Exploring age-related differences in metacognitive self-regulation: the influence of motivational factors in secondary school students

The study investigates whether adolescents’ metacognitive self-regulation and key academic motivation factors (self-efficacy, task value, mastery and performance goal orientations) decline across grades in lower secondary school, and whether changes in motivation explain age-related differences in metacognitive self-regulation. Prior literature shows mixed age-related trends in metacognitive skills—some studies report improvements during adolescence while others indicate declines. Motivation is theorized within self-regulated learning (SRL) models to precede and support metacognitive self-regulation, yet adolescent motivation often declines in secondary school. Within Greece, where lower secondary (Gymnasium) spans ages 12–15 and language lessons are compulsory, national concerns about performance underscore the importance of understanding these mechanisms. The study addresses: RQ1: whether measures are invariant across grades; RQ2: how latent means differ by grade; and RQ3: whether motivational factors independently mediate age (grade) differences in metacognitive self-regulation.

Evidence on age differences in metacognitive self-regulation is inconsistent. Some experimental and self-report studies indicate higher metacognitive performance in adolescence and plateau or decline by adulthood, with growth between ages 13–14 (Weil et al., 2013; van der Stel & Veenman, 2014; dos Santos Kawata et al., 2021). Other work finds adolescents aged 14–15 outperform 17–18-year-olds, and several studies document overall declines in secondary school (Bakracevic Vukman & Licardo, 2010; Ziegler & Opdenakker, 2018; Bardach et al., 2023; Sáez-Delgado et al., 2023). Despite metacognitive strategies being teachable and beneficial, declines remain perplexing. Academic motivation factors relevant to SRL—self-efficacy, task value, and achievement goals—often decline with age in adolescence. Longitudinal evidence suggests declines in self-efficacy (Caprara et al., 2008; Jacobs et al., 2002; Mozahem et al., 2021), though updated evidence is needed. Task value findings are mixed, with some declines reported (Lee & Seo, 2021) and other studies showing stability across domains (Guo et al., 2018; Part et al., 2023). Achievement goals typically decrease in adolescence, with drops observed in both mastery and performance goals (Duchesne et al., 2014; Ciani et al., 2011; Liu et al., 2023; Luo et al., 2023). Person–environment fit theory attributes declines to increasing academic demands and mismatches as students transition and progress through secondary school (Eccles & Roeser, 2009; Wigfield et al., 2015). SRL theory posits that motivation guides metacognitive processes across forethought, performance, and reflection phases (Zimmerman, 2008; Zimmerman & Moylan, 2009). While motivation precedes metacognitive self-regulation, interrelations among specific motivational constructs are debated. To disentangle their unique contributions, the study employs Cholesky decomposition in an SEM framework.

