Associations of Specific Indicators of Adult–Child Interaction Quality and Child Language Outcomes: What Teaching Practices Influence Language?

ABSTRACT

Research Findings: This study aims to extend our knowledge regarding contributions of educator–child interactions to child language outcomes by examining the extent to which specific dimensions of the CLASS observational tool of educator–child interactions are associated with child language abilities, utilizing data from an Australian longitudinal study of over 2,000 children attending formal Early Childhood Education and Care (ECEC). The analysis included a novel measurement model fitted to the data to allow each CLASS dimension to be modeled separately. Results showed that each CLASS dimension was associated with initial average language abilities. Small, negative effects of Emotional Support dimensions on growth of children’s average Understanding Directions score were found, but there were no associations between any of the dimensions and average growth in Verbal Ability. None of the Instructional Support dimensions (which are language focused) predicted growth in language abilities. These null findings are addressed in the discussion. Practice or Policy: Findings from this study illustrate that, typically, ECEC programs rate low on dimensions of quality developed to capture language-promoting educator–child interactions. Findings also suggest a selection effect related to equity of access to classroom quality with children with the highest initial language abilities in the highest quality classrooms.

There is a growing body of evidence to suggest that high-quality early childhood education and care (ECEC) can have a significant impact on children’s language development and predicts better achievement throughout childhood and into adulthood (Cabell et al., 2015; Ulferts et al., 2019; Yoshikawa et al., 2013). Quality in the ECEC context is multidimensional and often characterized by the dimensions of structural and process quality.

Structural quality encompasses the regulatable aspects of a program such as staff/child ratios and staff qualifications that are expected to support high-quality ECEC (Howes et al., 2008). Process quality refers to the quality of children’s day-to-day experiences and interactions with educators, other children and materials, and their participation in learning experiences that are associated with children’s learning and development (Howes et al., 2008; R. Pianta et al., 2005). Both theory and program evaluations, including the Abecedarian and Perry preschool programs, provide evidence supporting the positive effects of intensive high-quality ECEC settings on children’s language outcomes (Schweinhart et al., 2005; Slot, 2018).

While components from both structural and process quality have been linked to child language development and academic success, high-quality interactions between educators and children are a critical factor to children’s language outcomes (Curby et al., 2009; Sabol et al., 2013; Slot et al., 2018). It is widely acknowledged that children require an optimal language-rich environment to develop the necessary language skills to realize their full potential. Theoretical perspectives, both social interactionist theory (Bruner, 1981) and Vygotsky’s Zone of Proximal Development (ZPD) (Vygotsky, 1978), have contributed to understanding how child language develops within a social context that responds to children’s abilities. The concept of ZPD relates to adult linguistic input that is pitched slightly above the child’s level of development but is not beyond what the child can do. These well-supported theories of language development highlight that the social interactions children experience with those around them are crucial to their language development; that the nature of these interactions is reciprocal, with children playing an active role; and that the most supportive form of adult input is that which is responsive and focused on the child’s interests and needs (Rowe & Snow, 2020).

Bruner’s social interactionist theory states that as a child learns language to communicate and participate in the social world, they learn the linguistic code from the input given by adults (Fahim & Amerian, 2015). High-quality adult–child interactions offer meaningful learning contexts for the child to learn both linguistic and non-linguistic conversation skills (Grebelsky-Lichtman & Shenker, 2019). In contrast, environments of lower educational quality, which offer few language stimulation opportunities for children, can lead to developmental delays, including in language skills (April et al., 2018). Hence, the way adults support, respond to and extend the child’s communication attempts may impact their rate of language acquisition, which raises the need for effective educator practices in regard to language development in the early years.

Supportive Educator Practices for Language Development

There is extensive evidence which consistently identifies the components of language-promoting adult–child interaction shown to be associated with language learning (Levickis et al., 2014; Roberts et al., 2019) and which can be harnessed by educators to promote language development (Walker et al., 2020). Rowe and Snow (2020) recently conceptualized the features of caregiver input that facilitate language development in terms of three dimensions of input quality: interactive, linguistic, and conceptual. Interactive features include responsiveness and shared attention; linguistic features include levels of linguistic complexity and redundancy that are adapted to the child’s developmental stage; and conceptual features are those topics of conversation that provide appropriate challenges for the child’s level of development (Rowe & Snow, 2020). These features of language-promoting caregiver–child interaction have also been identified as important in educator–child interactions for supporting children’s language skills (Girolametto et al., 2006; Snow, 2014).

A recent systematic review of teacher language practices and child language outcomes in young children (ages 3–6 years) showed that the effectiveness of language practices within the three features proposed by Rowe and Snow (2020) will vary depending on the purpose, place, and participants involved (Hadley, Barnes, & Hwang, 2022). For example, the same combination of language-promoting strategies may not be as effective in a large group compared to a small-group setting. As an illustration of this, Hadley and colleagues found that interactive features varied due to the purpose and place of the interaction, for example, small group play was a setting for child-centered and interaction-promoting practices, such as following the child’s lead, purposefully pausing, and encouraging peer-to-peer talk. While the interactive features observed in large group shared book reading, included modeling, scaffolding, and affirmations, rather than those interactive features focused on extending conversations.

In a systematic review and meta-analysis, Hadley, Barnes, Wiernik, et al. (2022) further examined the relationships between teachers’ language practices and child language outcomes and tracked the variation of these practices across contexts. They extended on their first review by identifying two main registers of talk made up of a number of identified key types of teacher language practices: Emergent Academic Language (characterized by abstract talk and the use of linguistic features such as a greater variety of word types and longer sentences) and Bridge Language (charactertized by interactive language practices that support and extend child talk and use more concrete language practices to scaffold children’s language learning). Findings highlight that when planning language-rich activities, teachers may need to adopt a cluster of language practices that are most appropriate for the purpose of their talk, the place, or setting in which the practices are implemented and the people involved (Hadley, Barnes, Wiernik, et al., 2022). Limitations to these findings are the lack of information regarding classroom quality.

Small shifts in the quality of educator–child interactions within ECEC settings can benefit child language learning, particularly for children from socially disadvantaged circumstances and for those with less stimulating home learning environments (M. Burchinal et al., 2008). Educators who are responsive in interactions with children and support joint and sustained shared attention are more likely to strengthen children’s developmental pathways.

However, these responsive interactions and instructional support within ECEC programs are rare and less likely to occur in ECECs located in more disadvantaged communities (Cloney et al., 2016). In order to support children’s language development within ECEC settings, it is important to identify specific dimensions of educator–child interactions that can be promoted as part of educators’ professional development and reflective practice.

