Data Analysis Papers
All the Cows in Canada: Massive Mixed Modeling with the HPMIXED Procedure in SAS® 9.2
Wang, Tianlin; SAS Institute 2009
Abstract
Linear mixed models enable you to borrow strength for estimation, testing, and prediction by linking observations
through a covariance model. Many applications of mixed models lead to computational problems that can only be
described as huge. An example is the prediction of breeding value for a large population of livestock with a covariance
model that links animals through their pedigrees. The HPMIXED procedure fits large, sparse linear models by using a
number of specialized high-performance techniques, which include sparse matrix technology and algorithms for solving
the mixed model equations and for optimization. These techniques are tailored to very large systems. They can result
in significant savings of both memory and CPU resources and can actually make feasible the fitting of some models
that would be infeasible with other approaches. This paper introduces applications of large mixed models, discusses
the specialized techniques of the HPMIXED procedure to handle them, and demonstrates the utility of the procedure
with examples from agriculture and genomics.
