Abstract
The studies published in this issue of Language Teaching Research collectively demonstrate the vitality and diversity of current research in applied linguistics and language teaching. They address a wide range of topics, including learner engagement, emotions, mindset, grit, willingness to communicate, teacher well-being, corrective feedback, task-based instruction, and language aptitude. Together, these investigations provide valuable insights into the factors associated with successful language teaching and learning across diverse educational contexts.
In addition to their substantive contributions, this collection offers an opportunity for methodological reflection. A notable characteristic of many studies in this issue is the use of association-based statistical analyses. Of the 23 articles, 13 primarily focused on examining statistical relationships among variables using approaches such as correlation, regression, structural equation modeling (SEM), mediation, and moderation analyses, while the remaining 10 employed experimental, qualitative, conversation-analytic, or mixed-methods designs that focused on questions other than the statistical modeling of relationships among psychological variables.
The prevalence of association-based approaches reflects a broader trend in current second language (L2) research toward examining relationships among psychological, social, and educational variables (Nassaji, 2026). A substantial proportion of recent research in applied linguistics seeks to identify predictors of language learning outcomes. Researchers, for example, examine whether enjoyment predicts engagement, whether self-efficacy contributes to achievement, whether grit enhances performance, or whether classroom climate predicts willingness to communicate. Such investigations have enriched our understanding of language learning processes. However, they also raise an important methodological question: To what extent do findings based on statistical associations warrant causal interpretations? More specifically, are we sometimes over-interpreting correlational evidence as evidence of causation?
It is important to emphasize that correlational research is not a methodological weakness. Many constructs central to language learning and teaching, such as motivation, anxiety, enjoyment, beliefs, and identity, cannot easily or ethically be manipulated in experimental settings. Correlational research therefore plays an important role in theory building, construct validation, and hypothesis generation. Much of what we know about the psychology of language learning has emerged from carefully designed correlational investigations. The issue, therefore, is not the use of correlational or association-based methods but the interpretation of their findings. Indeed, several studies in this issue appropriately frame their conclusions in terms of association and theoretical explanation rather than definitive causation. Nevertheless, as researchers increasingly employ sophisticated statistical techniques to examine complex relationships among variables, it becomes especially important to distinguish between evidence of association and evidence of causation.
A salient feature of current L2 research, well represented in this issue, is the increasing use of statistical techniques to model complex networks of relationships among variables. SEM, regression, moderation, and mediation analyses are now widely used to examine theoretically derived relationships among learner, teacher, and contextual factors. Importantly, however, these techniques remain fundamentally association-based methods. Although they allow us to investigate multiple pathways simultaneously and evaluate complex theoretical models, their conclusions ultimately depend on patterns of association rather than direct evidence of causation. These approaches have contributed significantly to advancing our understanding of the complex relationships among variables. At the same time, the sophistication of these methods can sometimes create the impression that causation has been established when important questions about the direction of relationships and possible alternative explanations remain unanswered. Statistical sophistication should not be mistaken for causal evidence. A statistically significant path coefficient may indicate that variables are related in a manner consistent with a proposed model, but it does not by itself demonstrate that one variable causes change in another. Such analyses are valuable for theory development. However, establishing causality requires evidence from studies designed to test causal inference (Pearl, 2009; Pearl et al., 2016).
Several studies in this issue illustrate this challenge. They provide excellent examples of research that identify meaningful associations among psychological constructs. For example, control-value appraisals, academic emotions, engagement, growth mindset, self-efficacy, grit, motivation, and willingness to communicate are shown to be systematically related. These findings are valuable because they help us understand how different dimensions of language learning fit together. However, many of these variables are measured at the same time using self-report questionnaires. In such cases, it is difficult to determine which variable comes first, a necessary requirement for making causal claims. For example, engagement may increase enjoyment, but enjoyment may also increase engagement. Similarly, language anxiety may reduce willingness to communicate, while limited opportunities to communicate may increase anxiety. When variables are measured simultaneously, it is often difficult to distinguish among these competing explanations. In addition, when data on all variables are collected using the same method, observed relationships may be inflated because they share a common source of measurement, a phenomenon known as common method variance (Podsakoff et al., 2003).
For these reasons, caution is warranted both in interpreting findings and in the language used to describe them. Terms such as effect, impact, and influence as well as expressions such as lead to and result in, used in some of the studies in this issue, can imply stronger conclusions than the evidence supports. This observation should not be taken as a criticism of these studies, which make important theoretical contributions, but rather as a reminder that statistical relationships do not, by themselves, establish causal explanations.
