investigate the role of individual differences in this development. Results showed that learners’ speaking scores were predicted by time and prior L2 English receptive vocabulary knowledge, which was the main predictor of L2 speaking skills. Vocabulary knowledge furthermore significantly correlated with measures of out-of-school exposure and motivation. The results showed the key role of vocabulary in the early stages of L2 English learning.

1. Introduction

A significant line of research in second language acquisition (SLA) is concerned with the impact of individual-difference (ID) variables on second language (L2) learning. Individual learner variables can be both learner-internal, such as cognitive and motivational variables or learner-external, such as instruction time or out-of-school contact with the L2. Previous research with young and adolescent L2 English learners has shown that both internal and external individual-difference variables can play a role in language learning but that the contribution of each variable is dependent on the language learning context (De Wilde et al., 2021; Paradis, 2011; Unsworth et al., 2015). These findings align with complex dynamic systems theory (CDST), a process-oriented approach to second language development, which states that L2 development is driven by internal and external resources (De Bot et al., 2007; Larsen-Freeman, 1997). These resources are not static but dynamic, meaning that they can change depending on factors such as time, learning context and proficiency. Furthermore, these variables do not exist in isolation, but they are intertwined. Many studies have been devoted to the role of individual-difference variables in language learning but only few studies have looked into the role of individual-difference variables in language learning in young and adolescent learners (e.g., De Wilde & Goriot, 2022; Leona et al., 2021), and even fewer studies have focused on development and have adopted a longitudinal approach (Pfenninger, 2020). In the present study, I aim to investigate the relationship between various individual-difference variables, and investigate the possible impact of several internal and external ID variables on L2 English speaking development in adolescent learners, as previous research with this population has shown large differences in L2 English speaking skills between individual learners at the start of L2 English instruction (De Wilde et al., 2020). Furthermore, it has been shown that learners’ speaking skills had significantly increased after the onset of L2 English instruction (De Wilde et al., 2021). The study takes place at the start of secondary school, which is a transition period for all learners and thus, a period in which we might expect changes to take place (Nagle, 2021).

Below, I will first discuss individual-difference variables that have been shown to contribute to the L2 learning of young and adolescent learners and then discuss longitudinal studies that have investigated speaking development and individual-difference variables. The results of a study with 48 adolescent L2 English learners who were followed for a year will be reported. The study was set up to investigate how ID variables measured at the start of the study are interrelated and which ID variables play a role in language learning. The findings will be discussed in light of complex dynamic systems theory, and possible implications and recommendations for language teaching will be formulated.

1.1. Individual-Difference Variables and L2 Learning in Adolescent Learners

1.1.1. Cognitive Variables and L2 Learning

Research concerned with the role of individual-difference variables in L2 learning has looked at various dimensions. The research reviewed below focuses on studies conducted with young and adolescent learners. First, studies have shown that various cognitive differences can play an important role in language learning. Csapó and Nikolov (2009) and Nikolov and Csapó (2018) looked into the role of inductive reasoning in Hungarian adolescent L2 Learners of English and German. Csapó and Nikolov (2009) found that inductive reasoning significantly correlated with written productive skills (r = 0.47). Overall, findings on the impact of analytic reasoning ability on the development of receptive skills and writing showed moderate to strong correlations. However, the amount of variance explained by the variable was different within and across studies, depending on the age of the learners and the languages that were learned. Another cognitive variable that has been linked to L2 learning is working memory capacity. Researchers have explored the role of short-term memory storage and complex verbal memory (cf. Linck et al., 2014; Juffs & Harrington, 2011 for reviews) using various measures that tap into working memory capacity such as forward and backward digit span tasks and non-word repetition tasks. Results of studies exploring the role of working memory with adolescent learners have been mixed (Cheung, 1996; De Wilde et al., 2021; Gathercole & Masoura, 2005; Kormos & Sáfár, 2008; Michel et al., 2019; Service & Kohonen, 1995). Some studies have shown that working memory capacity has a considerable impact on L2 learning (e.g., Kormos & Sáfár, 2008), whereas other studies found no or only small effects (De Wilde et al., 2021; Michel et al., 2019). The mixed results could again be explained through differences in the instruction context, learners’ age and proficiency and/or the L2 under investigation.

