Many characteristics in organisms have a complex genetic basis, so statistical methods must be used to study their behavior under natural or artificial selection. Often, observations are longitudinal; for example, repeated measurements taken on a tree, whether or not an animal is infested with parasites at a number of episodes, litter sizes during reproductive history. This project will develop techniques for studying quantitative genetics of complex longitudinal characters using Bayesian inference and "robust" distributions. Bayesian analysis uses probability to describe uncertainty about unknowns, e.g., a future observation or the mathematical form of a model. "Robust" distributions aim to protect from making incorrect assumptions in statistical models. Focus will be on three types of longitudinal data: 1) continuous (exercise in mice, milk yield in sheep), 2) discrete (presence of a cattle disease) at each of a number of measurements, and 3) counts (litter size of pigs). Computer simulations will examine the performance of the robust models under several genetics scenarios.
Results will extend and improve the battery of statistical methods employed for genetic analysis of complex traits. The methods will draw exact inferences in small samples obtained from experiments or surveys generating longitudinal information.
