The effect of ECEC process quality on school performance and the mediating role of early social skills

ABSTRACT

Many studies have linked quality in early childhood education and care [ECEC] to school performance, but the mechanisms of how ECEC process quality affects children in ways that lead to improved school performance is unclear. In this study on 7431 children in Norway, we test the hypothesis that the relation between process quality in ECEC and later school performance is mediated by early social skills. Process quality was measured at age 5 years along two dimensions: ‘pedagogical practices’ and ‘structured activities’, and school performance was measured at age 11 years (5th grade) using mandatory national tests in math, reading, and English. The results show that the indirect effect of pedagogical practices on school outcomes through social skills was small but statistically significant. In contrast, there was no statistically significant indirect effect of structured activities on school performance through social skills.

KEYWORDS:

• ECEC
• child development
• social skills
• school performance
• MoBa

The vast majority of children within OECD countries attend early childhood education and care [ECEC] prior to primary school (OECD 2021b). Many studies have found positive effects of ECEC on children’s later school performance but that quality of care is of critical importance (see meta-analysis of natural experiments by Huizen and Plantenga 2018). However, less is known about the mechanisms of how quality affects children in ways that ultimately lead to increased school performance. One hypothesis is that attending high quality ECEC can lead to increased social skills that promote school readiness and in turn improve children’s later school performance.

The potential mediating role of early social skills has been indicated previously. For instance, social competence has been found to mediate between social understanding and later school achievement (Lecce et al. 2017), as well as mediate between teacher-instruction and academic outcomes (Ansari and Gershoff 2015). Similarly, self-regulation and pro-social behavior have also been suggested to mediate the relation between process quality in ECEC and academic outcomes (Sylva et al. 2020). Consequently, promoting social skills in the early years, apart from being important in and of itself, may be positive for children’s academic performance and later labor market participation (Deming 2017), in line with models of skill formation, where the acquisition of early skills lays the groundwork for later learning (Cunha and Heckman 2007a; 2007b).

Children are born social, as evident from the innate preference for faces over other sources of potential interest – a preference that increases throughout infancy (Frank, Vul, and Johnson 2009; Mondloch et al. 1999). As they grow, their social skills become increasingly sophisticated and elaborate. Social skills are manifestations of social-cognitive capacities which seemingly interlink with nearly all functions of the brain (Beer and Ochsner 2006). The tight integration of social and cognitive functions likely evolved in response to demands arising from living in groups which were built upon complex collaboration (Beer and Ochsner 2006). Social skills develop throughout the early years of life, a period marked by high brain malleability (Cantora et al. 2019). Consequently, early childhood is a highly sensitive period in children’s development where they are particularly susceptible to influence from their social environments (Fox and Levitt 2010; Nelson, Zeanah, and Fox 2019). During this sensitive period, many children spend a significant amount of time in ECEC. Since much of this time is spent interacting and playing with peers and staff (Storli and Sandseter 2019), ECEC could exert a strong influence on children’s social development, depending on the quality of care provided (Huizen and Plantenga 2018; Melhuish et al. 2015). Quality in ECEC is typically conceptualized along three dimensions: process quality, structural quality, and content quality. Process quality relates to the quality of the interaction between caretakers and the children (Edwards 2021); structural quality relates to the resources spent to ensure process quality (e.g. group size, teacher/child-ratios & teacher qualifications); and content quality relates to the substance of what is being learnt during daily activities (e.g. curriculum) (OECD 2011). While structural quality is seen as a prerequisite for process quality (Slot et al. 2018), process quality is thought to be the primary educational driver of social skills during the early years (Sylva et al. 2020).

Several aspects of early social skills have been linked to better school performance, such as approaches to learning (e.g. attentiveness, flexibility, organization), self-control (e.g. controlling temper, accepting peer ideas for group activities, responding appropriately to pressure), and interpersonal skills (getting along with people who are different, and comforting or helping other children). Conversely, higher levels of internalizing (e.g. anger, impulsivity) and externalizing (e.g. anxiety, loneliness, low self-esteem) problems have been shown to be inversely linked to academic performance (see Sung and Chang 2010). However, the overall empirical evidence is somewhat mixed, as others have found either only indirect effects of social skills on school performance by buffering against the adverse effects of mental health issues (Panayiotou, Humphrey, and Wigelsworth 2019), or no link between social skills in ECEC and later school performance (Claessens, Duncan, and Engel 2009; Duncan et al. 2007; Pagani et al. 2010).