1.1.2. The Role of Motivation

Another category of learner-internal variables is related to motivation. Many studies have been conducted to investigate the role of motivation in language learning. The two most frequently used frameworks are Gardner’s integrative and instrumental motivation (Gardner, 2010) and Dörnyei’s L2 motivational self-system (L2MSS, Dörnyei, 2009). Studies with young learners have looked into the role of motivation and cross-cultural contact (Csizér & Kormos, 2009), differences and similarities in motivation between learners in a content- and language-integrated learning (CLIL) instruction context (Mearns & de Jong, 2021) or students’ motivated behavior (Papi & Abdollahzadeh, 2012), but only Leona et al. (2021) investigated the relationship between motivation and language learning outcomes in young learners. The authors used a motivational questionnaire to measure several underlying constructs: desire to learn English, importance of communicating in a lingua Franca, linguistic self-confidence, self-advancement through learning English, attitude towards English-speaking people and willingness to communicate in English with peers. Only one variable, linguistic self-confidence, predicted young learners’ L2 English vocabulary.

1.1.3. L1 and L2 Prior Knowledge

Prior knowledge in the L1 and the L2 have also been shown to predict L2 learning. Jaekel et al. (2017) investigated the effects of early foreign language learning at two timepoints in secondary school. The authors found that young learners’ L2 proficiency at the start of the study (in year 5) was the largest predictor of their proficiency two years later. The second model (based on data gathered in year 7) explained 35% more variance compared with the model in year 5 because it included a measure of prior L2 knowledge. De Wilde et al. (2021) found that the main predictor of learners’ L2 English proficiency (receptive vocabulary knowledge and speaking skills) was their L2 English knowledge, which was measured two years earlier, before the start of formal English lessons. Studies with young L2 English learners (De Wilde et al., 2021; Puimège & Peters, 2019; Sun et al., 2018) have further shown that L1 vocabulary knowledge was also a positive predictor of L2 knowledge.

1.1.4. External ID Variables: Length of Instruction and Extramural English

As previously mentioned, there are also learner-external variables that can influence L2 learning. I will discuss previous findings related to instruction context and out-of-school exposure to the language. The latter is often referred to as ‘extramural English’ (EE; Sundqvist, 2009). Muñoz (2011, 2014) showed that children who received more L2 instruction generally performed better. She further showed that the amount of instruction was a better predictor for L2 learning than the age at which instruction had started. Similar results were found with young learners and adolescents. Unsworth et al. (2015) and Peters et al. (2019) showed that length of instruction was a positive predictor of L2 learning. Another type of input that has been studied in recent years is extramural English. Input received outside the classroom has been shown to be beneficial for language learning (cf. Kusyk et al. (2025) and Zhang et al. (2021) for reviews of previous studies). Sundqvist (2024) posited that extramural English was an important ID variable in L2 English learning. De Wilde et al. (2020) and Puimège and Peters (2019) found that this type of exposure could lead to large amounts of learning even before the start of formal English studies. Both studies found large individual differences between the participants, both in terms of amount of extramural English input and L2 English proficiency. De Wilde et al. (2020) found that not all types of out-of-school exposure were equally predictive of learning. The authors showed that types of exposure that entailed an element of production (such as gaming, using social media and speaking) were the best predictors for L2 English learning.

