Introduction

The ability to track decisions, cognitive states and behavioural responses, as well as to control one’s own cognitive performance, is referred to as metacognition (cognition about cognition). Metacognition has two important components: metacognitive control and metacognitive monitoring (Flavell, 1979; Schraw, 1998). Control and monitoring rely on partially overlapping neural networks (Fitzgerald et al. 2017; Fleming and Dolan, 2012; McCurdy et al. 2013; Molenberghs et al. 2016; Morales et al. 2018), and enhancement in one type of process may change the efficiency of the other type of process. Meta-monitoring manifests itself in the ability to estimate the degree of uncertainty associated with one’s decisions and behavioural responses. One of the approaches to objectively measure meta-monitoring is the analysis of how closely retrospective confidence ratings track the accuracy or appropriateness of a behavioural response. Efficient metacognition is reflected in higher confidence associated with correct responses than with incorrect responses (Yeung and Summerfield, 2012). The proposed underlying mechanisms are either (1) efficiency of error detection (detecting an error once it is committed) or (2) accurate estimation of the likelihood of making an error (detecting situations when the probability of committing an error is higher). Metacognitive processes that guide future behavioural responses given a particular level of uncertainty about the environment and one’s own cognitive states are meta-control processes. Metacognitive control is based on the efficiency of meta-monitoring processes or the evaluation of the decisions and behavioural responses that are already made. It allows living beings to adapt behaviour when the environment changes or when a new decision strategy leads to better performance (Kepecs et al. 2008). This can happen with (explicit metacognition) and without (implicit metacognition) conscious awareness, although underlying neurocognitive mechanisms of explicit and implicit metacognitive processes vary (Nicholson et al. 2019). We will leave implicit metacognition beyond the scope of this investigation.

Metacognition is affected by a variety of factors, including developmental (De Bruin and van Gog, 2012; Spiess et al. 2016), linguistic (Ordin et al. 2020; Polyanskaya et al. 2022), educational (De Jager et al. 2005; Yeager et al. 2019), learning (Carpenter et al. 2019), genetic (Greven et al. 2009), personality (Conway et al. 2020), mental states like anxiety and depression, as well as mental disorders (Rouault et al. 2018; Nicholson et al. 2019). Recently, researchers have proposed that not only individual differences but also group-level culture-pertaining differences may modulate metacognition by adapting behaviour to the social environment and its norms (Heyes et al. 2020).

The theory of culturally acquired differences in metacognition was tested in a low-level perceptual decision-making task (van der Plas et al. 2022). Chinese and British participants were asked to judge the direction of the dots movement on the screen, when 30 dots were randomly displayed on the screen for 300 ms. Some dots were moving either left or right, while other dots were moving randomly. Participants were asked to report in which direction the dots were moving. The proportion of dots moving directionally was calibrated to each individual to achieve 60%—for low accuracy; 75%—for medium accuracy; and 90%—for high accuracy performance. Once participants responded in which direction the dots were moving, they saw the dots moving in the same direction again, and the strength of evidence could vary (for example, in the initial presentation, the dots were moving leftwards, and the number of dots moving directionally corresponded to the individual perceptual performance of 75%, and in the post-decision presentation, the number of dots moving leftwards could correspond to the individual performance of 60%). After the post-decision presentation, participants were asked to report their confidence in their initial decision regarding the direction in which the dots were moving (on a scale from 0 to 100). The results showed that the Chinese participants revealed better metacognitive evaluations on error trials (i.e., when their response was incorrect, the Chinese participants, following the presentation of additional evidence, lowered their confidence more than did the British participants). The described procedure focuses on error detection (the detection of already committed errors) as a proximate mechanism of metacognition because participants, following their perceptual decision, are provided with more evidence to adjust their confidence in the initial perceptual decision. And Chinese participants were more willing to accommodate additional evidence and correct decisions. The authors suggested that this result pattern reflects the modulation of metacognitive efficiency by cultural differences related to susceptibility to social influence, which is greater in Chinese society than in Western European society (Korn et al. 2014; Lui, 2015; Mahmoodi et al. 2015; Mesoudi et al. 2015; Oeberst and Wu, 2015). Post-decision presentation was conceptualized as a proxy of social influence, and people who are more susceptible to social influence are more likely to re-evaluate their decisions, leading to better detection of trials with a higher probability of errors.

