Cognitive, affective and sociological predictors of school performance in mathematics Authors: Bernadett Svraka, Jordan Lasker, Péter Przemyslaw Ujma Published: 2024 ABSTRACT AbstractMathematics is a key school subject for some of the most lucrative and economically important careers. Low mathematics performance in school is associated with low psychometric intelligence, family socioeconomic status (SES), specific mathematical abilities, and high mathematics anxiety. We used a sample of Hungarian schoolchildren (N = 102, mean age = 12.3 years) to directly compare the predictive power of general intelligence, specific mathematical abilities measured by the Pedagogical Examination of Dyscalculia (DPV), mathematics anxiety, and socioeconomic status for mathematics grades. Mathematics grades correlated with IQ, specific mathematic ability, mathematics anxiety, and a composite measure of family SES. The WISC-IV showcased a manifest correlation of 0.62 and a latent correlation of 0.78 with the DPV and high manifest (r=-0.53) and latent (r=-0.59) correlations with mathematics anxiety. IQ alone accounted for 52% of the variance in mathematics grades. IQ, specific mathematical ability, family SES and mathematics anxiety jointly accounted for 56% of the variance in grades, with a non-significant contribution of specific mathematical ability and family SES over IQ and a marginal contribution of mathematics anxiety. Our results show that psychometric intelligence is the most important predictor of mathematics grades, while family SES and specific mathematical abilities are only associated with grades to the extent they reflect psychometric intelligence. The results, however, confirmed a small role of mathematics anxiety over intelligence in predicting grades.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-024-77904-7.Keywords: General cognitive ability, Family SES, Dyscalculia, Mathematics anxietySubject terms: Psychology, Mathematics and computing Introduction Mathematics is one of the most widely taught school subjects and improving performance in it is the goal of multiple nationwide efforts in various countries1–3. The significance of mathematics education is undergirded by the fact that mathematics knowledge is essential in STEM fields. Understanding sources of variation in mathematics performance is essential for both scientific understanding and policy, especially as interventions to improve school performance in mathematics and other subjects have been largely unsuccessful4. Numerous factors have been shown to predict school performance, including performance in mathematics. The goal of the current study was to utilize a unique dataset to directly, head-to-head test the magnitude of four possible influences: those of psychometric intelligence, family socioeconomic status, mathematics anxiety, and specific mathematical abilities, the last one using a dyscalculia screening instrument. Intelligence, by the definition of the American Psychological Association, is “the ability to derive information, learn from experience, adapt to the environment, understand, and correctly utilize thought and reason (https://dictionary.apa.org/intelligence). The concept of a single general intelligence is supported by very high convergent validity of not only various intelligence tests5, but also of tests nominally designed to measure non-intelligence abilities, such as educational tests6,7, university entrance tests8,9, or tests of executive functioning10. Such a domain-general ability to understand and solve cognitive problems is likely a large influence on mathematics performance in school. Indeed, in a meta-analysis11a correlation of 0.42 (uncorrected) to 0.49 (fully corrected) was observed between manifest intelligence test results and grades in mathematics. In another meta-analysis12, fluid intelligence exhibited a similar association with mathematics performance (r= 0.41). In yet another meta-analysis13, it was found that latent general intelligence correlated at r= 0.72 with mathematics performance. Once general intelligence is accounted for, specific cognitive abilities account for only a low but statistically significant amount of variance in school grades, including in mathematics (see14 for a recent study and a review of the previous literature). Socioeconomic status (SES) refers to the societal standing of students’ families, operationalized by various, relatively weakly correlated measures such as family income, parents’ educational level, the number of books at home, or composites of similar variable15. Various measures of students’ socioeconomic status are related to mathematics performance16. Some research, including the OECD’s Programme for International Student Assesssment (PISA, using a composite metric of various SES indicators) has naively attributed a causal nature to these correlations, suggesting that parents possessing higher socioeconomic status can directly improve their children’s performance through additional home tutoring or a more nurturing home environment17–19. In contradiction to this theory, however, more detailed studies have shown that once psychometric intelligence is accounted for, various measures of family socieconomic status are at best a minor contributor to school performance20,21. The between-family correlation of SES and school performance can also arise from genetic effects, wherein heritable characteristics such as intelligence or conscientiousness