1.2. Investigating L2 Speaking Development

Researchers have called for more longitudinal studies. Longitudinal studies are still scarce in SLA research, but they can give us an insight into the language learning process (Ortega & Iberri-Shea, 2005). Longitudinal studies looking into L2 speaking development have investigated various aspects of oral language proficiency. Researchers agree that various aspects play a role in speaking proficiency, which often involves interaction and thus, involves both productive and receptive knowledge (Huang et al., 2021). A number of studies have focused on learners’ pronunciation. In a review article on longitudinal L2 pronunciation research, Nagle (2021) analyzed 39 longitudinal L2 pronunciation studies. The author found that most studies covered a relatively short time period (mean study length = 11 months) and few data points. Twenty studies analyzed data from two or three data points. Most studies concerned adult learners who were either learning L2 English or L2 Spanish. When looking into the analyses, the author found that most analyses were group analyses, but a few studies also looked into individual cases. More recent studies have used mixed effects modeling as these analyses can take into account individual variation in group analyses (e.g., Nagle, 2018). Other longitudinal studies have focused on aspects of complexity, accuracy and fluency in L2 speaking development. Some of these studies have tried to uncover individual learning trajectories from a CDST perspective and have investigated single or multiple cases (e.g., Larsen-Freeman, 2006; Lowie et al., 2017; Polat & Kim, 2014; Yu & Lowie, 2020). Other studies have performed group analyses in which inter-individual variability was taken into account by using mixed effects models such as GAMMs (e.g., Kliesch & Pfenninger, 2021; Pfenninger, 2020). Finally, some studies have focused on general oral language proficiency. De Wilde et al. (2021) did a longitudinal study with young L2 English learners. The authors considered learners’ overall L2 English speaking development and used tasks from the Cambridge Young Learners English: Flyers test (Cambridge English Language Assessment, 2016) to investigate speaking development. Overall proficiency score was assigned using a rubric, which took into account vocabulary, grammar, pronunciation and interactive communication.

In the various strands of longitudinal L2 speaking research, most studies have focused on a group of learners, some studies have focused on individual learners and recently some studies have tried to account for individual differences in L2 speaking development, e.g., through the use of mixed effects models. CDST-inspired studies, which strongly focus on the learning process or trajectory, have typically collected data at more timepoints than other longitudinal studies, e.g., weekly measurements throughout a language course rather than two measurements, one at the start and one at the end of the course.

A few studies have considered speaking development and individual-difference variables. Pfenninger (2020) conducted a micro development study that investigated the development of L2 English speaking skills and also considered ID variables. In the study, the impact of age of onset on speaking development in a CLIL (content- and language-integrated learning) setting was investigated. The author used generalized additive mixed models (GAMMs) to model development. She found that learners who started CLIL slightly later (at age 7) showed similar development than learners starting at age 5, but the group with the oldest starting age (9 years old) had not caught up with the others by the end of primary school. The study also showed large improvements in the final years of primary school, which could be linked to increased engagement with English outside the classroom. Kliesch and Pfenninger (2021) examined cognitive and socio-affective predictors of L2 development in later life language learners. The participants in the study took part in a language course. Every week, they did a set of language tests (including an oral interview) and several cognitive tests, and they answered questions related to motivation and wellbeing. Results of the GAMMs showed that scores on all measures of L2 proficiency increased. Cognitive and socio-affective variables significantly predicted L2 proficiency and developmental patterns. The authors further found that daily variances in cognitive and socio-affective variables had little effect.

1.3. Aims and Research Questions

The present study looks into the relationships between several ID variables which have previously been shown to predict L2 learning in adolescent learners. I then move on to investigate whether and how these variables predict L2 English learners’ speaking scores over time. The study is guided by the following research questions:

RQ1: How are various internal and external individual-difference variables and L2 English speaking related?

RQ2: How do adolescent learners’ internal and external individual-difference variables impact their L2 English speaking scores during the first year of classroom instruction?

Based on previous findings from studies in similar contexts, various types of out-of-school exposure such as gaming and using social media are expected to be positively correlated with each other and with prior L2 knowledge (De Wilde et al., 2020; Puimège & Peters, 2019). As results for cognitive factors have been mixed, it remains to be seen how these variables interrelate with external ID variables and variables measuring L2 prior knowledge. Leona et al. (2021) further showed significant positive relationships between motivational measures and extramural English.

For the second research question, all the variables under investigation have been shown to impact L2 learning in some studies but results have been mixed. Studies looking into L2 English learning through out-of-school exposure have found large differences between individual learners, both in their L2 proficiency and the amount of input to EE (De Wilde et al., 2020; Puimège & Peters, 2019). External factors might thus have a larger impact than cognitive factors in our study. Based on findings by De Wilde et al. (2021) and Jaekel et al. (2017), I hypothesize that prior L2 knowledge will be a significant positive predictor of L2 speaking development. Finally, I expect learners’ speaking skills to improve over time and thus, I expect time to also positively predict speaking development.

2. Materials and Methods

2.1. Context

The study took place in Flanders, the northern, Dutch-speaking part of Belgium and in the Netherlands. These two regions are situated in Western Europe, and the lingua franca in both regions is English, similar to other parts of Europe and the rest of the world (Dewey, 2007). This means that there is easy access to English in society, since the language is often used in advertisements but also in popular media (De Wilde & Lowie, 2024).

