The topic of which factors affect academic performance is a long-standing debate in educational research (Guez et al., 2018). For decades, understanding academic achievement and its determinants has attracted considerable attention from researchers and practitioners. Numerous studies collect and analyze data on various factors that affect learning (O'Connell and Marks, 2022). This wealth of evidence has prompted multiple attempts to develop theories of academic achievement (Bloom, 1976; Carroll, 1963; Glaser, 1980). One of the most prominent factors determining academic achievement is the cognitive capacities of individuals. Cognition has a noteworthy impact on understanding, analyzing, and organizing the learning process. The cognitive capacities of individuals in different fields such as mathematics and science literacy can play a decisive role in academic performance (Deary et al., 2007; Roth et al., 2015). However, the complex relationships of these variables have not been comprehensively investigated in different samples. Moreover, the most frequently repeated determinants of learning outcomes are cognitive capacities, SES, and gender (Boman, 2023; Lee and Borgonovi, 2022; Marks and O'Connell, 2021).

Cognitive ability affects academic achievement (Deary et al., 2007; Shi and Qu, 2022; Sternberg et al., 2008). There is broad agreement regarding the positive relationship between cognitive ability and academic achievement (Chen et al., 2012). Studies have shown that academic achievement is related to basic cognitive processes (Tikhomirova et al., 2020). Cognition predicts academic achievement from elementary school through graduate school (Greene et al., 2018). Moreover, academic performance and cognitive abilities predict each other in the development process (Peng and Kievit, 2020). Cognitive ability accounts for from 51 to 75% of the variance in academic achievement alone (Rohde and Thompson, 2007). Specifically, later research such as Flores-Mendoza et al. (2021) and Pokropek et al. (2022) confirmed the strong relationships between performances on PISA and general cognitive ability. Therefore, to understand PISA scores, it is appropriate to examine cognitive constructs’ associations with PISA performance at the country level (Boman, 2022). Consequently, the effects of cognitive abilities on academic performance have been frequently investigated. The current study is different from previous research in addressing the relationship between cognitive constructs employed in reading texts and math and science performance. In addition, these relationships were analyzed for 71 countries overall and at each country level, which is another significance of the research.

Science, mathematics, and reading literacy are some of the most basic required skills in contemporary societies. Individuals’ capability to understand scientific and mathematical concepts, develop critical thinking skills, and solve complex problems is indispensable to adapt and succeed in today’s complex world (McConney et al., 2014; Zan and Di Martino, 2007). Science, mathematics, and reading literacy is a factor that affects not only personal lives but also societies and economies (Bybee and McCrae, 2011). Therefore, it is of great importance to understand the relationship between cognition and academic performance to develop science, mathematics, and reading literacy.

Besides cognitive capacities, socioeconomic status is an important factor influencing academic performance (Bradley and Corwyn, 2002; Davis-Kean, 2005). Indeed, socioeconomic status, family income, parent’s education level, and living environment can affect individuals’ access to educational resources and benefit from them for academic achievement (Hattie, 2009; Sirin, 2005). The impact of socioeconomic status on academic performance needs to be clarified across different countries. Furthermore, gender is an important variable affecting academic performance. Large-scale international or country-level investigations have shown trivial gender differences in this regard (Else-Quest et al., 2010; OECD, 2010). Previous studies have shown that gender may play a role in explaining differences in academic performance, but the causes and effects of these differences are still a matter of debate (Fryer and Levitt, 2010; Kenney-Benson et al., 2006; Lindberg et al., 2010).