Measuring Educator–Child Interaction Quality as it Relates to Language Development

There are a number of observational tools that can be used to measure the quality of educator–child interactions in ECEC settings, such as the Classroom Assessment and Scoring System (CLASS) Pre-K (R. C. Pianta et al., 2008). The CLASS Pre-K observation tool consists of three domains: Emotional Support, Classroom Organization, and Instructional Support, which are each made up of three to four dimensions.

Emotional support includes the following dimensions: Positive climate, Negative climate, Teacher sensitivity, and Regard for student perspectives. Positive climate is observed when there is an emotional connection, respect, and enjoyment demonstrated between teachers and children and among children. Negative climate reflects the level of negativity (such as anger, hostility, or aggression) expressed by teachers and/or children in the classroom. Teacher sensitivity captures teachers’ awareness of and responsiveness to children’s cognitive and social-emotional needs. Children are also comfortable enough to freely participate and take risks, seeking adult support and guidance when needed. Regard for student perspectives is characterized by teachers’ flexibility in incorporating children’s interests and ideas in their learning activities, encouraging children to express their ideas, and fostering independence and responsibility.

The dimensions of Teacher sensitivity and Regards for student perspectives in the Emotional Support domain align with the social interactionist theory of language learning that purports that the most beneficial adult–child interactions are those whereby the adult is responsive to the child’s interests and needs. There is mixed evidence for associations between the Emotional Support domain of CLASS and child language outcomes. In a study examining which dimensions within each CLASS domain were most strongly predictive of children’s academic learning, Curby and Chavez (2013) found that positive climate was the only dimension that predicted child language outcomes.

In another study, Mashburn et al. (2008) found no evidence of an association between the Emotional Support domain and child language outcomes, although they included domain scores rather than looking at individual dimensions (Mashburn et al., 2008). A recent study examining how changes in observed quality using CLASS were associated with changes in children’s cognitive development, including language skills, found that increases in Emotional Support were associated with improvement in children’s language outcomes (Rankin et al., 2022).

The Classroom Organization domain includes Behavior management, Productivity, and Instructional learning formats. High-quality Behavior management is characterized by classrooms where children are actively involved in activities, where there are clear rules and routines that the children follow, and little misbehavior. Productivity is achieved when teachers are successful in managing time so that children always have something to do. Teachers are effective in managing transitions from one activity or classroom to the next. Instructional learning formats are observed when teachers capture children’s interests, support their learning, and engage children in classroom activities, using a variety of modalities and materials. The Classroom Organization domain has been positively associated with children’s language and literacy skills in kindergarten and Pre-K (Carr et al., 2019) and shown to predict their engagement in the learning process (Castro et al., 2017).

Theory and evidence support the contribution of the Instructional Support domain to children’s language learning, which includes the dimensions of Concept development, Quality of feedback, and Language modeling. Concept development captures the teachers’ use of specific strategies during instructional discussions and activities to encourage children to think more deeply and expand their understanding. Another dimension is Quality of feedback, which is characterized by a teacher responding to what a child says or does by encouraging the child to keep thinking and trying, supporting children to clarify and expand their ideas. Language modeling is the third dimension and occurs when teachers participate in frequent meaningful conversations whereby the teacher asks open-ended questions, repeats, extends, and elaborates children’s responses.

Instructional support is thought to facilitate children’s language learning via feedback that expands on learning and meaning-based instructional discussions. These dimensions align with Rowe and Snow’s (2020) developmental theory of language learning, as these dimensions involve teacher’s use of interactive (e.g., engaging children in back-and-forth conversations), linguistic (e.g., exposing children to more complex syntax and sophisticated vocabulary through the use of syntactically complex explanations with cause-and-effect language), and conceptual (e.g., using decontextualized language to have extended discussions) features of linguistic input.

Research evidence also supports the association between the Instructional Support domain and child language outcomes. For example, the large, Australian longitudinal E4Kids study found that Instructional Support predicted children’s verbal abilities after adjusting for various child and family factors (Niklas & Tayler, 2018). However, there are other recent studies that have examined the CLASS Instructional Support domain, demonstrating only small associations with children’s language outcomes (Justice et al., 2018; Keys et al., 2013).

While we might expect, based on the literature, dimensions of Instructional Support to be most strongly predictive of language development, there are not only these “within-domain” links to children’s outcomes (whereby the domains conceptually align with the outcomes) but also “cross-domain” links to children’s outcomes. For example, children who experience responsive and warm educator interactions (Emotional Support domain) have demonstrated higher literacy and language outcomes (Downer et al., 2010). Therefore, when examining educator–child interactions that are likely to best support children’s language learning, it is important to look at all dimensions of classroom quality.

Variations and Predictive Power of Specific CLASS Dimensions

Although the aforementioned studies provide some evidence that each of the CLASS domains is associated with children’s language skills, it is important to note that educator’s scores vary across domains, particularly for Instructional Support. While observed Emotional Support and Classroom Organization quality is often in the mid to high range on CLASS, Instructional Support is generally located in the low range (Perlman et al., 2016; Slot, 2018). This trend is observed worldwide, with studies in North America (i.e., Hamre et al., 2013), Europe (i.e., Stuck et al., 2016), Asia (i.e., Hu et al., 2016), and Australia (i.e., Tayler et al., 2013) reporting similar results.

Moreover, the evidence demonstrating the associations between specific dimensions of educator–child interactions, such as those captured by the CLASS tool, and child language outcomes is mixed. One possible explanation for these mixed findings is selection bias, whereby families from more advantaged backgrounds enroll into higher-quality ECEC services. While studies may adjust for this by controlling for family demographic variables, there may be unobservable factors that impact on educator–child interactions and children’s language gains (Guerrero-Rosada et al., 2021; Weiland & Rosada, 2022). Another explanation for lack of predictive power may be due to CLASS observations being collected in one day, increasing the likelihood of measurement errors caused by factors such as rater effects (Styck et al., 2021).

In addition to finding limited evidence of associations between CLASS and growth in developmental skills, including language outcomes, researchers also suggest that systems-level classroom quality instruments, such as CLASS, may not adequately measure the critical proximal processes that are most likely to support children’s language development in ECEC settings (Justice et al., 2018; McDoniel et al., 2022). By examining the extent to which specific dimensions measured within the domains of Classroom Organization, Emotional Support, and Instructional Support predict child language outcomes, there is the potential to identify dimensions that may be used to inform educator practice for improving language abilities of children in ECEC settings.

