Type I Error Rates and Parameter Bias in Multivariate Behavioral Genetic Models.

Area of Science:

• Psychology
• Quantitative Genetics
• Biostatistics

Background:

• Multivariate twin models are widely used in behavioral genetics.
• Likelihood ratio tests (LRT) in these models often yield Type I error rates lower than theoretically expected.
• This conservatism increases with the number of variables, impacting statistical inference.

Purpose of the Study:

• To investigate Type I error rates in Cholesky decomposition and Correlated Factors models.
• To evaluate the performance of a direct symmetric approach for estimating variance components.
• To identify methods that ensure accurate statistical inference in multivariate twin studies.

Main Methods:

• Two simulation studies were conducted.
• Type I error rates were examined for Cholesky decomposition and Correlated Factors models.
• A direct symmetric approach was implemented for estimating variance-covariance matrices.

Main Results:

• Cholesky and Correlated Factors models showed significantly lower than nominal Type I error rates, especially with more variables.
• Parameter estimates were slightly biased in these constrained models.
• The direct symmetric approach yielded Type I error rates consistent with theoretical expectations and unbiased parameter estimates.

Conclusions:

• Model-implied boundaries in standard multivariate twin models can lead to inflated Type II errors and biased parameters.
• The direct symmetric approach offers a more statistically robust alternative, correcting error rates and bias.
• This method is computationally feasible and improves the reliability of genetic and environmental covariance matrix estimation.