Metacognition adapts behaviour at the individual level by encouraging introspection and revisions of one’s own decisions (van der Plas et al. 2022; Yeung and Summerfield, 2012). Such revisions are based on one’s own mental state, the state of the world, and the consequences of previous decisions. At the interpersonal level, behaviour is adapted by affecting interpersonal communications and collaborations (Bahrami, Olsen, Latham et al. 2010; Frith 2012). Improving interpersonal communication also requires better inferences about the mental states and intentions of other people (Fusaroli et al. 2012; Reddish et al. 2013). Individuals who are better at inferring the intentions of other agents in the communication process might also be better at developing meta-monitoring abilities (Heyes et al. 2020). Not surprisingly, developmental timelines for metacognition and Theory of Mind skills are correlated (Carruthers, 2009; Fleming and Daw, 2017; Lockl and Schneider, 2007). As societies differ in the perceived importance of understanding others’ states of mind and social collaboration (Cleeremans et al. 2020; Hofstede et al. 2010a; Markus and Kitayama, 2010), the precision of understanding others’ states of mind and intentions varies across globe (Heyes and Frith, 2014). Adapting behavioural responses to social norms, others’ expectations and smooth interpersonal communication may result in a better group performance at the expense of individual performance (Bang et al. 2017, 2014; Fusaroli et al. 2012). Therefore, societies that emphasize collective benefits rather than individualism will promote the development of metacognition at the individual level and further enhance metacognitive skills across communities through cultural learning mechanisms (Heyes et al. 2020).

In the current study, we wanted to explore cultural differences among a wider range of culturally divergent societies. To understand the effect of culture on metacognition, it is worthwhile to better define how we could operationalize cultural differences (Cohen, 2009). Cognitive processes and behavioural patterns might vary across populations with different cultural values, including differences in holistic versus analytic thinking, intrinsic versus extrinsic motivation, moral decisions, etc. (Fiske et al. 1998). In our study, we decided to use Hofstede’s model (Hofstede, 2001), when cultural differences are analysed along six dimensions: power distance, individualism, masculinity, uncertainty avoidance, and long-term orientation. Figure 1 shows the differences among Saudi Arabia, Portugal, China and the UK in terms of the six cultural dimensions proposed by Hofstede and Minkov (Hofstede, 2001; Hofstede et al. 2010a; Minkov, 2013). Although we do not collect primary data on the UK population for our study, we will refer to this data in the discussion, placing our results in a wider picture and drawing to the previous work by van der Plas et al. (2022), therefore we included UK data on this table for future reference. The values in Fig. 1 should only be considered as relational (for comparison), and the relative differences between societies are very resilient to changes over time because cultural values, especially those related to individualism and uncertainty avoidance, are self-sustaining even during societal transformations (Hamamura et al. 2021). Therefore, for the purposes of our investigation, we decided to operationalize cultural differences within this paradigm. This operationalization also allows us to move beyond typically exploited differences along the individualism–collectivism dimension and to incorporate other cultural differences that might influence metacognition.

Fig. 1: Differences between Saudi Arabia, Portugal, China and the UK along six cultural dimensions proposed by Hofstede and Minkov (Hofstede, 2001).
figure 1

The data was gathered by Hofstede et al. (2010a) and Minkov (2013) and is publicly accessible for noncommercial purposes as an online Country Comparison Tool, which was used to generate this image at (https://www.hofstede-insights.com/country-comparison-tool?countries=china%2Cportugal%2Csaudi+arabia*%2Cunited+kingdom, last accessed on 06/07/2023).

Full size image

Incorporating multiple measures of cultural dimensions is important in cross-cultural research because different measures might interfere. For example, San Martin et al. (2018) argued that individuals from Arabic and Western populations exhibit self-assertive behaviour combined with a strong commitment to the in-group membership, while in Western societies, self-assertion serves the independence of the group. It makes the socio-cultural profile in Arabic society to be distinct both from the Western profile and from the East Asian profile (in which self-assertion is considered to be a hindrance to in-group identity and in-group belonging).