In relation to the common occurrence of low levels of Instructional Support observed in ECEC settings and mentioned earlier, it may be that in order to see any impact on child language outcomes, Instructional Support needs to be above a certain threshold. Findings from a meta-analysis demonstrated that Instructional Support is related to larger gains in language and literacy skills in higher-quality versus lower-quality classrooms (Burchinal et al., 2016). Burchinal et al. (2010) also identified a minimum threshold for Instructional Support quality that is required in order to have an impact on child outcomes.

However, more recent studies examining threshold conditions have found that while it is evident that quality educator–child interactions are strongly associated with child outcomes, such as language skills, when specific aspects of these interactions are in the upper ranges of the distribution, there is no consistent evidence to support the use of a specific, universal cut-point for domains of quality (Hatfield et al., 2016; Weiland et al., 2013). Hence, there is a need to explore the predictive power of CLASS domains and dimensions regardless of recommended thresholds to understand the extent to which educator–child interactions impact early childhood language development.

The Current Study

We aim to examine, within an Australian context, the extent to which specific dimensions of the CLASS observational tool of educator–child interactions are associated with child language abilities, after controlling for classroom-level and child- and family-level factors. Specifically, we address the question, to what extent are CLASS dimension scores associated with children’s initial language scores and growth in language scores, adjusting for child, family, home, ECEC, and community characteristics?

Based on developmental theories of language development and empirical research, we hypothesize that all dimensions of CLASS will be associated with children’s language outcomes, but that dimensions within the Instructional Support domain will be the strongest predictors of children’s initial language scores and growth in language scores.

Unlike previous studies that have analyzed CLASS data using the average of the CLASS dimensions over repeated observations, we have taken a novel analytic approach by generating plausible values from a combined IRT and latent regression model. Typically, CLASS data are analyzed by taking the average of the dimensions over repeated observations (cycles), which assumes that there is no measurement error and that each cycle of observations contributes equally to the total rating on that dimension. However, this cannot be assumed to be an interval measure and does not take into account systematic variance, such as the fact that dimension scores decline over repeated cycles (i.e., it becomes harder to score more highly the longer a classroom is observed) (Cloney et al., 2017).

Therefore, rather than analyzing CLASS dimensions by taking the mean of the repeated cycles of each dimension, plausible values were used. In this item-as-factors model, each dimension was taken to be observed many times. This accounts for systematic variation (dimension scores going down over time, on average), measurement error (reliability), and item difficulty (some CLASS dimensions are harder to demonstrate despite them all being scored on the same 1–7 Likert scale), which addresses the limitations of previous studies by reducing the risk of bias.

Method

Sample

This paper utilized data from the Australian E4Kids longitudinal study, which aimed to determine the effects of mainstream ECEC programs on children’s learning, development, social inclusion, and well-being (Tayler, 2016; Tayler et al., 20162016). In 2010, ECEC programs were sampled from the states of Queensland and Victoria and were then randomly sampled from two metropolitan regions, one regional and one remote region. The services were randomly selected to include a range of ECEC programs from low and high socioeconomic areas. ECEC rooms that included a minimum of five children aged between 3 and 4 years were included and all children in the selected rooms were invited to take part.

The Australian ECEC sector is diverse and complex and made up of different types of delivery settings, including center-based long day care which delivers preschool programs, as well as education and care for children aged 0–5 years; kindergarten (also referred to as sessional preschool) which delivers preschool programs to children aged 3–5 years; family day care which delivers education and care to small groups in the educator’s home; and occasional care which delivers shorter, less formal sessions of education and care in a similar environment to long day care. The final sample included 2,494 children experiencing a range of approved ECEC programs, including center-based long day care, kindergarten, family day care, and occasional care (for further details on study sampling please see Tayler, 2016; Tayler et al., 2016). The current analyses include data from wave 1 (2010) and wave 2 (2011).

Procedures

In April–June 2010 and 2011, trained research assistants conducted direct assessments with children to measure children’s developmental outcomes, including language skills (i.e., child outcome data was collected at two time points). Parents provided informed written consent for their child to participate in the study. Parents of participating children completed surveys in April each year to provide relevant family and child information, such as demographics, the home learning environment, and ECEC attendance. Classroom observations were conducted between July and September 2010 by trained research assistants to measure the quality of educator–child interactions in ECEC services. Four to six repeated 20-min observations were carried out for each participating room.

Service directors and educators also completed surveys in May and June 2010 to capture room and service-level data, such as class size and service type. The E4Kids study received ethics approval from the University of Melbourne Human Research Ethics Committee (ID 0932660).

Measures

Quality of Educator–Child Interactions

The quality of educator–child interactions was measured using the Classroom Assessment and Scoring System Prekindergarten (CLASS Pre-K) version (R. C. Pianta et al., 2008). The CLASS Pre-K tool is a widely used, reliable measure designed for use in rooms of children aged 3 to 5 years. As described earlier, the CLASS-Pre-K tool measures three domains of educator–child interactions (Emotional Support, Classroom Organization, and Instructional Support) which each consist of a number of dimensions. Each dimension is rated on a 7-point scale, where a 1 is low quality and 7 is high quality. Scores of 3–5 are in the mid-range of quality. The rating for each dimension is determined using the CLASS manual, which provides detailed descriptors at each scale-point for each dimension. Research assistants completed initial reliability tests and re-tests throughout the duration of data collection to account for rater drift. Simple interrater agreement was greater than 90%. Reliability estimates (alpha) of the CLASS Pre-K domains determined from a number of US studies have been reported as ranging from .85 to .94 for Emotional Support, .76 to .89 for Classroom Organization, and .79 to .90 (R. C. Pianta et al., 2008). Estimates of internal consistency in the E4Kids study are reported as .87, .85 and .89, respectively (Cloney et al., 2016).

Language Outcomes

Child language ability was measured via direct assessment using tests from the Woodcock-Johnson III (WJIII) Tests of Cognitive Abilities and Tests of Achievement (McGrew & Woodcock, 2001), including Verbal ability and Understanding Directions. The WJIII is normed for children from age 2 (see Mather & Woodcock, 2001). The selection of specific tests from the WJIII was constrained by the need to administer the measure to children within an hour. W scores for each test were used in the analysis. The W score is an equal-interval scale, used to examine an individual’s growth in a skill or ability (Jaffe, 2009). An increase of 10 W units represents an individual’s ability to perform with 75% success on a task that they could previously perform with 50% success. This applies to any 10-unit increase on the W scale, regardless of the task level of difficulty or ability being measured.

