On Link Selection in Generalized Linear Models

  • Claudia Czado
Part of the Lecture Notes in Statistics book series (LNS, volume 78)

Abstract

Generalized Linear Models (GLM) are extended to include the choice of a parametric link transformation family to improve fit over the standard GLM analysis in some data sets. However, the additional estimation of the link parameter results generally in higher estimated variances of the parameter estimates and numerical instability compared to the case when the correct link is known apriori. This paper extends two ideas developed for binary regression with parametric link (Czado [3]) — standardization and parameter orthogonality — to GLM’s aimed at reducing the variance inflation and numerical instability. Simple standardized link families for GLM’s are introduced and their usefulness are illustrated by an example.

Keywords

Byssinosis 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. [1]
    Cox, D.R. and Reid, N. (1987). Parameter Orthogonality and Approximate Conditional Inference. J. Roy. Statist. Soc. (B) 49 1–39.MathSciNetMATHGoogle Scholar
  2. [2]
    Czado, C. (1989) Link Misspecification and Data Selected Transformations in Binary Regression Models Ph.D. Thesis, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, N.Y.Google Scholar
  3. [3]
    Czado, C. (1991) On Selecting Link Transformation Families in Binary Regression Analysis submitted.Google Scholar
  4. [4]
    Czado, C. and Santner, T. J. (1992) The Effect of Link Misspecification on Binary Regression Analysis. to appear in J. Statist. Plan. Inf.Google Scholar
  5. [5]
    Czado, C. and Santner, T. J. (1992) Orthogonalizing Parametric Link Transformation Families in Binary Regression Analysis. to appear in Canad. J. Statist.Google Scholar
  6. [6]
    Li, K.C. and Duan, N. (1989) Regression Analysis under Link Violation. Ann. Statist. 17 1009–1052.MathSciNetMATHCrossRefGoogle Scholar
  7. [7]
    Higgins, J.E. and Koch, G.C. (1977) Variable Selection and Generalized Chi-Square Analysis of Categorical Data applied to a Large Cross Sectional Occupational Health Survey. Int. Statist. Rev. 45, 51–62.Google Scholar
  8. [8]
    Stukel, T. (1988). Generalized Logistic Models. J. Amer. Statist. Assoc. 83 426–431.MathSciNetCrossRefGoogle Scholar
  9. [9]
    Taylor, J. M. G. (1988) The Cost of Generalized Logistic Regression. J. Amer. Statist. Assoc. 83 1078–1083.MathSciNetCrossRefGoogle Scholar

Copyright information

© Springer-Verlag New York, Inc. 1992

Authors and Affiliations

  • Claudia Czado
    • 1
  1. 1.York UniversityNorth YorkCanada

Personalised recommendations