result in higher SES in the parents and better school performance in their biological offspring, without a true causal effect from the former to the latter. Quantitative genetic studies can answer whether family SES is causal for school performance. Such studies are also important because they are not sensitive to precise definition of SES. This is because children growing up in the same family by definition share all of their families’ characteristics, so they also automatically share all possible indicators of family SES as well. Low effects of the family environment indicate that family SES, no matter how we define it, has little causal contribution to school performance. This is indeed the typical conclusion of behavior genetic studies, which report substantial genetic contribution to variation in school performance, with at most a minor causal role for the family environment, with a substantial genetic correlation between intelligence and general academic or mathematics performance22–25. Altogether, critical studies of the SES-school performance link suggest that children from socioeconomically better-off families do better in schoolmainly due to their higher average psychometric intelligence, which in turn is largely explained by genetic factors. However, the findings by no means fully exclude the contribution of family socioeconomic position for mathematics25,26. Socioeconomic status and psychometric intelligence are factors with widespread effects beyond the domain of mathematics performance. In addition to these general effects, at least two specific factors have also been hypothesized to have substantial effects on mathematics performance: mathematics-specific anxiety and specific deficits in the domain of mathematics (i.e., dyscalculia). Mathematics anxiety is “a feeling of tension, apprehension, or fear that interferes with math performance”27. Mathematics anxiety is a common complaint among students, and it disproportionately affects older children and females2. Mathematics anxiety is only moderately correlated with general anxiety2,28and it is sometimes separated into cognitive and affective components29. Higher levels of mathematics anxiety are related to lower levels of performance in mathematics2,30. Mathematics anxiety is also related to lower performance in non-mathematical tasks, although the correlation is generally weaker28,31. There are multiple non-exclusive pathways through which mathematics anxiety and mathematics performance could become related. First, it is possible that mathematics anxiety – as it is typical of anxiety – leads to avoidance behavior, and the affected students may, consequently, fail to obtain sufficient practice in mathematics27. Second, it has been hypothesized that mathematics anxiety affects performance by overloading working memory2. Several studies have found worse working memory performance in those with higher mathematics anxiety32–34. Third, it is possible that mathematics anxiety stems from students’ perceptions of their pre-existing mathematical difficulties35,36. Students with low general or specific mathematical ability may correctly perceive that they have trouble grasping concepts and procedures in mathematics and their anticipated failure may generate anxiety. The correlation between mathematics anxiety and non-mathematical abilities suggests that the former is related to perceptions of low ability in a more general sense31. This presents the possibility that mathematics anxiety is part of a vicious cycle35in which anxiety-amplifying performance deficits prevent students with poor initial performance from obtaining the practice they need to improve. Importantly, some studies demonstrated interaction effects between mathematics anxiety and motivation, with highly motivated students performing best in case of intermediate, not very low levels of anxiety37. In addition to the general intelligence discussed above, mathematics performance may be affected by specific mathematical abilities (or the lack thereof). Specific deficits of mathematical ability are of major clinical interest. “A pattern of difficulties characterized by problems processing numerical information, learning arithmetic facts”, “difficulties mastering number sense, number facts, or calculation” and “difficulties with mathematical reasoning” are diagnostic criteria for Specific Learning Disorder (with the “Dyscalculia” specifier if these specific abilities are affected) according to DSM-538. (See39for a brief review of other diagnostic systems.) Although dyscalculia is often interpreted40,41as a deficit of mathematical abilities with otherwise fully intact intellect, DSM-5 criteria about the latter are not strict, merely stating that “learning difficulties are not better accounted for by intellectual disabilities”, which is defined under Diagnostic Features as an IQ score above 70. There is no single standard diagnostic test to objectively define dyscalculia41, and existing screening tests42–49aim to identify dyscalculia based on low scores in tasks requiring mathematics skills or knowledge, usually without comparing these to performance in a standard intelligence test (but see44for a partial exception). Because the correlation between general intelligence and specific mathematics performance is high11–13, low performance on a mathematics test cannot be interpreted by itself as evidence for a specific (rather than general) deficit in abilities. While it