Even though the societal context is similar in both regions, the educational context is rather different. In the Netherlands, English is the first foreign language to be taught, and formal English instruction starts in primary school when children are ten years old or even younger, as there is a tendency to lower the starting age in Dutch primary education (Michel et al., 2021). In Flanders, French is the first foreign language to be taught, as it is one of the official languages in Belgium. French classes typically start when children are ten years old (in the two final years of primary school), and English classes start two or even three years later in the first or second year of secondary school. However, while formal L2 English classes start relatively late in Flanders, quite a few children have some knowledge of English, which they have gained from out-of-school exposure to the language (De Wilde et al., 2020).

2.2. Participants

The participants in this study came from three intact class groups, which were different in terms of instruction: one group had received L2 English instruction in primary school (n = 13); the second group had just started with formal L2 English lessons (n = 19), and the last group did not have any formal English lessons yet (n = 16). All 48 learners were in the first year of secondary school (11 to 13 years old) in a school with an academic rather than a vocational orientation. The sample consisted of 26 girls and 22 boys. Most learners spoke Dutch at home (n = 36), which is also the language of instruction in school. One in four learners (n = 12) reported they also spoke another language at home. The participants are a subgroup from the participants in the study by De Wilde and Lowie (2024), a study in which the authors explored the possibility of meaningfully grouping participants based on individual-difference variables via a data-driven approach.

2.3. Instruments and Procedure

At the start of the study the participants did multiple tasks looking into various ID variables. The participants took two tests to investigate their prior knowledge of English, the Peabody Picture Vocabulary Test (PPVT 4, Dunn & Dunn, 2007) and the Test for Reception of Grammar (TROG 1, Bishop, 1983). The PPVT 4 is a picture-based receptive vocabulary test that was originally intended for measuring L1 English vocabulary but has often been used in L2 studies. As in previous L2 studies (e.g., De Wilde et al., 2020), rather than applying the stopping rule used when testing L1 speakers, participants completed the first 144 items (12 sets of 12 items). On each trial, they listened to an English word, saw four pictures, and selected the picture that matched the word they heard. They received one point for a correct answer and zero points for a wrong answer. The maximum possible score was 144. The TROG (Bishop, 1983) measures receptive grammar knowledge. While taking this test, participants listened to a sentence and chose one of four visuals that corresponded to what was expressed in the sentence. The complete test can be found on https://osf.io/z8wbs/ (accessed on 16 August 2025). When they chose the correct visual, they received a score of one, when the chosen visual was incorrect, the score was zero. The first 12 items were skipped because they tested vocabulary knowledge. The maximum possible score was 68.

The participants also received four tasks from the Dutch version of the Wechsler Intelligence Scale for Children (WISC V NL, Wechsler, 2017). They did the Dutch vocabulary task in which participants were expected to provide a definition for each word (maximum score = 54). The participants also completed a matrix reasoning task in which they received a series of visuals. They had to complete the series with one out of five given visuals (maximum score = 32) to test analytic reasoning ability. Finally, the participants completed both a forward and backward digit span task. In the forward task, they were asked to repeat a series of numbers in the order they heard them, while in the backward task, they repeated the numbers in reverse order. These tasks measure phonological short-term memory and complex verbal memory, respectively. The maximum possible score for each task was 18.

The participants also filled in two questionnaires. The first questionnaire was taken from De Wilde et al. (2020). In this questionnaire, the participants provided personal and linguistic information (e.g., age, languages spoken at home) and information about extramural engagement with English (e.g., listening to music and using social media). The questionnaires asked about daily exposure to different types of media. For our analyses, I assigned a score to the different categories as follows: 0 = no exposure, 1 = 0–30 min, 2 = 30 min–1 h, 3 = 1 h–1 h 30 min, 4 = 1 h 30 min–2 h and 5 = more than 2 h. The second questionnaire was a motivation questionnaire. The questionnaire was adapted from Mearns and de Jong (2021), who used the questionnaire with Dutch adolescents in CLIL education. This questionnaire asked questions about the learners’ ideal L2 self, ought-to L2 self, motivation for English and L2 self-confidence. The questions related to their experiences with CLIL lessons were left out. Answers were given on a 6-point Likert scale (1 = totally disagree, 6 = totally agree). Most tests were performed in a group session. First, participants completed the vocabulary and grammar tests, which took approximately between 60 and 75 min. After the vocabulary and grammar test, learners did the matrix reasoning test and filled in both questionnaires. They could do this at their own pace (approximately 60 min were needed to complete these three tests). There was a break between the English tests and the other tests. The forward and backward digit span tasks were performed individually. Instructions were given in Dutch, the language of instruction.