Design: Cross-sectional survey study of Greek lower secondary (Gymnasium) students to examine grade-related differences in academic motivation and metacognitive self-regulation in language lessons and test motivational mediation mechanisms. Participants: N = 1,027 adolescents from 19 schools; ages 12–16 (M = 13.95, SD = 0.78); 46.71% male, 53.29% female. Grade distribution: A Gymnasium (n = 106; 10.30%), B Gymnasium (n = 376; 36.54%), C Gymnasium (n = 545; 52.96%). Procedure and ethics: Data collected Dec 2022–Apr 2023. Ethics approval: Faculty of Education, University of Cambridge (UK). Approval obtained from Greek Ministry of Education. Informed consent from parents/guardians; students informed about the survey. Measures (MSLQ-based, 1–7 Likert): - Metacognitive self-regulation in language lessons: 9 items adapted; 3 negatively worded items dropped due to method factor; ω = 0.85; item-total r = 0.49–0.61. - Academic self-efficacy (language): 9 items; ω = 0.92; item-total r = 0.61–0.74. - Mastery goal: 4 items; ω = 0.75; item-total r = 0.33–0.51. - Performance goal (extrinsic goals): 4 items; ω = 0.75; item-total r = 0.47–0.56. - Task value (language): 6 items; ω = 0.93; item-total r = 0.67–0.78. - Grade membership: ordinal 0 (A), 1 (B), 2 (C); also recoded to two dummies (BGYM, CGYM; A as reference) for mediation. - Sex: female vs. male. Analytic strategy: - Reliability: McDonald’s omega; item-total correlations. - Descriptives and latent bivariate correlations computed. - Multilevel necessity check: ICCs for school-level variance were < 5% across key variables (e.g., MCOG ICC = 0.036; SE = 0.016; MAST = 0.018; PERF = 0.042; TVAL = 0.045), indicating multilevel modeling not required. - Missing data: 9.72% missing overall; Little’s MCAR significant for key outcomes (p < 0.001); became non-significant after conditioning on sex, indicating conditional missingness. FIML used under MLR. - Measurement invariance across grades (A, B, C): Tested configural, metric, scalar invariance for each construct. Partial invariance pursued if needed using modification indices. - Structural modeling: Cholesky decomposition SEM (de Jong, 1999; Bentler & Satorra, 2000) to estimate unique contributions of motivational factors to metacognitive self-regulation while controlling intercorrelations via phantom/Cholesky factors. Entry order: (1) Mastery goals, (2) Performance goals, (3) Task value, (4) Self-efficacy; producing orthogonal Cholesky factors CH4→CH1 reflecting net effects of mastery, performance, task value, and self-efficacy, respectively. ΔR² for each factor computed from squared standardized paths. - Mediation: Grade dummies (BGYM, CGYM; A reference) predicting Cholesky factors, which predict metacognitive self-regulation; indirect effects estimated (MODEL INDIRECT). Sex included as covariate. Model fit assessed with CFI/TLI (~≥.95 good), RMSEA (<.06), SRMR (<.08); chi-square difference testing (Satorra–Bentler) and ΔCFI ≤ .01, ΔRMSEA ≤ .015 for invariance decisions. Software: Mplus 8.7 (robust MLR), R psych for omega.

Measurement invariance: - Scalar invariance achieved across grades for metacognitive self-regulation, self-efficacy, mastery goals, and performance goals. Task value achieved partial scalar invariance after freeing intercepts of two items for C Gymnasium. Model fit indices for invariance levels (examples from Table 2): - Metacognitive self-regulation: Configural CFI = 0.993, RMSEA = 0.029; Metric ΔCFI = 0.004, ΔRMSEA = 0.013; Scalar ΔCFI = 0.004, ΔRMSEA = 0.005. - Self-efficacy: Configural CFI = 0.950, RMSEA = 0.068; Scalar CFI = 0.941, RMSEA = 0.062. - Task value: Configural CFI = 0.985, RMSEA = 0.062; Scalar CFI = 0.966, RMSEA = 0.066; Partial scalar CFI = 0.979, RMSEA = 0.053. Latent mean differences (relative to A Gymnasium): - B vs A: small decreases in metacognitive self-regulation and performance goals; moderate decreases in task value and mastery goals. - C vs A: moderate decreases in metacognitive self-regulation, mastery, and performance goals; large decrease in task value. Overall, all motivational factors and metacognitive self-regulation decreased with higher grade. Direct effects-only model (grade → metacognitive self-regulation): - BGYM β = −0.161, p < 0.01; CGYM β = −0.268, p < 0.001. Full mediation model (Cholesky SEM): - Fit: CFI = 0.932, TLI = 0.923, RMSEA = 0.039 (90% CI [0.037, 0.042]), SRMR = 0.048. - Direct effects of grade on metacognitive self-regulation became non-significant: BGYM β = 0.082, p > 0.05; CGYM β = −0.022, p > 0.05. - Cholesky factors predicting metacognitive self-regulation: • CH4 (mastery goals net of others) β = 0.730, p < 0.001. • CH3 (performance goals net of task value & self-efficacy) β = 0.165, p < 0.001. • CH2 (task value net of self-efficacy) β = 0.251, p < 0.001. • CH1 (self-efficacy net of others): positive effect via B Gym only (see indirects); overall unique ΔR² contribution 2%. - Grade effects on Cholesky factors: • CH4 negatively predicted by B and C Gymnasium. • CH3 negatively predicted by C Gymnasium (B non-significant). • CH2 negatively predicted by both B and C Gymnasium. • CH1 positively predicted by B Gymnasium; not by C Gymnasium. Variance explained in metacognitive self-regulation: - Mastery goals explained 53% (CH4). - Adding performance goals explained +3% (CH3). - Adding task value explained +6.2% (CH2). - Adding self-efficacy explained +2% (CH1). - Total R² = 67.2%. Specific indirect effects (Table 3): - BGYM → CH1 → MCOG: β = 0.029 (SE 0.011), p = 0.007. - CGYM → CH1 → MCOG: β = 0.011 (0.008), p = 0.174 (ns). - BGYM → CH2 → MCOG: β = −0.060 (0.022), p = 0.006. - CGYM → CH2 → MCOG: β = −0.063 (0.022), p = 0.003. - BGYM → CH3 → MCOG: β = −0.011 (0.010), p = 0.264 (ns). - CGYM → CH3 → MCOG: β = −0.025 (0.012), p = 0.033. - BGYM → CH4 → MCOG: β = −0.210 (0.049), p = 0.000. - CGYM → CH4 → MCOG: β = −0.198 (0.049), p = 0.000. Overall pattern: Older grade membership is associated with reduced task value, mastery, and performance goals, which in turn reduce metacognitive self-regulation. A small positive pathway through self-efficacy appears only for B Gymnasium.