Traditionally, cultural comparative research focuses on differences between East Asian cultures and Westerners (typically represented by English-speaking populations from the USA or the UK). However, even representatives. of Western cultures may exhibit different cultural profiles (Na, Grossmann, Varnum et al. 2010; Kitayama et al. 2009). We wanted to use a Portuguese population representing Western culture, which is rarely explored in cross-cultural studies. Moreover, we wanted to go beyond the traditional East-West comparison (Kitayama and Salvador, 2024) by extending to a population with a different cultural and psychological profile—the Arabic population (San Martin, Sinaceur, Madi et al. 2018) from the Kingdom of Saudi Arabia. In addition to expanding a spectrum of culturally divergent societies, we wanted to explore whether the effect of culture on metacognitive monitoring will persist (a) in a higher-level cognitive task, and (b) if we allow for the engagement of another potential proximate mechanism of metacognition, which pertains to the estimation of the likelihood of an error. We tested metacognition in a mental rotation task (Shepard and Metzler, 1971), and each trial was immediately followed by confidence judgement. This means that both the error detection and estimation of an error probability may be engaged during metacognitive judgement. The test was administered in Saudi Arabia, China and Portugal. Finally, we attempted to make our samples gender-balanced to explore the potential effect of biological gender (sex) on metacognition in the task, in which cognitive performance is often reported to be higher in males than in females (Quinn and Liben, 2008). The exploration of sex differences in meta-monitoring during mental rotation tasks is particularly interesting given that males and females may potentially develop differential cognitive strategies to perform tasks (Hirnstein et al. 2009); hence, they need to evaluate the application of different cognitive processing strategies to achieve the same goal. We expect to replicate the finding that metacognition is higher in Chinese populations than in Western European (Portuguese) populations, extend the evidence that culture modulates metacognition to (a) previously unstudied culturally divergent populations and (b) to high-level cognitive tasks that engage cognitive mechanisms (that need to be monitored by metacognitive processes) that are distinct from those engaged by low-level perceptual tasks used in earlier experiments. Furthermore, we wanted to go beyond susceptibility to social influence and collectivism as the only culture-specific factor that affects metacognition and to explore a range of other cultural differences in metacognitive monitoring. The secondary purpose of our study—albeit exploratory—is to understand differences in which dimensions of culture affect metacognition. This will help us generate more precise hypotheses for future studies.

Methods

Participants

We recruited 226 students without an immigrant background: 84 from Saudi Arabia (52 females), 70 from China (35 females) and 72 from Portugal (39 females). In Saudi Arabia, participants were recruited separately on male and female campuses due to the national policy of sex segregation. Most participants were single (six were married) and between 18 and 30 years old (the age distribution was skewed towards the early 20s (mean—21.5 years, median—21 years). Experiments were performed in computer classes, with several participants working simultaneously on individual desktops.

Procedure and stimuli

The stimuli represented abstract shapes composed of ten grey square blocks against a black background, as shown in Fig. 2. On each trial, participants saw a pair of shapes, and they had to decide whether the two shapes were identical or different. On each trial, participants had to register—on a 4-point scale—how confident they were in their response (“4”—totally confident, “1”—not confident at all). For pairs with identical shapes, the second shape was rotated either around the X-axis (rolling) or the Y-axis (yawing). The rotation varied from 0° to 340°, with 20° steps. When two identical shapes were presented, mental rotation of the right shape could result in complete alignment of the right shape to the left shape. If the shapes were different, no kind of rotation could align them. In total, we had 18 pairs of identical shapes with rolling, 18 pairs of identical shapes with yawing, and 36 pairs of different shapes. Figure 2 shows examples of identical and different shapes.

Fig. 2: Stimuli example.
figure 2

On the left panel, the shapes are different, and no mental rotation can result in the alignment of these shapes. On the right panel, the shapes are identical, and alignment is possible by 60° yawing (rotating the second shape along the y-axis by 60°).

Full size image

Pairs were presented in randomized order for each participant. The experiment was administered in the native languages of the participants.

Analysis

We did not analyse the reaction time data because we could not be sure that the hardware was identical across the different experimental sites. In addition, performance in the mental rotation task was not the focus of the study. We were interested in how closely confidence ratings tracked accuracy (meta-monitoring).

To ensure that metacognition was engaged in this task, we calculated the average confidence per participant separately for correct and incorrect responses. Confidence ratings that are discriminative of correct and incorrect responses reflect the engagement of metacognitive monitoring on the particular task, and further analysis is justifiable subject to confidence ratings being higher on the trials in which correct responses are given than on the trials in which wrong responses are given. Average confidence is also analysed as a measure of metacognitive bias, allowing for comparisons of overconfidence and underconfidence between groups.

