Improving accuracy and precision of heritability estimation in twin studies through hierarchical modeling: Reassessing the measurement error assumption

Gang Chen ^* Dustin Moraczewski Paul Taylor

• National Institutes of Health (NIH), Bethesda, United States

In this study, we demonstrate the need for improvement in the conventional structural equation modeling (SEM) used for estimating heritability when applied to trait data with measurement errors. The critical issue revolves around an assumption concerning measurement errors in twin studies. In cases where traits are measured using samples, data is aggregated during preprocessing, with only a centrality measure (e.g., mean) being used for modeling. Additionally, measurement errors resulting from sampling are assumed to be part of the nonshared environment and are thus overlooked in heritability estimation. Consequently, the presence of intra-individual variability remains concealed. Moreover, recommended sample sizes are typically based on the assumption of no measurement errors.We argue that measurement errors in the form of intra-individual variability are an intrinsic limitation of finite sampling and should not be considered as part of the nonshared environment. Previous studies have shown that the intra-individual variability of psychometric effects is significantly larger than the inter-individual counterpart.Here, to demonstrate the appropriateness and advantages of our hierarchical linear modeling approach in heritability estimation, we utilize simulations as well as a real dataset from the ABCD (Adolescent Brain Cognitive Development) study. Moreover, we showcase the following analytical insights for data containing non-negligible measurement errors:i) The conventional SEM may underestimate heritability.ii) A hierarchical model provides a more accurate assessment of heritability.iii) Large samples, exceeding 100 observations or thousands of twins, may be necessary to reduce imprecision.