In this context, the growing use of mediation analysis deserves particular attention. Mediation models are appealing because they move beyond simple associations and suggest potential mechanisms linking variables. However, mediation provides the strongest evidence when we can establish the order of events over time or manipulate variables experimentally. When predictors, mediators, and outcomes are measured simultaneously, a significant indirect effect may support the proposed pathway, but it does not show that the pathway actually operates in that way. This distinction is important because mediation analyses are sometimes discussed as if they reveal exactly how one variable leads to changes in another. Such causal claims, however, require stronger evidence, typically from longitudinal, experimental, or multi-method research designs (Cole & Maxwell, 2003; Maxwell & Cole, 2007).
As noted earlier, many psychological constructs are measured through self-report questionnaires. A related consideration concerns the interpretation of the resulting data. Self-report measures provide valuable insights into learners’ and teachers’ perceptions, beliefs, and experiences. However, these perceptions do not necessarily correspond to actual behavior. Consider engagement, for example. Questionnaires can capture learners’ perceptions of their engagement, but these perceptions may not always align with observable manifestations of engagement, such as classroom participation, attentiveness, persistence, or interaction with others. A similar consideration applies to willingness to communicate. Learners’ reported readiness to speak may not necessarily correspond to their actual communicative behavior. Therefore, self-report data should not be assumed to reflect learners’ actual behavior. For this reason, such data should, wherever possible, be complemented by other sources of evidence, including classroom observations, interactional data, learning analytics, and performance-based measures (Nassaji, 2025). Combining multiple sources of evidence can provide a richer understanding of language learning processes and strengthen research conclusions. Several studies in this issue employ observational, qualitative, and mixed-methods approaches, illustrating the value of drawing on diverse sources of evidence to investigate second language learning and use.
Beyond these methodological considerations, the studies in this issue also invite reflection on a broader conceptual trend in current L2 research: the growing emphasis on psychological, individual-difference, and learner-related variables as explanations for language learning and teaching outcomes. This trend has generated important insights into why learners and teachers differ in their experiences and outcomes. At the same time, it has led to a proliferation of constructs. Examples include self-efficacy, self-confidence, and identity; grit, resilience, and academic buoyancy; engagement, agency, and autonomy; as well as affective constructs such as enjoyment, anxiety, boredom, burnout, well-being, and other positive and negative emotions. While this reflects the field’s creativity and expanding interest in psychological dimensions of language learning and teaching, it also raises an important question: How distinct are these constructs, and how much do they overlap with one another? This issue is not entirely new. Long (2007) argued that SLA faced a problem of theory proliferation and called for greater attention to the evaluation of comparative theories. Today, a related challenge may be emerging at the level of constructs, as new constructs are often introduced faster than researchers can determine how they differ from existing ones or whether they contribute uniquely to explaining language learning outcomes. Future progress may therefore depend not only on creating new constructs but also on integrating existing ones into more coherent theoretical frameworks (Nassaji, 2025). From this perspective, a key challenge for the field is to build a more cumulative, parsimonious, and theoretically integrated understanding of the key variables involved in language learning and teaching.
In addition to association-based research discussed earlier, several contributions in this issue use experimental or quasi-experimental designs to investigate instructional interventions, including cognitive apprenticeship, corrective feedback, online planning, vocabulary-focused exercises, drama-based instruction, and task repetition. Although such studies are often more difficult to conduct than survey-based investigations, they provide opportunities to examine causal hypotheses and evaluate instructional effectiveness. At the same time, they should be viewed as complementing rather than replacing correlational research, as each approach contributes valuable but distinct insights into second language education.
Overall, the studies published in this issue make significant contributions to our understanding of L2 learning and pedagogy. They demonstrate the growing sophistication of current L2 research in identifying factors associated with successful learning and teaching. At the same time, they remind us that identifying relationships among variables is only one stage of scientific inquiry. As researchers, we should continue to value correlational evidence while remaining cautious about causal interpretations that exceed the limits of our research designs. Statistical techniques such as SEM, mediation, and moderation analyses provide powerful tools for examining complex patterns of association and testing theoretically motivated models. However, even the most sophisticated association-based procedures cannot on their own establish causal relationships or determine with certainty whether one factor leads to another. Stronger causal claims require research designs that allow for the examination of the order of events, rule out alternative explanations, and, where possible, involve experimental manipulation. Recognizing these limitations does not diminish the value of association-based research; rather, it helps us interpret its findings appropriately and situate them within a broader program of cumulative inquiry.