The result of multi-group path analysis by country and subject, overall sample, and overall sample with country fixed effects are reported in Table 3. The values in the table are standardized regression coefficients, their standard errors (in parenthesis), the total R2 value for the full model (all variables in the table), and the ΔR2, which is the unique contribution of the subscales of reading compared to the demographic-only model (FEMALE and ESCS). Although results from the overall sample and the overall sample with country fixed effects are similar, we report results of the overall sample with country fixed effects. Considering the contextual effect of countries, females scored substantially lower on math and science despite controlling for socioeconomic status and reading. They scored 0.23 (p < 0.001) and 0.18 (p < 0.001) standard deviation below males in math and science, which is equivalent to dragging an average student from the 50th percentile to 41st and 43rd percentile, correspondingly (calculated using PowerUpR; Bulus et al., 2021). The benefit of coming from a more well-to-do family was minuscule after controlling for students’ gender and reading ability. A standard deviation increase on the ESCS was associated with a measly 0.07 (p < 0.001) standard deviation increase in math and 0.04 (p < 0.001) standard deviation increase in science. This is an important finding because what we know about the socioeconomic gap seems to be mainly driven by a lack of resources to prepare students for reading literacy.

In the overall sample with country-fixed effects, an increase in the cognitive constructs employed reading (RCLI, RCUN, and RCER) was associated with an increase in math and science after controlling for gender, socioeconomic status, and text structure subscales of reading. If a student is one standard deviation above the average on all of the RCLI, RCUN, and RCER, most of the increase in math and science will be due to RCER, followed by RCUN, and finally RCLI. In comparison, RCER plays a more significant role in math and science achievement. This makes sense since RCER is a more complex cognitive process than RCUN, and RCUN is a more complex cognitive process than RCLI.

RCER affected math achievement in 58 out of 71 countries and science achievement in 69 out of 71 countries. On average, one standard deviation increase in RCER was associated with 0.17 (p < 0.001) and 0.20 (p < 0.001) standard deviation increase in math and science. We also found that RCUN affected math achievement in 27 out of 71 countries and science achievement in 55 out of 71 countries. On Average, one standard deviation increase in RCUN was associated with an additional increase of 0.13 (p < 0.001) and 0.17 (p < 0.001) standard deviation in math and science. Finally, RCLI affected math achievement in 33 out of 71 countries and science achievement in 26 out of 71 countries. On average, one standard deviation increase in RCLI was associated with a further increase of 0.10 (p < 0.001) and 0.07 (p < 0.001) standard deviation in math and science. In other words, a student one standard deviation above the mean on RCER, RCUN, and RCLI, cumulatively, will have a math score roughly 0.17 + 0.13 + 0.10 = 0.40 standard deviation above an average student in math and 0.20 + 0.17 + 0.07 = 0.44 standard deviation above an average student in science.

An increase in the text structure subscales of reading (RTSN and RTML) was associated with an increase in math and science after controlling for gender, socioeconomic status, and cognitive reading processes. We found that RTSN affected math achievement in 58 out of 71 countries and science achievement in 71 countries. On average, one standard deviation increase in RTSN was associated with 0.21 (p < 0.001) and 0.23 (p < 0.001) standard deviation increase in math and science. We also found that RTML affected math achievement in 67 out of 71 countries and science achievement in 69 out of 71 countries. On average, one standard deviation increase in RTML was associated with an additional increase of 0.18 (p < 0.001) and 0.20 (p < 0.001) standard deviation in math and science. In other words, a student with one standard deviation above the mean on RTSN and RTML will have a math score of roughly 0.21 + 0.18 = 0.39 standard deviation above an average student in math and 0.23 + 0.20 = 0.43 standard deviation above an average student in science.

Cognitive constructs employed in reading subscales (RCLI, RCUN, RCER, RTSN, and RTML) explain 56% of the variance in math on average (as low as 0.43 [Tatarstan (RUS)], as high as 63% [Albenia]) and 63% of the variance in science on average (as low as 52% [Tatarstan (RUS)], as high as and 70% [Albenia]). These values are the incremental increase in R2 value, in other words, what is left after removing the explanatory power of gender and socioeconomic status. This means, regardless of students’ gender and their socioeconomic background, good readers do substantially better in math and science. While this is still the case, it is still unknown why most of those who are good at math are also good at reading, but vice versa does not hold. Interestingly, adding country-fixed effects does not improve R2beyond what is already in the model. This tells us that the conditional mean of math and science across countries (after controlling for gender, socioeconomic status, and subscales of reading) does not vary substantially. This was also confirmed via checking their regression coefficients for fixed effects. This provides evidence against the idea that some cultures are better at math and science. They might be better because they are good readers or come from well-to-do families.

