Titles and abstracts for records were initially screened and retained if they: (a) were written in English; (b) were published in peer-review journals; (c) described a longitudinal, cohort or population-based quantitative study; (d) were based on UK data (from any of the four UK nations); (e) reported measurement of factors associated with academic attainment during middle childhood (ages 6–12 years); and (f) reported academic attainment as an outcome at the end of compulsory mainstream schooling (ages 15–17 years).

Following initial screening, full text articles were obtained and assessed for eligibility for review according to the following set of expanded inclusion criteria.

Literature searches were conducted in March 2020 and an updated search in September 2021. The following electronic databases were used: Social Science Citation Index, the Education Collection, Social Science Database, Sociology Collection, PsycInfo, and Proquest Applied Social Sciences Index and Abstracts. The search strategy for electronic databases was based on the combination of four lists of words: list 1—search terms related to study design (e.g., longitudinal OR prospective OR cohort OR predict OR follow-up^*); list 2—search terms related to participants (e.g., child^* OR pupil^* OR student^*); list 3—search terms related to factor time point (e.g., primary school OR Key Stage 2); list 4—search terms related to outcome time point (e.g., secondary school OR Key Stage 4), educational outcome (e.g., academic outcome OR academic attainment OR educational outcome OR educational attainment OR exam attainment OR exam outcome^* OR qualification^* OR performance measure^* OR GCSE OR General Certificate of Education OR BTEC OR National Vocational Qualification OR Progress 8 OR National 4 OR National 5 OR Intermediate exam^*), and educational subject headings (e.g., Welsh OR English OR Irish OR Scottish OR Mathematics OR Science OR Literacy OR Numeracy OR Design Technology OR Information Technology OR History OR Geography OR Modern Foreign Languages OR Religious Studies). The four lists were then combined using AND (see example search string in Supplementary Material).

Database searches were supplemented with searches of gray literature including national level reports from the four UK nations' governmental websites, Department for Education and statistics webpages, as well as websites for the National Foundation for Educational Research, Center for Longitudinal Studies (1970 British Cohort Study, 1958 National Child Development Study, Millennium Cohort Study, Next Steps) and the University College London Institute for Education, Cambridge Assessment, AQA Research and Policy Analysis, National Audit Office, Institute for Fiscal Studies, Nuffield Foundation, Sutton Trust, and Center for Multilevel Modeling. Reference lists of included papers were also reviewed as well as publications from birth cohorts not included on the Center for Longitudinal Studies website (Avon Longitudinal Study of Parents and Children, and British Household Panel Survey). The keyword “academic attainment” was used in the searches of gray literature.

One author (MEW) conducted the initial screening of records based on information from titles and abstracts. Another author (SC) reviewed the first 2,500 (35.8%) potentially relevant abstracts to check for accuracy. Interrater agreement was 99.7%. All full texts were appraised by MEW and any queries discussed with SC.

Literature searches were conducted in March and April 2020 with an updated search in September 2021. There were no date limits to the searches and only peer-reviewed articles published in English or Welsh were included. The only exception was the gray literature search which often included non-peer reviewed content (e.g., government reports) but may have included pertinent information for this review. The initial database search yielded 14,088 articles and an additional 1,715 in the updated search. Following the removal of duplicates, the titles and abstracts of 6,622 articles were screened against the eligibility criteria. Forty-one articles were selected for full-text review with a further 183 identified through supplemental means. A total of 90 studies met the inclusion criteria (see Figure 1 for a flow diagram of study selection) and were retained for the review.

For the 90 retained studies, information was extracted on sample information, methodology, factors and timing, attainment outcomes and timing, and study findings using a bespoke data extraction tool designed for this review. Supplementary Table 1 summarizes the extracted information from the 90 studies.

The methodological quality of each included study was assessed using items from existing quality criteria lists (Bohensky et al., 2011; National Institute for Health, 2013; Critical Appraisal Skills Programme, 2018). The list of 10 items are categorized into four dimensions: 1) study population and participation (3 items); 2) data collection (3 items); 3) reporting (2 items); 4) data analyses (2 items). Each item had three possible responses, either “yes,” “no,” or “unclear.” The maximum points given to any item was 1. For each study, a total methodological quality score was calculated by summing the items. A quality score of 8 or above, i.e., 80% of items scored positively, was considered a high-quality study. Lower scores were considered low quality.