It is also unclear to what degree ECEC influences social skills. Studies tend to favor parts of process quality relating to teacher–child interactions being most predictive of children’s social skills, such as teachers being emotive when interacting with children (Broekhuizen et al. 2016; Mashburn et al. 2008), or showing engaged support for learning (Salminen et al. 2022). One Norwegian study, however, found that process quality only had modest positive effects on some aspects of social skills in the short-term (age 3 and 5 years) (Løkken et al. 2018). Furthermore, lower quality ECEC may also have a negative impact on children’s social skills. For instance, a Finnish study found that children in presumably lower quality ECEC often report experiencing negative peer-interactions, feeling lonely and lacking friends, or opposition to forced daily routines or activities (Pihlainen et al. 2020).

The present study

In this study, we test the hypothesis that the relation between ECEC process quality and school performance at age 11 (fifth grade) is mediated by children’s social skills in ECEC. We do this on a sample of 7431 Norwegian children attending ECEC in a context of (near) universal access. Norway provides a good opportunity to assess the importance of developing social skills on school performance for several reasons. First, Norwegian ECEC places a strong focus on social development and play-based learning in the national framework plan which governs the tasks and content in ECEC centers (Norwegian Directorate for Education and Training 2017). Norwegian ECEC is therefore strongly rooted in a European tradition of ECEC with a holistic approach to learning (The Norwegian Directorate for Education and Training 2017), as opposed to having more traditional curriculums with specific learning outcomes. Second, Norwegian children attend ECEC from a very young age. The vast majority (81%) enroll before they are 2 years old (The Norwegian Directorate for Education and Training 2021), providing long exposure to ECEC settings. Third, structural quality is high. Norwegian ECEC child groups are led by teachers trained in ECEC pedagogy (bachelor’s degree or higher), with a minimum requirement of 1 ECEC teacher per 7 children for children under 3 years, and 1 ECEC teacher per 14 children over 3 years (Forskrift om pedagogisk bemanning og dispensasjon i barnehager 2017). Norway also has among the highest per child expenditures for children aged 0–6 in ECEC of all countries – spending over twice the OECD average (OECD 2021a). Although Norwegian – and Nordic ECEC in general – has high structural quality, a study using standardized measurement tools (ITERS-R) to assess observed quality found overall quality to be variable and moderate on average, at least for infants and toddlers (Bjørnestad and Os 2018). Process quality has, however, been found to be adequate to good for basic interactions relating to sensitivity, responsiveness, respect of autonomy, structuring, and limit setting, but inadequate for educational interactions relating to verbal communications, developmental stimulation and fostering positive peer interactions (Baustad and Bjørnestad 2022; Bjørnestad et al. 2020).

We test our hypotheses using structural equation models, modeling social skills as a latent variable, with national test scores in mathematics, reading, and English in fifth grade (age 11 years) as outcomes.

Methods

Sample and procedure

We used data from a sub-cohort of participants in the Norwegian Mother, Father and Child Cohort (MoBa). MoBa is a prospective, nationwide, population-based, pregnancy cohort, and in the sub-cohort of children that we study, questionnaire data were collected from preschool teachers and linked with registry data at child age 11 from the National Education Database (NUDB), collected by the Norwegian Directorate for Education, with access provided by Statistics Norway (SSB). The study is conducted by the Norwegian Institute of Public Health (Magnus et al. 2016). Women pregnant in their second trimester were recruited from all over Norway from 1999 to 2008, with a participation rate of 40.6%. Written informed consent was obtained from all participants. In total, the full cohort includes 114,500 children, 95,200 mothers and 75,200 fathers. Participants have been followed up by questionnaires (in Norwegian) during pregnancy and after birth. Pregnancy and birth records in MoBa have been obtained from the Medical Birth Registry of Norway (MBRN) (Irgens 2009). When the children of participating families had turned 5 years, ECEC teachers of the children born between 2006 and 2009 were invited to evaluate the ECEC quality as well as the children’s functioning and development in an ECEC questionnaire (Q-Cc). The teacher response rate was approximately 40%. Our analytic sample included 7431 children, born between 2006 and 2009 whose ECEC teacher returned the Q-Cc, and with available linkage to mandatory national test scores.