is known that deficits in other domains are common in those with dyscalculia50, it is currently unknown to what extent single-component screening tests of dyscalculia capture unique mathematics deficits or low general intelligence. To our knowledge, no previous study has explicitly compared the predictive power of general (intelligence, family socioeconomic position) and specific (specific mathematics ability, mathematics anxiety) predictors of mathematics performance in school. We aimed to bridge this gap with a unique sample of school children who had measurements of all these factors. Specifically, we were interested in how well individual predictors (intelligence, SES, specific mathematical abilities, and mathematics anxiety) predict mathematics grades, as well as how much their predictive power is shared. Methods Participants and ethics Participants were 6th -grade elementary school students in Budapest (N = 102), recruited by convenience sampling. All parents signed a consent form. This study was approved by the ethics committee of the ETT TUKEB, permit number: 35643-5/2017/EKU. The gender ratio was equally distributed: 52 boys and 50 girls participated in the study. The average age was 12 years 3 months, i.e., 12.28 (± 0.41) years, the extreme values were 11.42 years and 13.80 years. Collecting data in the 6th grade was motivated by previous observations that mathematics anxiety, while present before this time, is amplified when the mathematics curriculum gets more difficult around this time51. In this sample, we collected data about mathematics grades as the outcome variable of interest, WISC-IV IQ as a measure of general cognitive ability, scores on a dyscalculia screening tool as a measure of specific mathematical abilities, family socioeconomic status as a possible additional predictor of school performance, and scores on a standardized mathematics anxiety scale to assess the role of this factor. Mathematics grades We collected participants’ 6th grade end-of-year mathematics grades from school records. Intelligence testing All participants completed the core subtests of the Hungarian version of the Wechsler Intelligence Scale for Children (WISC-IV)52,53. IQ scores were calculated based on the WISC-IV’s standards. WISC-IV is an intelligence test whose structure corresponds to the Cattell-Horn-Carroll hierarchical model53,54 with broad abilities at the intermediate and a general factor (g, general intelligence) at the highest level. We used IQ scores derived from WISC-IV as a single estimate of general intelligence. Socioeconomic status Socioeconomic status (SES) was calculated from the following variables: maternal and paternal education levels, household size, and student reports about whether they had their own room. Education levels were originally reported as a categorical variable, but they were recoded as years of education according to ISCED-2011 criteria55. Observed education levels were vocational education (coded as 11 years in education), high school diploma (coded as 12 years), 3-year college degree (coded as 15 years) or 5-year college degree (coded as 17 years). We expressed family SES as the first unrotated principal component score of our SES indicators, but we also explored the roles of our individual indicators on their own. Specific mathematical abilities Specific mathematical abilities were examined using the Pedagogical Examination of Dyscalculia (Diszkalkulia Pedagógia Vizsgálata, DPV), a test developed to measure mathematical abilities and deficits47. This test is widely used by educational testing services in Hungary to screen for dyscalculia, but it has not been described in the English language. Therefore, we provide a detailed description of the test here. The DPV is designed to not only measure numerical skills, but to also provide a comprehensive picture of a child’s mathematical skills and abilities. The study seeks to separate mathematical underachievement and backwardness caused by inadequate education or environmental disadvantage, as well as, according to degree of severity, mathematics learning difficulty and developmental dyscalculia as a specific learning disorder. The main goal of the test is to assess number concepts (abstract and discrete semantic representation), the understanding of mathematical operations, and basic mathematical functions and sub-skills. The measurement procedure examines the separate hypothetical systems of internal representation, the analogue quantity system (comparison, approximate calculation, estimation), the Arabic number format (symbolic system of Arabic numerals), the verbal system (storing and retrieving arithmetic facts, e.g. multiplication tables), the basic output and input modalities (number processing, e.g. writing, reading numbers). During the examination, among the components of calculation operations, the observation of calculation procedures (operational procedures) and conceptual knowledge (arithmetic rules and principles, e.g. interchangeability, groupability, inversion) is a prominent aspect. The main basic functions, the spatial-visual and central executive system, as well as working memory and linguistic aspects are also emphasized in the measurement procedure. The Dyscalculia Pedagogical Examination measures several ability domains, typically consisting of multiple subtests. Scoring is criterion-oriented, and points are given if errors do not occur. The structure of the DPV is reported