To analyse meta-monitoring, we used the signal detection approach (Galvin et al. 2003; Maniscalco and Lau, 2012, 2014). When correct responses (hits and correct rejections) with high confidence are conceptualized as meta-hits, wrong responses (false alarms and misses) with high confidence are conceptualized as meta-false alarms, correct responses with low confidence are conceptualized as meta-misses, and wrong responses with low confidence are conceptualized as meta-correct rejections. These conceptualizations were used to estimate a pseudod’ that would perfectly fit confidence ratings if a particular individual was an ideal metacognitive observer and what his/her d’ would be if the confidence ratings tracked accuracy most precisely. This pseudod’ is referred to as meta-d’, and it reflects metacognitive sensitivity or the discriminability of confidence ratings.

Meta-d’ is independent of the individual’s tendency to assign higher or lower ratings overall. Importantly, meta-d’ is dependent on a person’s “sensitivity to the signal” (d’). The meta-d’ scales with performance because better performance means better sensitivity to the signal and hence more information for making confidence judgements (Fleming and Lau, 2014). If one person exhibits a high level of task performance and another person exhibits a low level of task performance, the meta-d’ of the former will tend to be higher than that of the latter because the former has more “signal” for metacognitive judgement. Thus, we also added the M-ratio (meta-d’ to d’ ratio) to estimate metacognition given an individual level of task performance. The M-ratio is known as metacognitive efficiency. Metacognitive sensitivity is the sensitivity to the signal that is subject to metacognitive judgement (i.e., the amount of signal that is provided by the cognitive system for metacognitive judgement). Metacognitive efficiency is how effectively an individual can operate with the signal that his or her cognitive system makes available for metacognitive judgements. The M-ratio allows for comparing individuals and groups that differ in task performance.

To model meta-d’, we adopted a hierarchical Bayesian estimation proposed by Fleming (2017). This approach was preferred because it allows for robust estimation given the limited number of trials (we have 72 trials per participant); it does not require data padding (e.g., when participants give zero responses at a particular confidence level); it reduces the effect of outliers on the group results, which means that the full data set can be used; and it also increases the reliability of comparing groups with different sample sizes. We calculated d’ and estimated meta-d’ separately for each sample. Using these scores, we calculated metacognitive efficiency as a ratio of meta-d’ to d’ (M-ratio).

It is important to acknowledge that metacognitive processing does not necessarily engage conscious awareness (Jachs et al. 2015; Kentridge and Heywood, 2000; Logan and Crump, 2010). Although the proposed method is frequently used for measuring overall metacognitive efficiency, we want to emphasize that participants need to consciously evaluate their mental states in order to assign corresponding confidence ratings for each trial. Hence, measuring the efficiency of metacognitive monitoring using retrospective confidence is biased toward explicit metacognition (Fleming and Daw, 2017; Ordin and Polyanskaya, 2021), which is based on conscious representations that are stored in working memory used by cognitive control processes (Heyes et al. 2020). The effect of culture, however, may create unconscious social biases in behaviour. In this analytic approach, such biases are weighted lower than the contribution of conscious awareness to the evaluation of one’s own post-decision mental states. The results presented below are, therefore, more indicative of individuals who are conscious of the social norms regulating culture-specific behavioural patterns.