Cognitive Test 1: Verbal Ability

The test measures the broad ability of Comprehension Knowledge (Gc) and incorporates the narrow abilities of Lexical Knowledge and Language Development. This test has four subtests: Picture Vocabulary (e.g., identify the picture of a horse from a set of pictures), Synonyms (e.g., another word for “angry” is?), Antonyms (e.g., the opposite of “no”?), and Verbal Analogies (e.g., “eye is to see, as ear is to…”). Each subtest starts with a sample item to demonstrate how to approach the question and for the researcher to provide feedback to the child. No feedback is provided beyond the sample item. Together, these subtests measure different aspects of children’s acquired vocabulary skills, language development in the form of spoken language skills that do not require reading, and lexical reasoning. For 4-to 6-year-old children, estimated reliabilities on Verbal Ability range from 0.89 to 0.90.

Achievement Test 4: Understanding Directions

The test measures the broad ability of Comprehension Knowledge (Gc) and the narrow abilities of Listening Ability and Language Development.

Classroom-Level Variables

A set of classroom-level variables, known to be associated with child language outcomes and ECEC quality, were included as potential confounding factors. Factors captured via service director and educator surveys included service types (long day care, family day care, stand-alone kindergarten, and occasional care), and group size.

Service-level socioeconomic status (SES) was measured using the Australian census-based Socio-Economic Indexes for Areas (SEIFA) Index of Relative Socio-Economic Advantage and Disadvantage (IRSAD) at the postcode level (mean of 1,000, SD ±100; Australiab Bureau of Statistics [ABS], 2008), with higher scores indicating in general a relative lack of disadvantage and greater advantage.

We also adjusted for the average verbal ability of the classroom using the 2010 verbal ability scores detailed earlier. Evidence suggests that average cognitive abilities of children attending the same ECEC room may impact a child’s individual competencies (Niklas & Tayler, 2018).

Family- and Child-Level Variables

Family- and child-level variables identified as predictors of child language outcomes were included in the current analyses and captured via parent survey. Items included in the survey were child sex, prior ECEC experiences (hours in formal and informal care the year prior to study participation), and the child’s primary language (a question in the survey regarding the child’s main language spoken in the home included as a covariate as language abilities were assessed in English).

The age of the child at first direct assessment was included, as well as days between first assessment and CLASS observation. As with service SES, Australian SEIFA scores were used as an indication of SES of study children.

A proxy measure of the quality of the home learning environment was captured with a question in the parent survey asking how frequently a family member read to the study child from a book in the last week (0–7 days). Frequency of reading to children in their preschool years is demonstrated to predict better later language outcomes (Eadie et al., 2022; Raikes et al., 2006; Yu & Daraganova, 2015).

Maternal mental health was determined using the Kessler Non-specific Psychological Distress Scale (K-10) (Kessler et al., 2002). The K10 consists of 10 questions about psychological distress to measure prevalence and severity. Each question was scored on a 5-point Likert scale from 1 (none of the time) to 5 (all of the time), with scores summed up to provide a total K10 score. The lowest possible score is 10, and the highest possible score is 50. The cutoff scores used for both the 2000 Health and Wellbeing Survey (conducted in Western Australia) and the (ABS, 2001) National Health Survey used an approach to reporting K10 results using four levels of psychological distress. The cutoff scores for level of psychological distress were as follows: 10–15 low; 16–21 moderate; 22–29 high; and 30–50 very high.

Child social, emotional, and behavioral difficulties and prosocial behavior were measured using the parent-reported Strengths and Difficulties Questionnaire (SDQ) (Goodman, 1997). The SDQ produces a Total Difficulties score (possible range 0–40) and Prosocial Behavior score (possible range 0–10). Children with behavioral problems and poorer social skills have been shown to have lower language scores than their peers without behavioral problems and good social skills (Kaiser et al., 2000).

Analysis

First, descriptive statistics were generated, and a correlation matrix of raw CLASS scores was computed to examine the correlations between each of the 10 CLASS dimensions. The next stage of analysis involved calculating plausible values (Davier et al., 2009) of each CLASS dimension.

The analysis was conducted in three steps:

1. We conducted a 10-dimensional item calibration, which allows for the estimation of CLASS dimension scales using data from all time points (nested within classrooms). The calibration decomposes the thresholds of the CLASS response categories (1 to 7, low to high) into a difficulty component (the location of the threshold on the scale whereby a classroom is equally likely to be rated two rather than one, three rather than two and so on) and some average deviation from that difficulty at each time point (the average change in scores within classrooms over repeated observations). This calibration yields item parameters to be anchored and treated as fixed (known parameters) in subsequent analysis.

2. We combined the item response and population model. Treating the item parameters as fixed (anchored), we then added a set of regression variables represented at the classroom level (classroom SEIFA, class size, classroom type [Long day care, kindergarten, family day care], and average Verbal Ability score) to the population model. The same set of regression variables also go into the secondary analysis – such an approach is necessary to yield unbiased parameter estimates (Wu, 2005) in the growth models fit last. This is consistent with the methods used in educational surveys (e.g., OECD PISA) approaches to “conditioning” – that is, providing more information in the latent regression model to ensure the plausible values (PVs) appropriately account for variables to be included in later secondary analysis (e.g., produce unbiased estimates) and accurately approximate the underlying latent distribution of classroom quality (Marsman et al., 2016).

3. Secondary analysis. At the child level, we fitted a group of mixed-effects model to model the change in Verbal Ability and Understanding Directions (separate models) over time. The secondary analysis was adjusted according to the child weights provided in the data set to account for the number of children each participant represents in the population (Tayler et al., 2016).

Item Response Theory (IRT) Model

Up to four repeated observations of each classroom were used in the calibration to yield item parameters. We estimate a multidimensional one parameter item response model (1PL) – the many facets model (facet model; Linacre, 1993), an extension of the Partial Credit Model (PCM; Masters, 1982). An appropriate latent regression model (or population model) is fit to ensure that the extracted plausible values are appropriate for secondary analysis (Davier et al., 2009; Marsman et al., 2016). All item response models are estimated using ACER ConQuest (Adams, Wu, et al., 2020). For details on the estimation routines, see Adams, Cloney, et al. (2020). The full derivation of the IRT models can be found in supplementary material (Appendix A).

