Cultural influence on metacognition: comparison across three societies

Metacognition refers to the ability to track and evaluate one’s own decisions, cognitive states, and behavioral responses, comprising metacognitive monitoring (e.g., confidence judgments that track accuracy) and metacognitive control (adjusting future behavior based on uncertainty). Prior work shows metacognition is shaped by developmental, linguistic, educational, learning, genetic, personality and clinical factors, and may also be modulated by cultural values via social norms and collaboration. A previous cross-cultural study found higher metacognitive adjustment on error trials in Chinese vs. British participants in a perceptual decision task, attributed to greater susceptibility to social influence. The present study tests whether cultural differences modulate metacognitive monitoring in a higher-level cognitive task (mental rotation) across three culturally divergent societies—Saudi Arabia, Portugal, and China—moving beyond the typical East–West comparison and incorporating broader cultural dimensions using Hofstede’s framework. The authors hypothesize that metacognition varies with cultural values (especially individualism and uncertainty avoidance), expect Chinese metacognition to exceed Western European, and examine sex differences given known male advantages in mental rotation performance.

• Metacognition involves monitoring (confidence tracking of accuracy) and control, with partially overlapping neural substrates (Flavell, 1979; Schraw, 1998; Fitzgerald et al., 2017; Fleming & Dolan, 2012; McCurdy et al., 2013; Molenberghs et al., 2016; Morales et al., 2018). Retrospective confidence typically increases for correct vs. incorrect responses (Yeung & Summerfield, 2012), reflecting mechanisms such as error detection and estimation of error likelihood.
• Influences on metacognition include development, language, education, training, genetics, personality, affective states and disorders (De Bruin & van Gog, 2012; Spiess et al., 2016; Ordin et al., 2020; Polyanskaya et al., 2022; De Jager et al., 2005; Yeager et al., 2019; Carpenter et al., 2019; Greven et al., 2009; Conway et al., 2020; Rouault et al., 2018; Nicholson et al., 2019).
• Cultural modulation: van der Plas et al. (2022) reported that Chinese participants showed larger confidence reduction after additional evidence on error trials vs. British, suggesting higher metacognitive adjustment linked to social influence and collectivism (Korn et al., 2014; Lui, 2015; Mahmoodi et al., 2015; Mesoudi et al., 2015; Oeberst & Wu, 2015). Metacognition also supports interpersonal coordination and theory of mind, which vary across cultures (Frith, 2012; Fusaroli et al., 2012; Reddish et al., 2013; Carruthers, 2009; Fleming & Daw, 2017; Lockl & Schneider, 2006, 2007; Heyes & Frith, 2014).
• Cultural dimensions: Hofstede’s six dimensions (power distance, individualism, masculinity, uncertainty avoidance, long-term orientation, indulgence) offer a framework to operationalize cultural differences (Hofstede, 2001; Hofstede et al., 2010a; Minkov, 2013). Societal profiles can be nuanced: Arab cultures show self-assertive interdependence distinct from Western and East Asian profiles (San Martin et al., 2018). Western populations themselves differ (Na et al., 2010; Kitayama et al., 2009).

Participants: 226 university students without immigrant background: Saudi Arabia n=84 (52 females), China n=70 (35 females), Portugal n=72 (39 females). Ages mostly 18–30 (mean ~21.5, median 21). In Saudi Arabia, male and female campuses were separate due to policy. Sessions were run in computer classrooms on individual desktops; experiments were conducted in participants’ native languages.

Stimuli and task: Abstract 3D-like shapes of ten gray square blocks on black background. Each trial presented a pair of shapes; participants decided whether shapes were identical or different. For identical pairs, the second shape was rotated around the X-axis (roll) or Y-axis (yaw) by 0°–340° in 20° steps. There were 18 identical-roll trials, 18 identical-yaw trials, and 36 different-shape trials, for a total of 72 trials per participant. After each decision, participants rated confidence on a 4-point scale (1=not confident at all, 4=totally confident).

