One complication in expounding an integrated model of EF and SR growth is diversity in conceptions, definitions, and operationalisations of key constructs and the relations between them. For instance, EF is generally considered a collection of cognitive control capacities to activate and maintain mental information (working memory), resist contrary urges (inhibition), and flexibly shift attention (cognitive flexibility; Miyake et al., 2000). While the number and organisation of EFs are debated (Lehto et al., 2003), a prominent taxonomy distinguishes between these three core EFs, which are the basis for more complex, higher order cognition (Miyake et al., 2000). Further complicating this picture, there is evidence that the ability to disaggregate these EFs changes over the preschool to early primary school years (Zelazo et al., 2013).

Even greater diversity exists in conceptualizations of SR (Burman et al., 2015), although definitions tend to emphasize its role in exerting control over manifest behaviors, emotional reactions, and social interactions (some conceptions term this “self-control,” due to the need to override unwanted impulses; Hofmann et al., 2012). Whereas EFs enable control over mental activity, SR has been distinguished as enabling control over in situ, and often emotionally laden manifest responses. This is not to say that SR applies solely to the control of emotions; indeed, factor analysis of SR measures often identifies separate emotional, cognitive, and behavioral SR factors (e.g., Howard and Melhiush, 2017; Howard et al., 2019). Whereas cognitive SR is often considered to be concerned with attentional and higher order cognitive control (Blair, 2016), behavioral SR is often used to describe children’s ability to control their actions in everyday contexts (Howard and Melhiush, 2017). In early childhood, SR thus has broad and multiple applications such as waiting one’s turn despite the impulse to act now, overcoming strong emotions in order to respond adaptively, and remaining within the rules and requirements of the setting (e.g., at preschool vs. home).

Despite conceptual overlap between EF and SR, the extent to which they are related and the nature, mechanisms, and shifts in such relations across childhood remain unclear. For instance, one model to explain self-regulatory success posits EFs as the capacity to overcome obstacles and contrary impulses to reduce a discrepancy between actual and goal states (Baumeister and Heatherton, 1996). In this model, EFs are necessary but not sufficient for successful SR. Also essential are selection and maintenance of goals, and motivation to continually invest effort until goal achievement. By contrast, in the bi-directional model of EF and SR (Blair and Ursache, 2011; Blair, 2016) EFs are top-down mechanisms by which an individual can direct attention and manage arousal (Ochsner and Gross, 2005) for the purposes of goal-directed action. From a bottom-up perspective, the mobilisation of EFs is influenced by activity in stress, emotional, and attentional systems (Blair and Dennis, 2010). That is to say, within this model EFs are necessary for successful SR but may be impaired by particularly high or low levels of arousal (i.e., reactivity). In this sense, the capacity for SR maps well on to inverted U-shaped Yerkes-Dodson curve, wherein EF proficiency is maximized at moderate levels of emotional and attentional reactivity, but is undermined at overly high or low reactivity levels (Arnsten, 2009).

While both models described above envision an interaction between EF and SR, the bi-directional model uniquely includes mechanisms by which EF and SR may be mutually influential throughout development (Blair, 2016). For example, exposure to chronic stress – requiring frequent and effortful SR – can release neurochemicals affecting activity and development of the prefrontal cortex, thereby influencing EF development (Cerqueira et al., 2007; Liston et al., 2009). Reciprocally, children with higher EFs have a wider allostatic range in which they can self-regulate under conditions of heightened arousal, and thus can engage and extend their SR abilities across a wider range of challenge. The bi-directional model, which implies that causal influences on development stem from both EF and SR processes, has important implications for identifying plausible targets for education, prevention, and intervention. For example, this model requires us to ask whether EF should in fact be targeted (as is common) to achieve real-world SR benefits; the possibility at least needs to be examined that SR change should be addressed directly and, furthermore, whether such changes in SR produce EF benefits. In sum, the bi-directional model potentiates greater clarity regarding how EF and SR develop, and how the relationship between them underpins important features of later development (i.e., academic success, mental health, physical health, and social competence).