in Table 1. Table 1A description of the subtests of the Dyscalculia Pedagogical Examination (DPV).Ability domainSubtestTypical mistakesExample tasks1. OrientationOrientation on one’s own body, in space, in a plane, in time, between numbersCreating relations in space, time and numbers is incorrect.The relational vocabulary/grammar in time and numbers is incorrect.Left-right differentiation on his/her own body/in a plane is incorrectTell me what time it is on the clock?What time of day is it?Write your name in the upper left corner.Mark the place of the numbers on the number line.2. Counting in words2.1. Counting in the number range < 10 000The knowledge of numerals is incorrect.Consistent skipping or swapping of numbers.The counting is correct, but the rhythm is stuttering, and the tempo is slow.Count one by one from 2003, in descending order, until I tell you to stop!2.2. Number memoryNumber memory is weak compared to what is standard at the candidate’s ageI will say numbers, listen and repeat them in the same order.4-7-32.3. Numeral-digit matching in a number range of 10 000Content error when writing/reading the full five-digit number.Writing numbers as they are dictated.2.4. Quantitative relations in a number range of 10 000; Numerical neighboursIncorrect interpretation (creation) of quantitative relations.Content identification of the relational sign is incorrect.The determination of number neighbours is incorrect.I will say numbers, decide which is bigger or smaller.2.5 Naming a positional notation in the 100 000 range / crossing the 100 000 rangeIncorrect naming of a place name.12 345There are 5 in the ones place, 4 in the tens place, 3 in the hundreds place, 2 in the thousands place and 1 in the tens of thousands place.3. Counting in writing3.1. Replacement, division, addition, subtraction in a number range of 10s, 20s; Arithmetic facts, rulesThe interpretation of the mathematical operation is incorrect.Transition to another operation/ sticking to the previous operation/disorientation while performing an operation.Calculate the problem out loud, I want to know how you think? 3 + 7=?10 − 7=?14 − 7=?3.2. Addition, subtraction in a number range of 100, 1000; Analogies; Approximate calculationThe content identification of the operation sign is incorrect.The interpretation/understanding of the operation is incorrect.The recognition and application of analogies is incorrectCalculate the problem out loud, I want to know how you think?1580 − 240=?If 8 + 7 is 15, then 18 + 7 is…?3.3. Multiplication, divisionThe interpretation of the operation is incorrect.The interpretation of the relationship between multiplication and division is undeveloped/incorrect.How much is 6 × 4, 36 ÷ 6=?3.4. Written operationsThe recording of the written operation is incorrect.Use of tools is needed.The interpretation of the concept of the remainder is incorrect. 9 9 6+ 9 7 84. Math word problemsComposedA leading question is required due to the unsuccessful highlighting of the point.A leading question is required due to incorrect selection of the operation.I will tell you a word problem, please, repeat it! Write it down and calculate it! 15 boys signed up for the camp, and the number of girls is two less than that of boys. How many girls signed up for the camp?5. Mathematical – logical rulesPerforming logical tasksThe creation of the rule is incorrect.Compliance with the rule is incorrect.Use of tools can be observed while performing the task.I will show you a series of numbers, find the rule and continue the series.6. Higher level mathematical concepts; Sequence of operations6.1. Quotient fractionsThe interpretation of the concept of a fractional number is incorrect.Creation of relations is incorrect.The operation with fractional numbers is incorrectColour 3/6 of the circle!Which is smaller: 3/2 or 3/6?Substracting two fractionsCalculate it and then explain how you thought!6.2. Negative numbersThe concrete interpretation of the concept of negative number is incorrect (debt, thermometer)Calculate the problem out loud, I want to know how you think?-13-(+)8=?7. Practical knowledge7.1. Units of measurementThe content (conceptual) identification of the units of measurement is incorrect.Do you think this room might be 2 km long?7.2. Perimeter/areaConcept of perimeter/area is incorrect.What is the perimeter and area of the rectangle if its sides are 3 m and 10 m? A description of the subtests of the Dyscalculia Pedagogical Examination (DPV). Creating relations in space, time and numbers is incorrect. The relational vocabulary/grammar in time and numbers is incorrect. Left-right differentiation on his/her own body/in a plane is incorrect Tell me what time it is on the clock? What time of day is it? Write your name in the upper left corner. Mark the place of the numbers on the number line. The knowledge of numerals is incorrect. Consistent skipping or swapping of numbers. The counting is correct, but the rhythm is stuttering, and the tempo is slow. I will say numbers, listen and repeat them in the same order. 4-7-3 Incorrect interpretation (creation) of quantitative relations. Content identification of the relational sign is incorrect. The determination of number neighbours is incorrect. 12 345 There are 5 in the ones place, 4 in the tens place, 3 in the hundreds place, 2 in the thousands place and 1 in the tens of thousands place. The interpretation of the mathematical operation is incorrect. Transition to another operation/ sticking to the previous operation/disorientation while performing an operation. Calculate the problem out loud, I want to know how you think? 