Secondary Analysis: Mixed-Effects Models of Children’s Growth, Given Classroom Quality

MODEL 1: In the secondary analysis, a set of growth models are built up, starting from an “empty” model and then adding predictors, including measures of classroom quality, and contextual factors.

Model 1 is a random intercept, fixed slope effects (growth) model mixed:

Level 2:

${\pi }_{0\mathit{i}}={\beta }_{00}+u_{0\mathit{i}}$

Level 1 (substituting level 2 into level 1):

$Y_{\mathit{ti}}={\beta }_{00}+{\beta }_{10}\mathit{tim}{e}_{\mathit{ti}}+u_{0\mathit{i}}+{\epsilon }_{\mathit{ti}}$

where $Y_{\mathit{ti}}$ is the developmental outcome (either VA or UD) of child $\mathit{i}$ at time $\mathit{t}$, ${\beta }_{00}$ is the grand mean at time 0, ${\beta }_{10}$ is the fixed effect for time (linear) where all $\mathit{i}$ have some deviation from the grand mean at time 1 (that is, time is months between assessments, with the date of the first assessment in the data coded to 0). and ${\epsilon }_{\mathit{ti}}$ is the level 1 (random) residual variance $\sim \mathit{N}\left(0,{\sigma }^2\right)$.

MODEL 2: We then added a set of predictors, on the random intercept term that are taken from our CLASS analysis (that is, estimate the effect of some set of CLASS predictors on the intercept) and on the main effect (fixed effect) for time (average growth):

Level 2:

${\pi }_{0\mathit{i}}={\beta }_{00}+\sum _{\mathit{k}=1}^m{\beta }_{0\mathit{k}}{x}_i^k+u_{0\mathit{i}}$

Level 1 (substituting level 2 into level 1):

$Y_{\mathit{ti}}={\beta }_{00}+\sum _{\mathit{k}=1}^m{\beta }_{0\mathit{k}}{x}_i^k+{\beta }_{10}\mathit{tim}{e}_{\mathit{ti}}+\sum _{\mathit{k}=1}^m{\beta }_{1\mathit{k}}{x}_i^k\mathit{tim}{e}_{\mathit{ti}}+u_{0\mathit{i}}+{\epsilon }_{\mathit{ti}}$

where $\sum _{\mathit{k}=1}^m{\beta }_{0\mathit{k}}{x}_i^k$ represents a set of $\mathit{m}$ fixed regression effects at level 2 (e.g., CLASS scale scores)

MODEL 3: In addition to the CLASS predictors, in model 3, we added a set of level 2 predictors relating to child, family, home, ECEC, and community characteristics. This is a fully conditional model. Note that this is the same specification as above, simply with a greater number of predictors in the model (that is, the m predictors in model 2 is a subset of the m predictors in model 3).

All secondary analysis was conducted in R (R Core Team, 2020), using the library lme4 (Bates et al., 2015). Robust standard errors are obtained using merDeriv to account for clustering within classrooms (Wang & Merkle, 2018). Visualization is done with the ggplot2 library (Wickham, 2016).

Results

Descriptive Statistics and Correlation of CLASS Dimensions

Of the original E4Kids cohort who provided data in 2010 (n = 2,487), 2,038 (81.95%) completed a direct assessment in 2011 and were included in the current analyses. Table 1 shows the descriptive characteristics of children who were provided data at both time points compared to those who did not. Services included in the current analyses had on average higher SEIFA scores than those who were recruited but were not included in the current analyses (1025.15 versus 998.31). The average family SEIFA score was also higher for those included in the current analyses compared with those not in the current analyses (1021.60 versus 995.05). A slightly higher proportion of those not included had LOTE as their main language compared with those included in the current analyses (8.59% versus 5.11%).

Table 1. Descriptive statistics for child and family-level and classroom-level variables.

Variable	N = 2038 Included	N; min-max	N = 449 Not included	N; min-max
Classroom level
Service type; n (%)		2028		447
Family day care (FDC)	72 (3.55)		28 (6.26)
Kindergarten (K)	696 (34.35)		122 (27.29)
Long day care (LDC)	1,250 (61.34)		292 (65.32)
Occasional childcare (OCC)	10 (0.49)		5 (1.12)
Group size; Mean (SD)	21.43 (5.13)	2013; 3–30	22 (6.16)	441; 4–30
Verbal Ability 2010; Mean (SD)	451.21 (13.54)	1754; 407–495	446.72 (14.05)	369; 407–483
Understanding Directions 2010; Mean (SD)	454.16 (16.03)	1,760; 400–495	450.02 (17.14)	369; 399–481
Verbal Ability 2011; Mean (SD)	463.88 (14.63)	1,994; 400–506	-	-
Understanding Directions; Mean (SD)	460.08 (14.21)	1,967; 407–500	-	-
Service SEIFA; Mean (SD)	1025.15 (77.78)	2014; 848–1175	998.31 (75.27)	435; 848–1167
Child and family level
Child sex, female; n (%)	990 (48.58)	2038	206 (45.88)	449
LOTE is main language; n (%)	87 (5.11)	1704	25 (8.59)	291
Age at first assessment (months); Mean (SD)	48.39 (6.83)	1834; 25–72	47.45 (7.46)	386; 26–73
Hours per week in informal care 2009	5.77 (10.10)	1153; 0–97	5.40 (8.74)	185; 0–60
Hours per week in formal care 2009	19.62 (15.47)	1228; 0–90	17.23 (13.85)	202; 0–56
Frequency of reading to child; Mean (SD)	5.36 (1.94)	1475; 0–7	4.96 (2.10)	239; 0–7
Family SEIFA; Mean (SD)	1021.60 (75.60)	1995; 802–1193	995.05 (73.77)	436; 841–1214
Maternal mental health: Mean (SD)	14.52 (4.31)	1448; 10–41	15.36 (5.28)	235; 10–36
Total difficulties on the SDQ: Mean (SD)	12.47 (3.26)	1442; 5–31	12.50 (3.42)	230; 5 27
Prosocial skills on the SDQ: Mean (SD)	7.93 (1.81)	1455; 0–10	7.79 (1.76)	233; 2–10

Note. SEIFA = Socioeconomic Indexes for Areas, Index of Relative Socio-economic Advantage and Disadvantage; LOTE = Language Other Than English; SDQ = Strengths and Difficulties Questionnaire.