The area where EF and SR growth have received the most attention is in relation to school readiness and academic success. Independent investigation of links between EFs and academic achievement show preschool EF abilities accounting for substantial variability in later academic achievement (Welsh et al., 2010; Fuhs et al., 2014). For instance, Bull et al. (2008) showed that high preschool EF provided an immediate head start in mathematics and reading that persisted across the first 3years of schooling. This association is robust, with numerous studies showing that relationships between EF and later academic achievement (i.e., mathematics and reading) remain even after controlling for general cognitive abilities (Espy et al., 2004; Clark et al., 2010), baseline measures of academic ability (McClelland et al., 2007), and general intelligence (Fitzpatrick et al., 2014). Furthermore, there is some evidence that this sequence is causal, with Ribner et al. (2017) showing that children with better EFs were able to catch up to peers who initially had better early mathematics abilities. Similarly, SR abilities in preschool have also been independently linked with school readiness and later academic achievement. For instance, early SR predicts status and change in academic abilities (i.e., literacy, vocabulary, and mathematics skills) across the final preschool year (McClelland et al., 2007) and during the transition to school (McClelland and Wanless, 2012). Self-regulation at age 4 has also been linked with academic achievement at age 7 (i.e., in mathematics and reading) with effects persisting into early adulthood (McClelland et al., 2013).

To explain these associations, researchers have proposed distinct and complementary ways in which EF and SR may influence academic achievement. With regard to the effects of EF on academic achievement, researchers have emphasized the direct facilitative role EFs play in learning (e.g., to remain focused, hold information in the mind, resist distraction), as well as specific links between EF processes and the inherent requirements of learning tasks (Blair et al., 2015). By contrast, proposed mechanisms for the direct impact of SR on academic achievement have tended to focus on broader contextual determinants of engagement in education. For example, given that lower SR is linked with poorer teacher–child relationships and heightened conflict (Hamre and Pianta, 2001; Valiente et al., 2011), SR may support or constrain adaptive engagement with learning environments and/or with educators more directly.

These explanations are consistent with Blair and Raver’s (2015) extension of the bidirectional model of EF and SR to school readiness, in which children are ready to start school (i.e., they are well-positioned to benefit from its structures for teaching and learning) when they are sufficiently able to regulate their arousal and attention to sustain engagement with learning experiences. While efforts have been made to independently link EFs and SR to school readiness, limited empirical research investigates the longitudinal and potentially bi-directional associations between EF and SR, and their unique/shared prediction of important developmental outcomes. Yet these insights are important for establishing a developmental model that integrates EF and SR, and for theoretical models of change that underpin EF and/or SR growth.

The current study thus sought – in a sample of preschool-aged children assessed three times over nearly 2years – to investigate the bidirectional relationships between EF and SR longitudinally, as well as the independent and cumulative associations of EF and SR with school readiness over the transition to school period. In line with predictions of Blair and colleagues’ bi-directional model (Blair and Ursache, 2011; Blair, 2016), it was expected that EF and SR would be distinct yet related, with bi-directional associations due to their theorized reciprocal influence. Further, it was expected that EF and SR would independently and cumulatively predict early academic skills on entry to school, given their expected unique contributions to engagement in and acquisition during early learning experiences. If supported, such a model would suggest potential benefit for an integrated approach to promoting EF and SR growth.

