Inter-individual differences in intelligence predict a variety of important life outcomes, such as life satisfaction, mortality, and educational achievement (EA)^1,2,3, showing correlations higher than 0.7 with achievement^4,5. There has been considerable controversy on the source of these differences. Decades of research have shown intelligence to be highly associated with genetic differences (heritability estimates around 0.6) but also impacted by experiences and context (called here “environmental factors”)^{6,8,9,10,11,12}.

Schooling is an important environmental factor, having a large impact on intelligence in children^13,14,15. This has been shown in longitudinal studies controlling for prior intelligence¹⁶ and in studies evaluating the cognitive effects of policy changes regarding compulsory schooling¹⁷. A third method to evaluate the impact of schooling (called regression discontinuity designs) exploits the fact children are put in a grade based on an arbitrary age cut-off, and allows us to separate the effect of chronological age from months of schooling¹⁸. Studies using this method also replicate the findings that schooling affects intelligence^19,20. Thus, many lines of research provide converging evidence that schooling can change abilities often thought to be “fixed”, such as fluid intelligence and working memory, with estimates of one year of additional schooling benefitting cognitive abilities somewhere between 1 to 5 IQ points, or 0.07 to 0.3 SD^13,19,21. What is less clear is how the impact of schooling interacts with the environments that children experience and with their genetic predispositions—in other words, do school settings amplify preexisting differences (i.e., “rich-gets-richer”), or conversely, weaken existing differences between children (i.e., “catch up”).

Much of a child’s environment is captured by socioeconomic status (SES), a summary measure usually comprised of household income, parental education, and neighborhood quality²². While commonly implicated in initial differences in intelligence, SES has also been found to widen existing differences throughout development^23,24,25. Recently, research on SES has been criticized for neglecting the role of genetics—as parents not only hand down environments but also genes^26,27. For example, a genetically informed twin study found a strong genetic influence on SES and its association with intelligence²⁸. While there is a large body of research looking at how preexisting SES differences interplay with years of schooling²⁹, none of these previous studies have incorporated genetics.

Fortunately, genome-wide association studies (GWAS) with extremely large sample sizes and the viability of polygenic scores, have made it possible to incorporate genetically informative measures into a study. A polygenic score is an index that combines thousands/millions of DNA regions (each with only a tiny effect on the trait of interest) and gives a value to each individual representing their genetic propensity. Relevant to us here, a multi-trait cognitive polygenic score (cogPGS) was recently shown to predict 7–10% of the variance in cognitive performance³⁰. This and similar polygenic scores correlate moderately (r ~0.3) with SES^31,32, but little is known about their unique contributions to different domains of intelligence. And even less known if, or how, they interplay with schooling.

Gene-by-environment (GE) interplay is often proposed as an explanation of how intelligence can show high heritability alongside malleability^33,34,35,36. To date, there are only a few studies on GE interplay using polygenic scores. Of these studies, most have focused on SES as the environmental variable of interest and none have used the environmental variable of schooling, which is particularly relevant to intelligence. Unfortunately, GE-interplay results with SES have been inconsistent for educational achievement/attainment—some positive, negative, or null^{37,38,39,40,41}. While there is a high correlation between educational achievement and intelligence, educational achievement is a broader concept thought to also include personality characteristics such as consciousness and openness to experience^42,43. Furthermore, we are not aware of any research examining if schooling moderates’ genetic effects on intelligence.

Schooling, SES, and genetics thus represent three substantial contributors to intelligence, but it is unclear to what extent their contributions are (1) unique and (2) interact with one another. In a sample of 3rd to 5th grade children, we estimated the unique contributions of a year of schooling, SES, and cogPGS on crystallized intelligence (cIQ), fluid intelligence (fIQ), and working memory (WM). We chose these domains of intelligence since they are particularly important for educational outcomes^4,16,44. WM was included since WM has shown to be heavily impacted by schooling²¹, malleable⁴⁵, and potentially relevant for GE-interplay⁴⁶. Our main aim was to examine if GE-interplay is present for these domains of cognition. Specifically, we tested if the effect of (1) schooling or (2) SES is moderated by cogPGS, and (3) a three-way interaction between schooling, SES, and cogPGS. These interactions allowed us to test if schooling enhances or compensates for preexisting genetic and environmental inequity.

We included data from 6567 children (mean age = 9.88, range 8.92–11.00; Supplementary Table 1) recruited by the ABCD consortium to be representative of the United States in sex, race, ethnicity, SES, and urbanicity⁴⁷. SES was defined as the first component of a probabilistic PCA, capturing 65% of the variance in total household income, highest parental education, and neighborhood quality. Due to modeling constraints, we had to exclude 1,086 children that were missing DNA data. This group had