L2 English speaking skills were tested weekly over a period of 10 months (one school year). No data was collected during holidays or exam weeks. There were 25 weeks when the learners completed speaking tasks. Each speaking test consisted of two similar tasks: a picture narration task that was based on sample tasks from the Cambridge English Young Learners—Flyers test and a task in which they had to talk about an everyday topic that was based on sample tasks from the Pearson Test of English Young Learners. Fifteen different prompts per task type were selected that were suitable for the learners. After 15 weeks, the learners started again with the task from week 1. There might have been a possible learning effect that resulted from repeating the task type or the exact same task (cf. Nitta & Baba, 2018), but this is difficult to avoid in longitudinal studies. The learners received the instructions in English and in Dutch from the researcher or the teacher during class. They then left the classroom to do the speaking task in a quiet room where they recorded themselves with their mobile phones and sent the recordings to the researcher. The tasks were scored with a rubric (cf. De Wilde et al., 2021) that considered grammar, vocabulary, pronunciation and fluency and communication. The rubric can be found on https://osf.io/ad7kf/ (accessed on 9 September 2025). All tasks were scored by the same rater. To check the reliability of the scores, a second rater scored over 600 tasks. The inter-rater agreement was very high (r = 0.97). The maximum score for the speaking tasks was twenty. The minimum score indicated that the speaking task was performed at absolute beginner level; the maximum score was given when the learner’s speaking proficiency was at Common European Framework of Reference for languages (CEFR) level A2. This level was chosen because it is the level expected by the Flemish government at the end of the second year of secondary school. As we know from previous research (e.g., De Wilde et al., 2020), there are large differences between learners’ speaking abilities in this learning context. Thus, there might be a ceiling effect for some learners but giving a more difficult speaking task might have been too challenging for other learners and might have resulted in floor effects, as the Flemish participants had not had any English instruction prior to the study. Therefore, it was decided to align the level with the Flemish curriculum objectives. Forty-eight participants took part in the speaking activities more than 15 times (total number of speaking tasks = 998). All learners participated throughout the study. Missing values were often due to illness or quarantine measures because of the COVID-19 crisis.

2.4. Analysis

To obtain better insight into the relationships between various ID variables, I calculated Spearman correlations. I then built a model to investigate how speaking scores were predicted by ID variables. As the participants did a speaking activity every week, I also added time as a variable. Since CDST stresses that language development can follow a non-linear trajectory, which can vary across individuals, generalized additive mixed models (GAMMs) were used to investigate the role of internal and external differences in speaking development as these analyses are flexible and can model non-linear patterns (Winter & Wieling, 2016). All analyses were performed in R (version 4.4.1.; R Core Team, 2024). To answer research question 1 about the relationships between the ID variables, I used the following packages: corrplot version 0.98 (Wei et al., 2017), ggplot2 version 3.5. (Wickham, 2011) and Hmisc version 5.1-3 (Harrell & Harrell, 2019). To answer research question 2, GAMMs were modeled with the mgcv version 1.9-1 package (Wood & Wood, 2015) and results were visualized using the itsadug version 2.4.1 package (Van Rij et al., 2015). Data and R code can be found on https://osf.io/ad7kf/ (accessed on 9 September 2025).

3. Results

To obtain better insights into the data, I first report descriptive statistics for the various ID variables (

Table 1. Descriptive statistics for ID variables (measured at time 1).