3 + 7=? 10 − 7=? 14 − 7=? The content identification of the operation sign is incorrect. The interpretation/understanding of the operation is incorrect. The recognition and application of analogies is incorrect Calculate the problem out loud, I want to know how you think? 1580 − 240=? If 8 + 7 is 15, then 18 + 7 is…? The interpretation of the operation is incorrect. The interpretation of the relationship between multiplication and division is undeveloped/incorrect. The recording of the written operation is incorrect. Use of tools is needed. The interpretation of the concept of the remainder is incorrect. 9 9 6 + 9 7 8 A leading question is required due to the unsuccessful highlighting of the point. A leading question is required due to incorrect selection of the operation. The creation of the rule is incorrect. Compliance with the rule is incorrect. Use of tools can be observed while performing the task. The interpretation of the concept of a fractional number is incorrect. Creation of relations is incorrect. The operation with fractional numbers is incorrect Colour 3/6 of the circle! Which is smaller: 3/2 or 3/6? Substracting two fractions Calculate it and then explain how you thought! Calculate the problem out loud, I want to know how you think? -13-(+)8=? DPV is available is several versions depending on the the age of the students to be tested. The number and complexity and tasks increases with higher age. The version described above and used in the study was designed for sixth-grade students (~ 12 year olds). In the measuring procedure, the standard questions were designed to be both objective and to allow a dynamic evaluation, enabling the examiner to also gather information about the child’s thinking processes, metacognitive skills, and non-intellectual factors accompanying task solving (e.g. motivation, frustration tolerance)56. There is limited overlap between the content of the DPV and the WISC-IV. The DPV task “number memory” is very similar to the forward digit span task in the WISC-IV as both involve the memorization and repetition of numbers in a certain order, but there is no other clear similarity between the content of the two tests. The DPV showed excellent reliability/internal consistency (McDonald’s ω = 0.909, Cronbach’s α = 0.840). Mathematics anxiety We used the Hungarian version56of the Mathematics Anxiety Rating Scale (MARS57), . This scale contains 40 self-report items about mathematical anxiety, each of which is scored on a seven-point Likert-type scale (1 = not at all typical; 7 = completely typical). Example items are “When I have to solve math problems, I am very nervous”, “My stomach convulses when I think about mathematics”, and “Mathematics is the most disgusting subject for me out of all of them”. We used the total score on MARS as an indicator of mathematical anxiety. MARS showed excellent reliability/internal consistency (McDonald’s ω = 0.976, Cronbach’s α = 0.975). Statistical analyses Bivariate correlations and multivariate regressions were calculated in MATLAB 2021a and with R. Structural equation models were fitted using the lavaan R package. Our code and raw data are available at https://zenodo.org/record/7935031. A detailed description of analyses, including correlation matrices used to fit models, is available at https://rpubs.com/JLLJ/DPVDysc. Results Bivariate correlations All indicators (IQ, DPV scores, MARS scores, and family SES) were correlated with mathematics grades in the expected directions (Table 2). Notably, mathematics grades exhibited the strongest correlation with WISC IQ (r = 0.72), and DPV total score (0.50). Even though the DPV was designed to measure specific cognitive deficits, it had a high correlation with WISC IQ (0.62). Parental education and SES were associated with better grades, but not as strongly as IQ. Finally, MARS scores were associated not only with worse mathematics grades, but also with lower IQ. Table 2Correlation matrix of all variables, with means and standard deviations. For the sake of simplicity, we marked correlations which did not pass the conventional significance threshold of p < 0.05 with †.MeanSDGradesIQDPVMARSMaternal educationPaternal educationHousehold sizeOwn roomOwn phoneGrades3,780.940IQ108,0013,4840.721DPV252,1827,5690.4970.624MARS117,3758,025-0.532-0.501-0.645Maternal education5,790.9160.5350.7370.433-0.419Paternal education5,761.0160.5470.7710.498-0.4270.745Household size3.870.951-0.031†-0.078†0.006†0.008†-0.042†0.081†Own room0.790.4060.090†0.2750.233-0.2060.2580.145-0.530Own phone0.860.3460.152†0.2980.281-0.2330.191†0.189†-0.2640.431Composite SES010.5560.7820.447-0.4360.9120.9360.019†0.2030.168Each variable had n = 102 observations. Correlation matrix of all variables, with means and standard deviations. For the sake of simplicity, we marked correlations which did not pass the conventional significance threshold of p < 0.05 with †. Each variable had n = 102 observations. Latent correlations between the WISC-IV and DPV Next, we fitted a structural equation model in lavaan to estimate the correlations between the common variances of the two cognitive tests. For the WISC-IV, we followed the structure of latent variables described by the test manual, with the ten subtests being indicators for four group factors, which in turn loaded