Project topics and materials
Get Complete Project Material File(s) Now! »
What shapes early cognitive development2
Cognitive development is the result of an intricate combination of genetic and environmental factors. Decades of research in psychology and epidemiology have highlighted the extent to which these factors predict variance in cognitive abilities and traits (1.1.1), and made considerable advances in uncovering the complexity of their relationships, from disentangling their unique contributions (1.1.2) to mapping out their convoluted web of interactions (1.1.3).
Main predictors of cognitive development
Cognition is a vast array of abilities and traits, including a variety of domains – such as language, motor skills, reasoning, working memory or attention, but also social, emotional and behavioral skills. Providing a detailed picture of the multitude of factors that influence these various areas is no easy task. We do not intend here to be exhaustive, but rather to report the main factors which have stood out in the literature, and for which solid evidence – from meta-analyses when possible, and large cohort studies otherwise – has been provided. In doing so, we chose not to restrict our scope to some specific domains of cognition in order to illustrate the diversity of influences that predictors can have on different areas of cognitive development.
Sex
Sex differences in cognitive development have been the focus of a wide range of studies in psychology. Although male and female children are largely similar, they show some differences in the developmental trajectories of certain cognitive functions. For instance, while there is no sex difference in general intelligence (Deary et al., 2007), robust small differences have been found in specific cognitive abilities. Hence, meta-analytic evidence shows that girls have better verbal skills than boys (Hyde & Linn, 1988), and boys perform better in mental rotation tasks (Maeda & Yoon, 2013; Voyer et al., 1995). Besides, boys tend to be more at risks of having neurodevelopmental disorders (May et al., 2019) such as ASD (male-to-female ratio equal to 3:1, Loomes et al., 2017) and ADHD (3:1, Willcutt, 2012).
Prenatal exposure
Exposure to certain elements during pregnancy may have detrimental effects on the child’s cognitive development. For instance, children of epileptic mothers who have been exposed in utero to valproic acid, an antiepileptic drug, have on average lower general cognitive abilities (Banach et al., 2010) and are at higher risks of developing ASD (Christensen et al., 2013) and various neurodevelopmental problems (Blotière et al., 2020). Prenatal alcohol exposure also has negative consequences for cognitive development: meta-analytic studies suggest that moderate alcohol intake during pregnancy (3-6 drinks per week) is negatively associated with child behavior outcomes (d=-0.15; Flak et al., 2014), while binge drinking (more than 4 drinks per occasion) and heavy drinking (more than 2 drinks per day) are negatively associated with general cognitive development (respectively d=-0.13 for binge drinking, Flak et al., 2014; and d=-0.53 for heavy drinking, Testa, 2003).
Birth factors
Cognitive development is also associated with several birth characteristics, such as gestational age, birth weight and the Apgar score (which evaluates a newborn’s clinical status). Thus, preterm children and those with low birth weights experience a variety of cognitive deficiencies including linguistic, intelligence, sensory, and motor difficulties, compared to term children with normal birth weights (Aarnoudse-Moens et al., 2009; Barre et al., 2011; Beauregard et al., 2018; Courchia et al., 2019; de Kieviet et al., 2009; Nguyen et al., 2018; Twilhaar et al., 2018). Cognitive declines range from 0.2 to 0.3 SD for preterm (<37 weeks) and early term children (37–38 weeks; Beauregard et al., 2018) and up to 0.86 SD for very preterm children (<32 weeks) and/or with very low birth weight (<1500 g) (Twilhaar et al., 2018). Lastly, a low Apgar score (<7) is additionally associated with lower cognitive abilities (Ehrenstein, 2009; Razaz et al., 2016) and teacher-rated hyperactivity and inattention (Guhn et al., 2020; Razaz et al., 2016).
