Atlas: An improved algorithm for fitting factor analytic variance models by Alison Smith

Atlas home || Conferences | Abstracts | about Atlas

Australasian Genstat Conference 2002
December 4-6, 2002

Busselton, Western Australia, Australia

Organizers
Jane Speijers - Convenor Organising Committee, Peter Clarke - Chairman Programme Committee

An improved algorithm for fitting factor analytic variance models
by
Alison Smith
NSW Agriculture
Coauthors: Robin Thompson (Rothamsted Research), Brian Cullis (NSW Agriculture), Arthur Gilmour (NSW Agriculture)

Factor Analytic (FA) variance structures have been proposed for genotype by environment effects in mixed model analyses of data from series of plant variety trials. Smith et al. (2001) use an FA structure to model the genetic variance matrix across environments whilst simultaneously estimating a separate spatial correlation structure for the errors for each trial. The most commonly used software for fitting these models is ASReml or GenStat (which accesses ASReml routines for REML estimation of variance parameters). The original method of estimation for FA parameters in ASReml involved relatively dense mixed model equations, resulting in substantial computing times for models with large numbers of trials or several terms in the FA model. Additionally the (commonly occurring) case of FA variance structures with less than full rank was not explicitly addressed. In this talk we present a computationally superior algorithm for estimation of FA parameters (see Thompson et al., 2002). The algorithm exploits the regression underpinning the FA model thereby facilitating substantial time savings. It also accommodates reduced rank variance models. This type of model may be useful in the more general setting of multivariate analysis. The algorithm has been implemented in ASReml (accessed via the "XFA" variance model) and will soon be available in GenStat.

References

Smith, A.B., Cullis, B.R. and Thompson, R. 2001. Analysing variety by environment data using multiplicative mixed models. Biometrics, 57: 1138-1147.

Thompson, R., Cullis, B.R., Smith, A.B. and Gilmour, A.R. 2002. A sparse implementation of the Average Information algorithm for factor analytic and reduced rank variance models. ANZJS, Submitted.

Date received: September 20, 2002