Background The relevance of neighbourhoods for children’s cognitive development, while receiving increasing attention, is still subject to debate. One mechanism that has been proposed is via institutional resources such as schools. Another hypothesis is that neighbourhood aspects influence maternal psychological well-being and parenting practices. However, the contribution of schools to the neighbourhood effect is understudied, and findings regarding the role of maternal mental health and parenting have so far been inconsistent. The aim of this research was twofold: first, to partition the variability in children’s cognitive test performance between neighbourhoods, schools and families, and second, to examine whether maternal psychological distress and parenting practices were on the pathway between neighbourhood characteristics and these outcomes.

Methods We analysed data from the UK Millennium Cohort Study. Outcome measures were cognitive test performance at age 7 in reading, maths and a pattern construction test. Data were analysed for 9,412 children who had not changed school and lived in the same neighbourhood since age 5, and for whom information was complete. Neighbourhoods were defined as Lower Layer Super Output Areas. Cross-classified multilevel models were used to simultaneously estimate the variability in cognitive test outcomes between neighbourhoods and schools before and after adjusting for family-level covariates; and to examine the contribution of maternal psychological distress and parenting practices.

Results At age 7, the percentage of the unadjusted variance in cognitive test performance between neighbourhoods was about 5% for reading, 9% for maths and 7% for pattern construction. The unadjusted variance between schools was about 10% for reading, 12% for maths, and 6% for pattern construction. After allowing for a range of family background characteristics, the neighbourhood variance for reading was no longer statistically significant, while a school effect of about 7% remained. For maths, the remaining variance attributable to neighbourhoods and schools was 7% and 10% respectively, while for pattern construction, this was about 3% at both levels. Reading and math test performance were independently associated with neighbourhood median household income, and pattern construction scores with the IMD education domain in the expected direction. Our data did not support the hypothesis of a mediating effect via maternal psychological distress and parenting practices.

Conclusion Our findings highlight the importance of including both neighbourhoods and schools when aiming to determine the contribution of contextual factors to children’s cognitive test performance. Differences between neighbourhoods and schools appear to contribute to inequalities in children’s cognitive development already at this early age.