The moderating role of language self-efficacy between Japanese acquisition stress and learning persistence: a survey and quasi-experimental evaluation of an intervention mechanism

Introduction Japanese learners face distinctive stressors involving kana and kanji, honorific selection, and rapid-speech comprehension. This study examined whether language self-efficacy moderates the association between acquisition stress and learning persistence and whether an instructional intervention can improve efficacy and persistence. Methods Study 1 surveyed 548 university learners sampled from sixteen intact classes at a Chinese university. Moderated regression assessed persistence using behavioral intention and platform-logged self-study time. Study 2 was an eight-week quasi-experiment involving 96 learners in four intact classes, with two classes per condition. Four classroom modules addressed mastery experience, vicarious modelling, attributional feedback, and arousal regulation using CC0 and CC-BY materials. Outcomes were assessed using repeated-measures ANOVA and exploratory mediation analysis. Results Stress negatively predicted persistence, whereas efficacy positively predicted persistence. Their positive interaction explained an additional 3.3% of variance. Simple slopes were −0.421 at low efficacy and −0.128 at high efficacy. The Johnson-Neyman threshold was 4.06 on the five-point scale, with a bootstrap interval of 3.78−4.40. Buffering was strongest for self-regulated learning management efficacy. The intervention produced group-by-time interactions for efficacy, persistence, and stress, with the largest effect on efficacy. Exploratory mediation suggested that efficacy gains accounted for part of the persistence gain. Robustness analyses returned consistent estimates. Discussion Language self-efficacy, particularly confidence in managing one’s own learning, may buffer the stress-persistence association. Low-cost classroom instruction can improve efficacy and persistence. However, the survey was correlational, the intervention involved only four classes, and changes in the moderation effect itself were not tested. The threshold is sample-specific and should not be treated as a universal cut-off.