Findings address the research questions by demonstrating that measures of self-efficacy, mastery and performance goals, task value, and metacognitive self-regulation are largely comparable across grades (scalar or partial scalar invariance), enabling valid latent mean comparisons. Results reveal consistent declines across grades in both motivation and metacognitive self-regulation, indicating that older lower-secondary students report less value for language lessons, reduced mastery/performance orientations, and lower self-efficacy, alongside lower metacognitive self-regulation. Crucially, when motivational factors are modeled simultaneously using Cholesky decomposition, age-related differences in metacognitive self-regulation are fully accounted for by motivation: mastery goals emerge as the dominant predictor, with additional unique contributions from task value, performance goals, and a smaller unique contribution from self-efficacy. Thus, the study clarifies a key mechanism: declines in specific aspects of academic motivation—especially mastery goals and task value—propagate the observed declines in adolescents’ metacognitive self-regulation in language lessons. This aligns with SRL theory, highlighting the forethought phase (goals, value, efficacy) as foundational for effective planning, monitoring, and control strategies.

This study shows that among Greek lower secondary students, metacognitive self-regulation and academic motivation (self-efficacy, task value, mastery and performance goals) decline with grade level. Measurement invariance analyses validate cross-grade comparisons. Structural modeling indicates that the observed age-related decreases in metacognitive self-regulation are indirectly driven by declines in motivation, predominantly through mastery goals, with additional contributions from task value and performance goals, and a smaller unique contribution from self-efficacy. The work advances understanding by isolating unique motivational pathways via Cholesky decomposition, explaining two-thirds of the variance in metacognitive self-regulation. Future research should employ longitudinal designs (growth curves, cross-lagged models) and leverage online/behavioral assessments to test causal pathways and refine interventions that bolster mastery orientation and task value to sustain metacognitive regulation across adolescence.

- Cross-sectional design limits causal inference; grade differences may reflect cohort or unmeasured confounds. Longitudinal growth and cross-lagged models are needed to confirm mechanisms. - Sample, while large and multi-school, is not nationally representative; findings are specific to Greek lower secondary language lessons and may not generalize across subjects or contexts. - Self-report measures may be subject to method effects and common-rater bias (partly addressed by dropping negatively worded items in the metacognitive scale and testing invariance). - Missing data were conditionally missing (MCAR violated before conditioning); handled via FIML, but residual bias cannot be entirely ruled out. - Partial scalar invariance for task value (two item intercepts freed in C Gymnasium) suggests minor measurement noninvariance that should be considered in interpretation.