We used the tenth version of the quality-assured dataset, which was released for research in 2017 (Norwegian Institute of Public Health 2022). The research project is approved by the Regional Committees for Medical and Health Research Ethics (REK) (2015/1324). The MoBa cohort is regulated by the Norwegian Health Registry Act. (for complete details, see Magnus et al. 2016; Norwegian Institute of Public Health 2022).

Measures

School performance. For school performance, we used test results from Norway’s National Education Database in 5th grade (ca. age 11 years). Tests are mandatory with 96% of all students in Norway taking them (students with special needs and those following introductory language courses may be exempt). The national tests were first introduced in primary school in 2004, and subsequently revised and modernized in 2014 through the use of Item Response Theory to ensure measurement validity (see Bjørnsson 2018; Yang and Kao 2014). The tests are intended to reflect children’s basic skills within three different domains: math, reading and English, in accordance with the national competency goals. The tests are standardized, from 0-100, with a mean of 50 and a standard deviation of 10. Test scores are normally distributed (Bjørnsson 2018).

Social skills at age 5 years. Children’s social skills at age 5 years was modeled as a latent variable from 9 ECEC teacher-rated questionnaire items. The items were selected from the School Readiness Questionnaire (SRQ) (Prior et al. 2000). We included the following teacher-rated items on the question ‘how do you find the child is coping in the following areas?’: ‘settling into the child-care center, ‘co-operation with other children’, ‘relationship with teacher’, ‘confidence’, ‘speaks in groups of children, ‘follows instructions’, ‘agreeableness’, ‘adaptation to child-care center’. Items were rated on a 5-point scale, from 1 = ’has considerable difficulties’, to 5 = ’very well’. The social skills at age 5 years measure exhibited good to acceptable fit on the following robust goodness of fit indicators: CFI = 0.959, TLI = 0.945, RMSEA = 0.146, SRMR = 0.067.

ECEC process quality. ECEC quality was modeled along 2 dimensions: pedagogical practices and structured activities. These are related (r = 0.338, p < 0.001) but measure different aspects of process quality during ECEC. One important consideration was to only include variables on process quality that did not relate specifically to children individually, but rather the entire group. We did this to avoid unwanted correlation between ECEC process quality and social skills, considering that teachers rated both.

Pedagogical practices relates mostly to the pedagogical choices that are made during daily activities such as during free play and how staff engages and interacts with the children. It was modeled as a latent variable from the following 9 teacher-rated items on whether ‘the children can mostly play undisturbed’, ‘the adults are actively looking for opportunities to guide the children in play’, ‘children take part in decisions when we plan projects in the child-care center’, ‘children take part in planning daily activities in the child care center, ‘if the children’s play is very good, we drop planned activities, ‘we have a strong focus on giving the children the knowledge they need’, ‘play groups are initiated by children themselves, ‘we challenge the children by sometimes facilitating activities that are more difficult than what they are used to’, and ‘we prioritize good conversations with the children even at the expense of other planned activities’. The items had a 6-point scale from 1 = ‘very uncommon practice’ to 6 = ‘very common practice’. The pedagogical practices measure exhibited good to acceptable model fit on the following robust goodness of fit indices: CFI = 0.947, TLI = 0.924, RMSEA = 0.075, and SRMR = 0.042.

Structured activities relates to structured and mostly planned activities with specific learning-outcomes in mind. It was modeled as a latent variable and consists of 12 items from the Early Childhood Environment Rating Scale [ECERS-R] (Harms, Clifford, and Cryer 1988), as adapted by the The Behaviour Outlook Norwegian Development Study [BONDS], on how often the teacher facilitates learning related to ‘scribbling’, ‘exploring letters’, ‘practicing word pictures’, ‘writing whole words’, ‘exploring geometry, shapes, patterns or other mathematical concepts’, ‘understanding numbers’, ‘sensory-motor and physical play, ‘culture and distinctiveness’, ‘creative activities’, ‘outdoor activities focusing on environmental knowledge’, ‘playgroups focusing on role play’, and ‘computers [ICT]’. The items had a 5-point scale from 1 = ’once a month’ to 5 = ’daily’. The structured activities measure exhibited good to acceptable fit on the following robust goodness of fit indices: CFI = 0.898, TLI = 0.876, RMSEA = 0.116, and SRMR = 0.072.

