Delayed tracking and inequality of opportunity: Gene-environment interactions in educational attainment

The study examines whether delaying the age at which students are sorted into secondary school ability tracks reduces inequality of opportunity by shifting the determinants of educational attainment from family background to children’s potential. Traditional approaches conflate family influences and ability, or rely on incomplete measures of each. Using a behavioral genetics framework, the authors decompose variance in educational attainment into genetic (A), shared environmental (C; family background), and non-shared environmental (E) components and test how these depend on tracking age in the Dutch system. Two research questions guide the study: (1) How do genetic and shared environmental influences on secondary educational attainment (track level) depend on whether tracking is immediate (end of primary school) or delayed (after 1–3 years in mixed-level classes)? (2) Does this moderation by tracking age differ across students’ performance levels? Hypothesis H1 predicts stronger genetic and weaker shared environmental influence with delayed tracking. To distinguish mechanisms, the authors discuss primary effects (family background operating via performance) and secondary effects (differences in track placement net of performance). If early tracking mainly hampers disadvantaged high performers, effects should be strongest at high performance (H2a). If early tracking mainly compensates advantaged low performers, effects should be strongest at low performance (H2b).

Prior cross-national and reform studies generally find that earlier tracking is associated with stronger effects of family background on attainment without gains in average performance. Standard SES and IQ measures incompletely capture family and ability effects, and sibling similarity conflates genetic and shared environmental influences. Behavioral genetics research shows substantial heritability of cognitive ability and educational performance, with heritability often increasing with age as autonomy enables gene–environment correlation processes (active/evocative) to amplify genetic proclivities. In the Netherlands, educational performance (CITO) is associated with SES, and genetic influences on cognition and performance do not vary by parental SES, unlike some U.S. evidence. Teacher recommendations and parental strategies can induce secondary effects net of performance, potentially through hampering (underplacement of disadvantaged students) or compensation (overplacement of advantaged students). This study leverages twin designs to separate A, C, and E, and a bivariate model to partition influences shared with performance (primary pathways) versus unique to attainment (secondary pathways).

Data: Twins from the Young Netherlands Twin Register (NTR), birth cohorts 1986–1999. Analytic N = 8847 twins from 4941 pairs (MZ and DZ), with repeated surveys of parents, teachers, and twins. Measures: (1) Educational attainment is operationalized as the last reported secondary track level (VMBO-b=0, VMBO-k=1, VMBO-g/t=2, HAVO=3, VWO/Gymnasium=4), harmonized across survey years; treated as continuous. (2) Educational performance is the standardized national CITO test score (approx. mean 535, SD 9–10; sample mean 538, SD 8.44). (3) Tracking timing: Immediate if students start secondary school in homogeneous-track classes; Delayed if they start in heterogeneous mixed-track classes and are definitively tracked 1–3 years later. Covariates: sex and year of birth (mean-centered). Zygosity determined by questionnaire (≈97% accuracy) and biological data in subsets. Modeling: Genetic structural equation modeling in OpenMx (R). First, an ACE model estimates variance components for attainment and performance overall and by tracking group. A multigroup framework distinguishes pairs where both twins are immediate, both delayed, and discordant, plus a separate group with missing tracking info; parameters (a, c, e) are allowed to differ by group and constrained for tests of equality (likelihood ratio tests). Second, a bivariate Cholesky ACE model decomposes attainment variance into components common with performance (primary pathways) and unique to attainment (secondary pathways), with moderation by performance level (continuous moderator) and with moderation effects allowed to differ by tracking timing. Full Information Maximum Likelihood handles missing outcomes; cases missing the performance moderator are excluded (co-twins retained when possible). Assumptions of twin models (equal environments, additivity, no assortative mating, no gene–environment correlation) are discussed; potential bias from passive gene–shared environment correlation is considered likely small for the delayed group and modest for the immediate group. Selection checks: Parental education (SES proxy) and performance associations with delayed tracking are examined to assess potential confounding.

