Parental education and children's depression, anxiety, and ADHD traits, a within-family study in MoBa

The study addresses whether the commonly observed associations between lower parental socioeconomic position—particularly lower parental educational attainment—and greater childhood emotional and behavioral difficulties (depression, anxiety, and ADHD) are causal. Socioeconomic indicators are interrelated and subject to confounding by parental mental health and other environmental factors, and associations may also be influenced by genetic transmission because educational attainment is partly heritable. Prior work suggests substantial genetic influences on education and potential genetic nurture effects, complicating causal interpretation. The authors aim to disentangle environmental effects of parental education from direct genetic effects on children by using a within-family Mendelian randomization (MR) approach in genotyped mother–father–child trios from the Norwegian MoBa cohort, estimating the independent effects of maternal and paternal education on children’s depressive, anxiety, and ADHD traits at age 8 while adjusting for children’s own genetic liability to educational attainment.

Previous research documents social gradients in child mental health across parental education and income. Family-based genetically informed designs (e.g., children-of-twins) indicate that once genetic and shared environmental confounding are accounted for, maternal and paternal education may influence children’s ADHD traits but not depressive traits. Genomic analyses of trios have reported evidence of parental (genetic nurture) influences on depressive and ADHD traits but not anxiety, though without isolating specific parental phenotypes. Some studies report little evidence that parental educational attainment influences children’s ADHD once accounting for children’s own genetic liability for education. Mendelian randomization has been proposed to estimate causal effects of parental exposures intergenerationally, but requires large genotyped samples of parents and offspring, which are scarce. The current study builds on and extends prior work by using within-family MR to separate maternal and paternal effects of education from direct genetic transmission, and by assessing pleiotropy with two-sample MR methods.

Design and sample: The study used 40,879 genotyped mother–father–child trios from the Norwegian Mother, Father and Child Cohort Study (MoBa), a population-based pregnancy cohort recruited 1999–2008 (participation rate 41%). Genetic family ID (via KING) was used to cluster related observations. Ethical approvals and data sources (including linkage to national registries) are described. Exposures: Maternal and paternal educational attainment (years) were derived primarily from Statistics Norway administrative data (99% coverage), with questionnaire data used to impute missing values. Educational categories were mapped to years using ISCED conventions. Outcomes: Child depressive traits (Short Mood and Feelings Questionnaire, SMFQ, 13 items), anxiety traits (Short SCARED, 5 items), and ADHD traits (RS-DBD total and subscales for inattention and hyperactivity) were reported by mothers at child age 8. Summary scores were prorated when ≥80% items present. Parental depressive/anxiety traits (selected items from the Hopkins Symptom Checklist) and parental ADHD traits (Adult ADHD Self-Report Scale, 6 items) were also measured. Genetic instruments: Polygenic indexes (PGIs) for educational attainment were constructed for mothers, fathers, and children using genome-wide significant SNPs (p < 5×10⁻⁸) from the largest GWAS (Okbay et al., 2022; including 23andMe), weighted by GWAS effect sizes. After excluding unavailable SNPs and clumping (r² = 0.01, 10,000 kb), 1,729 independent SNPs were included. Genotyping covariates included center, chip, and 20 principal components of ancestry for each trio member. Imputation: Phenotypic missing data were addressed via multiple imputation by chained equations in Stata v16, generating 50 datasets. The imputation model included parental education, all outcomes, child sex and birth year, parental smoking during pregnancy, parental depressive/anxiety and ADHD traits, maternal parity, birth registry variables (marital status, previous pregnancies, parental ages), earlier child ADHD (age 5; Short-Form Conners), and genetic covariates and PGIs on the right-hand side. Continuous variables were imputed with predictive mean matching (knn=10); ordered categoricals with ordered logistic regression. Analytic strategy: - Observational models: Linear regressions of child outcomes on maternal and paternal years of education, adjusting for child sex, birth year, and the same genotyping covariates as genetic models; then further adjusted for likely confounders (parental depressive/anxiety and ADHD traits, parental smoking during pregnancy, and maternal parity). - Mendelian randomization (intergenerational): Two-stage least squares (2SLS) using parental education PGIs as instruments for maternal and paternal education, mutually adjusting for the other parent’s PGI, child sex and birth year, and genotyping covariates. - Within-family MR: Multivariable 2SLS where mothers’ and fathers’ PGIs instrument their respective educations, additionally adjusting for the child’s education PGI to block direct genetic transmission pathways, plus child sex, birth year, and genotyping covariates. Robust standard errors were used. - Instrument strength: First-stage F-statistics were 591.6 (mothers) and 600.9 (fathers); conditional R² ≈ 1.4% variance explained in years of schooling for both parents. Sensitivity analyses: Sex-stratified analyses; outcome square-root transformations; complete-case analyses (unimputed) for trios with complete data (N≈6,300 per outcome); ADHD subscales analyses. Two-sample MR for pleiotropy assessment: SNP-level associations with parental education from the GWAS and SNP-outcome associations estimated in MoBa (mutually adjusting for mother, father, and child genotypes and covariates). Methods included IVW, MR-Egger, MR-Median, and MR-Modal using all 1,729 SNPs and a stringent subset of 510 SNPs (clumping r²=0.001). MR-Egger intercepts were examined for evidence of horizontal pleiotropy.