Almost two-thirds of children in the included sample were attending long day care (61.34%), with just over a third attending kindergarten (34.35%). The average group size of ECEC rooms the children were attending was 21 (range: 3–30). The proportion of female (48.58%) and male children was similar, and the average age of children at their first assessment was 48 months (range: 25–72 months). Most maternal psychological distress scores were in the low to moderate range (10–21), with 7.31% (123/1683) mothers reporting high to very high levels of psychological distress. Children’s average total difficulties score on the SDQ were in the low range (M = 12.47, SD: 3.26), while the average prosocial skills score was in the high range (M = 7.93, SD: 1.81).

The average (SD; min–max) raw scores across the four observations for each CLASS dimension were as follows: Positive Climate: 5.07 (1.32; 1.75–7); Reverse Negative Climate: 6.38 (0.87; 2.25–7); Teacher Sensitivity: 4.56 (1.25; 1–6.75); Regard for Student Perspectives: 4.66 (1.25–7); Behavior Management: 4.87 (1.25; 1.75–7); Productivity: 5.11 (1.10; 2–7); Instructional Learning Formats: 3.87 (1.12; 1–6.25); Concept Development: 1.98 (1.01; 1–5.5); Quality of Feedback: 2.62 (1.15; 1–5.75); and Language Modeling: 2.67 (1.14; 1–6). Correlations between the 10 CLASS dimensions are presented in Table 2. Moderate to strong correlations are evident between most of the 10 CLASS dimensions. The weakest correlations were found between Negative Climate and the Instructional Support dimensions of Concept Development, Quality of Feedback and Language Modeling (r = 016, 0.36, and 0.30, respectively).

Table 2. Latent correlation matrix of CLASS dimensions.

Dimension	1	2	3	4	5	6	7	8	9	10
(1) Positive Climate		1.15	1.84	1.00	1.47	0.86	1.12	1.23	1.35	0.96
(2) Reverse Negative Climate	0.72		0.78	0.42	1.01	0.64	0.18	0.38	0.42	0.35
(3) Teacher Sensitivity	0.96	0.61		0.97	1.21	0.76	0.99	1.01	1.23	0.84
(4) Regard for Student Perspectives	0.68	0.43	0.81		0.66	0.48	0.54	0.51	0.76	0.54
(5) Behavior Management	0.80	0.82	0.82	0.58		0.93	0.58	0.70	1.01	0.58
(6) Productivity	0.64	0.72	0.70	0.58	0.91		0.34	0.42	0.64	0.44
(7) Concept Development	0.65	0.16	0.71	0.50	0.43	0.35		1.08	1.35	0.65
(8) Quality of Feedback	0.77	0.36	0.78	0.52	0.57	0.48	0.93		1.27	0.64
(9) Language Modeling	0.64	0.30	0.72	0.58	0.63	0.54	0.88	0.90		0.73
(10) Instructional Learning Formats	0.85	0.46	0.91	0.76	0.67	0.70	0.78	0.84	0.72
Variance	2.37	1.07	1.56	0.91	1.41	0.75	1.27	1.07	1.87	0.54

Note. All correlations are p < .05.

Figure 1 shows the distribution of CLASS dimensions (EAP is the expected a posteriori estimator, which can be thought of as the mean of the plausible values), grouped by the domains of Emotional Support, Classroom Organization, and Instructional Support. The dimensions of Concept Development, Quality of Feedback, and Language Modeling, which make up the Instructional Support domain, were on average, much lower than the dimensions in each of the other domains. This was also reflected in the average raw scores for each dimension, reported earlier.

Figure 1. Distribution of CLASS by the domains of emotional support, classroom organisation and instructional support.

Note: EMOT = Emotional Support; CORG = Classroom Organization; INST = Instructional Support; EAP = expected a posteriori estimator; PC = Positive Climate; RNC = Negative Climate; TS = Teacher Sensitivity; RSP = Regard for Student Perspective; BM = Behavior Management; PD = Productivity; ILF = Instructional Learning Formats; CD = Concept Development; QF = Quality of Feedback; LM = Language Modeling.

Linear Mixed-Effects Models to Examine Associations Between CLASS Dimensions and Child Language Outcomes

Tables 3 and 4, and Tables B.1 to B.18 in Appendix B show the results of linear mixed-effects modeling to predict later language outcomes whereby CLASS dimensions are included in Model 2 and 3 as the primary predictor (key research question). All dimensions were positively associated with children’s initial average language abilities (both Understanding Directions and Verbal Ability) after adjusting for classroom-level variables (service type, group size, classroom verbal ability, service SEIFA). However, estimates were small, ranging from 0.83 to 2.54 W units. Results show a small negative effect of Positive Climate, Negative Climate, Teacher Sensitivity, Regard for Student Perspective, and Instructional Learning Formats on children’s growth scores for Understanding Directions. No effects were evident for any of the dimensions on children’s growth scores for Verbal Ability.

Table 3. Association between positive climate and understanding directions.

Predictors	Base model		+ service-level predictors		+ family/child predictors
Positive climate	Coefficient (S.E.)		Coefficient (S.E.)		Coefficient (S.E.)
Intercept	453.00***	(0.31)	448.35***	(2.93)	453.20***	12.89
Year	10.40***	(0.34)	14.88***	(3.28)	−3.87	13.75
Months between assess.	0.27*	(0.13)	0.37**	(0.14)	0.42*	0.19
Age	0.89***	(0.04)	0.84***	(0.04)	0.85***	0.05
Group size			0.16*	(0.07)
Service SEIFA			6.11***	(1.02)
Dimension PV			0.83***	(0.24)	0.64*	0.30
Dimension PV slope			−0.52.	(0.27)	−0.77*	0.35
LDC service type slope			−5.35.	(3.22)
Gender					−1.88*	0.82
First language ≠ English					−21.35.	12.70
Prosocial score					0.49*	0.23
Family SEIFA					7.37***	1.38
Family SEIFA slope					−3.15.	1.61

*** = p = 0; ** = p < .001; * = p < .01; . = p < .05.

Note. only variables with a significant effect are retained in the table. For the full list of variables in the model refer to Table 1. SEIFA = Socioeconomic Indexes for Areas, Index of Relative Socio-economic Advantage and Disadvantage; PV = Plausible Value; LDC = Long Day Care.

Table 4. Association between positive climate and verbal ability.