Parental and social factors
Parental and social factors after birth also explain individual differences in children’s cognitive abilities and traits. For instance, breastfeeding is associated with higher offspring’s general cognitive abilities (Horta et al., 2015: 3.44 more IQ points) and with lower risks of developing certain behavioral problems such as ADHD symptoms (Tseng et al., 2019: Odds ratio for non-breastfeeding = 3.71). Other parental characteristics such as paternal and maternal age, socio-economic status, and maternal depression, have also been found to be associated with children’s cognitive outcomes in large scale cohort studies. Younger maternal age is negatively linked with general cognitive abilities (less than 25 years old: β = -0.13 to -0.17), while the reverse is true for older age ranges (35-39 years old: β = 0.10) (Goisis et al., 2017). However, advanced maternal age is positively associated with an increased likelihood of internalizing problems (OR=1.06), but negatively linked with externalizing problems (OR=0.88) (Saha et al., 2009). On the contrary, paternal age is likely to increase the risks of developing externalizing problems (OR=1.12) but not internalizing problems (Saha et al., 2009). Both advanced maternal and paternal age (older than 40) are associated with augmented risk of developing ASD (Reichenberg et al., 2006; Sandin et al., 2017). Besides, parents’ socio-economic status is positively associated with a wide range of cognitive outcomes – general cognitive ability, executive functions, behavioral outcomes, language development –, as supported by meta-analytic evidence (Lawson et al., 2018; Letourneau et al., 2013; Scaff & Cristia, in prep.). Lastly, education is an obvious contributor to children’s cognitive abilities, with an increase of, on average, 3.4 IQ points for 1 year of education (Ritchie & Tucker-Drob, 2018).
Genes
A survey of the factors influencing cognitive development would not be complete without a mention of genetic factors. The collective effect of genes on cognition has been investigated in heritability studies, which determine the share of variance in phenotypes that is due to genetic variance. Historically, such studies have relied on the comparison between mono- and di-zygotic twins (Bartels et al., 2002; Bishop et al., 1995), but have also exploited other situations such as adoption at birth and more generally trait correlations between relatives of varying genetic and environmental similarity (Plomin et al., 1997). Across all cognitive traits, heritability has typically been found to lie between 20 and 80% (Plomin et al., 1994), making the genome the single most important factor in cognitive development (although each individual genetic variant only has a minute effect on cognition).
Since the beginning of the 21st century, new molecular genetic methods have complemented twin and family studies. Genome-wide complex trait association (GCTA) studies use whole-genome analysis to estimate the proportion of phenotypic variance that can be explained by genetic variance, directly measured across dozens or hundreds of thousands of single-nucleotide polymorphisms (SNPs) (Yang et al., 2011). For reasons that are well understood, they show systematically lower heritability estimates than twin studies (Trzaskowski et al., 2013), but do confirm the substantial influence of genes on most cognitive traits. Such genome-wide association results are now being used to compute polygenic scores, which cumulate the predictive power of thousands of SNPs that are most strongly associated with the phenotype of interest. Current polygenic scores may account for up to 10% of the variance in cognitive performance (Lee et al., 2018).
Beyond documenting the contribution of genes to cognitive development, perhaps the most interesting contribution of such genetic studies is to enrich studies of environmental factors by allowing one to consider interactions between genetic and environmental factors, and providing a way to adjust for the confounding effects of genetic factors on environmental ones.
The importance of controlling for confounding variables
As one can imagine, many of the various predictors of cognitive development are correlated with each other. It is therefore often necessary to measure as many factors as possible, and adjust them on one another to identify the specific contribution of each one (see Figure 1). For instance, parental education is correlated with family income, with the quality of medical care, and with parent/child interactions. Failing to measure and control for any of these factors may lead to overstating the influence, or misattributing a causal role, to the others. We develop and illustrate two main types of such confounding: confounding due to omitted environmental variables, and to genetic factors (i.e. gene-environment correlations).
Figure 1: Confounding variables. Omitting to account for a confounder (a variable correlated with both the predictor of interest and the cognitive outcome) results in a biased estimate of the association between this predictor and the outcome.
Environmental confounders
Environmental factors are the vast array of conditions to which a child can be exposed during or after pregnancy. It is often the case that environmental predictors of cognitive development are correlated with one another. In order to identify their unique contributions and compare their relative influence, it is therefore important to control for the influences of potential confounders. One example of such confounded relationship is that between breastfeeding and maternal IQ. Breastfeeding has been purported to have a positive influence on cognitive development due to the particular composition of maternal milk. However, when controlling for maternal IQ, the association between breastfeeding and the child’s intelligence falls from 3.44 IQ points to 2.62 (Horta et al., 2015). Similarly, after matching breastfed children with non-breastfed children on a range of individual and parental characteristics, the difference in general cognitive outcomes considerably shrinks and becomes non-significant (Girard et al., 2018). These results suggest that a large part of the association between breastfeeding and the child’s cognitive development may stem from richer mother-child interactions, rather than nutritional benefits. In a similar fashion, maternal smoking during pregnancy has long been believed to be associated with decreased cognitive outcomes. However, large scale studies which controlled for a wide range of factors thought to be correlated both with maternal smoking and cognitive outcomes, such as maternal education, found no evidence for such association (Batty et al., 2007; Gilman et al., 2008). Therefore, maternal smoking in itself does not seem to be detrimental to the infant’s cognitive development.