- Observational associations (minimally adjusted): Each additional parental year of education was associated with small reductions in child depressive traits (−0.02 SD per year for both parents; p<0.001) and ADHD traits (maternal: −0.02 SD, paternal: −0.03 SD; both p<0.001). Maternal education was modestly associated with lower child anxiety traits (−0.01 SD per year; p=0.008); paternal education showed no clear association with anxiety. - Observational associations (adjusted for confounders): Effects attenuated substantially. For depressive traits, paternal education remained associated (−0.01 SD per year; p=0.002), while maternal education was consistent with null. For ADHD, associations were reduced and in some cases consistent with null (e.g., maternal β≈−0.01 SD; p=0.15). - MR without adjusting for child genotype: Estimates were less precise. Depressive traits: maternal −0.02 SD (95% CI −0.06, 0.03; p=0.44); paternal −0.04 SD (−0.08, 0.00; p=0.05). ADHD traits: paternal −0.06 SD (−0.10, −0.02; p=0.003); maternal not clearly associated. - Within-family MR (adjusting for child’s PGI): Little evidence that either maternal or paternal education causally affected child depressive, anxiety, or ADHD traits. Depressive traits: maternal 0.00 SD (−0.04, 0.05; p=0.94); paternal −0.02 SD (−0.06, 0.02; p=0.39). ADHD traits: paternal effect attenuated to −0.02 SD (−0.06, 0.02), maternal effect near null. - Child’s own PGI for educational attainment: Independently and negatively associated with child traits. Depressive traits: −0.04 SD per 1 SD PGI (p=0.001). Anxiety traits: −0.02 SD (p=0.05). ADHD traits: −0.08 SD (95% CI ≈ −0.09, −0.06; p<0.001), ≈0.6 points on a 54-point scale. Subscale results mirrored the total ADHD effect. - Instrument strength: First-stage F-statistics > 590 for both parents; PGI explained ≈1.4% of variation in parental schooling even in within-family models. - Sensitivity and complete-case analyses: Sex-stratified and transformed-outcome analyses were broadly consistent but less precise. Complete-case within-family MR showed some unexpected positive associations (e.g., maternal education with child depressive and ADHD-inattention traits), likely reflecting type I error and selection bias given the ≈15% subsample. - Two-sample MR: IVW suggested inverse associations of parental education with child depressive and some ADHD outcomes, and a positive maternal effect on ADHD-inattention, but alternative MR methods (MR-Egger, Median, Modal) were largely null and MR-Egger intercepts did not indicate strong pleiotropy. Overall, these results were interpreted cautiously and in the context of the main within-family MR findings.

The central question was whether parental educational attainment causally shapes children’s depressive, anxiety, and ADHD traits at age 8. Within-family MR, which mitigates environmental confounding and blocks bias from direct genetic transmission by conditioning on the child’s genotype, found little evidence for causal effects of maternal or paternal education on these outcomes. Although observational associations were present, they attenuated with confounder adjustment and disappeared in genetic models, suggesting residual confounding and/or genetic transmission explain much of the observed gradients. The robust negative associations of children’s own educational attainment PGI with depressive and ADHD traits (and marginally with anxiety) imply that direct genetic factors related to educational attainment—or correlated traits such as cognition, non-cognitive skills, or overlapping genetic architecture—may underlie part of the phenotypic correlation between parental education and child mental health. Effect sizes were small, consistent with polygenic influences and complex developmental pathways. Context matters: Norway’s comparatively equal socioeconomic environment and high tertiary education participation may dampen the observable mental health consequences of parental education relative to other contexts. The findings align with prior genetically informed studies showing limited genetic nurture for these mental health traits and stronger roles for children’s own liability, while contrasting with some literature suggesting stronger effects of parental income. The study underscores that policies increasing adult education alone may have limited impact on child emotional and behavioral difficulties in similar contexts; targeted supports and environmental interventions (e.g., enhancing school connectedness) could be more pertinent. Two-sample MR suggested some associations under IVW but lacked consistency across pleiotropy-robust estimators, reinforcing cautious interpretation and the primacy of within-family MR estimates for causal inference in this setting.

Using one of the largest samples of genotyped mother–father–child trios, the study provides little evidence that maternal or paternal educational attainment causally affects 8-year-old children’s depressive, anxiety, or ADHD traits. Observational associations appear largely attributable to confounding and direct genetic transmission. Children’s own genetic liability for educational attainment is modestly protective for depressive and ADHD traits. These results suggest that improving parental education alone is unlikely to yield substantial benefits for child emotional and behavioral outcomes in contexts like present-day Norway. Future research should leverage larger and more representative family-based genetic datasets, examine different developmental stages (including adolescence), evaluate contexts with greater socioeconomic variability, assess clinically diagnosed outcomes, and use multivariable MR to disentangle correlated parental phenotypes (e.g., income, mental health, parenting behaviors) and child traits that may mediate observed associations.

- Generalizability and selection: MoBa participation (41%) and the genotyped trio subsample (~36% of MoBa) overrepresent higher education and better health, potentially biasing estimates and limiting generalizability to Norway or other countries. - Missing data and attrition: Approximately 54% missingness for child outcomes at age 8 required multiple imputation; although preferable to complete-case analysis, imputation may not fully recover the true distributions. Complete-case analyses were small and potentially biased. - Measurement: Child outcomes were mother-reported and subject to reporting bias. Results may differ for clinically diagnosed outcomes. - Precision: Within-family MR estimates, while robust to key confounders, were imprecise; small causal effects (e.g., <0.08 SD per parental education year) cannot be ruled out, particularly for ADHD. - Genetic instrument limitations: PGIs capture only SNP-based liability and may miss other genetic architecture; residual horizontal pleiotropy cannot be excluded despite sensitivity analyses. - Context dependence: Findings may not extend to settings with greater educational and economic inequality or different school and healthcare systems.