Consistency of noncognitive skills and their relation to educational outcomes in a UK cohort

Noncognitive skills, encompassing personality traits, goals, motivations, and preferences, are considered complementary to cognitive abilities. While often linked to socioeconomic outcomes like educational attainment and job performance, the evidence is inconsistent. Many studies rely on cross-sectional data, hindering the assessment of longitudinal consistency. Heterogeneous effects across studies might stem from within-individual variation, measurement error, or the lack of robustness of noncognitive skill constructs. Previous research has shown varying test-retest reliability for different noncognitive skills, generally lower than for cognitive skills. Genetic data can help explore measurement consistency and the robustness of associations with socioeconomic outcomes. While personality traits have shown substantial heritability and temporal stability, the genetic architecture of other noncognitive skills remains less clear. This study uses data from the Avon Longitudinal Study of Parents and Children (ALSPAC) to investigate the longitudinal consistency of nine noncognitive skill measures over 18 years and their associations with educational and labor market outcomes. The research questions are: 1) How consistent are noncognitive skills measured in a large cohort study over 18 years? 2) How strongly do different noncognitive skills associate with socioeconomic outcomes?

The literature on the relationship between noncognitive skills and socioeconomic outcomes is diverse and inconsistent. While associations between the Big Five personality traits and outcomes are well-replicated, many other noncognitive skills show small and heterogeneous effects. Few studies have measured a wide range of noncognitive skills or adjusted for cognitive ability, which strongly attenuates noncognitive-socioeconomic outcome associations. Heterogeneous effects may arise from within-individual variation over time, measurement error, or noncognitive skills not being robust constructs. Limited longitudinal data makes it difficult to assess the consistency of noncognitive skill measurement. Studies on heritability of noncognitive skills yield varied results, with personality traits showing the most consistent evidence of heritability. The genetics of multiple noncognitive skills within the same sample is also rarely explored.

This study uses data from the Avon Longitudinal Study of Parents and Children (ALSPAC), a UK birth cohort study. The sample included 14,899 children, with 7988 having genetic data. Genetic data was obtained from blood, cell line, and mouthwash samples and underwent quality control procedures, including imputation to the Haplotype Reference Consortium (HRCr1.1, 2016) panel. Nine noncognitive skills were assessed using various questionnaires and assessments throughout childhood and adolescence (age 6 months to 18 years): behavioral problems (SDQ), social skills, communication, self-esteem, persistence, locus of control, empathy, impulsivity, and personality (Big Five). Cognitive skills (IQ) were measured at ages 8 and 15. Socioeconomic outcomes included educational achievement (Key Stage assessments at ages 11, 14, and 16, A-levels at 18), employment at 23, and income at 23. Phenotypic correlations were estimated using Pearson's correlation coefficient, adjusted for false discovery rate. Heritability was estimated using genomic-relatedness-based restricted maximum likelihood (GREML) in GCTA. Genetic correlations were also estimated. Regression analyses investigated the associations between noncognitive skills and socioeconomic outcomes, controlling for sex, month of birth, and cognitive ability.

Longitudinal measurement consistency was high for behavioral and communication skills (SDQ), with correlations ranging from 0.36 to 0.73 for parent-reported SDQ and 0.53 for teacher-reported SDQ. Consistency decreased with greater elapsed time. Correlations for other noncognitive skills were generally low. Consistent non-zero heritability estimates and genetic correlations were observed only for behavioral difficulties (SDQ). Aggregate measures of each skill over time showed phenotypic correlations and heritability (h² = 0.1–0.2) for behavior, communication, self-esteem, and locus of control. Associations between noncognitive skills and educational outcomes were observed for skills measured in mid-to-late childhood, but these were at most one-third the size of IQ-education associations. Phenotypic correlations between different noncognitive skills were generally low (mean r = 0.11), except for the SDQ which correlated consistently with other skills. Associations between SDQ measures and labor market outcomes were consistently strong. Many noncognitive and cognitive skills were weakly negatively correlated with non-response. Heritability was strong for parent-reported SDQ at multiple occasions, communication, self-esteem, and locus of control. Limited evidence was found for genetic correlations between different noncognitive skills, except for parent-reported SDQ and communication at age 10. Strong genetic correlations were observed between educational achievement and SDQ, communication, self-esteem, and some Big Five personality subscales.

The findings demonstrate considerable heterogeneity in the consistency of noncognitive skill measures over time. The relatively high consistency of behavioral and communication skills, particularly using the SDQ, may reflect the reliability of the measures employed. Aggregate measures showed improved precision compared to individual measures, suggesting that combining longitudinal data reduces measurement error. However, this approach also averages out potentially important developmental differences. The weak phenotypic correlations between different noncognitive skills support the notion that they represent distinct psychosocial phenomena rather than a single underlying factor. The generally weak associations between noncognitive skills and socioeconomic outcomes contradict some previous findings but support a recent systematic review. The stronger associations of teacher-reported SDQ with educational achievement suggest a potential role of teacher perceptions and responses. The weaker than expected associations with labor market outcomes may be due to the timing of assessment. Study limitations include potential measurement error and low precision of genetic correlation estimates due to low heritability of some noncognitive skills. The results also highlight potential biases from uneven linkage disequilibrium, residual population structure, and assortative mating. The negative correlation between non-response and some skills highlights potential biases in cohort studies.

This study reveals significant heterogeneity in the consistency and predictive power of noncognitive skills. While some measures, like the SDQ, showed strong internal consistency, others exhibited low temporal stability. Associations with socioeconomic outcomes were generally weak compared to cognitive abilities. Future research should focus on improving measurement methods, utilizing multi-source multi-method approaches, and considering the potential for developmental changes and measurement error in noncognitive skills.

Several limitations affect the study's interpretation. First, the possibility of high measurement error in noncognitive measures is acknowledged. However, the measures used were widely validated, and results using aggregate measures minimized this issue. Second, low precision in estimating genetic correlations is attributed to low heritability of many skills. Larger sample sizes are needed to improve these estimates. Third, uneven linkage disequilibrium, residual population structure, and assortative mating may have biased genetic associations, despite applying statistical controls. Fourth, the assessment of labor market outcomes soon after entry may reflect institutional effects rather than long-term career success. Finally, some measures might reflect parental rather than child genes, especially when parent-reported.