Verbal ability. We included a measure of verbal ability as a proxy for cognitive development. The measure was modeled as a latent variable and included 8 items from the 20 Questions About Language Skills scale (Ottem 2009) on whether the child ‘forgets words s/he knows the meaning of’, ‘mixes up words with similar meaning’, ‘has difficulties in understanding the meaning of common words’, ‘has difficulties in answering questions as quickly as other children’, ‘is often searching for the right words’, ‘uses incomplete sentences’, ‘uses short sentences when s/he is answering questions’, and ‘has difficulties retelling a story s/he has heard’. The items had a 5-point scale from 1 =  ‘describes the child well / completely correct’, to 5 =  ‘does not describe the child well / completely incorrect’). The verbal ability measure also included 5 items from the Child Developmental Inventory [CDI] scale (Ireton 1992) on whether the child ‘talks in long, complex sentences, ten words or longer’, ‘uses plurals correctly, for example, says men, not mans, mice, not mouses’, ‘tells where s/he lives, naming the town or city’, ‘When asked “What is a … ?”, [s/he] talks about the group it belongs to. For example, “a horse?”: “an animal”, “an orange?”: “a fruit”’, and ‘uses the words today, yesterday, and tomorrow, correctly’. The CDI items had a scale of 1 = ’no’ and 2 = ’yes’. The verbal ability measure exhibited good to acceptable model fit on the following robust goodness of fit indices: CFI = 0.968, TLI = 0.962, RMSEA = 0.062, and SRMR = 0.08.

Mother’s educational attainment. We also included mother’s educational attainment as a covariate, which was measured during pregnancy (age 15 weeks). Educational attainment was measured ordinally on a 7-point scale but treated as a continuous predictor in our models.

Statistical approach

All analyses were conducted using R open-source statistical software package. We used confirmatory factor analysis to model latent factor variables that served as both our predictor measures. Due to the categorical (ordinal) nature of our questionnaire data, the models were fitted using the diagonally weighted least squares (DWLS) estimator which is robust to violations against multivariate normality (Li 2016). All latent variables were subsequently z-score standardized, with a mean of zero and a standard deviation of one, in all regression models to make them more easily interpretable given the differences in scales.

Missing data. In total, 26% of the cases contained at least some missing data. Missing values were therefore imputed using full information maximum likelihood (FIML), which has shown to both be more efficient than more common approaches to dealing with missing data, such as listwise deletion, and yielding unbiased estimates under ignorability assumptions (Enders and Bandalos 2001). FIML has also in many cases been shown to outperform other imputation techniques such as multiple imputation by chained equations (MICE) (Xiao and Bulut 2020).

Goodness of fit. Our measurement models were evaluated using root mean square error approximation [RMSEA], standardized root mean square residual [SRMR], comparative fit index [CFI], and Tucker-Lewis Index [TLI]. We followed recommended guidelines by Hooper, Coughlan, and Mullen (2008); however, we recognize that guidelines for fit indices for DWLS estimated models on categorical outcomes are less established and often inflated compared to maximum likelihood (ML) estimated models (see Nye and Drasgow 2011).

Mediation analyses. Structural Equation Modeling [SEM] is a framework for simultaneous testing of a set of structural equations such as regressions (see Stein et al. 2017). Using SEM, we specified a set of mediation models, testing the indirect effects of ECEC quality on school performance at age 11 through social skills at age 5 years. Mediation models are understood differently depending on which theoretical framework you adhere to. For instance, some have argued that a direct effect is a prerequisite for mediation: without a direct effect, no indirect effect should be tested for (e.g. Judd and Kenny 1981). However, such views have later been reconsidered (see O’Rourke and MacKinnon 2018), and arguably lack logical basis. As is easy to demonstrate, the absence of a direct effect can be the result of two competing indirect effects of the same magnitude, effectively canceling each other out. Alternatively, the total effect may be underpowered, while the indirect effect may not (Kenny and Judd 2014). Consequently, we did not estimate the direct effects prior to testing our fully specified structural equation mediation models. Although mediation models are causal hypotheses, causal inference is only valid under strict assumptions of ignorability, meaning no unmeasured confounders. In most cases, such as ours, mediation models are a means to decompose the variance in a relation between x and y through m (see Kaufman 2010).

