Generalized least squares, taylor series linearization and fisher's scoring in multivariate nonlinear regression

Edward F. Vonesh*, Hao Wang, Dibyen Majumdar

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

19 Scopus citations

Abstract

In this article, we consider a general multivariate nonlinear regression setting in which the marginal mean and variance–covariance structure share a common set of regression parameters. Estimation is carried out via iteratively reweighted generalized least squares (IRGLS) that entails repeated application of Taylor series linearization and estimated generalized least squares (EGLS). Under normality, this IRGLS procedure is equivalent to Fisher's method of scoring and hence maximum likelihood estimation (MLE). However, estimates from this procedure are also shown to minimize a bias-corrected generalized least squares objective function that does not require the assumption of normality. Under fairly mild regularity conditions, the resulting estimates are consistent, asymptotically normal, and–under normality assumptions–asymptotically efficient. The estimates are compared against those obtained as solutions to the usual generalized estimating equations (GEE) using both simulation and numerical examples.

Original languageEnglish (US)
Pages (from-to)282-291
Number of pages10
JournalJournal of the American Statistical Association
Volume96
Issue number453
DOIs
StatePublished - Mar 1 2001

Keywords

  • Least squares
  • Multivariate nonlinear regression
  • Objective function
  • Unbiased estimating equations

ASJC Scopus subject areas

  • Statistics and Probability
  • Statistics, Probability and Uncertainty

Fingerprint

Dive into the research topics of 'Generalized least squares, taylor series linearization and fisher's scoring in multivariate nonlinear regression'. Together they form a unique fingerprint.

Cite this