It is not easy to make sense of results by country. Standardized effects and their 95% Confidence Intervals (CI) are reported in Supplement Figures S1, S14. Countries are ordered by their effect size. For math and science subjects, there was a moderate to large negative correlation between the coefficient of RCLI and coefficients of RCUN (r = −0.42 and − 0.49, p < 0.001) and RCER (r = −0.55 and −0.53, p < 0.001) across 71 countries. The smaller coefficient of RCLI in a country, the larger coefficients of RCUN and RCER. For example, in math subjects, the Netherlands, Korea, and Latvia were within the bottom five countries on the effect of RCLI, but all three countries are within the top five on the effect of RCER (see Supplement Figures S3, S5). Similarly, in the science subject, Netherlands and Denmark were within the bottom five on the effect of RCLI, but Denmark was within the top five on the effect of RCUN (compare Supplement Figures S10, S11), and the Netherlands was within the top five on the effect of RCER (compare Supplement Figures S10, S12).

For math and science subjects, there was a moderate to a large negative correlation between the coefficient of RCUN and the coefficient of RCER (r = −0.49 and −0.43, p < 0.001). The larger coefficient of RCUN, the smaller the coefficient on RCER. For example, in math subject, Dominican Republic, Thailand, and Indonesia were within the top five countries on the effect of RCUN, but all three countries were within the bottom five on the effect of RCER (compare Supplement Figures S4, S5). For math subject only, there was a moderate positive correlation between the coefficient of RCUN and the coefficient of RTSN (r = 0.30, p = 0.01). The larger coefficient of RCUN, the larger the coefficient of RTSN.

Finally, for math and science subject, there was a very large negative correlation between the coefficient of RTSN and the coefficient of RTML (r = −0.94 and −0.95, p = 0.001). The smaller coefficient of RTSN, the larger the coefficient on RTML. For example, in math subjects, countries Canada and New Zealand were within the bottom five on the effect of RTSN. However, they are both within the top five on the effect of RTML. Similarly, in the science subject, countries Korea, Macao, the United Kingdom, and Denmark were within the bottom five on the effect of RTSN; however, they are all within the top five on the effect of RTML.

This study conducted a multi-group path analysis to examine the relationships between cognitive constructs, gender, socioeconomic status, and math and science achievement across 71 countries. The results are as follows: First, increases in the cognitive constructs employed in reading (locating information, understanding, evaluation and reflection, single thinking, and multiple thinking) were associated with increased math and science achievement. “Evaluation and reflection” had the most significant impact, followed by “understanding” and then “locating information.” A student with higher scores on all three cognitive constructs had higher math and science scores. Moreover, “single thinking” and “multiple thinking” substantially impacted math and science scores. Second, the combined effects of cognitive constructs employed in reading explained a significant portion of the variance in math and science scores. On average, they explained 56% of the variance in math and 63% of the variance in science, after accounting for gender and socioeconomic status. Third, adding country-fixed effects did not substantially improve the model’s explanatory power. This suggests that the conditional mean of math and science achievement does not significantly vary across countries when controlling for gender, socioeconomic status, and reading abilities. Finally, there were various inter-country relationships among the coefficients of cognitive constructs employed in reading. For instance, countries with lower “locating information” coefficients often had higher “understanding” and “evaluation and reflection” coefficients. Similarly, countries with lower “single thinking” coefficients tended to have higher “multiple thinking” coefficients.