Gene-environment correlations
While it is commonplace in social science and epidemiological research to measure and control as many potentially confounding factors as possible, this approach is often restricted to environmental factors. Yet, genetic factors are also often intertwined with environmental factors. This has been known for a long time, with the paradoxical discovery of the heritability of environmental factors, also known as “the nature of nurture” (Plomin
& Bergeman, 1991). For instance, the very exposure to life events (accidents and trauma), an unambiguous environmental factor, is more concordant between monozygotic than between dizygotic twins, hence has a non-null heritability. This can be understood as reflecting genetic influences on cognitive traits such as risk-taking or impulse control, or less directly, genetic influences on intelligence which in turn has an effect on the likelihood of understanding and following basic safety recommendations. Gene-environment correlations can take different forms (Pingault et al., 2018; Rutter, 2007). They can be passive, such as when parents with language skills both genetically transmit these predispositions to their children, and provide a richer linguistic environment for these children to grow up in. They can be evocative, such as when children with good language learning predispositions talk more and better, and therefore elicit richer language input in return. They can also be active, such as when children with good language learning predispositions actively seek peers with good verbal skills, books, and challenging linguistic environments. In all cases, studies may measure the association between the linguistic environment and children’s language abilities, and make incorrect (or inflated) causal inferences if they don’t control for genetic transmission.
Nowadays, molecular genetics offers a way to directly measure and control genetic influences. It has indeed been shown that certain polygenic scores are significantly correlated with some environmental factors known to have an effect on cognitive development. For instance, a child’s genome-wide polygenic score (GPS) for educational attainment is correlated with parental education, income, and age at the child’s birth, with number of books in the home, with breastfeeding duration, with smoking during pregnancy, with whether the TV is usually on, with smacking or slapping (Krapohl et al., 2017). Although parental SES is one of the main predictors of educational achievement, this relationship may be to a large extent accounted for by genetic variance (Trzaskowski et al., 2014).
From simple associations to complex relationships
More often than not, it is plausible to imagine that predictors and cognitive outcomes may be related in more complex ways than portrayed so far. What is more, cognitive skills are also related to each other due to the dynamic nature of cognitive development – it is the idea that “skills beget skills” (Cunha & Heckman, 2007). More sophisticated statistical models can be used to understand the particular mechanisms through which a predictor ultimately affects cognitive outcomes: is the effect mediated through a third factor? Does the effect depend on particular circumstances? To what extent do two factors exert reciprocal influences on each other?
Mediation effects
In order to have a more complete picture of the effect of one factor on cognitive development, one can look at the potential mediators of such relationship. Mediation effects designate a relationship (thought as causal) between one distal factor, such as parental education, one proximal factor, such as breastfeeding, and one outcome, such as verbal cognitive ability (see Figure 2): the positive association between parental education and a child’s verbal skills is partly explained by the fact that higher educated mothers breastfeed more, which is itself associated with higher verbal outcomes (Peyre, et al., 2016). Statistical models allow to estimate to what extent the effect of the distal factor is mediated through the proximal factor. Different methods can be used, depending on the nature of the variables. When the relationships between variables is linear, that the variables are normally distributed and that there are no interactions, structural equation models (SEM) are one efficient way to estimate mediation effects, even when multiple mediators are present (Preacher & Hayes, 2008). When such assumptions are not reasonable, SEMs can be used in an exploratory fashion to generate hypotheses, but will often need to be followed by more rigorous analysis strategies. Causal mediation analysis, a method based on counterfactual reasoning, provides a more rigorous framework for estimating such relationships (VanderWeele, 2016).
Figure 2: Mediation relationships. The underlying mechanism through which a predictor ultimately influences cognitive outcomes may be captured by a mediation relationship, whereby the predictor’s influence is partly (or fully) explained by its effect on a third, mediator variable which is also associated with cognitive outcomes.
Moderation/interaction effects
Given exposure to similar environmental factors, distinct individuals may react differently. This may be due to different developmental history, sex, or genes conferring different vulnerability or potential. This phenomenon is known as moderation of the effect of one factor by another, or interaction effect between the two predictors: i.e., when the effect of one factor depends on the presence or the value of another (see Figure 3). For instance, the negative influence of prenatal alcohol exposure on the child’s executive functions is greater when the mother is older (Burden et al., 2005; Chiodo et al., 2010).
