Quantitative genetic modeling and inference in the presence of nonignorable missing data

Area of Science:

• Quantitative genetics
• Evolutionary biology
• Statistical modeling

Background:

• Natural selection often acts before traits can be measured, complicating population parameter inference.
• Standard models can yield inaccurate results when missing data are related to the trait of interest.

Purpose of the Study:

• To develop a joint modeling approach that accounts for nonrandom missing data in quantitative genetic analyses.
• To provide a method for accurately estimating additive genetic variance and natural selection when data is missing early in life.

Main Methods:

• Proposed a shared parameter model combining an animal model for phenotypic data and a logistic model for the missing data process.
• Linked the models using additive genetic effects and employed a Bayesian approach with integrated nested Laplace approximations for inference.
• Validated the approach through a simulation study and application to real data from Swiss barn owls (Tyto alba).

Main Results:

• Assuming missing data are missing at random can severely bias estimates of additive genetic variance.
• The joint model correctly estimated genetic parameters and selection in the presence of nonrandom missing data.
• Analysis of barn owl data suggested selection acts on genes before the trait (black spots) is expressed.

Conclusions:

• The proposed joint modeling approach is crucial for accurate estimation of genetic variance and natural selection when data is missing nonrandomly.
• It highlights the importance of accounting for the missing data process, especially when selection occurs before trait measurement.
• This method provides a robust tool for understanding evolutionary processes in wild populations.