This was a longitudinal observational study of EF, SR, and academic skills with data collected at three time points: T₁ at the beginning of children’s final preschool year; T₂ at end of the final preschool year; and T₃ 1-year later, at the end of the first year of school. It leveraged data from a cluster randomized controlled trial (RCT) intervention evaluation (Howard et al., 2020), collecting data at a third time point from a geographically constrained subset of participating children who had transitioned to formal school in the following year. Longitudinal studies tracking children between preschool and school in Australia are complex because there is relatively little continuity between preschool and school contexts; children from a single preschool often disperse to many geographically distributed schools (up to nine, in the current sample) depending on jurisdiction. For this reason, a geographically defined constraint was used when recruiting children for T₃ data collection. The characteristics of the geographically constrained sample are described in detail below. The initial RCT (see Howard et al., 2020 for a description) revealed little difference between groups, which justified collapsing across groups for the purposes of the current study. Nevertheless, possible longitudinal impacts of the intervention were examined in relation to key T₃ measures to further ensure this strategy was justified, analyses of which showed no significant difference between groups on any modelled T₃ measure.

Children who had transitioned to 160 schools within the geographic radius were eligible and invited to participate in the follow-up, yielding an eligible sample of 316 children. Of these children, data were collected at T₃ (i.e., end of first year of formal schooling) for 199 children. Reasons for non-participation at this time point were parental non-consent (n=53 unable to contact, n=31 declined) or schools declining participation (impacting participation of n=33 children). The mean age of the final sample, based on baseline demographics, was 4.46years (SD=0.35, range 3.65–5.24), with 51.8% girls. Children who were identified as of Aboriginal or Torres Strait Islander descent comprised 3.0% of the sample, which is in line with population estimates for this age (Australian Institute of Health and Welfare, 2012). Family income was diverse: 6.5% of families qualified for full childcare benefit subsidies (low income); 53.3% of families qualified for some childcare benefit subsidy (low-middle to middle-high income); 23.6% of families did not qualify for any childcare benefit subsidy (very high income); while 16.6% of families declined to disclose. The highest level of maternal education attained was also diverse: 6.0% did not complete high school; 8.0% completed high school; 22.1% had completed diploma, trade, certificate; 35.2% completed a tertiary degree; 12.6% a post-graduate qualification; and 16.1% did not disclose. The 112 participating schools spanned Public (n=82), Catholic (n=28) and Independent School systems (n=2). The mean number of children per participating school was 1.78 (SD=1.21, range=1–8). These children derived from 41 preschool services, which were diverse in their socio-economic decile for catchment area (M=6.63, SD=2.44, range=1–10) and statutory quality assessment rating (i.e., n=20 Exceeding, n=19 Meeting, n=1 Working Toward, n=1 unrated against the National Quality Standard).

This study was approved by the University’s Human Research Ethics Committee, Social Sciences (2018/536), and participants were those who provided verbal assent and their parents provided informed written consent to participate.

At T₁ and T₂, all tasks were administered to children in a quiet area of their preschool centre in five sessions across the same day, to maximize children’s attention and minimize fatigue. Measures were administered in the same order to all children, as follows: (1) Bracken School Readiness Assessment (not used for the purposes of this analysis, given its use only at earlier time points and performance near ceiling by T₂); (2) PRSIST curiosity boxes and HTKS; (3) Mr. Ant and Go/No-Go; (4) PRSIST memory; and (5) Card Sort. Each session took 10–20min to complete. For T₃ data collection in schools, data were collected in a quiet space (e.g., office, library) across three sessions in the same day, as follows: (1) Mr. Ant and Go/No-Go; (2) Card Sort and HTKS; and (3) Numeracy, Expressive Vocabulary and Shape Trail (the latter not used for purposes of this study). Children’s Kindergarten teachers also reported on children’s SR using the CSBQ at the same time.

To evaluate the bi-directional associations between EF and SR, and longitudinal relations to early academic skills, cross-lagged panel models were run using AMOS (Version 25, IBM Corp Armonk NY, United States). In line with theoretical predictions of EF and SR abilities as distinct – i.e., they are reciprocally influential and independently account for unique variance in development and outcomes – three models were evaluated. These evaluated a three-timepoint (start and end of preschool year, end of first year of school) cross-lagged panel model (Model 1, Figure 1), which subsequently added early numeracy skill as the outcome (Model 2, Figure 2) or expressive vocabulary as the outcome (Model 3, Figure 3). Given PRSIST was not possible to conduct at T₃ due to a limited number of participating students per school, the T₃ SR variable was a composite of HTKS and two school-teacher-reported CSBQ subscales: behavioral self-regulation and cognitive self-regulation. The stability of the SR factor over time, as a consequence of this change, was also evaluated through these models.