Predictors	Base model		+ service-level predictors		+ family/child predictors
Positive climate	Coefficient (S.E.)		Coefficient (S.E.)		Coefficient (S.E.)
Intercept	449.80***	(0.26)	446.79***	(2.41)	436.40***	10.57
Year	9.93***	(0.24)	9.86***	(2.31)	12.62	10.06
Months between assess.	0.75***	(0.10)	0.82***	(0.10)	0.75***	0.14
Age	0.95***	(0.03)	0.90***	(0.03)	0.94***	0.04
Dimension PV			0.93***	(0.20)	0.66*	0.26
Service SEIFA			6.50***	(0.86)
Service SEIFA slope			1.42.	(0.81)
Proxy measure of literacy					−1.70***	0.19
Family SEIFA					5.70***	1.19
Previous formal ECEC use					−0.04.	0.02
Prosocial score					0.57**	0.20
Previous informal ECEC use slope					0.06.	0.03
Maternal mental health slope					0.23**	0.09

*** = p = 0; ** = p < .001; * = p < .01; .= p < .05.

Note. only variables with a significant effect are retained in the table. For the full list of variables in the model refer to Table 1. PV = Plausible Value; SEIFA = Socioeconomic Indexes for Areas, Index of Relative Socio-economic Advantage and Disadvantage; ECEC = Early Childhood Education and Care.

In the fully conditional model including the CLASS predictors as well as the set of predictors relating to child and family characteristics, Positive Climate, Teacher Sensitivity, Regard for Student Perspective, Behavior Management, Instructional Learning Formats, Concept Development, Quality of Feedback, and Language Modeling were associated with children’s initial average Understanding Directions scores.

The same set of dimensions, excluding Language Modeling and including Negative Climate, were associated with children’s initial average Verbal Ability scores. Estimates for all dimensions and language outcomes, while significant, were small.

After adjusting for child and family characteristics, Instructional Learning Formats (β = 1.32, p < .05) was positively associated with average growth in Understanding Directions. Positive Climate (β = −0.77, p < .01), Teacher Sensitivity (β = −1.03, p < .01) and Regard for Student Perspectives (β = −1.60, p < .01) were negatively associated with average growth in Understanding Directions. There was no evidence that any of these dimensions affected average growth in Verbal Ability.

For Understanding Directions, family SEIFA score (indication of SES) was negatively associated with average growth scores (β =−3.29 to −2.97). Long day care service type was negatively associated with average growth in Understanding Directions, but only in the models where Positive Climate and Negative Climate were the primary predictors. Previous ECEC use and maternal mental health were both positively associated with average growth in Verbal Ability, but estimates were small.

Discussion

Process quality, the interactions children experience within ECEC settings, is expected to support children’s learning and development, including language skills. The CLASS is one of the most widely used tools for measuring process quality in ECEC. However, when exploring the relationship between CLASS measured quality and outcomes, researchers often include the CLASS domains, rather than the individual dimensions. We hypothesized that specific CLASS dimensions would predict children’s language outcomes, with those dimensions from the Instructional Support domain being most strongly predictive and providing insight into the strategies that educators could focus on to support children’s language learning.

In this large, Australian sample of preschool-aged children, all CLASS dimensions were associated with initial Verbal Ability and Understanding Directions scores, but these attenuated once child and family factors were included in the models. When examining growth in child language outcomes, CLASS dimensions had little or no effect on the growth of Understanding Directions scores, and there was no evidence of an effect of CLASS dimensions on the growth of Verbal ability scores. Results showing that the CLASS dimensions, and specifically those that were developed to capture interactions that support language growth (i.e., Concept Development, Quality of Feedback, and Language Modeling), were not associated with growth in children’s language outcomes align with recent studies demonstrating null findings (Justice et al., 2018; McDoniel et al., 2022).

In contrast, of the few studies that have looked specifically at the associations between individual CLASS dimensions and language outcomes, Curby and Chavez (2013) found Positive Climate, Productivity, and Concept Development predicted children’s language skills (receptive vocabulary and understanding and use of spoken language). While the Curby and Chavez (2013) study included an older version of CLASS that did not include Language Modeling, as with other international studies, the average Instructional Support domain scores were low. Although these three dimensions (Positive Climate, Productivity, and Concept Development) were found to predict better language skills, the size of effect was small (e.g., the average PPVT was 99, while the estimate was −0.72) and potentially lacks practical relevance (Thorpe et al., 2020). A similar result was found in the current study, i.e., Positive Climate was one of the few dimensions associated with growth in Understanding Directions, but again, the magnitude of the estimate was small (estimate of −0.77, with average UD score of 453, SD = 16).

As with previous studies, the current study also illustrates relatively low quality for the dimensions that make up the Instructional Support domain (Hu et al., 2016; Stuck et al., 2016). In their meta-analysis of CLASS as a measure of educator–child interaction quality in ECEC settings and child outcomes, Perlman et al. (2016) demonstrated that of the CLASS domains, Instructional Support was most closely linked with child outcomes but was also the lowest scoring domain across studies included in their analysis. While Slot (2018) found that all CLASS domains were associated with children’s language outcomes, Instructional Support was the weakest predictor, with a low average score with limited variation. It may be that to achieve significant benefits to child language outcomes, the change in educator practice during educator–child interactions needs to reach minimum levels of intensity and be sustained over time (Eadie et al., 2019). This is further supported by findings from a meta-analysis demonstrating that Instructional Support is related to larger gains in language and literacy skills in higher versus lower-quality classrooms (Burchinal et al., 2016).

While studies have attempted to identify minimum thresholds for Instructional Support quality required to impact child outcomes (e.g., Burchinal et al., 2010), a minimum threshold for quality has not consistently been identified. Further research examining threshold conditions, and within different contexts, such as mid- to high-quality classrooms, rather than just high versus low, is necessary (McDoniel et al., 2022). This is demonstrated in the recent Rankin et al. (2022) study, illustrating that increases in Emotional Support over time were associated with improved child language skills, supporting the notion that the greatest benefits for children’s outcomes are achieved when children are exposed to higher thresholds of process quality that are sustained over time. However, it is important to note that unlike the current study, the Rankin et al. (2022) study used CLASS raw scores, which do not account for measurement errors.

