Cognitive and Academic Skills in Two Developmental Cohorts of Different Ability Level: A Mutualistic Network Perspective.

Challenges of the Unidirectional One-to-One Mapping Perspective

Traditional theoretical models of learning difficulties have assumed that cognitive abilities provide the foundation for academic development. For example, phonological processing deficits observed among children with reading difficulties provide the basis for the hypothesis that reading difficulties are caused by poor phonological processing ( Bishop & Snowling, 2004 ; Clayton et al., 2020 ; Perfetti, 2007 ). Similarly, working memory deficits observed among children with specific maths problems provide support for the hypothesis that memory systems are critical for maths ( Szucs et al., 2013 ). Such theories are appealing due to the simplicity of assuming a “core deficit” ( Astle & Fletcher-Watson, 2020 ) and their direct implications for intervention (e.g., phonological interventions for struggling readers, Bowyer-Crane et al., 2008 ).

Core deficit theories are challenged by evidence suggesting that similar learning difficulties may stem from different causes (i.e., equifinality). For example, reading difficulties are not wholly explained by phonological deficits. Some struggling readers have difficulties in applying letter-sound correspondences to decode words, the primary role ascribed to phonological processing in reading ( Castles & Friedmann, 2014 ). However, other poor readers have no difficulties with decoding, but instead, struggle with reading comprehension, an aspect of reading assumed to be supported by skills such as working memory (e.g., Cain et al., 2004 ). Similarly, data-driven explorations of the links between cognitive and academic skills suggest that children can arrive at similar profiles of learning impairment through multiple etiological routes: Children with comparable difficulties in both reading and maths can have different cognitive profiles, such as relatively more severe problems with phonological processing or working memory ( Astle et al., 2019 ). Consistent with equifinality, multiple deficit theories suggest that risk factors spanning multiple levels (genes, brain, cognition, and behaviour) contribute probabilistically to neurodevelopmental difficulties ( Pennington, 2006 ) and that shared risk factors contribute to comorbidity. This idea is supported by evidence that processing speed explains some of the comorbidity between dyslexia, dyscalculia, and Attention Deficit Hyperactivity Disorder (ADHD), while language difficulties contribute to comorbidity between dyscalculia and dyslexia ( McGrath et al., 2011 ; Peterson et al., 2017 ).

Bidirectional Dynamics

A challenge to both single and multiple deficit models comes from studies showing that while cognitive skills influence academic skills, academic skills also shape cognitive development. Take our earlier example of the association between phonological processing and reading. Phonological awareness contributes to reading development, but it also benefits from reading instruction and exposure to text ( Huettig et al., 2018 ; Nation & Hulme, 2011 ). Similar reciprocal effects, where the development of a given academic skill predicts growth in cognitive performance and vice-versa, have been reported between working memory and reading and maths ( Miller-Cotto & Byrnes, 2020 ), IQ and reading ( Ferrer et al., 2007 , 2010 ), nonverbal reasoning and vocabulary ( Kievit et al., 2017 ; Kievit, Hofman, & Nation, 2019 ), and executive functions and maths ( Schmitt et al., 2017 ; Van der Ven et al., 2012 ). These examples suggest that difficulties in one domain may have downstream effects on other abilities and challenge the use of analytic strategies in which cognitive deficits are uniquely modelled as predictors and academic skills as outcomes.

An alternative view of the relationship between cognitive and academic skills, consistent with evidence for reciprocal benefits, is mutualism. Mutualism proposes that different abilities interact bidirectionally to reinforce one another during development ( Peng & Kievit, 2020 ; Van Der Maas et al., 2006 ). In other words, the mastering of a skill supports and amplifies the development of other abilities. The mutualism model is neuro-constructivist in nature, acknowledging that specialised abilities likely emerge developmentally through a process of multidirectional interactions between genes, brain, cognition, and the environment ( Karmiloff-Smith, 2009 ; Kievit, 2020 ).

In the context of cognitive-academic coupling, mutualistic transactions might be driven by experience, and particularly by educational experiences. Fundamental cognitive resources support the development of academic skills while performing academic tasks uses and trains cognitive abilities, and over time, these relationships become mutually beneficial ( Peng & Kievit, 2020 ). For example, fluid reasoning skills aid the use of analogies and abstract schema in academic tasks, while concrete knowledge (e.g., verbal skills) supports the decomposition of complex reasoning tasks ( Kievit et al., 2017 ). The type and strength of these transactions might be moderated by experience. Weaker or absent bidirectional relations have been reported in children with learning difficulties (e.g., Ferrer et al., 2010 ; Quinn et al., 2019 ). This could be because difficulties with specific skills operate as the bottleneck to the development of the wider cognitive system and/or because struggling learners may choose to avoid exercises that tax their area(s) of weakness. In the latter case, the net result could be that poor learners may engage in fewer activities that develop positively reinforcing associations, which over time might constrain both cognitive and academic development.