Results

We ran an analysis of variance (ANOVA) to understand how performance in the mental rotation task (measured as d’) is modulated by sex (male vs. female) and group (Arabic vs. Portuguese vs. Chinese). The analysis revealed that group, F(2,220) = 36.545, p < 0.001, η2p = 0.249, and sex, F(1,220) = 116.625, p < 0.001, η2p = 0.07, significantly affected performance. As expected from earlier studies, males were generally better at performing the task than females were. However, the simple main effect of sex modulated by the group showed that this pattern held in the Portuguese population, F = 5.324, p = 0.022, and in the Chinese population, F = 9.433, p = 0.002, but no difference was observed between males and females in the Arabic population, F = 2.699, p = 0.102. Pairwise comparisons (with the Bonferroni correction) showed that performance in the Arabic population was lower than that in the Portuguese population (for females: ∆d’ (difference in d’) = 1.036, t = 4.999, p < 0.001; for males: ∆d’' = 1.208, t = 4.98, p < 0.001) and in the Chinese population (for females: ∆d’ = 1.041, t = 4.869, p < 0.001; for males: ∆d'’ = 1.398, t = 5.845, p < 0.001). The differences in performance between the Chinese and Portuguese populations were not significant for either females or males (all p values were equal to 1.0 after Bonferroni correction). Although performance in the Arabic population was relatively low (d’ = 0.55 for females and d’ = 0.88 for males), it was still significantly greater than what would be expected by chance. The performance in the Portuguese and Chinese populations was rather high (d’ > 2.0 for males and d’ > 1.5 for females). Overall, the differences in d’ modulated by sex were smaller than those modulated by group. The pattern is shown in Fig. 3a.

Fig. 3: Cognitive and metacognitive performance in the mental rotation experiment in Arabic (Ar), Chinese (Ch) and Portuguese (Pr) populations.
figure 3

A Accuracy in reporting whether two images display rotated and identical, or different shapes (measured as d'). B Metacognitive efficiency in the mental rotation task (measured as M-ratio and reflecting how well participant can track whether their cognitive–accuracy–responses are likely to be correct or wrong). Error bars show 95% CI.

Full size image

We then ran a repeated-measures ANOVA to understand how confidence ratings are modulated by response accuracy (correct vs. incorrect, introduced as a within-subject factor), sex (male vs. female) and group (Arabic vs. Portuguese vs. Chinese). The analysis showed that the correct responses attracted overall higher confidence ratings than did the incorrect responses, F(1,218) = 170.145, p < .001, η2p = .348. The effect of sex, F(1,218) = 11.903, p < 0.001, η2p = 0.052, and the effect of group, F(1,218) = 5.809, p = 0.003, η2p = 0.051, were also significant. The effect size of accuracy was stronger than that of group and sex, suggesting that the confidence ratings were modulated by metacognitive processes to a greater extent than by sex and group. A simple main effect of sex modulated by group showed that males overall were more confident than females in the Portuguese (F = 6.676, p = 0.012) and Chinese (F = 11.984, p < 0.001) populations. In the Arabic population, no significant difference was observed in metacognitive bias between sexes (F = −0.07, p = 0.79). Pairwise comparisons (with the Bonferroni correction) revealed that although correct responses attracted higher confidence ratings across all samples, the difference in confidence rating was significant in the Chinese (males: ∆ = 0.23, t = 5.03, d = 0.48, p < 0.001; females: ∆ = 0.28, t = 6.17, d = 0.6, p < 0.001) and Portuguese (males: ∆ = 0.39, t = 8.2, d = 0.82, p < 0.001; females: ∆ = 0.29, t = 6.78, d = 0.615, p < 0.001) populations but not in the Arabic (∆ = 0.12, t = 2.62, d = 0.26, p = 0.346; females: ∆ = 0.11, t = 2.86, d = 0.22, p = 0.181) population. This proves that metacognition manifests itself in retrospective confidence in the former two populations but not in the latter. The pattern is shown in Fig. 4.

Fig. 4: Mean confidence in correct and wrong responses in Arabic (Ar), Chinese (Ch) and Portuguese (Pr) populations.
figure 4

A Female groups. B Male groups. Error bars show 95% CI.

Full size image

To further probe metacognition, we estimated meta-d’ and calculated M-ratio as a measure of metacognitive efficiency. We subjected the M-ratio to ANOVA with group and sex as factors. The effect of group, F(2,220) = 35.922, p < 0.001, η2p = 0.246, was significant, but the effect of sex was not significant, F(1,220) = 1.803, p = 0.181, η2p = 0.008. As the effect of sex and the interaction between the factors were not significant, F(2,220) = 2.69, p = 0.07, η2p = 0.02, we explored whether differences between populations were significant by comparing the groups pairwise (with the Bonferroni correction) without looking at the modulatory effect of sex. Metacognitive efficiency was higher in the Chinese population than in the Arabic population (∆ = 0.08, t = 4.6, p < 0.001, d = 0.755). In the Portuguese population, we observed significantly higher metacognitive efficiency than in the Chinese population (∆ = 0.07, t = 3.705, p < 0.001, d = 0.623). The pattern is shown in Fig. 3b.