	Min	Max	Median	Mean	SD
English vocabulary knowledge (Max = 144)	57	140	106	104.4	19.79
English grammar knowledge (Max = 68)	22	68	60	56.25	10.45
Matrix reasoning task (Max = 32)	7	29	21	20.23	3.65
Dutch vocabulary knowledge (Max = 54)	15	40	28	27.29	5.14
Forward digit span task (Max = 18)	4	16	8.5	8.42	2.09
Backward digit span task (Max = 18)	4	13	9	8.86	1.92
Watching TV with L1 subtitles (Max = 5)	0	5	2	2.15	1.75
Watching TV with EN subtitles (Max = 5)	0	5	0	1.04	1.35
Watching TV without subtitles (Max = 5)	0	5	1	1.52	1.76
Listening to English music (Max = 5)	1	5	4	3.58	1.44
Reading in English (Max = 5)	0	5	0	0.46	0.90
Gaming in English (Max = 5)	0	5	3	2.73	1.97
Using social media in English (Max = 5)	0	5	3.5	3.33	1.60
Speaking in English (Max = 5)	0	5	1	1.33	1.31
Ideal L2 self (Max = 6)	2	6	4.10	4.09	1.03
Ought-to L2 self (Max = 6)	1	5.17	2.25	2.35	1.04
Motivation for English (Max = 6)	1	6	4	4.13	1.25
Linguistic self-confidence (Max = 6)	1	6	4.25	3.88	1.28

). The results show that most learners already have quite some knowledge of English grammar and vocabulary at the start of the first year of secondary school but that there are large differences in prior L2 English knowledge between the learners. For the vocabulary test for example, the lowest score was 57 out of 144 and the highest score was 140 out of 144. All other variables also show a broad range, but some overall tendencies can be discerned as related to extramural English engagement and motivation. When it comes to out-of-school exposure to English, the most popular activities are listening to music and using social media, followed by gaming and watching television with subtitles in the home language. The least frequent activity is reading in English. Results from the motivation questionnaire showed that overall learners’ motivation was quite high, with the highest median scores for linguistic self-confidence, ideal L2 self and motivation for English.

As research question 1 concerns the relationship between various ID variables, I then calculated correlations between the ID variables under consideration. Because some variables were measured with Likert scales, I calculated Spearman’s rho, as this type of correlation is more suitable for ordinal data. Figure 1a shows all correlations, and Figure 1b shows significant correlations only. The results showed significant moderate to strong correlations between prior L2 vocabulary and grammar knowledge (r = 0.86), some types of out-of-school exposure, where the strongest correlation was observed between gaming and social media (r = 0.68), and various measures of motivation, where I observed significant correlations between ideal L2 self and all other motivation variables and a correlation of 0.60 between motivation for English and linguistic self-confidence. Furthermore, various significant correlations were observed between L2 measures and measures of out-of-school exposure with correlations around 0.5 between L2 knowledge and speaking, using social media and gaming. L2 knowledge was also significantly correlated with motivation, especially with linguistic self-confidence. Finally, some aspects of motivation were also moderately to strongly correlated with specific types of out-of-school exposure. Motivation for English and linguistic self-confidence showed the strongest correlation with gaming and using social media. Fewer significant correlations were found between cognitive variables (matrix reasoning, working memory) and Dutch vocabulary knowledge, and the other variables and these correlations were somewhat lower overall. No significant correlations were found between the scores for the matrix reasoning task and any other variables.

To answer the second research question, I fitted a GAMM, which modeled how ID variables predict speaking development over time. I started with a base model containing a variable for time and factor smooths (which are similar to random effects in a linear mixed effects model) for the task and participant. In a next step, I also allowed for variable trajectories per participant over time. I then added the ID variables related to prior knowledge, input (instruction and extramural input), motivation and cognition. The variables were added one by one. Model fit was tested with the AIC-function in which I compared the fit of the new model with the previous best-performing model. The best model contained the variables from the base model, one ID variable, prior L2 vocabulary knowledge and an interaction between time and prior vocabulary knowledge. AIC values clearly dropped when adding time and the factor smooth for Time and ID. The drop in AIC was far smaller after adding prior vocabulary knowledge, but in this model, there was a large drop in the F statistic explaining differences per participant over time. Finally, there was a further (albeit small) decrease in AIC after adding the interaction term between prior vocabulary knowledge and time. This is the code for the best-performing model:

bestmodel <- bam(score ~ s(Time, k = 10) + s(PPVT) + ti(Time, cPPVT, k = c(10, 10)) + s(Time, ID, k = 10, bs = ‘fs’, m = 1) + s(Task, k = 7, bs = ‘fs’, m = 1), data = mydata)

A summary of the model can be found in

Table 2. Summary of the GAMM.