- Twin similarity: Attainment correlations ρMZ=0.80, ρDZ=0.48; Performance ρMZ=0.81, ρDZ=0.45, supporting ACE decomposition. Overall variance decomposition (all tracking types): Attainment ≈58% genetic, 22% shared environment, 19% non-shared; Performance ≈70% genetic, 11% shared environment, 19% non-shared. About 57% of genetic influence on attainment is shared with performance; 43% is unique to attainment. - Impact of tracking timing (Model 1b vs constrained Model 1a): Delayed tracking is associated with larger genetic and smaller shared environmental contributions to attainment. Standardized components: Immediate—A=55% (0.72/1.31), C=27% (0.36), E=18% (0.24); Delayed—A=74% (0.69/0.93), C=2% (0.02), E=24% (0.22). Differences are statistically significant (χ²(22)=312.1, p<0.001). - Primary vs secondary pathways: Components common to performance differ strongly by tracking timing: Immediate—common A=53%, C=40%, E=6%; Delayed—common A=88%, C=0%, E=11%. Unique-to-attainment components differ modestly: Immediate—A=57%, C=5%, E=38%; Delayed—A=66%, C=3%, E=31%. This indicates that delayed tracking mainly reduces primary effects (family background operating via performance), with smaller changes in secondary effects. - Moderation by performance (Model 2): The increase in unique genetic influence on attainment with delayed tracking is most pronounced among low performers; differences are negligible among high performers (supporting H2b, contrary to H2a). The unique shared environmental effect does not vary by performance between tracking groups; however, non-shared environmental influence is larger for immediate than delayed tracking among low performers. Overall (common+unique), shared environment is higher for immediate tracking at all performance levels. Model 2 fits better than a model without differential moderation by tracking timing (χ²(18)=304.2, p<0.001) and better than a model constraining moderation to be equal across timing (χ²(6)=13.0, p=0.044). - Selection checks: Delayed tracking is not associated with parental education (delayed ≈23.5–29.3% across SES; Rao–Scott corrected χ², p=0.111). Delayed tracking increases with performance level (except at the top due to ceiling; p<0.001). After modeling performance moderation, differences in A and C between tracking groups persist, suggesting SES and performance do not confound the main associations.

Findings indicate that delaying definitive tracking reallocates variance in educational attainment away from family-shared environmental factors toward genetic influences, consistent with greater opportunities for children to realize their potential. The primary mechanism appears to be a reduction in primary effects: with delayed tracking, performance reflects genetic differences more and family background less, which then shapes track placement. Secondary effects are reduced to a lesser degree overall, though moderation results show that delayed tracking especially benefits lower-performing students by increasing the role of genetic factors and reducing non-shared (potentially idiosyncratic) influences relative to immediate tracking. High performers show strong genetic influence regardless of timing, suggesting early tracking particularly enables compensatory processes for advantaged low performers. The results align with transactional models where older age and autonomy allow environments to be selected and evoked in line with genetic predispositions. They complement non-genetic evidence on delayed tracking and highlight that a substantial share of genetic influence on secondary attainment is not captured by primary-school performance, underscoring the value of reassessment in secondary school.

The study contributes genetically informed evidence that delaying tracking in secondary education is associated with greater equality of opportunity: genetic influences on attainment rise and shared environmental influences fall, largely by reducing primary effects of family background via performance. The benefits are especially notable for lower-performing students, for whom delayed tracking increases the expression of genetic potential. Policy implications include postponing definitive track placement and/or reassessing placement during early secondary school to minimize mismatches between student potential and track. Future research should triangulate twin designs with molecular genetic data (e.g., polygenic scores) and include richer measures of family processes and school decision-making to unpack mechanisms behind primary and secondary effects.

- Observational design: Timing of tracking is not exogenously assigned; unmeasured confounding remains possible despite checks for parental SES and performance. - Potential gene–shared environment correlation and assortative mating could bias ACE estimates; simulations suggest minimal bias for the delayed group and modest for immediate, but differences could be overstated. - Measurement limitations: Attainment is last reported track, not always final credential; track coding involves harmonization and averaging in some cases; performance reports from multiple sources. - Sample representativeness: Twin samples and the NTR slightly over-represent higher SES and non-migration-background families; generalizability may be affected. - Power and precision: Few low performers experience delayed tracking, making moderation estimates at low performance less precise. - Twin design assumptions (equal environments, additivity) are untestable in full; violations could affect parameter estimates.