Procedure: Trials were randomized per participant. Confidence was recorded after every response. The experiment focused on metacognitive monitoring via retrospective confidence.

Analysis: Reaction times were not analyzed due to potential hardware differences across sites and because mental rotation performance per se was not the focus. Engagement of metacognition was first verified by comparing per-participant mean confidence for correct vs. incorrect responses (also indexing metacognitive bias). For metacognitive sensitivity, a signal detection theoretic approach was used: mapping confidence×correctness into type-2 SDT categories (meta-hits, meta-false alarms, meta-misses, meta-correct rejections) to estimate meta-d′ (Maniscalco & Lau, 2012, 2014; Galvin et al., 2003). Because meta-d′ scales with task sensitivity (d′), metacognitive efficiency was computed as M-ratio = meta-d′/d′ (Fleming & Lau, 2014). Meta-d′ was estimated using hierarchical Bayesian modeling (HMeta-d; Fleming, 2017), suitable for limited trials, zero counts at some confidence levels, and unequal group sizes. Group comparisons used ANOVA: for task performance (d′) with factors sex (male, female) and group (Arabic, Portuguese, Chinese); for confidence a repeated-measures ANOVA with within-subject factor accuracy (correct vs. incorrect) and between-subjects factors sex and group; and for M-ratio with factors sex and group, followed by Bonferroni-corrected pairwise tests. The approach emphasizes explicit metacognition via retrospective confidence and acknowledges that implicit processes may be underweighted.

• Task performance (d′): Significant main effects of group and sex. • Group: F(2,220)=36.545, p<0.001, η²=0.249. Sex: F(1,220)=116.625, p<0.001, η²=0.07. • Males outperformed females overall; sex effect held in Portugal (simple main effect F=5.324, p=0.022) and China (F=9.433, p=0.002), but not in Saudi Arabia (F=2.699, p=0.102). • Pairwise comparisons (Bonferroni): Arabic performance lower than Portuguese (females Δd′=1.036, t=4.999, p<0.001; males Δd′=1.208, t=4.98, p<0.001) and lower than Chinese (females Δd′=1.041, t=4.869, p<0.001; males Δd′=1.398, t=5.845, p<0.001). China vs. Portugal differences were not significant (all Bonferroni p=1.0). • Approximate performance levels: Arabic d′≈0.55 (females), 0.88 (males); Portuguese and Chinese d′>1.5 (females) and >2.0 (males); Arabic d′ still above chance.

• Confidence vs. accuracy (evidence of metacognitive monitoring and bias): Repeated-measures ANOVA showed higher confidence for correct than incorrect responses, indicating metacognitive modulation. • Accuracy effect: F(1,218)=170.145, p<0.001, η²=0.348. • Sex effect: F(1,218)=11.903, p<0.001, η²=0.052. Group effect: F(1,218)=5.809, p=0.003, η²=0.051. • Sex×group: Males more confident than females in Portugal (F=6.676, p=0.012) and China (F=11.984, p<0.001); no sex difference in Saudi Arabia (F=−0.07, p=0.79). • Correct vs. wrong confidence differences significant in China (males Δ=0.23, t=5.03, d=0.48, p<0.001; females Δ=0.28, t=6.17, d=0.60, p<0.001) and Portugal (males Δ=0.39, t=8.2, d=0.82, p<0.001; females Δ=0.29, t=6.78, d=0.615, p<0.001), but not in Saudi Arabia (males Δ=0.12, t=2.62, d=0.26, p=0.346; females Δ=0.11, t=2.86, d=0.22, p=0.181).

• Metacognitive efficiency (M-ratio): ANOVA showed a significant group effect, no sex effect. • Group: F(2,220)=35.922, p<0.001, η²=0.246. Sex: F(1,220)=1.803, p=0.181, η²=0.008. Interaction: F(2,220)=2.69, p=0.07, η²p=0.02 (ns). • Pairwise (Bonferroni): China > Saudi Arabia (Δ=0.08, t=4.6, p<0.001, d=0.755); Portugal > China (Δ=0.07, t=3.705, p<0.001, d=0.623). Overall: Portugal highest, Saudi Arabia lowest; no sex differences in M-ratio.