Due to the differences in study methodologies, time points, factor and outcome variables, a formal meta-analysis was not possible. Therefore, a narrative synthesis was conducted. Individual middle childhood factors were grouped by general type (e.g., student, socioeconomic, school) and analyzed in terms of the number of times explored in the research, range of academic outcomes with which the factor is associated, consistency of direction of effect, magnitude of effect, and quality of the studies.

To summarize the results and ease the drawing of conclusions regarding level of evidence, best evidence synthesis was used (see Belmon et al., 2019). Levels of evidence were defined as follows: Strong evidence—consistent findings in multiple (2 or more) high quality studies; Moderate evidence—consistent findings in one high quality study and at least one low quality study, or consistent findings in multiple (2 or more) low quality studies; Inconsistent evidence—inconsistent findings in multiple studies (2 or more); Insufficient evidence—only one study available. If there were two or more high quality studies showing consistent results, the low-quality studies were disregarded in the narrative synthesis and conclusions.

Characteristics of the included studies are shown in Supplementary Table 1. Sample sizes in the studies ranged from 59 (Toivainen et al., 2017) to 2,740,182 students (Goldstein and Leckie, 2016). Twenty-nine of the studies used the UK Department for Education National Student Database as the sole data source, 10 used local Education Authorities' databases, six used education surveys, four used exam board databases, 17 were based on UK cohorts, and 24 studies used multiple data sources. Factors were measured at ages 6–12 years, with 67 of the studies measuring factors at only one time point, 18 studies at two time points, three at three time points, one at four time points, and one at seven time points. The most frequently used outcome measure at ages 15-17 was GCSE attainment (n = 65), whilst nine used GCE or O-Level attainment, eight used a KS4 points score, five used the Scottish Certificate of Education (SCE) or O-grade, and three used the Attainment 8 and/or Progress 8 indicators. Four studies reported results for each factor separately: one by gender, two by cohort year, and one by city.

Results for the critical appraisal of study quality are summarized in Supplementary Table 1 (a more detailed appraisal based on individual items can be seen in Supplementary Table 2). Overall, the risk of bias was moderate with only 19 studies rated as high quality. The most common limitations were not reporting rates of missing data for the variables used and low participation rate (although many of these were rated as unclear due to insufficient information). A total of 54 of the studies reported an estimate of effect size.

A summary of the results for each middle childhood factor can be found in Table 2 with detailed descriptions across all included studies in Supplementary Table 3.