Beyond environmental factors interacting with each other, the child’s sex seems to moderate the effects of certain environmental factors on cognitive development. For instance, low birth weight is a long-term risk factor for depression in adolescent girls, but not in boys, and only in conjunction with other childhood risk factors (Costello et al., 2007). As another example, the well-known male advantage in spatial skills has been found to emerge only at middle/high SES, but not at low SES, thus constraining the possible explanations for this sex difference (Levine et al., 2005).
Genetic makeup has also been shown to interact with environmental factors. Understanding such interactions may throw new light on well-established environmental effects. For instance, it has long been known that childhood maltreatment is associated with conduct disorder and with later antisocial personality behavior. This may be interpreted as reflecting a form of learning by imitation. However, not all maltreated children become maltreating parents. In a landmark study, Caspi et al. (2002) showed that a particular polymorphism of monoamine oxydase a (MAO-A) interacted with childhood maltreatment, such that carriers of the low protein expression variant were more at risk of developing conduct disorder if they were maltreated (but not if they were not). This result, strengthened by meta-analytic evidence (Byrd & Manuck, 2014), suggests that the learning-by-imitation interpretation is at best incomplete. Thus, given the variations in the response to environmental factors, it is important to consider genetic factors as one possible source of this variability.
Figure 3: Moderation/interaction effects. The effects of two predictors on cognitive outcomes can be multiplicative, when the effect of a predictor A varies with the values of a second predictor B: the influence of A is moderated by B. The bottom panel illustrates a moderation/interaction relationship where the positive relationship between a continuous predictor A and the outcome is more or less strong depending on the value of a categorical predictor B.
Reciprocal relationships
When two variables which evolve in time are correlated, it is often hard to determine which one causes the other, or, if each has a causal effect on the other, which one has the larger effect (see Figure 4). For example, language abilities and behavioral problems are two cognitive outcomes which are correlated, but for which the direction of the relationship is not obvious: it is possible that early behavioral problems impair language development, but also that early language difficulties prevent children from properly regulating their behavior. It is also possible that there is no causal link between the two outcomes, but that both are caused by a third, potentially unobserved, factor, which creates a correlation between them. Cross-lagged panel models are a kind of structural equation models which can help disentangling such longitudinal relationships. In these models, the two variables are measured at different points in time and are simultaneously regressed on past values of themselves and on past values of the other one. When measures are available at more than two time points, more sophisticated models can be used, which allow distinguishing between-person from within-person variance; for example, models including a random intercept (Hamaker et al., 2015). These models are able to estimate to what extent a variable A affects the within-person change in variable B, and vice-versa. Cross-lagged panel models examining the relationships between language abilities and ADHD symptoms have thus shown that better early language skills prevent the development of ADHD symptoms, but that early ADHD symptoms do not impair language acquisition (Petersen et al., 2013; Peyre et al., 2016).
Similar methods can be applied when exposure to a risk factor varies with time, and its relationships with cognitive outcomes are thus unclear. For example, exposure to screens is correlated with children’s cognitive abilities (Madigan et al., 2020; Walsh et al., 2018). However, we do not know a priori if this correlation conceals a causal relationship from screen time to cognitive abilities (for example, if watching TV or playing a video game deters children from doing activities more beneficial to cognitive development), a causal relationship from cognitive abilities to screen time (for example, if children with lower cognitive abilities are more attracted to screens), or is simply due to external factors (for example, if children from lower socioeconomic backgrounds both have lower cognitive abilities and are more exposed to screens). Going beyond simple associations by using a random intercept cross-lagged panel models showed that there is a small negative link from screen time to general cognitive development, but not the reverse (Madigan et al., 2019).
Figure 4: Bidirectional relationships. Disentangling the links between two correlated cognitive outcomes or risk factors which evolve in time (from variable A at T1 to variable B at T2, and vice-versa) requires the use of longitudinal data and more complex statistical models such as the random-intercept cross-lagged panel model.
Cognitive development is thus a dynamic and complex process, shaped by the influences and synergies of a large array of environmental, biological and genetic factors. The set of abilities thus formed constitutes the foundations of academic skills learning.