• Cultural modeling: A simple linear cumulative model using Hofstede scores for uncertainty avoidance (x) and individualism (y) yields inequalities whose common solution space is consistent with observed ordering Portugal > China > Saudi Arabia in metacognitive efficiency and also aligns with prior China > UK findings (van der Plas et al., 2022). Trend: metacognition increases with decreasing individualism and increasing uncertainty avoidance.

• Metacognitive bias: Overall confidence (bias) higher in Chinese participants than Arabic and Portuguese, more evident in males, consistent with higher cultural masculinity. Bias appeared independent of metacognitive sensitivity/efficiency.

Findings show robust cultural modulation of metacognitive monitoring in a high-level cognitive task (mental rotation). While cognitive performance (d′) varied by sex in Portugal and China, sex differences did not appear in metacognitive efficiency (M-ratio) in any group, indicating that metacognitive monitoring can be decoupled from known sex-based differences in mental rotation performance. Cross-societal differences in metacognitive efficiency were largest between Portugal and Saudi Arabia and smaller between China and Saudi Arabia, with Portugal exhibiting the highest M-ratio and Saudi Arabia the lowest. The results go beyond a single cultural dimension (e.g., collectivism/individualism) and suggest interacting cultural influences: lower individualism and greater uncertainty avoidance jointly predict higher metacognitive efficiency. A simple linear cumulative model using Hofstede’s cultural scores for individualism and uncertainty avoidance accounted for the observed cross-societal pattern and was consistent with earlier reports of enhanced metacognitive adjustment in Chinese vs. UK samples. Additionally, cultural masculinity appeared to modulate confidence bias (overall confidence level) without affecting metacognitive sensitivity, with Chinese participants showing higher confidence, particularly among males. These patterns underscore the social nature of metacognition: societies emphasizing coordination, norm adherence, and avoidance of ambiguity may foster better recognition of uncertainty and error likelihood, influencing metacognitive processes that guide behavior. The societal implications are notable: because confidence influences social persuasion and group decisions, groups with lower metacognitive discrimination may place undue weight on confidently stated but less reliable judgments, shaping collective outcomes differently across cultures.

This study demonstrates that culture modulates metacognitive monitoring across culturally diverse societies and in a high-level cognitive domain, extending prior perceptual findings. Metacognitive efficiency was highest in Portugal and lowest in Saudi Arabia, with China intermediate; sex differences in mental rotation performance did not translate to differences in metacognitive efficiency. A parsimonious model implicates interacting effects of lower individualism and higher uncertainty avoidance in enhancing metacognition, while cultural masculinity relates to confidence bias rather than sensitivity. These findings suggest metacognition is, at least in part, socially shaped and culture-dependent. Future work should test refined (potentially non-linear) models across more culturally diverse samples, include individual-level cultural value measures to parse personality vs. culture effects, examine contexts (e.g., mixed-gender vs. segregated settings) that may modulate confidence bias, and explore domain-generality of cultural effects on metacognition.

• Group-level cultural metrics: Hofstede dimensions compare societies and are not designed to describe individuals; results generalize to groups, not necessarily individuals. Individual adherence to cultural values varies, and personality traits may confound cultural effects.
• Method emphasizes explicit metacognition: Retrospective confidence judgments bias measurement toward conscious, explicit monitoring; implicit metacognitive processes may be underrepresented.
• Reaction times were not analyzed due to heterogeneous hardware across sites.
• Sampling constraints: University students (ages mainly 18–30); Saudi participants tested on segregated campuses, which may influence confidence bias and limits comparability across sexes and contexts.
• Cultural modeling used a simple linear cumulative assumption for illustrative viability; more complex interactions or additional cultural dimensions may be relevant.
• No primary UK data in this study; UK comparisons are indirect via prior literature.