Using transformation algorithms to estimate (co)variance components by REML in models with equal design matrices

Abstract

The reduction of computational demands on Restricted Maximum Likelihood (REML) procedures by a diagonalization approach is extended to multiple traits by the use of canonical transformations. A computing strategy is developed for use on large data sets employing two different REML algorithms for the estimation of (co)variance
components. Results from a simulation study indicate that (co)variance components can be estimated efficiently at a low cost on a wide range of computer systems.

Die vermindering van rekenaarbehoeftes vir Beperkte Maksimum Aanneemlikheid(REML)-prosedures deur 'n diagonalisasie benadering word uitgebrei na meervoudige eienskappe deur die gebruik van kanoniese transformasies. 'n Berekeningstrategie is ontwikkel vir gebruik op groot datastelle deur die aanwending van twee REML-algoritmes vir die bepaling van (ko)variansie-komponente. Resultate met 'n simulasiestudie dui aan dat (ko)variansie-komponente doeltreffend en teen lae koste op 'n wye reeks rekenaarstelsels beraam kan word.