Results

The sample consisted of 7431 children, of which 50.3% were male. Table 1 shows descriptive statistics of all included variables in the structural equation models. Figure 1 shows the specification of our SEM mediation model where each path represents an estimated regression coefficient. Specifically, we jointly estimated regression coefficients for the following paths: ECEC process quality (pedagogical practices & structured activities) → social skills → school performance (mathematics, reading & English), while adjusting for verbal abilities (serving as a proxy for cognitive development). We also allowed for residual direct effects between ECEC process quality and school performance, adjusting for the indirect effect through social skills (partial mediation).

Figure 1. Structural equation mediation model.

Table 1. Descriptive statistics of included variables.

	N = 7431
Variable	Summary statistics
Pedagogical practices (age 5 years)	M = 0.00 (0.294)
Structured activities (age 5 years)	M = 0.00 (0.612)
Social skills (age 5 years)	M = −0.024 (0.73)
Verbal ability (age 5 years)	M = −0.126 (0.60)
Maternal education	M = 6 (1.045)
Mathematics (age 11 years)	M = 53.86 (9.57)
Reading (age 11 years)	M = 53.85 (9.49)
English (age 11 years)	M = 51.72 (9.94)

Note: M = mean, with standard deviations in parentheses. Summary statistics of latent variables are shown unstandardized.

From our SEM model results (Table 2), we see that there were small but significant indirect effects of pedagogical practices through social skills on mathematics and reading, albeit not statistically significant for English. In contrast, there were no significant indirect effects of structured activities through social skills on any of the outcomes. There were also no significant total effects of either pedagogical practices or structured activities on either outcome.

Table 2. Regression models.

		Social skills (m)		Math (y1)	Reading (y2)	English (y3)
	Path		Path	β	β	β
Pedagogical practices	a1	0.137 (0.011)***	c’1	−0.017 (0.116)	0.005 (0.129)	0.005 (0.138)
Structured activities quality	a2	0.023 (0.011)	c’2	−0.011 (0.115)	−0.003 (0.132)	−0.000 (0.142)
Social skills			b	0.079 (0.121)***	0.096 (0.122)***	0.017 (0.131)
Verbal ability	u1	0.369 (0.011)***	u2	0.038 (0.011)***	0.039 (0.012)***	0.037 (0.012)
Mother’s educational attainment	u3	0.040 (0.010)***	u4	0.211 (0.104)***	0.189 (0.113)***	0.126 (0.121) ***
Total effect of Pedagogical practices			c’1 + (a1b)	−0.006 (0.115)	0.008 (0.128)	0.017 (0.133)
Total effect of structural quality			c’2 + (a2b)	−0.009 (0.115)	−0.005 (0.133)	−0.003 (0.140)
Indirect effect of Pedagogical practices			a1b	0.011 (0.017)***	0.013 (0.021)***	0.003 (0.018)
Indirect effect of structured activities quality			a2b	0.002 (0.009)	0.002 (0.012)	0.001 (0.004)

Note: XY-standardized regression estimates. Standard errors in parentheses. *p < 0.05; **p < 0.01; ***p < 0.001.

Discussion

In this study on the indirect effects of ECEC process quality on school performance through early social skills, we found that social skills was a significant mediator between pedagogical practices and school outcomes, but not between structured activities and school outcomes. There was no significant total effect between either of the quality measures on either of the school outcomes.