The current research was one of the first to examine the relationships between five cognitive constructs employed in reading (locating information, understanding, evaluating and reflecting, single and multiple thinking), gender, ESCS, and science and mathematics literacy performances across 71 different countries. Overall findings for 71 countries indicated that “locating information,” “understanding,” “evaluating and reflecting,” “single thinking,” and “multiple thinking,” gender, and ESCS explained the total variance of mathematics 75% and science 80% after controlling for country effects (see Table 3). At the country level, cognitive constructs; locating information, understanding, evaluating and reflecting, and single and multiple thinking explained the total variance ranging from 52 to 70% for science and from 43 to 63% for mathematics across 71 countries (see Table 3). Moreover, ESCI was linked with science and math overall, and across 71 countries except for Brunei Darussalam, Kazakhstan, Philippines, and B-S-J-Z (China). Findings related to cognitive constructs; locating information, understanding, evaluating and reflecting, single and multiple thinking associated with science and mathematics performance for most of the 71 countries. Single and multiple thinking were related to science and mathematics performance for all countries except Costa Rica, in which multiple thinking was not related to mathematics (see Table 3).

This study emphasized the influence of cognitive constructs employed in reading (locating information, understanding, evaluating and reflecting, single and multiple thinking), gender, and ESCS on science and mathematics literacy performance across the 71 countries that highlighted the significance of national cultural context on these relationships. Thus, it was suggested gender, ESCS, “locating information,” “understanding,” “evaluating and reflecting,” “single thinking,” and “multiple thinking” have culturally specific relationships with science and mathematics performance across the 71 countries (see Table 3). As a result, the relationships of these cognitive constructs with science and mathematics performance indicated differences as well as similarities across 71 countries. A follow-up random-effects meta-analysis indicated that effects varied across countries from small to moderate degrees (see Table 4). Among the cognitive constructs examined, understanding appears to be the most consistent across countries, with only small variations observed. In contrast, locating information and evaluating and reflecting show moderate degrees of variation across countries.

Especially, understanding the relationships among these variables overall and for each country level would provide valuable insights for local and global researchers, policymakers, and educators. Specifically, country-level findings regarding the relationship between cognitive constructs and math and science performance would provide projections for policymakers and educators on which relationships should be improved.

Based on the findings, cognitive constructs; locating information, understanding, evaluating and reflecting, single and multiple thinking were found to be facilitative to science and mathematics literacy for 15 years old students, which was verified by previous studies on the effect of cognition on academic performances (Deary et al., 2007; Hopfenbeck et al., 2018; Roth et al., 2015; Zhang and Bae, 2020). Moreover, locating information, understanding, evaluating and reflecting, single and multiple thinking and academic performance relationships varied substantially across 71 countries, these findings are consistent with the assertion that culture shapes cognition (Markus and Kitayama, 1991). Furthermore, this result was also consistent with Nisbett (2004) proposal, that culture exerts an impact on the development of cognition, and cognitive outcomes, including academic performance.

This study enriches prior research on cognition which has primarily compared a few Western and Eastern cultures and examined only limited cognitive variables. Another critical significance is exploring comprehensive cognitive variables related to academic performances across 71 different countries, which makes it more comprehensive compared to prior studies that only examined limited countries. However, the current study has limitations in analyzing the relationships amongst socioeconomic status, gender, cognitive constructs, and academic performance across 71 countries. The fact that this research contained too many findings from 71 countries makes it impossible to evaluate and discuss these findings at the country level within the scope of the article. In coming research, these relationships could be investigated at the individualism vs. collectivism (Hofstede and Hofstede, 2005) and different cultural levels such as grouped by King et al. (2024) Africa and the Middle East, Confucian, East-Central Europe, East Europe, English Speaking, Latin America, Southeast Asia, West Europe, etc.