on a general intelligence factor. As no psychometric model was available for the DPV, we conducted a parallel analysis58. This is one of the most common methods for determining the number of factors to extract in a factor analysis. The steps of it are to generate random data, factor analyze the random and real data, and to compare the eigenvalues of randomly generated data with the real data. If the eigenvalues for the real data are not substantially greater than the eigenvalues for the factors from the random data, they are not retained. This procedure directed us towards extracting just one factor that explained 54% of the variance in the DPV and no other reliable factors could be identified.” We fitted this model using diagonally weighted least squares as DPV scales were either ordinal or binary. A graphical depiction of our model can be found in Fig. 1. A psychometric model of the WISC-IV and the DPV dyscalculia screening test. DPV subtest appear left to right in the same order they are enumerated in Table 1. For WISC-IV, wm: Working memory, vc: Verbal Comprehension, PS: Processing Speed, PR: Perceptual Reasoning, bt: Letter-number Sequencing, szmt: Digit span, szk: Vocabulary, kz: Similarities, al: Comprehension, kd: Coding, szmb: Symbol Search, kp: Picture Concepts, mz: Block Design, mt: Matrix Reasoning. This model fit the data extraordinarily well (TLI = 1.066, CFI = 1.00, RMSEA = 0.000), presumably due to the modesty of our statistical power. The latent correlation between WISC and DPV was very high at 0.777 (SE: 0.032). Multivariate analyses We fit a series of HC5 robust regressions to estimate the unique and joint association of IQ, DPV scores, mathematics anxiety and family SES with mathematics grades. We controlled for sex and age in all models, although we note that these were not significantly associated with grades in any of the models (R2 = 0.2%). Results are depicted in Table 3. Table 3Associations between mathematics grades, intelligence (WISC-IV IQ), dyscalculia risk (DPV total score), family socioeconomic status (SES), and mathematics anxiety (MARS total score). We reported unstandardized OLS regression coefficients, which are to be interpreted as expected change in mathematics grades (on a scale of 1 to 5) per 1-point change (in case of SES, 1 standard deviation) in the predictor. Standard errors are shown in parentheses.Model 1Model 2Model 3Model 4Model 5Model 6Model 7Model 8WISC-IV IQ0.05 (0.004)***0.047 (0.005)***0.05 (0.007)***0.04 (0.005)***0.047 (0.008)*** DPV score 0.017 (0.004)*** 0.003 (0.002)-0.002 (0.002) SES 0.18 (0.02)*** -0.01 (0.04)-0.02 (0.04) MARS score -0.008 (0.001)*** -0.004 (0.001)** -0.004 (0.001)** R 2 52.03% 24.70% 30.95% 28.76% 52.39% 52.04% 56.00% 56.31% Associations between mathematics grades, intelligence (WISC-IV IQ), dyscalculia risk (DPV total score), family socioeconomic status (SES), and mathematics anxiety (MARS total score). We reported unstandardized OLS regression coefficients, which are to be interpreted as expected change in mathematics grades (on a scale of 1 to 5) per 1-point change (in case of SES, 1 standard deviation) in the predictor. Standard errors are shown in parentheses. All four predictors were significantly associated with mathematics grades when considered alone. Variance accounted for ranged from 24.70% (DPV) to 52.03% (IQ). However, when all the variables were entered in a single model, incremental variance accounted for only modestly improved over the IQ-only model (to 56.31%), and only two predictors of school grades remained significant: IQ and mathematics anxiety. Notably, regression coefficients for IQ barely changed when other predictors were also included in the model. The largest drop was observed at co-inclusion with MARS (20%). In contrast, MARS effects, while significant even over the effects of IQ, were reduced by 50–60% when estimated net of intelligence. The effects of specific mathematics ability and family SES were reduced to insignificance (and in the latter case, they changed their sign) once jointly included in the model with IQ. In other words, IQ is expected to predict school grades among students with identical measured SES, DPV scores and anxiety almost as strongly as in the total population. Conversely, among students with identical IQ scores we would expect no correlation between mathematic grades, DPV scores and family SES and a strongly reduced association with mathematics anxiety. At a reviewer’s suggestion, we re-conceptualized SES as parental education, given the simplicity of this definition and the high correlation with the original composite SES metrics. Results from these alternative models (Model 3a-c and Model 6a-c) are shown in Supplementary table S1. Both higher maternal and paternal education were associated with better mathematics grades, accounting for somewhat less variance than composite SES. However, when entered in the same model as IQ, both lost their significance, indicating that parental education is only indicative of school performance as a proxy of student intelligence. Discussion Our study fits into a voluminous literature14,20,21,59,60in demonstrating the primacy of general intelligence in predicting school outcomes. We found that with both correlations and regressions intelligence emerged as the single strongest predictor of school grades. The bivariate association with intelligence and school grades was stronger than the associations between school