Metacognition and engagement in second language task-based interaction: A comparative interrupted time series study

Abstract

Research in second language acquisition underscores the crucial impact of the language environment, especially conversational interaction, on learners' L2 development (Long, 1996, 2015; Mackey, 2012, 2020). While conversational interaction offers opportunities for comprehensible input, modified output, and corrective feedback through meaning negotiation, learners often do not fully utilize these opportunities during peer interaction in foreign language contexts (Adams, 2007; Foster & Ohta, 2005; Moranski et al., 2024; Toth, 2008; Ziegler et al., 2025). Therefore, researchers have explored teaching learners to better exploit the developmental opportunities available from peer interaction under the metacognitive instruction (MCI) framework (Fujii et al., 2016; Moranski et al., 2024; Sato & Loewen, 2018; Sippel, 2024, 2019). However, findings on the efficacy of such instruction have been mixed. While some studies report improved corrective feedback and grammar learning (Sato & Loewen, 2018; Sippel, 2019, 2024), others found no impact on interaction or task engagement, particularly in technology-mediated environments (Moranski et al., 2024; Ziegler et al., 2025). These discrepancies may stem from substantive issues (e.g., MCI quantity/quality, outcome estimates) or methodological issues (e.g., estimation methods, statistical models), which underscores the importance of rigorous research designs and analytic strategies. In Applied Linguistics, there remains a pressing need for systematic comparisons of intervention frameworks, estimation techniques, and statistical models to generate reliable, generalizable evidence for classroom practice (Gass et al., 2021; Plonsky, 2013, 2024)This dissertation examines both the impact of a metacognitive instruction on L2 learners’ task engagement and the methodological consequences of alternative statistical model specifications in a quasi‐experimental context. Thirty‐four dyads (n=68) from two intact university‐level language classes in Vietnam completed a series of communicative homework tasks over a 15‐week semester. Beginning in Week 11, the treatment group received metacognitive instruction on how interaction, tasks, and corrective feedback aid learning, while the control group continued with standard instruction. Using a comparative interrupted time series (CITS) framework, I estimated intervention effects on time-on-task (an operational proxy for task engagement) under multiple mixed-effects model specifications, including level-only, trend-adjusted, and fully flexible interaction formulations. Results indicate a small, non-significant increase in time-on-task following metacognitive instruction (? = 0.18, 95% CI [-2.25, 2.62], t(497) = 0.15, p = 0.882). Methodological analyses demonstrate meaningful variation in effect estimates and inferential conclusions across model forms, underscoring the critical role of specification choice in CITS studies with short time series. These findings contribute to the ongoing methodological innovations in instructed second language acquisition (ISLA) by (a) empirically evaluating the efficacy of metacognitive instruction, and (b) identifying optimal analytic approaches for causal inference in short time series, longitudinal intervention research. Implications for applied practice and future research on both metacognitive instruction and robust quasi-experimental modeling are discussed.