Examining the relationships between cognitive load, anxiety, and story continuation writing performance: a structural equation modeling approach

L2 writing is cognitively demanding, drawing on limited attentional resources for formulation, execution, and monitoring while also engaging affective responses such as anxiety. Prior work has documented negative effects of L2 anxiety on writing, but cognitive load in L2 writing, particularly within integrated reading–writing tasks, remains underexplored. The story continuation writing task (SCWT) may both scaffold language and impose additional cognitive demands. This study investigates how cognitive load and L2 writing anxiety interact and contribute to SCWT performance, positing that cognitive load may directly or indirectly affect performance via anxiety. Four hypotheses were advanced: H1 cognitive load directly affects L2 writing anxiety; H2 cognitive load directly affects SCWT performance; H3 L2 writing anxiety directly affects SCWT performance; H4 cognitive load indirectly affects SCWT performance through L2 writing anxiety.

Cognitive load theory distinguishes intrinsic and extraneous load but can be treated holistically for task-based research. L2 writing often increases working memory demands due to linguistic and cultural processing, potentially elevating cognitive load. Evidence suggests cognitive load may raise anxiety, and that anxiety often predicts poorer L2 writing outcomes, though task type and measurement matter. SCWT, an integrated reading–writing task used in Chinese assessments, may reduce output pressure via alignment with source text, but empirical evidence on cognitive load within SCWT is lacking. Prior findings show cognitive load can negatively relate to performance in independent writing and online settings; the role in integrated tasks is unclear. Attention control theory posits anxiety impairs executive functions (inhibition, shifting), diverting resources from task processing. Thus, anxiety may mediate the load–performance link. The study formulates H1–H4 to jointly model these relationships.

Design: Cross-sectional, variable-oriented study employing structural equation modeling (SEM). Participants: 197 Grade 12 students from a public high school in central mainland China; after exclusions, N=182 (mean age 17.21, SD=0.648), native Chinese speakers with approximately 6.74 years of English study. They had 12 English classes (9 h/week) and practiced SCWT weekly; no prior SCWT instruction before high school. Ethics: Approved by the IRB of the Faculty of English Language and Culture, Guangdong University of Foreign Studies; informed consent obtained from students and guardians. Instruments: (1) L2 Writing Cognitive Load Scale adapted from Paas (1992) and Lee (2020); initial 8 items across six dimensions (overall difficulty, linguistic demand, time pressure, mental effort [reading, thinking, writing], frustration, and reverse-coded performance) on a 9-point Likert scale (1–9). (2) Short L2 Writing Anxiety Scale (Cheng, 2017), 9 items across three subscales: cognitive anxiety, somatic anxiety, avoidance behavior; 5-point Likert (strongly disagree to strongly agree). (3) SCWT: Based on a 309-word excerpt about ballet dancer Marie; students wrote two paragraphs (~150 words total) continuing the story using provided opening sentences. Scoring: School’s 5-band rubric (connection to source, content, language, structure, coherence), total 25 points; scores obtained from school records. Procedure: Chinese-language printed questionnaire (demographics, cognitive load, anxiety) proofread by experts. Under a trained research assistant’s supervision, students completed the SCWT in 30 minutes, followed by the questionnaire in 15 minutes; confidentiality assured. Data analysis: SPSS 27.0 for assumptions (outliers, normality, multicollinearity, homogeneity), EFA (principal components; retain loadings >0.4; drop cross-loadings >0.32), reliability; AMOS 28.0 for SEM with model fit indices (χ², χ²/df, RMSEA, CFI, TLI, SRMR); bootstrapping (95% CI) for direct/indirect effects. Measurement evaluation: Normality acceptable (skewness −0.306 to 0.323; kurtosis −0.592 to 0.707); no multicollinearity (tolerance=0.953, VIF=1.049). EFA yielded a 14-item questionnaire: 5-item cognitive load subscale and 9-item anxiety subscale (KMO=0.724); item loadings 0.666–0.879 (p<0.001); total variance explained 65.116%. Reliability: Cronbach’s α cognitive load=0.782; L2 writing anxiety=0.771. CFA composite reliability: cognitive load=0.704; L2 writing anxiety=0.918.