At the country level, cognitive constructs demonstrated an interesting reconciliatory pattern with academic performances. There were various inter-country relationships among the coefficients of cognitive constructs employed in reading. For instance, countries with lower “locating information” coefficients often had higher “understanding” and “evaluation and reflection” coefficients. Similarly, countries with lower “single thinking” coefficients tended to have higher “multiple thinking” coefficients. When students search for answers to an item, they use “locating information,” “understanding,” “evaluation and reflection,” “single thinking,” and “multiple thinking” employed in reading at different levels across 71 countries. These findings showed that 15-year-old students used various cognitive constructs at different levels when searching for the right choice across countries. We can see this as an indication that different cognitive paths are used to achieve the same goal across 71 countries. It is a common finding of many studies that cognitive constructs and development differ across cultures (Markus and Kitayama, 1991; Nisbett, 2004). Cultural differences in the cognitive constructs would also lead to differences in the variables associated with these cognitive constructs. Specifically, “locating information,” “understanding,” and “evaluating and reflecting’s” association with academic performance have reflected more apparent cultural differences amongst countries. Furthermore, the minuscule differences in explanatory power between models with and without country-fixed effects tell an interesting story. While the cultural context plays a minor role in predicting outcomes above and beyond SES, gender, and cognitive constructs, the effects of cognitive constructs vary depending on the culture. This pattern suggests a reconciliatory relationship between cultural factors and cognitive constructs in explaining the observed outcomes.

Overall findings for 71 countries indicated that “locating information,” “understanding,” “evaluating and reflecting,” “single thinking,” and “multiple thinking,” gender, and ESCS explained the total variance of mathematics 75% and science 80% after controlling for country effects. On the other hand, cognitive constructs employed in reading explained 71% of the variance in math and 78% of the variance in science. This finding shows that “locating information,” “understanding,” “evaluation and reflection,” “single thinking,” and “multiple thinking” utilized in reading literacy are important determinants of mathematics and science performance and explain majority of the effect of cultural differences on cognitive outcomes. One of the prominent findings of this research is that mathematics and science achievement gaps across countries decrease when controlling country effects. This finding suggests that differences in reading literacy between countries might explain the mathematics and science achievement gap across countries. Science, mathematics, and reading literacy are not independent each of other; performance in science and mathematics is determined by the cognitive constructs employed in reading. Therefore, educational policymakers intending to increase science and mathematics performance should consider cognitive constructs employed in reading literacy. So, policymakers should focus on not only science and mathematics content but also reading literacy to improve math and science performance. Through reading literacy, both cognitive skills could be developed and learned how to use these skills in different contexts. From this point of view, PISA participant countries intending to increase PISA academic performance should give priority to the development of these cognitive constructs; “locating information,” “understanding,” “evaluation and reflection,” “single thinking,” and “multiple thinking.” Country-level education policies on the development of PISA performances should be organized in a way that ensures more effective use of cognitive constructs especially in the fields of reading, science, and mathematics literacy.

To summarize the implications for practice: The first implication, strengthening cognitive constructs; “locating information,” “understanding,” “evaluation and reflection,” “single thinking,” and “multiple thinking” should be a priority in educational programs. Improved cognitive constructs employed in reading not only enhance literacy but also contribute to better performance in math and science literacy. Second, schools can adopt cross-curricular strategies that integrate reading, math, and science education. Fostering connections between these subjects might help students transfer cognitive skills across domains. Third, while country-specific differences exist, the impact of cognitive constructs employed in reading on math and science achievement seems consistent. This suggests that interventions focusing on these cognitive skills could have a broad impact across different cultural contexts. Fourth, further research could investigate the specific mechanisms through which cognitive constructs employed in reading contribute to math and science achievement. Understanding these mechanisms could inform more targeted educational interventions. Last implication, policymakers should consider these findings when designing curriculum frameworks and assessments, ensuring that both reading and cognitive skills are adequately developed to support math and science learning.

In conclusion, the study underscores the importance of cognitive constructs employed in reading; “locating information,” “understanding,” “evaluating and reflecting,” “single thinking,” and “multiple thinking,” gender, and socioeconomic status in explaining math and science achievement across various countries. 15-year-old adolescents’ cognitive constructs positively relate to science and math performances, which implies that using cognitive constructs in school settings is crucial for students’ academic performances. Moreover, as a reflection of country-level relationships, cognitive constructs have correlations with academic performances at different significance levels across 71 countries. These findings provide valuable insights for educators, policymakers, and researchers to design effective interventions and strategies that promote student success in these critical subjects for each of countries.