As CLASS is a global measure and dimensions are ratings on a Likert scale, it may be that taking a more fine-grained approach to measuring responsive educator–child interactions could provide further insight into specific educator behaviors that predict children’s language skills. A fine-grained approach, such as detailed coding of educator–child conversations, is potentially more sensitive to capturing the impact of language-promoting educator–child interactions on children’s language development (Cabell et al., 2015; Houen et al., 2019). For example, CLASS does not capture educator’s linguistic responsivity (e.g., facilitating peer-to-peer communication or using a slow conversation pace) or information-providing features of educators’ talk (e.g., educators’ sentence length) (Guerrero-Rosada et al., 2021; Justice et al., 2018). It may also be that the context in which the educator–child interactions are taking place needs to be accounted for, given that clusters of particular educator language practices may be more or less effective in supporting children’s language acquisition, depending on contextual factors, such as activity settings (i.e., book-reading, play, and mealtime) and group size (Hadley, Barnes, & Hwang, 2022; Hadley, Barnes, Wiernik, et al., 2022).

The finding in the current study that CLASS dimensions were associated with initial ability but were not predictive of growth in child language skills may be a result of higher achieving children enrolled in higher-quality rooms at baseline, i.e., there is a selection effect. Children from low socioeconomic (SES) backgrounds often access lower-quality ECEC and the socioeconomic gradient is most apparent for Instructional Support (Cloney et al., 2016). Therefore, children from low SES backgrounds are likely to be missing out on the high-quality, responsive educator–child interactions that support children’s language development. If CLASS is highly related to initial ability (that is, high achieving children are in higher-quality rooms at the start of the study, and these children tend to come from the most advantaged backgrounds), this is an important finding that highlights the issue of equity.

While the policies and frameworks of ECEC globally acknowledge the importance for all children to have access to high-quality ECEC programs, evidence suggests that this is often not the case, with children from less disadvantaged backgrounds more likely to access higher-quality ECEC programs (Alexandersen et al., 2021). Further research into barriers to access for families’ from low SES backgrounds, as well as investing in professional learning for those staff working in services in low SES areas is needed. Further research examining the interaction effects of family SES status and quality in predicting children’s language skills could also be explored further, given research shows that children experiencing higher levels of social disadvantage are likely to show the most gains from exposure to high-quality ECEC (Burchinal et al., 2002). Given that lower ECEC quality is over-represented in lower SES areas, conducting research on high-quality services in pockets of disadvantage could help delineate further those aspects of educator–child interactions and language teaching that most influence the language abilities of young children experiencing disadvantage.

Strengths and Limitations

The strength of the current study is that it is based on a large sample of Australian children. As such, our results are likely to be fairly representative of preschool children who do attend ECEC in Australia. Moreover, although CLASS is widely used, its validation is inconsistent. Validation studies have often relied on complex models, including models with correlated residual variances to meet criterion measures of fit (Cloney et al., 2017; Pakarinen et al., 2010) and large within-classroom variations have been observed, including linear declines in observed quality over repeated observations (Cloney et al., 2017). Recent work has also applied more complex scoring models to yield quality scores than have been used in the past (traditionally, raw scores averaged across repeated observations are used assuming no uncertainty due to measurement error and potentially biasing estimates) (Styck et al., 2021).

Our approach to analyzing CLASS by creating plausible values allowed us to overcome these limitations typical of studies associated with CLASS scores. The results obtained are thus more likely to be representative of classroom quality by reducing the risk of bias. The reduced risk of bias also allows for better generalization of the data to a larger population and improves the robustness of the results and findings. This paper also demonstrates an appropriate way of conditioning in a latent regression model within the IRT framework to ensure that secondary analysis (e.g., the mixed-effects models presented here) are unbiased. Analysis using factor scores (e.g., MLEs or EAPs) has been shown to yield biased variance estimates (Wu, 2005).

This study also made innovative contributions in applying an IRT model to the CLASS data. This is an important contribution, as the distribution of the CLASS domains is highly variable: it is much more likely that a classroom will be rated high on Positive Climate than on Language Modeling. In the IRT framework, this is captured by showing that it is much more difficult, all else equal, for a classroom to demonstrate the aspects of quality explained by the first few categories of the Likert scale for Language Modeling than it is Positive Climate: even though both items are rated 1, 2, 3, and so on, a middling score on the former is much more likely to differentiate a classroom as being of quality than the latter.

It is also important in a statistical sense, as analysts typically want to use linear methods (e.g., regression and correlation) with CLASS data and averaging raw scores from Likert scales cannot be described as being true interval measures. Further, extraneous factors like the increasing difficulty of demonstrating quality over the day go un-modeled in studies that use raw scores. In a study where the number of repeated observations is not constant for all classrooms, those classrooms with fewer observations are likely to be erroneously rated higher.

There were also limitations, including that this is a nonrandom design where quality and outcomes are observed as they happen in the everyday market – given the strong socioeconomic gradient, there are few observations of the most vulnerable children receiving the highest quality programs (and indeed, on average all the programs observed are low to medium quality). Furthermore, the socio-economic areas of both the ECEC services and families included in the current analyses were less disadvantaged compared to those services and families who were part of the original sample. Results from the current study may not be representative of ECEC services, families, and children from highly disadvantaged areas.

There are limitations in the use of the Woodcock Johnson subscales as a measure of children’s language abilities. However, the E4Kids study was a large community-based study, focused on language development, which required the inclusion of measures that could capture a range of developmental outcomes without being too time intensive. While they are not exhaustive measures of language, Verbal Ability captures expressive language, including vocabulary and syntax, and Understanding Directions is a receptive language measure. The included tests are also appropriate for use with young children as they do not require any reading.

Conclusion

Findings from this study illustrate that, typically, ECEC programs rate low on dimensions of quality developed to capture language-promoting educator–child interactions. It may be that complementary measures of educators’ use of language promoting strategies are necessary, as a way of monitoring implementation of strategies and for self-reflection. There is a need for measures that can capture specific teacher practices, as well as children’s individual experiences in ECEC settings (Guerrero-Rosada et al., 2021). Further understanding of specific features of teacher talk might guide the aspects of practice that need to be the focus of professional learning to enhance children’s language development. Perhaps most importantly, it is critical to pursue efforts to ensure equitable access to quality ECEC services for all children, and to ensure lifelong impacts on educational success.

Acknowledgments

The authors acknowledge and express their gratitude to all those involved in the E4Kids study. E4Kids was a project of the Melbourne Graduate School of Education and was conducted in partnership with Queensland University of Technology. E4Kids was funded by the Australian Research Council Linkage Projects Scheme (LP0990200), the Victorian Government Department of Education and Early Childhood Development, and the Queensland Government Department of Education and Training. E4Kids was conducted in academic collaboration with the University of Toronto Scarborough, the Institute of Education at the University of London and the Royal Children’s Hospital in Melbourne.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Supplementary Data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/10409289.2023.2193857

Additional information

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.