Statistics and Computing

, Volume 6, Issue 3, pp 289–296

Unconstrained parametrizations for variance-covariance matrices

  • José C. Pinheiro
  • Douglas M. Bates
Papers

DOI: 10.1007/BF00140873

Cite this article as:
Pinheiro, J.C. & Bates, D.M. Stat Comput (1996) 6: 289. doi:10.1007/BF00140873

Abstract

The estimation of variance-covariance matrices through optimization of an objective function, such as a log-likelihood function, is usually a difficult numerical problem. Since the estimates should be positive semi-definite matrices, we must use constrained optimization, or employ a parametrization that enforces this condition. We describe here five different parametrizations for variance-covariance matrices that ensure positive definiteness, thus leaving the estimation problem unconstrained. We compare the parametrizations based on their computational efficiency and statistical interpretability. The results described here are particularly useful in maximum likelihood and restricted maximum likelihood estimation in linear and non-linear mixed-effects models, but are also applicable to other areas of statistics.

Keywords

Unconstrained estimationvariance-covariance components estimationCholesky factorizationmatrix logarithm

Copyright information

© Chapman & Hall 1996

Authors and Affiliations

  • José C. Pinheiro
    • 1
  • Douglas M. Bates
    • 2
  1. 1.Department of BiostatisticsUniversity of Wisconsin-MadisonMadisonUSA
  2. 2.Department of StatisticsUniversity of Wisconsin-MadisonMadisonUSA