Parametric Coefficients	Estimate	SE	t	Pr (>|t|)	Edf	Ref.df	F	p
(Intercept)	12.63	0.53	23.73	<0.001 ***
s(Time)					6.29	6.89	7.94	<0.001 ***
s(PPVT)					3.98	3.99	29.08	<0.001 ***
ti(Time, PPVT)					14.40	18.64	1.12	0.33
s(Time, ID)					192.19	478	10.80	<0.001 ***
s(Task)					8.44	14	2.39	<0.001 ***

*** p < 0.001.

. The edf values show that all variables predicted the speaking scores in a non-linear manner (edf > 1). All variables, except for the interaction term, further significantly impacted the speaking scores. The F value shows that learner’s prior vocabulary knowledge at the start of the study was the best predictor for the score followed by the by-person differences over time and a significant main effect of time. Task had a smaller but also significant effect, indicating that a learner’s performance was also affected by the task they were given. Figure 2 shows the effect of prior vocabulary knowledge on learners’ speaking scores. The plot shows that learners need a certain amount of vocabulary knowledge to be able to do the speaking task. From a certain vocabulary level onwards, there was a rather steep increase in the score, which then seemed to level off once a certain level of vocabulary knowledge was reached. Figure 3, which shows the effect of time on speaking scores, shows an overall, albeit non-linear, increase in speaking scores and wide confidence intervals, indicating that large differences in scores remain over time. This can also be observed in Figure 4, which shows the by-person differences over time. Large differences could be observed from the start of the study, which mostly remained over time. From Figure 3, it might seem as if there was little development, but we should keep in mind that the holistic score ranged from absolute beginner to A2-level, which is a wide proficiency range and indicates that increases that seem small could still be meaningful. The interaction term between time and prior vocabulary knowledge was not significant, indicating that the rate of development did not depend on learners’ prior knowledge.

4. Discussion

One of the aims of the study was to shed more light on the relationships between various variables. As expected, the results showed that individual-difference variables looking into prior L2 knowledge, some types of out-of-school exposure and aspects of motivation were strongly correlated amongst each other. But there were also significant moderate to strong positive correlations between variables measuring different types of ID variables, especially related to prior knowledge, motivation and certain extramural activities (gaming, using social media and speaking English outside the classroom). A study by De Wilde and Lowie (2024), which looked into data-driven grouping of learners with this dataset, also found that through a cluster analysis three learner types could be discerned who were different in terms of L2 prior knowledge, motivation and out-of-school exposure. The results of the present study further showed that productive types of out-of-school engagement were more strongly related to L2 measures than other types of extramural English, a finding that confirms earlier research (De Wilde et al., 2020; Sundqvist, 2009). Overall, the study also showed that adolescent learners were highly motivated to learn English. The results concerning the relationship between out-of-school exposure and motivation were also in line with previous research (Leona et al., 2021), as significant positive relationships between these ID variables were found. The correlation matrix clearly shows various significant relationships between internal and external ID variables. This finding is in line with CDST, which posits that ID variables are linked and cannot be considered in isolation. In the present study, fewer and lower significant correlations were found between cognitive variables and other ID variables. These findings could possibly be explained by the fact that all learners were in similar educational tracks and might have had similar cognitive profiles (even though some differences can be observed in Table 1). Second, large differences in L2 input outside the classroom could have increased the impact of external variables compared with internal cognitive variables (cf. Sun et al., 2018). As ID variables were only measured at the start of the study for reasons of feasibility, it was unfortunately not possible to investigate whether and how relationships between these variables changed over time.