grades and SES, mathematics anxiety, or specific mathematics ability. IQ scores alone accounted for over half of the variance in mathematics grades, replicating previous results13. In fact, the association between IQ and school grades was only modestly changed when we adjusted for specific mathematics ability, socioeconomic status, or mathematics anxiety (Table 3). Socioeconomic status is frequently claimed to causally contribute to students’ performance in school21,61. One mechanism for this might be through educated parents emphasizing the role of learning in the household or through better-off parents being able to afford additional training for their children. This interpretation of the correlation between family SES and school grades is challenged, however, by behavior genetic studies showing that school performance is substantially heritable with a comparably smaller role for the common environment22–25, and studies which show that net of intelligence, the relationship between family characteristics and school performance is small to nonexistent20,21. Our results support this more critical, non-causal interpretation of the family SES-school grades correlation. While socioeconomic status was also reasonably strongly correlated with mathematics grades in our sample, in multivariate analyses we found that net of intelligence these associations were trivialized. In other words, children from higher-SES families only do better in school to the extent their IQs are higher: in a sample of children with identical intelligence, based on our findings we would not expect those from better educated families to do better in mathematics. Theoretically it would be possible that family SES causally influences IQ scores in a way that is not captured at all by parental education and contributes to school grades through its effect on intelligence. A substantial causal influence of family SES on intelligence is unlikely, however, as even at this age pedigree studies show minimal shared environmental effects on intelligence62,63( see also60 for a broader overview of the causal effect of SES on intelligence.) Only a few previous studies46,48,64–66studied the psychometric properties of tests designed specifically to measure mathematical abilities or deficits thereof, and ours is the only study which examines the psychometric properties of the Pedagogical Examination of Dyscalculia, a tool widely used by testing services in Hungary. (See also some other studies44,48 for comparisons of manifest scores on dyscalculia screening tests and intelligence tests). Confirmatory factor analysis confirmed that a single-factor model in which all components of this test loaded on a single factor (‘dyscalculia risk/specific mathematics ability’) fit reasonably well. However, further analyses highlighted issues with the interpretability of the test. First, the DPV exhibits high manifest (r = 0.624) and especially high latent correlations (r = 0.777) with the WISC-IV. This suggests that although the DPV is intended to be labeled as a test of domain-specific abilities, in fact what it measures is very similar to a test of domain-general cognitive abilities (intelligence). Second, net of IQ, DPV did not predict mathematics grades in school, further calling into question its ability to reflect domain-specific deficits in mathematics ability, at least in the general population without diagnosed dyscalculia. The failure of the DPV to exhibit a large degree of discriminant validity from the g factor of the WISC-IV can be considered emblematic of the jangle fallacy6767. It is also predicted by Spearman’s hypothesis of the indifference of the indicator68. As with many previous tests designed to measure domain-specific abilities or deficits6,9,69–72, the DPV to a large degree indexes general intelligence, and its residual variance seems to have poor criterion validity for what it was designed to assess. This issue is very important for studies purporting to measure something specific such as spatial or mathematical ability, as tools like our dyscalculia screening tool, spatial tests, examinations of specific aspects of achievement like writing examinations or vocabulary tests, etc. clearly indicate the g factor. Models based on them cannot be assumed to measure something besides g unless g variance is explicitly parceled out through a model in which g accounts for variance in that assessment instrument. Because g is general, papers failing to model it as a driver of performance across assessments, for example, by regressing out an IQ score or a factor score computed from a separate battery of tests, are incomplete until the g factor is computed based on all available applicable assessments. Our results highlight the importance of defining dyscalculia specifically as a discrepancy between general and specific mathematical abilities, and to test mathematical abilities as part of a complex cognitive test battery. Despite high content validity, low scores on a mathematical ability test do not necessarily indicate a deficit of mathematical abilities: low general intelligence is an equally possible cause. Our findings specifically call into question the utility of the DPV as a stand-alone dyscalculia screening tool, as scores on this test added no explanatory power to an intelligence test towards school performance. Education specialists should be cognizant of the high correlation of performance in various cognitive domains