By far the most frequently examined factor was prior attainment. Strong evidence of an association with later academic attainment was found for numerous measures of prior attainment in middle childhood. For example, 12 high quality studies found that students with higher overall prior attainment were more likely to have higher academic attainment at age 15–17 years (Jenkins et al., 2006; Levacic and Marsh, 2007; Coe et al., 2008; Leckie, 2009; Department for Education, 2010, 2016; Fletcher et al., 2015; Hampden-Thompson and Galindo, 2017; Schoon and Ng-Knight, 2017; Gorard and Siddiqui, 2019; Hansen and Henderson, 2019; White et al., 2021). Students with higher English, mathematics, and science scores at age 11 (Jenkins et al., 2006; Department for Education, 2010; Sammons et al., 2014; Fletcher et al., 2015) were more likely to have better GCSE attainment. Students with higher verbal reasoning scores at the end of primary school were also more likely to have better academic attainment at age 15–17 based on consistent findings from one high quality study (Deary et al., 2007) and 18 low quality studies (Willms, 1985, 1986; Nuttall et al., 1989; Garner and Raudenbush, 1991; Jesson and Gray, 1991; Paterson, 1991; Goldstein et al., 1993; Daly, 1995, 1996; Sammons, 1995; Sammons et al., 1995; Thomas and Mortimore, 1996; Daly and Shuttleworth, 1997; Goldstein and Sammons, 1997; Thomas et al., 1997a,b; Strand, 2006; Deary et al., 2007; Thomas et al., 2007; Sullivan et al., 2010), as well as students with higher cognitive ability at age 11 based on two high quality studies (Deary et al., 2007; Larose et al., 2021). There was moderate evidence showing that students with higher quantitative reasoning, reading and creative expressiveness in middle childhood were more likely to have better attainment at age 15–17, although based on evidence from low quality studies only (Quantitative reasoning: Thomas and Mortimore, 1996; Strand, 2006; Thomas et al., 2007; Reading: Garner and Raudenbush, 1991; Goldstein et al., 1993; Sammons, 1995; Sammons et al., 1995; Goldstein and Sammons, 1997; Thomas et al., 1997a,b; McNiece et al., 2004; Brown and Taylor, 2008; Sullivan et al., 2010; Mourgues et al., 2016; Creative expressiveness: Toivainen et al., 2017, 2021). There was inconsistent evidence for measures of non-verbal ability during middle childhood with four low quality studies showing that students with higher non-verbal ability in middle childhood had lower O-level attainment (Thomas and Mortimore, 1996; Strand, 2006; Thomas et al., 2007; Sullivan et al., 2010), whilst one study found that students with higher non-verbal ability at age 7 were less likely to achieve CSE science passes (Sullivan et al., 2010). Logic also showed inconsistent results based on two low quality studies, one reporting no association with later academic attainment (Toivainen et al., 2021), and one reporting that students with higher logic scores at age 9 had higher GCSE English attainment (Toivainen et al., 2017). Logic was measured as part of a story writing task and consisted of ratings of the degree to which story events were logical, the straightforwardness of the story, and the grammatical correctness of the written story. KS2 writing skills and executive function had insufficient evidence for an association with later academic attainment.

An important strength of this systematic review is the inclusion of non-peer reviewed reports, as well as peer-reviewed articles. The overall quality of the included studies was relatively poor with only 19 studies rated as high quality. The common weaknesses within these longitudinal studies were a failure to report missing data for the variables used and low participation rates (although many reports were unclear regarding this).

There are also some limitations to consider. First, only papers that were published in English were reviewed. No eligible studies published in other languages in the UK (e.g., Welsh) were eligible. Second, we did not attempt to perform a meta-analysis, which would allow a quantitative analysis of factors related to academic attainment, due to the differing research methods of studies included in this systematic review. Third, many papers were based on the same data (e.g., over half used the Department for Education National Pupil Database). This suggests that caution should be taken when interpreting the results even though there may be advantages to analysis based on large numbers across the UK. Fourth, there is also an issue with relationships between factors that can affect interpretation. For example, Strand (2014) found that the attainment gap at 16 years associated with social class was larger than for gender and ethnicity, although there were substantial interactions between the three factors. A similar relationship can be seen for deprivation (see Parsons, 2016). This suggests that complex relationships exist and that examining such factors in isolation may be misleading. In this review, only 54% of the included studies adequately controlled for confounding variables. Finally, the search was limited to factors in middle childhood (6–12 years). This may have excluded factors from early and later childhood which could affect academic attainment at KS4, but such studies were out of the scope of the current review and could not be assumed to apply to decisions relating to actions that may be taken based on middle childhood factors.

The current findings have several implications for policy and interventions targeting educational inequality. For example, Cognitive Ability Tests, particularly verbal reasoning, provide schools with a powerful tool to estimate students' academic attainment (see https://www.gl-assessment.co.uk/products/cognitive-abilities-test-cat4/; GL Assessment, 2012). Studies in this review suggest that verbal reasoning is strongly associated with later attainment and the Strand (2006) study suggests that cognitive ability scores show stronger relationship to later attainment than KS2 scores. In the UK, many schools use cognitive ability estimates during primary phase (age 7–11 years) and/or at the start of secondary school (age 11–12 years), although the use of these data is not compulsory. Where schools use cognitive ability estimates effectively, it helps teachers to adjust provision and plan interventions for students to help them make expected progress (Estyn, 2019). Encouraging school leaders to make the formative use of cognitive ability estimates may be a policy priority to ensure that more students achieve their potential.