Cognitive and Academic Skills: New Approaches

To address the challenges outlined above, it is necessary to rethink the recruitment and analysis strategies typically used to study learning difficulties. The prevailing sampling approach involves recruiting highly selective samples of children based on the presence of a specific diagnosis or difficulty, with comorbidity often treated as a confound (e.g., Szucs et al., 2013 ). This runs counter to a wealth of evidence showing that disorders are highly comorbid, heterogeneous, and explained by multiple causes ( Astle et al., 2019 ; Peters & Ansari, 2019 ). Overstating the “purity” of learning problems at the point of recruitment biases outcomes towards core-deficit accounts. To avoid such biases, learning-related problems are increasingly studied using transdiagnostic approaches ( Astle et al., 2021 ; Casey et al., 2014 ; Holmes et al., 2019 ; Mareva & Holmes, 2019 ). These approaches aim to understand the processes and causes of difficulties that occur across individuals irrespective of diagnosis or group membership.

In terms of methods, unidirectional associations between cognition and academic performance are typically modelled using latent variable approaches whereby cognitive and academic factors are derived separately, and the relations between them are mapped as directional paths from cognitive to academic skills ( Peterson et al., 2017 ). One alternative, which can better accommodate the possibility of equifinality without assuming causational directionality, comes from network science. Network models are relatively new to cognitive science but have already proved useful in tackling similar challenges in the field of psychopathology ( Borsboom, 2016 ). In simple terms, network models use partial correlations to depict how each observed variable relates to all other observed variables. In this way, they offer a modelling framework that is more consistent with mutualism and the possibility that reciprocal mutual transactions are the generating process behind the relationships observed across cognitive and academic skills. Applied to cross-sectional data, these models offer a tool to explore or test specific hypotheses about whether and how the complex developmental interplay between academic and cognitive skills may differ across time points and/or groups of children.

Objectives

The aim of the current study was to apply networks models to explore and compare the interrelations between cognitive and academic abilities in a community sample and a sample of struggling learners. A heterogeneous cohort of children identified by practitioners as having school-related difficulties was included to represent struggling learners. The comparison cohort was an age-matched group of children selected as being nationally representative. For academic skills, the focus was on literacy and maths. For cognitive skills, we included assessments of processing speed, working memory, executive function, and nonverbal reasoning, all of which have been previously linked to academic performance in both typical and atypical learners ( Altemeier et al., 2008 ; Booth et al., 2010 ; Gathercole et al., 2004 ; Geary, 2011 ; Green et al., 2017 ; Holmes et al., 2020 ; Mayes & Calhoun, 2007 ; Peng & Fuchs, 2016 ; Taub et al., 2008 ; Yeniad et al., 2013 ). To our knowledge, this is the first application of network science to cognitive-academic interrelationships in learners of different abilities. Consistent with mutualism and equifinality, we anticipated there would be multiple direct links between cognitive and academic skills in both cohorts. We were agnostic as to whether and how task interrelationships would differ across cohorts.

Matrix Reasoning

The Matrix Reasoning subtest of the Wechsler Abbreviated Scales of Intelligence II (WASI-II; Wechsler, 2011 ) was administered to children in both cohorts. Children are presented with increasingly complex nonverbal analogical reasoning problems in 2x2 matrices and are asked to select from a range of alternatives a shape that completes the pattern. The number of correctly solved matrices was scored.

Motor Speed

Children in both cohorts completed the Motor Speed test of the Delis Kaplan Executive Function System (D-KEFS; Delis et al., 2001 ). This measured time (in seconds) to trace a dotted line as quickly as possible. Completion time in seconds was scored.

Tower

The Tower subtest of the D-KEFS, completed by both cohorts, required children to move disks of different sizes around pegs from a start position to an end state shown on a picture following a set of rules. The number of correctly completed towers was scored.

Trails Number-Letter Sequencing

Both cohorts completed the Trails Number-Letter Sequencing task of the D-KEFS, which required children to connect letters and numbers in a progressive alternating sequence (e.g., 1-A, 2-B, etc). Completion time in seconds was scored.