- Descriptive statistics: SCWT scores M=15.99, SD=2.00 (range 9–21); cognitive load M=5.09, SD=1.19; L2 writing anxiety M=2.68, SD=0.72. Reliability acceptable (α cognitive load=0.782; α anxiety=0.771). - Correlations (Pearson): Cognitive load positively correlated with L2 writing anxiety (r=0.216, p<0.01); L2 writing anxiety negatively correlated with SCWT performance (r=−0.310, p<0.01); cognitive load not significantly correlated with SCWT performance (r=−0.082, p>0.05). Supports H1 and H3; refutes H2. - SEM model fit: χ²=156.357 (p<0.01), χ²/df=1.954, RMSEA=0.073, CFI=0.912, TLI=0.885, SRMR=0.076; variance explained: 15.6% in L2 writing anxiety, 10.8% in SCWT performance. - Structural paths (standardized, bootstrapped 95% CI): Cognitive load → L2 writing anxiety β=0.396, 95% CI [0.138, 0.639], p=0.003 (H1 supported). L2 writing anxiety → SCWT performance β=−0.289, 95% CI [−0.635, −0.001], p=0.048 (H3 supported). Cognitive load → SCWT performance β=−0.080, 95% CI [−0.310, 0.182], p=0.491 (H2 not supported). Indirect effect cognitive load → anxiety → SCWT performance β=−0.114, 95% CI [−0.374, −0.003], p=0.032 (H4 supported). Overall pattern: full mediation by anxiety. - Subcomponent insight: Cognitive load mainly influenced cognitive anxiety (R²=86.2%) within L2 writing anxiety; avoidance behavior explained relatively little variance in cognitive load–performance relations (R²=14.8%). - Additional analyses: No significant differences in cognitive load or anxiety between long- vs short-duration English experience groups (top vs bottom 30%). Low-anxiety students outperformed high-anxiety students on SCWT (t=3.64, p<0.01, d=0.67).

Findings confirm that higher perceived cognitive load in SCWT elevates L2 writing anxiety, and anxiety in turn reduces SCWT performance, with anxiety fully mediating the load–performance link. The absence of a direct cognitive load effect on SCWT performance may reflect modality differences (writing allows pausing and revision, lessening immediate working memory pressure compared to listening/speaking), the SCWT’s alignment-based scaffolding that can ease linguistic production demands, and limited sensitivity of a global writing score to working-memory-related effects relative to CAF measures. The strong association between cognitive load and cognitive anxiety aligns with attention control theory, wherein anxiety impairs central executive functions (inhibition, shifting), diverting resources to worry and threat monitoring rather than task processing. Contextual factors such as competitive, exam-oriented educational culture may heighten anxiety and its detrimental impact. Ancillary analyses indicate that lower anxiety relates to better performance, suggesting that managing severe anxiety while harnessing potentially motivating moderate arousal may optimize outcomes.

This study integrates cognitive load theory with affective factors to model L2 writing performance in an integrated SCWT context. It demonstrates a positive link between cognitive load and L2 writing anxiety, a negative link between anxiety and SCWT performance, and a full mediation of the cognitive load effect on performance by anxiety. Contributions include extending cognitive load theory to L2 writing within integrated tasks and employing SEM to disentangle direct and indirect effects. Pedagogically, educators should monitor learners’ cognitive states, segment complex tasks, and use interaction or collaboration to reduce overload while helping students regulate excessive anxiety yet leverage moderate arousal for motivation. Future research should: use more sensitive performance metrics (CAF), incorporate objective cognitive load indices (e.g., eye-tracking, EEG/fMRI), examine longitudinal trajectories, test different modalities and task complexities, and explore roles of proficiency and motivation.

Generalizability is limited by a single cultural/educational context (Chinese K-12). Data were collected in a high-stakes mock-exam environment, potentially inflating cognitive load and anxiety relative to classroom settings. Inter-rater reliability for writing scores was not disclosed due to school regulations. The global writing score may be insufficiently sensitive to detect cognitive load effects compared to CAF metrics. Cross-sectional design precludes developmental inferences. Cognitive load was measured via subjective self-report; future work should include objective physiological/behavioral measures.