From the results, we gain some insight into the developmental pathways between ECEC process quality and later school performance through increased social skills. Given that pedagogical practices seem to be more important than planned activities for both social skills and later school outcomes, it might matter less what you do than how you do it during everyday life in ECEC. Although previous studies have not tested our mediation model specifically, the different parts of our model can nevertheless easily be compared to the existing literature. The first part of our model, relating process quality to social skills, has previously been investigated with inconsistent findings, where some have found similar associations (e.g. Broekhuizen et al. 2016; Løkken et al. 2018), where others have found no associations (e.g. Aguiar et al. 2019; Hu et al. 2017). The second part of our model, relating social skills in ECEC to school outcomes such as mathematics, reading, and English, runs contrast to several previous studies finding no such associations (e.g. Claessens, Duncan, and Engel 2009; Duncan et al. 2007; Pagani et al. 2010). The third part of our model, relating process quality to school outcomes, aligns somewhat paradoxically with previous studies finding no long-term effects, considering that the total effect of process quality on school performance was not statistically significant, even though indirect effects through social skills were. This is due to the direct effects being imprecisely estimated (high standard errors), as opposed to the indirect effects which were precisely estimated and consequently adequately powered for the detection of small but significant effects. This is evident from how the total effects and the indirect effects were approximately of the same magnitude, but only indirect effects were statistically significant – serving as a reminder that indirect effects may be present even in the absence of significant total or direct effects (see Kenny and Judd 2014; Rucker et al. 2011). In sum, the findings, albeit very weakly, fit with recent models of skill formation, where skills acquired early in life serve as foundational building blocks for later learning (see Cunha and Heckman 2007a; 2007b).

It is, however, important to note that the effect sizes are very small. There may be several reasons for this. First, it may simply be the case that ECEC in the longer term only explains a small proportion of children’s later development. Between ECEC at age 5 years and 5th grade (age 11 years), 6 years have passed and other more recent factors in children’s lives may explain more of the variance in children’s national test scores. The phenomenon of diminishing or vanishing effects of early life events over time (fade-out) is frequently seen in the early intervention literature (Bailey et al. 2020).

Second, Norwegian children often come from relatively highly educated families while the largest effects of ECEC are often found for disadvantaged children from lower socioeconomic backgrounds (Huizen and Plantenga 2018). The MoBa-study, which this study is a subsample of, has due to sampling and attrition resulted in a sample with a skewed socioeconomic profile, with higher educated families being overrepresented. This is also indicated by how children in our sample on average scored between 10% and 30% of a standard deviation above the population mean on our outcome measures, the national tests (see Utdanningsdirektoratet 2021). Consequently, children from families of higher socioeconomic backgrounds may be both more robust against the impact of lower levels of ECEC quality while also being less dependent upon higher ECEC quality due to the likelihood of benefitting from a high-quality home environment.

Third, there is somewhat limited variance in our ECEC process quality measures. Specifically, this relates to modest differentiation in the lower end of the quality spectrum. In comparison, some previous Norwegian studies have found considerably lower levels of process quality in ECEC (see Baustad and Bjørnestad 2022; Bjørnestad et al. 2020). Although not directly comparable due to the different measures used (observed vs. self-reported), this could nevertheless indicate that the participating centers in this study were of higher quality than what is commonly the case. However, it may also be the case that teachers who self-report may be overly optimistic on their own behalf with regards to the quality of care that they provide for the children. Although the true relation between the level of quality ECEC teachers report and the level of quality they provide is unknown, we may suspect that self-reporting tends to be somewhat biased in the direction of being overly optimistic, at the expense of the predictive value of our statistical model.

It is, however, important to note that although mediation models are often viewed as means of testing causal hypotheses, depending on which theoretical framework one adheres to, causal interpretations generally depend on strict assumptions of ignorability being met (see Imai et al. 2014; Imai, Keele, and Tingley 2010; Keele 2015; Pearl 2014). Although our statistical models are built upon a theoretical framework positing social skills as a plausible mediating variable between ECEC quality and school performance, our research design cannot rule out the possibilities of omitted variables potentially biasing results.

In summary, this study sheds some light on how ECEC process quality relates to school performance through increased social skills, providing some indication that implementing practices which further strengthen and develop children’s social skills in ECEC may lead to longer-term academic. It is, however, important to note that most of the variance in children’s abilities within mathematics, reading and English was unexplained by our two measures of process quality (pedagogical practices and structured activities). This could partially be explained by weaknesses in the way process quality was measured but is also likely due to early social skills being just one part of children’s development to predict academic outcomes in the longer-term.

Acknowledgements

We are grateful to all the participating families in Norway who take part in this on-going cohort study.

Disclosure statement

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

Additional information

Funding

The Norwegian Mother, Father and Child Cohort Study is supported by the Norwegian Ministry of Health and Care Services and the Ministry of Education and Research.