This review also has implications for current education policy and guidance. For example, the findings for the socioeconomic factors suggest that providing support to disadvantaged students early to reduce the attainment gap may lead to improved attainment at the end of secondary education. Schools in the UK are given additional funds to support disadvantaged students who are identified as being eligible for free school meals based on measures related to household income. The results of this review partially support this policy, since parent income and parent social class (which is associated with income) in middle childhood were all found to have moderate evidence for an association with later academic attainment, although eligibility for free school meals had inconsistent results and some have questioned its suitability as a proxy for disadvantage (see Ilie et al., 2017).

Guidance for schools on how to best support disadvantaged students across the UK (Estyn, 2012; Department for Education, 2015a) identifies seven key principles of effective practice, three of which are supported by the results of this review: (1) developing a whole-school ethos of high attainment for all—this review suggests on the whole that students from primary schools with higher mean attainment are more likely to have higher later academic attainment; (2) using student data, including cognitive abilities tests, to inform teaching approaches—prior attainment was the factor with the strongest association identified in this review as well as elements of cognitive ability which were all positively related to later academic attainment; and (3) ensuring students acquire early language and literacy skills—language and literacy skills are important for all students and the current review shows that prior attainment is positively related to later academic attainment. For the other four UK recommendations on supporting disadvantaged students (high quality teaching for all; addressing attendance and behavior; deploying staff effectively; and clear and responsive leadership), there were no supporting studies in the current review, mainly because the factors had not been measured in middle childhood. For example, although Department for Education (2016) showed that attendance is negatively associated with GCSE attainment, the study only evaluated the attendance of students during KS3. Also, there was lack of research on school-based factors with most of those identified in this review coming from one study (Gorard and Siddiqui, 2019). Further research needs to be conducted on school-based factors including teaching and leadership quality and staff ratio as suggested by the Department for Education recommendations.

Another implication of the current review relates to the paucity of research addressing middle childhood to end-of-school association with academic attainment since most of the factors identified had very few supporting studies (the exception being academic factors). Given that it is crucial to ask what could be done during primary education, or at the start of secondary education, to increase the chances of young people leaving school with better academic attainment, the paucity of direct research evidence is a serious concern. Therefore, we call for a step change in longitudinal educational attainment research. A second problem is that not all of the relevant research includes metrics that might directly inform policy makers (i.e., estimates of effect size). The reporting of effect sizes was variable within the studies with 54 providing an effect size estimate, thus limiting meaningful interpretation of results. Effect sizes are important since they tell us the magnitude of an effect whilst statistical testing only tells whether an effect exists (Fan, 2001); but many studies particularly in education neglect to include effect sizes (Sun et al., 2010).

A final significant limitation of the extant literature, and thus the findings of the current review, is the lack of an explicit framework to guide what middle childhood factors might be examined as potential associated factors of end-of-school academic attainment. Several of the factors examined in studies included in the current review may be important but appeared to follow the particular interests of researchers or the availability of certain variables in existing databases. A focus on physical activity and body mass index (Booth et al., 2014a,b) is a case in point. Although potentially important factors, it is unclear why these factors and not others might have been examined. Thus, there are potential biases overall in the findings of the current review relating to researcher priorities.

What is needed is an agreement about how to categorize potentially important variables, to measure them in education databases, and to explore them in future research. One option, prompted by the focus of the studies included in the current review, would be to consider a broader child development framework such as Bronfenbrenner's bioecological approach (Bronfenbrenner and Morris, 2006; Darling, 2015). What is notable about the studies included in the current review is that very few broader contextual factors that may affect a young person's academic attainment from middle childhood have been examined in research. For example, experience at school or the quality of schooling in middle childhood have not been examined as factors associated with later outcome. Such factors are important for a more wholistic understanding of a young person's later academic attainment, but also because they may be amenable to change through their interaction with different practices in primary schools or early in secondary school education.