Estimating parameters of a singular variance matrix in ASReml
by
Arthur Gilmour
NSW Agriculture
Coauthors: Robin Thompson (Rothamsted Research)

The talk will describe some of the variance models available for estimating variance parameters in a multitrait context. Our focus will be to discuss how well the models cope when the data supports the matrix being singular. We will demonstrate the situation with two animal breeding examples. In these examples, we are estimating a Sire variance component matrix and an Error Variance component matrix for four or six traits where some traits are highly correlated at the genetic level. The Cholesky and Factor Analytic models appear best in this situation.

Date received: September 11, 2002