Acquiring academic skills: building blocks and determinants
In contrast to other cognitive abilities such as language or reasoning, literacy and numeracy need to be explicitly taught to children. However, the success of this instruction hinges on the child’s initial cognitive skills, which are the building blocks of reading and mathematics (1.2.1), as well as on her socio-emotional skills, which may affect her approach to learning and behavior in the classroom (1.2.2). In addition, other socio-demographic factors come into play, likely exerting an influence on academic skills learning on top of their early influence on early cognitive development (1.2.3). We now delve into how these three aspects govern the acquisition of academic skills.
The role of cognitive abilities
Both reading and mathematics are cumulative processes that build on some key cognitive abilities. We first present the links between general cognitive development (intelligence) and academic skills, before focusing on specific cognitive domains (language, visuospatial and motor development, and executive functions).
General cognitive ability, or intelligence
Intelligence tests were initially designed with the explicit purpose of predicting children’s future educational success (Binet & Simon, 1904). Similarly, it is from the observation that multiple school examination scores were all positively correlated that Charles Spearman extracted the first measure of general intelligence (the ‘g’ factor) (Spearman, 1904). Therefore, it should come as no surprise that IQ is one of the best predictors of academic achievement – if not the best, depending on the outcome measure used. Thus, the correlation between intelligence test score and academic skills lies between 0.5 and 0.8 (Deary et al., 2007; Rohde & Thompson, 2007; Roth et al., 2015). Nowadays, several standardized tests have been developed by psychologists to measure human intelligence; the most widely used for children being the Wechsler Preschool and Primary Scale of Intelligence (WPPSI) for those aged 3 to 7, and the Wechsler Intelligence Scale for Children (WISC) for those aged 6 to 16. These tests measure the principal cognitive functions of an individual: processing speed, working memory, verbal comprehension, fluid reasoning, and visual spatial skills, summed up in a total score: IQ (standardized with a mean of 100 and standard deviation of 15). The full-scale IQ score and the five subcomponents are thought to correspond, respectively, to the g factor and five broad abilities in the Cattell-Horn-Carroll (CHC) intelligence theory, in which human intelligence is modelled as a hierarchical structure with the g factor at the top stratum, hypothesized to be at the core of all broad abilities in the stratum beneath (Schneider & McGrew, 2012). General intellectual ability as measured by IQ tests can also be broken down into two more comprehensive components: verbal intelligence, and non-verbal intelligence, or crystallized (gc) and fluid intelligence (gf) (Cattell, 1963; Horn & Cattell, 1966). While both gc and gf are well correlated with academic performance, crystallized intelligence seems to have a higher predictive power than fluid intelligence – which makes sense, since crystallized intelligence encompasses acquired knowledge, reflecting prior learning (correlation of 0.36 to 0.65 for crystallized intelligence versus 0.26 to 0.40 for fluid intelligence; Postlethwaite, 2011).
We now look beyond these general standardized measures of cognition to understand how various components of cognitive ability support the acquisition of academic skills.
Language abilities
Language abilities are the backbone of learning to read, as one can easily imagine; but they are also essential in learning mathematics. Several aspects of language are crucial in the acquisition of literacy. The first one is phonological processing, which is the ability to perceive, store, access and manipulate speech sounds. A particularly useful component of phonological processing is phonological awareness (being aware of and manipulating speech sounds), which enables children to map graphic symbols to the sounds of spoken words (at a sublexical level), and hence plays an important role in decoding and spelling. Phonological awareness is the best predictor of word recognition (Melby-Lervåg et al., 2012) and a good predictor of spelling (Landerl & Wimmer, 2008; Lervåg & Hulme, 2010). The second aspect of early language ability that is crucial in literacy acquisition is language comprehension. This includes vocabulary (mapping phonological representations onto semantic representations), which is essential for reading comprehension (Hjetland et al., 2020; Ouellette, 2006). Beyond vocabulary, grammar (the implicit knowledge of syntax and morphology) (Durand et al., 2013; Hjetland, 2018; Hjetland et al., 2020; Lehrl et al., 2020; Muter et al., 2004; NICHD Early Child Care Research Network, 2005; Su et al., 2017) and conceptual knowledge (the understanding of concepts and classifications) (Hjetland et al., 2020; National Early Literacy Panel, 2008; Storch & Whitehurst, 2002) play an important role in reading comprehension.