After looking into the relationships between ID variables, I investigated whether and how speaking development could be predicted by ID variables. Previous studies (e.g., Pfenninger, 2020) have shown speaking skills increase over time. Furthermore, Pfenninger (2020) found a steeper increase in the final years of primary school, which could be linked to out-of-school exposure. The present study was different from Pfenninger’s study since it lasted only one year, and her study spanned eight years. However, time also positively predicted speaking skills in the present study. At the same time, the results indicated that the increase in speaking skills over time followed a non-linear path, indicating that the growth in speaking skills was not necessarily a steady trajectory but rather a process with ups and downs and different amounts of learning in different periods. The results further showed that there was a large confidence interval when modeling the effect of time and a significant by-person effect of time, indicating that large differences between individual learners existed at the start of the study and remained throughout the study, an observation which is in line with previous findings (De Wilde et al., 2021). The results also align with CDST premises that L2 learning is a non-linear process with large intra-individual variation (De Bot et al., 2007). The differences between individuals can partly be explained by differences in learners’ receptive L2 English vocabulary knowledge at the start of the study. This variable was the strongest predictor for learners’ L2 English speaking scores. Further, the effect of prior L2 vocabulary on speaking scores was non-linear. The results suggest that a basic threshold of receptive vocabulary knowledge is necessary to be able to start developing speaking skills and a certain amount of receptive vocabulary knowledge suffices to perform speaking tasks at the A2 level. Further research could investigate how much receptive vocabulary knowledge is necessary to initiate growth in productive skills and whether and where this threshold can be situated for a specific type or level of speaking task. At the initial stages of L2 English speaking development, receptive vocabulary knowledge seems to be a better predictor for learners’ speaking skills than receptive grammar knowledge. This does not come as a surprise, as we know that vocabulary development generally precedes grammar development, also in L1 learning (e.g., Bates & Goodman, 1999). When an interaction between time and L2 vocabulary knowledge was added, the model slightly improved, but the interaction was not significant. This finding shows that the amount of prior L2 vocabulary knowledge did not impact the rate at which speaking skills developed. L2 English speaking scores were also influenced by task. As the type of task remained the same throughout the study, the effect of task was likely an effect of topic which might have made the speaking task easier or more difficult for some learners. In a longitudinal study with dense measurements this is difficult to avoid. None of the other ID variables (related to out-of-school exposure, instruction, motivation and cognitive differences) significantly improved the model. This means that productive knowledge is strongly predicted by prior L2 receptive knowledge and that other ID variables are not significant when prior L2 vocabulary knowledge is considered in the model (cf. also De Wilde et al., 2021; Jaekel et al., 2017). However, in the correlation matrix, we saw that vocabulary knowledge was significantly and positively correlated with aspects of motivation and extramural input to English, and previous studies have shown that these variables are important for language learning, especially in contexts where there are a lot of opportunities for extramural English input (e.g., De Wilde et al., 2020; Puimège & Peters, 2019). In the present study, the impact of prior L2 vocabulary (often acquired through out-of-school input) clearly overshadowed the impact of the other ID variables.

This finding has important pedagogical implications. As L2 vocabulary knowledge seems to be key in the development of initial L2 speaking skills, it should be language teachers’ priority to increase their learners’ vocabulary knowledge at the initial stages of formal L2 English learning, as this can help them to learn and use English productively and is a first step towards being able to communicate at the A2-level and become basic users of English.

Finally, this study also has some limitations. The sample size was rather small, which is often the case in dense longitudinal studies as it is not feasible to frequently collect and analyze linguistic data, especially in tasks measuring development of productive skills, from a large sample. Second, I was only able to measure individual-difference variables at the start of the study, whereas we know that ID variables can be dynamic (Dörnyei & Ryan, 2015). If EE input and motivation, for example, changed throughout the study, this could not be captured. It was, however, not feasible to do all the tests on a weekly basis. Future studies could use shorter tests and questionnaires and measure a selection of ID variables more frequently.

5. Conclusions

This study investigated the impact of various internal and external individual differences on the development of adolescent learners’ L2 English speaking skills. It was found that learners’ L2 receptive vocabulary knowledge at the start of the study was a key variable in the development of L2 English speaking skills. The study further showed that a basic level of receptive vocabulary knowledge is necessary before productive skills can develop. Finally, even though vocabulary knowledge was the only variable to predict speaking skills, the correlation matrix showed that vocabulary was significantly correlated to other ID variables measuring extramural input and aspects of motivation, clearly showing that various ID variables were interrelated. The findings are in line with some of the premises of CDST, as we found clear relationships between internal and external ID variables, large differences between individual learners and evidence of non-linear development.