Discussion

The data showed that Portuguese and Chinese males outperformed females in the mental rotation task, while sex-based differences were not observed in the Arabic population. Overall, performance was lower in the Arabic population than in the Chinese and Portuguese populations, with no observable differences between the latter populations. Metacognitive efficiency, on the other hand, was the highest in the Portuguese population and the lowest in the Arabic population, with no observable differences between males and females in any population. Further work is necessary to understand why sex-based differences in cognitive performance in mental rotation tasks do not emerge at the metacognitive level.

In an earlier study, Chinese participants exhibited higher metacognitive efficiency in a low-level visual perception task (Van der Plas et al. 2022), and the proposed explanation was based on a greater emphasis on collaboration and shared goals in Chinese than in Western European society. The tendency toward collaboration leads to conformity to social norms (Lau et al. 2012; Lui, 2015; Mann et al. 1998; Markus and Kitayama, 2010) and the ability to correctly infer others’ intentions and states of mind (Bahrami et al. 2010; Cleeremans et al. 2020). Such tendencies may converge and lead to the emergence of collectivism as a shared cultural value.

Collectivism and conformity to social norms are indeed greater in the Chinese population than in the UK population (Korn et al. 2014; Hamamura et al. 2021; Lui, 2015; Mahmoodi et al. 2015; Mesoudi et al.2015; Oeberst and Wu, 2015), which explains the data pattern in the study by Van der Plan et al. (2022). The data we present here, however, cannot be accounted for by differences along a single cultural dimension. In terms of individualism (the reverse of collectivism), Chinese society indeed places less weight on individualism than Western societies (see Fig. 1), and according to Van der Plas et al. (2022) and Heyes et al. (2020), this should lead to lower metacognitive efficiency in the Chinese population. Comparing China, Saudi Arabia and Portugal (see Fig. 1), we should expect the lowest metacognition to occur in the Saudi population (which is indeed observed in the data) and the highest metacognitive efficiency to occur in the Chinese population. Contrary to this prediction, in the observed data, we observed higher metacognition in the Portuguese population than in the Chinese population.

It is worth noting that enhanced metacognition in the Chinese population was observed only in error trials (van der Plas et al. 2022). This is in line with the hypotheses that the proximate mechanisms of metacognitive enhancement are the detection of trials when the likelihood of an error is increased and the detection of recently committed errors (Ordin et al. 2020; Yeung and Summerfield, 2012). These mechanisms rely on the correct estimation of uncertainty in one’s own cognitive states and information about the environment or stimuli, which might facilitate detecting error-prone trials or responses. Societies, in which avoiding errors or ambiguous, error-prone situations is important may encourage the development of metacognitive ability. We suggest that the differences along the cultural dimension related to avoiding uncertainty—at the group level—might correlate with intercultural differences in metacognitive efficiency. Figure 1 shows that the tolerance of ambiguity and uncertainty in Portugal is substantially lower than that in Saudi Arabia and China. Intolerance to ambiguity hones the ability to detect cases of high uncertainty, promoting metacognitive abilities, at least in social settings, which might also be transferred to other domains. In an attempt to reduce ambiguity in social settings, fixed social rules and rituals are established (Hofstede et al. 2010a). Although this does not mean that such rules are always followed (Hofstede et al. 2010a), the very existence of such rules and the need to clarify whether rituals are followed during each interaction promote interpersonal coordination, reducing the manifestation of individualism. We propose that differences in metacognition that are observed in our data can be explained by the interacting influence of cultural differences along at least two dimensions: uncertainty avoidance and individualism.

This is an exploratory observation that can be used to generate further precise hypotheses for confirmatory studies. Before using this explanation to generate further precise hypotheses for confirmatory studies, we first decided to test its viability, internal consistency and whether it could account for the results of previous studies reported in other experiments (e.g., van der Plas et al. 2022). For this, we assumed that the interaction between intercultural differences and their effect on metacognition is linear and cumulative. This is the simplest interaction suitable only for rough modelling to verify whether the hypothesis is generally feasible. We used the values from Hofstede’s database (see Fig. 1) as the strength coefficients for the effect of uncertainty avoidance (x) and individualism (y). If metacognition is modulated by the linear cumulative effect of these two factors, we have a system of inequalities:

27y + 99x > 20y + 30x (the cumulative effect of individualism and uncertainty avoidance on metacognitive efficiency in the Portuguese population is stronger than that in the Chinese population);and20y + 30x > 48y + 64x (the cumulative effect of individualism and uncertainty avoidance on metacognitive efficiency in the Chinese population is stronger than that in the Saudi Arabian population).

