The study addressed a specific gap in second language motivation research. Although prior longitudinal work has examined self-efficacy growth, reciprocal links between self-efficacy and flow, and change in future L2 selves, limited evidence existed on the joint three-wave relationships among ought-to L2 self, self-efficacy, and a behavioral-emotional-cognitive engagement composite in postgraduate English learning. The theoretical framework combined Dörnyei’s L2 Motivational Self System, in which the ought-to L2 self concerns perceived obligations and the avoidance of negative outcomes, with regulatory focus theory and Bandura’s social cognitive theory. Under this account, obligation-based standards may encourage participation to meet requirements, while self-efficacy beliefs about one’s capacity to complete tasks can support persistence and be revised through learning experience.

Participants were drawn from five schools: Mechanical Engineering (n = 39), Mathematics (n = 40), Materials Science and Engineering (n = 46), Humanities and Social Sciences (n = 22), and Economics and Management (n = 27). Of the 174 students in the final analytic sample, 98 were female and 76 were male, and most were between 22 and 25 years old. All were non-English majors required to pass a one-year English course for graduation. Data were collected during the first semester of the 2025–2026 academic year through voluntary online questionnaires completed during regular class hours, each taking about 15 minutes. University identification numbers were used to match responses across the three waves. Twenty-two of the original 196 students did not complete all three waves and were excluded from the longitudinal analyses.

Measurement quality was assessed with confirmatory factor analysis at each wave. Standardized factor loadings ranged from 0.75 to 0.98, Cronbach’s alpha ranged from 0.903 to 0.947, composite reliability ranged from 0.891 to 0.954, and average variance extracted ranged from 0.657 to 0.873. Longitudinal measurement invariance testing showed acceptable configural and metric invariance; the scalar model was rejected by the chi-square difference test, but changes in approximate fit indices were small, so the scalar model was retained as an approximate basis for longitudinal comparison. Self-efficacy was measured with five adapted items, engagement with 12 items covering behavioral, emotional, and cognitive dimensions, and ought-to L2 self with four items adapted from Papi and colleagues, all rated on five-point Likert scales. An a priori Monte Carlo power analysis indicated that a complete-case sample of 170 would yield power of 0.916 for a pre-specified focal cross-lagged effect of β = 0.25.

The growth models revealed differentiated trajectories. The linear model for ought-to L2 self showed no statistically significant mean linear change across the semester, indicating that average perceived obligations remained relatively stable. In contrast, latent-basis models, which allowed change to differ between the two eight-week intervals, yielded significantly negative mean growth factors for self-efficacy and engagement. The observed means of both constructs declined from the first to the second wave and partly recovered at the third. Significant intercept variances for ought-to L2 self and self-efficacy indicated between-person differences in initial levels, and significant growth-factor variances for ought-to L2 self and engagement indicated that students differed in their patterns of change. For self-efficacy, the growth-factor variance was not significant, meaning the model did not detect reliable between-person variation in change even though the mean change was significant. For engagement, a positive and significant intercept-growth covariance indicated that students with higher initial engagement tended to have more positive growth-factor scores.

The cross-lagged panel model, which fit the data acceptably, examined prospective associations conditional on each construct’s prior level and the other lagged predictors. Ought-to L2 self and engagement showed significant positive autoregressive paths across both intervals, indicating continuity in their relative standing. Self-efficacy’s autoregressive path was not significant from the first to the second wave but was positive and significant from the second to the third. In the first interval, ought-to L2 self at Week 4 positively predicted engagement at Week 12 (β = 0.17, p = 0.02) and negatively predicted self-efficacy at Week 12 (β = −0.26, p < 0.01), while Week 4 self-efficacy positively predicted Week 12 engagement (β = 0.25, p < 0.01). In the second interval, Week 12 self-efficacy positively predicted Week 20 engagement (β = 0.31, p < 0.01) and negatively predicted Week 20 ought-to L2 self (β = −0.30, p < 0.01), and Week 12 engagement positively predicted Week 20 self-efficacy (β = 0.21, p < 0.01).

These patterns are consistent with reciprocal prospective associations between capability beliefs and participation over the latter half of the semester, but the authors caution that they do not establish causal effects or the internalization of obligations. The conventional cross-lagged panel model combines stable between-person differences with occasion-specific variation, so its paths should not be interpreted as isolated within-person causal effects. The negative prospective associations between ought-to L2 self and self-efficacy in both directions are compatible with a theoretical possibility drawn from the framework: obligations can make a gap between expected performance and perceived capability salient, while confidence may change how strongly learners orient themselves toward perceived duties. However, anxiety, regulatory focus, and internalization were not measured as separate variables, so these processes remain explanations for future work to test rather than demonstrated mechanisms.

The authors also note interpretive limits on the construct measures. Engagement was analyzed as a composite of behavioral, emotional, and cognitive dimensions, so the reported paths concern overall involvement rather than separate effects for each dimension. The ought-to L2 self measure was modeled as one composite without separate estimates for own-standpoint and other-standpoint obligations, so the findings should not be attributed exclusively to external expectations. Contemporaneous correlations among the three composites were small and nonsignificant at the mid-semester wave, contrasting descriptively with the wider ranges at the first and third waves, though differences between waves were not formally tested. The authors note that these small mid-semester correlations warrant cautious, context-sensitive interpretation and do not contradict the significant cross-lagged paths, which concern associations across different occasions.

The findings carry pedagogical implications that the authors frame as suggestions to be evaluated rather than interventions tested in the study. Because self-efficacy and engagement declined at mid-semester before partly recovering, the results suggest that support should continue beyond the opening weeks of a course. The authors suggest pairing clear expectations with achievable tasks and progress-focused feedback throughout the semester, offering regular low-stakes opportunities to contribute such as prepared discussion questions or rotating group roles, and providing feedback that connects current performance with earlier work so students can judge progress from specific evidence. Teachers might also invite students to reflect on whose expectations influence their English learning and how those expectations relate to their academic goals. Such approaches acknowledge obligations without assuming that greater pressure will improve both participation and confidence.

Several limitations bound the conclusions. The design covered only three waves within one semester in a single compulsory course at one university, so it may not reflect longer-term motivational restructuring across thesis writing, internships, or job searches. The findings describe completers and may be affected by attrition-related selection, and they may not generalize to other institutions, regions, academic stages, or learner groups. The observed-score cross-lagged model did not explicitly account for measurement error, and future work could apply random-intercept cross-lagged panel models, latent change score models, or dynamic structural equation models to separate trait-like stability from within-person fluctuation. Experimental or intervention-based designs would provide stronger evidence for causal mechanisms. Future research should also examine engagement dimensions and obligation standpoints separately and directly measure regulatory focus, mastery experiences, and internalization to evaluate the proposed mechanisms.

Subject of Research: Longitudinal relationships among ought-to L2 self, self-efficacy, and engagement in Chinese postgraduate English learners

Article Title: A longitudinal study on relationships of Chinese postgraduates’ ought-to L2 self, self-efficacy and engagement

Article References: Huo, W., Deng, S., & Wang, X. (2026). A longitudinal study on relationships of Chinese postgraduates’ ought-to L2 self, self-efficacy and engagement. Frontiers in Psychology, 17, Article 1936128. https://doi.org/10.3389/fpsyg.2026.1936128

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1936128

Keywords: ought-to L2 self, self-efficacy, L2 engagement, longitudinal study, cross-lagged panel model, latent growth curve modeling, postgraduate learners, Chinese EFL, L2 Motivational Self System, social cognitive theory, educational psychology, second language motivation