and specifically test if deficits are domain-general or domain-specific in nature. Our other findings confirm the role of mathematics anxiety as an independent contributor to school outcomes. Scores on the MARS exhibited a reasonably strong correlation with IQ, contradicting previous claims2 that mathematics anxiety is derived from a domain-specific as opposed to a general perceived deficit in cognitive performance (Latent correlations rg-MARS=0.594, rdpv-MARS=-0.719, see also contradicting results from28,31). While we are unable to causally account for this association, it is possible that students with lower general cognitive ability correctly perceive that studying mathematics will be an especially challenging task for them35,36. However, only about a quarter of mathematics anxiety variance is accounted for by IQ, and higher mathematics anxiety is related to lower grades even net of IQ. Our findings are in line with at least one previous study31which also found that intelligence only partially mediates the relationship between mathematics anxiety and mathematics performance. Our study cannot conclusively answer if mathematics anxiety is domain-specific or a reflection of a broader underlying trait of anxiety-proneness or neuroticism2. However, in either case, our results identify anxiety as a potentially important factor in low grades. Interventions designed to lower student anxiety – for example, by directly targeting anxiety through psychological interventions73–75, or by making mathematics education less daunting and more accessible through a focus on practical problems and explanation methods designed to fit the students’ preferences – are promising targets for improving mathematics grades. Our work suffers from some limitations. One is that we had no direct measure of household income, rendering our estimate of family SES somewhat incomplete. However, previous analyses21have concluded that parental education, which we could measure, is a better predictor of children’s outcomes than income. Thus, we believe that our SES estimate – especially as it was complemented by three indirect measures of income and wealth, each relatively weakly related to school performance – reasonably captured the most important, putatively causal components of family socioeconomic position. A further limitation is the limited size of our sample. Marks’21recent study did find a small but significant correlation between family characteristics and school performance in the NSLY79 and NLSY97 even net of IQ, but we may have had insufficient power to detect a similar, small association. Thus, while our findings clearly showed that most of the relationship between family SES and school grades was mediated by IQ, they do not show beyond doubt that IQ-independent effects of SES are absent. Importantly, close to half of school grade variance was not accounted for by intelligence, showing that unmeasured student characteristics likely including motivation or teacher quality also likely play an important role in determining school performance. Our sample had above average intelligence (mean IQ = 108), which may affect the generalizability of findings. An issue related to above-average ability in our sample was that DPV exhibited a ceiling effect, with a high proportion of students achieving close to maximal scores. Finally, we only assessed a single – albeit important – indicator of mathematics performance, school grades at a specific grade level. Previous meta-analytic evidence suggests that the relative strength of the predictors of math performance vary across grade levels and test type76, which may affect the generalizability of our findings. In sum, our study demonstrates that the most important predictor of mathematics grades in this sample of Hungarian schoolchildren is simply psychometric intelligence. Despite strong causal claims in the scientific and popular literature, we found no evidence that family SES predicts grades beyond intelligence. We found that scores on a screening test for dyscalculia were highly correlated with IQ, and that these scores does not account for additional variance in mathematics grades. Mathematics anxiety, however, was an independent predictor of school grades beyond intelligence, and it is thus a more promising target for psychological interventions than attempts to improve mathematics performance in school. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Acknowledgements This publication has been supported by the National Research, Development and Innovation Fund of the Ministry of Innovation and Technology (grant ID: OTKA PD 138935). This work was supported by the Research Programme for Public Education Development of the Hungarian Academy of Sciences. National Laboratory for Social Innovation project (RRF-2.3.1-21-2022-00013), within the framework of Hungary’s Recovery and Resilience Plan, with the support of the Recovery and Resilience Facility of the European Union. Author contributions CRediT author statementBernadett Svraka: Conceptualization, Funding Acquisition, Investigation, Resources, Writing - Original Draft, Writing - Review & Editing; Jordan Lasker: Methodology, Formal Analysis, Writing - Original Draft, Writing - Review & Editing, Visualization; Péter P. Ujma: Conceptualization, Methodology, Formal Analysis, Writing - Original Draft, Writing - Review & EditingAll authors reviewed the manuscript. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3