Figure 5a shows the solution space to the first inequality, which is in the half-plane above the blue line, and the solution space to the second inequality is in the half-plane below the green line. The solutions to the system of inequalities are at the intersection of half-planes (shaded area in Fig. 5a).

Fig. 5: Solutions to inequalities.
figure 5

Common solution space for all systems of inequalities confirms the plausibility of the approach to identify cultural differences in at least two cultural dimensions—individualism and uncertainty avoidance—as interacting factors in modulating metacognition.

Full size image

Now, we will add the solutions for the following system:

27y + 99x > 48y + 35x (the cumulative effect of individualism and uncertainty avoidance on metacognition in the Portuguese population is greater than that in the Saudi Arabian population);

And

20y + 30x > 48y + 64x (the cumulative effect of individualism and uncertainty avoidance on metacognition in the Chinese population is stronger than that in the Saudi Arabian population).

Figure 5b represents solutions to this system as a shaded area. The shaded area in Fig. 5a is within the shaded area in Fig. 5b, showing a larger expected difference in metacognitive efficiency between the Portuguese and Arabic populations than between the Chinese and Arabic populations, which is in line with the experimental data.

This system can embrace the primary data we have collected, but a good model should be generalizable over different tasks and cultures. We put our model to the test by checking whether it can also incorporate the results of other studies (if the model is study-specific, i.e., can account for the results of one study but not other related studies, the generalizability and hence the validity of the model is questionable). To validate the model, we tested whether the cumulative effect of individualism and uncertainty avoidance on metacognition in the Chinese population is stronger than that in the UK population, as postulated by van der Plas et al. (2022). We added to the system the following inequality:

20y + 30x > 89y + 35x

Adding this inequality to the system will tell us whether our explanation that metacognition is modulated by cultural differences along two dimensions can also account for the results reported by Van der Plas et al. (2022). Figure 5c adds a black line, with solutions to the new inequality defined as a half-plane below the black line (shaded darker). The shaded area in Fig. 5c (corresponding to the shaded area in Fig. 5a) defines solutions to the system of all four inequalities, suggesting the feasibility of the hypothesis that the interaction of intercultural differences along two dimensions—uncertainty avoidance and individualism—affects intersocietal differences in metacognitive efficiency reported in several studies. The general tendency is that metacognition increases with decreasing individualism and increasing uncertainty avoidance, and differences in one dimension may enhance or inhibit differences in the other dimension. This conclusion might be subject to further rectifications and elaborations (we used the simplest model based on linear cumulative effects) once more data across societies that are more diverse in cultural values become available. Taken together, the study shows that the ability to evaluate one’s own cognitive performance—the ability that allows adapting behavioural responses based on estimated uncertainty and error probability—is culture-determined and social in nature, at least to some extent.

Another interesting observation is related to the cultural dimension of masculinity, or competitiveness, and the importance of open evaluation, ranking in the community and completing work in the best way possible. Anecdotally, Chinese students place considerable weight on exam notes and achieving success. This corresponds to higher masculinity in Chinese society than in Saudi Arabian and Portuguese societies, which is in agreement with the findings of Hofstede et al. (2010b) and Minkov (2013), as shown in Fig. 1. Cross-cultural differences along the masculinity dimension manifest themselves in metacognitive bias: the Chinese participants were more confident overall in their judgements than the Arabic and Portuguese participants were. This pattern is more evident in males than in females, which is in line with more openly ascribed gender roles in societies that are characterized by higher masculinity scores. This finding agrees with earlier studies showing that confidence in one’s decisions tends to be higher in the Chinese population than in the US and Western European populations (Moore et al. 2018; van der Plas et al. 2022; Yates et al. 1989, 1998). It should be noted that metacognitive bias is a general tendency towards over-confidence or under-confidence, both for correct and incorrect responses. Hence, masculinity is not necessarily related to metacognitive efficiency and sensitivity or the power of confidence ratings to discriminate between correct and wrong responses.

