Psychometric Modelling of Longitudinal Genetically Informative Twin Data

Area of Science:

• Behavioral Genetics
• Psychometrics
• Quantitative Genetics

Background:

• Longitudinal studies assess trait changes over time, enabling analysis of genetic and environmental influences on covariance.
• Traditional univariate A(C)E models can be extended longitudinally, but using sum-scores for self-report data may bias heritability estimates.
• Previous research indicates modeling raw item data is superior to sum-scores in univariate settings, but this hasn't been fully explored in longitudinal twin models.

Purpose of the Study:

• To investigate the bias introduced by using sum-scores in longitudinal A(C)E twin models.
• To evaluate the effectiveness of a latent state twin A(C)E model, incorporating item response theory (IRT), in overcoming sum-score bias.
• To compare the performance of sum-score and IRT-based approaches in longitudinal genetic analyses.

Main Methods:

• Developed a latent state twin A(C)E model combining genetic twin modeling with item response theory (IRT) within a Bayesian framework.
• Conducted two simulation studies to assess bias in heritability and covariance estimates under sum-score and IRT approaches.
• Analyzed empirical data from a two-wave twin study (N=8,016) on social attitudes to illustrate model differences.

Main Results:

• Sum-score approaches in longitudinal A(C)E models led to underestimated heritability and biased covariance estimates.
• The IRT-based latent state twin model also exhibited bias, but to a significantly lesser degree than the sum-score approach.
• Bias increased in both frameworks in the second simulation (ACE model), with the IRT approach consistently showing less bias.

Conclusions:

• Using sum-scores in longitudinal twin studies can lead to biased genetic and environmental parameter estimates.
• The latent state twin A(C)E model integrating IRT is a less biased alternative for analyzing longitudinal phenotypic data.
• Researchers are advised to adopt the IRT approach for more accurate estimation of genetic and environmental influences in longitudinal twin studies.