The three-timepoint model (Figure 1) provided good fit to the data: χ²(4)=13.42, p=0.009, CFI=0.99, RMSEA=0.07, AIC=59.42. Each of EF and SR showed moderate prediction of that same ability from each time point to the next (βs ranging from 0.36 to 0.57). Patterns of bi-directional association were also evident: EF and SR at T₁ were moderately correlated (r=0.57) and there were small and similar loadings from EF to SR, and from SR to EF, from one time point to the next (βs ranging from 0.20 to 0.28). Lower correlations between EF and SR at later time points are to be expected due to variance accounted for by autoregressive paths – that is, each subsequent correlation between EF and SR accounts for prior levels of these abilities – and thus should not be interpreted as point-in-time reductions in correlation. As such, they are not further interpreted or presented in depictions of subsequent models.

The addition of early numeracy as an outcome showed slightly reduced, albeit still good, model fit: χ²(8)=41.38, p<0.001, CFI=0.96, RMSEA=0.09, AIC=95.38. In terms of prediction of early academic skills, EF (β=0.30) and SR (β=0.43) at T₃, both independently predicted numeracy scores (Figure 2A). Constraining these paths to equivalence did not significantly alter model fit, which suggests that the path loadings onto early numeracy were comparable. R² statistics indicated that this model accounted for: at T₃, 38% of the variance in early numeracy skills, 31% of the variance in EF and 28% of the variance in SR; and at T₂, 37% of the variance in EF and 53% of the variance in SR. Modification indices suggested model improvement with inclusion of additional paths from T₁ and T₂ EF and SR to T₃ numeracy skills, which were added sequentially by order of their strength and evaluated in Model 2b.

The final revised model (Figure 2B) provided better relative fit to the data: χ²(7)=19.28, p=0.007, CFI=0.99, RMSEA=0.07, AIC=76.33. While all paths from the initial Model 2 remained significant, there was an additional significant path from T₁ EF to T₃ numeracy (β=0.30), which slightly reduced the strength of paths from T₃ EF (β=0.22) and SR to numeracy (β=0.35). This model provided better explanation of early numeracy as well, R²=0.43. No further modification indices were indicated after addition of this path.

The three-timepoint model integrating expressive vocabulary (Figure 3A) also provided good fit to the data: χ²(8)=37.99, p<0.001, CFI=0.97, RMSEA=0.09, AIC=91.99. In terms of prediction of early academic skills, EF (β=0.17) and SR (β=0.17) at T₃ both equally and independently predicted expressive vocabulary scores (Figure 2A). R² statistics indicated that this model accounted for: at T₃, 8% of the variance in expressive vocabulary skills, 31% of the variance in EF and 28% of the variance in SR; and at T₂, 37% of the variance in EF and 53% of the variance in SR. Modification indices suggested model improvement with inclusion of additional paths from T₁ and T₂ EF and SR to T₃ expressive vocabulary, which were added sequentially by order of their strength and evaluated in Model 3b.

The final revised model (Figure 3B) provided better relative fit to the data: χ²(7)=20.18, p=0.005, CFI=0.98, RMSEA=0.06, AIC=76.18. While all paths from the initial Model 1 remained significant, there was an additional significant path from T₁ EF to T₃ numeracy (β=0.33), which rendered non-significant the paths from T₃ EF (β=0.06) and SR to numeracy (β=0.07). This model provided better explanation of expressive vocabulary as well, R²=0.15. No further modification indices were indicated after inclusion of this path.