Intriguingly, differences in overall confidence between males and females are not observed in the Saudi population, in contrast to Chinese and Portuguese ones. Noteworthy, Saudi was the only group where male and female participants attended different campuses (due to the higher-education governmental policy of segregation based on the biological gender of students). Maybe, the confidence bias is modulated stronger in mixed groups by the societal expectations, strengthening the stereotypical phenotypes, while segregation reduces the stereotype threats. Potentially, we could then expect that the differences in confidence between males and females will not emerge either in societies, placing stronger weight on and promoting gender equality, in the absence of the social expectations regarding phenotypes modulated by biological gender. Empirical evidence is required to investigate how overall confidence bias emerges and is maintained in mixed societies varying in gender-equalities and in societies that follow segregation practices.

Our study showed that culture may modulate metacognitive monitoring not only in low-level perceptual tasks that were used in previous empirical studies, but also in high-level cognitive task that engages very different mechanisms for decision-making, and the effect of culture is not limited to traditional West–East comparison, it is much more nuanced. Culture might have a pervasive effect on metacognitive efficiency across multiple domains. As enhanced metacognition is related to lower confidence assigned to those decisions when the likelihood of an error is higher (e.g., Fleming and Lau, 2014; Kunimoto et al. 2001; Maniscalco and Lau, 2012; Ordin and Polyanskaya, 2021; Persaud et al. 2011; Schwiedrzik et al. 2011), and people tend to assign higher credibility to those opinions that are expressed with a higher level of confidence (Ais et al., 2016; Mann et al. 1998; Zarnoth and Sniezek, 1997), people in ambiguous situations tend to believe the opinions of those who are less able to discriminate between potentially more and less erroneous decisions. Also, metacognition might affect decision-making processes (Yeung and Summerfield, 2012). Therefore, differences in metacognition between groups might have severe societal ramifications, especially in the context of group decision-making. Differences in metacognition might also account for why, in different societies, different conclusions might be drawn based on the same information, and different decisions are made at the group level. Therefore, it is of paramount importance to study what might influence metacognition. Our study showed that in addition to educational and linguistic factors (Ordin et al. 2020; Polyanskaya et al. 2022), metacognition might also be modulated by intergroup cultural differences, social norms and expectations. In line with San Martin et al. (2018), we argue that these differences are more subtle than the predominant East-West opposition.

This study is built on the premise that cultural values are shared by all society members, and this assumption represents a limitation on the transferability of the results from the group to the individual levels. Individuals, however, might differ in how strongly they subscribe to the cultural values of their communities. We present the results at the group level because the tool (Values Survey Module: Hofstede, 2001; Hofstede and Minkov, 2013; Minkov, 2013) was designed with the purpose of comparing societies in several cultural dimensions rather than describing societies (in other words, the tool is not designed to say that individualism is a shared cultural value in Saudi society and collectivism is a shared cultural value in Chinese society; rather, it is designed to say that individualism in Saudi Arabi society is expressed more strongly than in Chinese society). The data are only valid at the group level. Importantly, Hofstede and McCrae (2004) showed that mean personality scores (using the Five-Factor Model of personality) in culturally divergent societies correlate with culture dimension scores. Individualism is positively correlated with extraversion, masculinity is positively correlated with openness to experience and neuroticism and negatively correlated with agreeableness, and uncertainty avoidance is negatively correlated with agreeableness and positively correlated with neuroticism. Moreover, “independent”, “unique”, “individual” (descriptors that are typically claimed to be characteristic of Western societies) and “interdependent”, “collective” (descriptors that are typically claimed to be characteristic of non-Western societies) measurable images of personalities can and do exist in each individual (Singelis, 1994). Hence it might be useful to add the scores on cultural dimensions for individuals. Individual scores might tell us about the culture an individual is brought up in and also about personality, which either enhances or inhibits the strength, with which the cultural values of society are shared by an individual. Although it is not possible to disentangle individual personality and culture-specific aspects from individual scores, such scores can be used as covariates to regress out personality contribution to metacognition from culture-based modulation of metacognition. This direction for future research may provide explanations for why meta-sensitivity differs between individuals even within culturally homogeneous societies (Ais et al. 2016; Conway et al. 2020; Fleming et al. 2010; Lockl and Schneider, 2006; 2007).