In parallel, language abilities also play multiple roles in the development of numeracy skills. Indeed, children need to associate the rote-learnt number words with the quantities they represent (Geary, 2013). Besides, simple arithmetic facts such as multiplications seem to be stored and retrieved from long-term verbal memory (Dehaene & Cohen, 1995). Lastly, in order to solve an arithmetic problem presented in sentences, children need to use their vocabulary and language comprehension abilities to understand the problem and translate it into an equation (Fuchs et al., 2010), and often keep the elements of the problem in verbal working memory. Thus, language skills have been found to predict arithmetic abilities as well (Durand et al., 2005; Fuchs et al., 2010; Träff et al., 2018; Zhang et al., 2017).
Visuospatial abilities
Visuospatial abilities are important in the acquisition of both reading and mathematics. On one side, visuospatial skills are necessary to identify letters and segment written words into graphemes (letter or combination of letters transcribing phonemes). Few studies have examined the role of visuospatial skills in non-pathological reading, but the National Early Literacy Panel (2008) reported low univariate correlations with reading comprehension and word identification (around 0.2). In particular, deficits in visual attention have been proposed to account for the occurrence of developmental dyslexia in some children (Facoetti et al., 2010; Vidyasagar & Pammer, 2010). However visuospatial impairments could be a consequence rather than a cause of reading disorders (Ramus, 2003), and hence not be an early predictor.
On the other side, visuospatial abilities are an important foundation of numeracy acquisition. Indeed, children’s arithmetic abilities partly lie on the development of an accurate linear mental representation of quantity (Siegler & Booth, 2004). In addition, spatial processing helps to solve complex arithmetic problems which require multistep calculations (Dehaene & Cohen, 1995). Lastly, in arithmetic word problems3, visuospatial abilities may support the construction of a visual schematic representation of the problem, which in turn may improve performance (Boonen et al., 2013). Thus, visuospatial abilities have been found to be correlated with higher results in arithmetic concurrently (Hawes et al., 2019; Reuhkala, 2001; Träff et al., 2018) and longitudinally (Yang et al., 2019; Zhang et al., 2014, 2017).
Table of contents :
Chapter 1 – General introduction
1.1 What shapes early cognitive development
1.1.1 Main predictors of cognitive development
1.1.2 The importance of controlling for confounding variables
1.1.3 From simple associations to complex relationships
1.2 Acquiring academic skills: building blocks and determinants
1.2.1 The role of cognitive abilities
1.2.2 The role of social, behavioral and emotional skills
1.2.3 The role of socio-demographic factors
1.3 Studying the acquisition of academic skills in France
1.3.1 Learning in France: an outlook
1.3.2 Data: two French cohort studies
1.3.3 Objectives and research questions of this dissertation
Chapter 2 – Intelligence and academic skills
2.1 Predictors of the IQ-achievement gap
2.1.1 Abstract
2.1.2 Introduction
2.1.3 Method
2.1.4 Results
2.1.5 Discussion
2.1.6 Appendix A: Data
2.1.7 Appendix B : Results
2.2 Are high-IQ students more at risk of school failure?
2.2.1 Abstract
2.2.2 Introduction
2.2.3 Method
2.2.4 Results
2.2.5 Discussion
Chapter 3 – Early predictors of arithmetic skills
3.1 Cognitive and environmental predictors of problem solving skills
3.1.1 Abstract
3.1.2 Introduction
3.1.3 Method
3.1.4 Results
3.1.5 Discussion
3.1.6 Appendix
3.2 Cognitive predictors of multiplication, addition and subtraction
3.2.1 Abstract
3.2.2 Introduction
3.2.3 Method
3.2.4 Results
3.2.5 Discussion
Chapter 4 – Early predictors of literacy skills
4.1 Cognitive and environmental predictors of reading and spelling
4.1.1 Abstract
4.1.2 Introduction
4.1.3 Method
4.1.4 Results
4.1.5 Discussion
4.1.6 Appendix
Chapter 5 – Sex differences in academic skills
5.1 Sex differences are modulated by evaluation type
5.1.1 Abstract
5.1.2 Introduction
5.1.3 Method
5.1.4 Results
5.1.5 Discussion
5.1.6 Appendix
Chapter 6 – General discussion
6.1 Cognitive and socio-emotional foundations of academic skills
6.1.1 Cognitive skills
6.1.2 Socio-emotional skills
6.2 Environmental and individual influences on academic achievement
6.2.1 Parental socio-economic and cultural factors
6.2.2 Sex
6.2.3 Pre-natal and birth factors
6.3 General limitation
6.3.1 Genetic confounding
6.3.2 Correlation and causality
6.4 Practical implications and conclusion
References
