论文信息 - A Comparative Study of Analyses of 2 n Factorial Experiments with a Poisson Distributed Response Variable

A Comparative Study of Analyses of 2 n Factorial Experiments with a Poisson Distributed Response Variable

Many industrial experiments involve a Poisson response where a square root (SQRT) or log transformation (LOG) is employed to obtain properties needed to use analysis of variance (ANOVA). These approaches have many drawbacks. Generalized linear models (GLM) are an attractive alternative. This article attempts a comparative study among the above-mentioned approaches, for 2 n full factorial experiments with a Poisson response. Each of these methods gave very poor performances when the expected response was small. A modification is suggested to improve this behavior. The modified GLM method turns out to be uniformly best amongst all approaches.

H. V. Kulkarni | V. V. Patil | H. Kulkarni | V. V. Patil

[1] R. H. Myers,et al. A TUTORIAL ON GENERALIZED LINEAR MODELS , 1997 .

[2] D. Firth. Bias reduction of maximum likelihood estimates , 1993 .

[3] R. H. Myers,et al. Examples of Designed Experiments with Nonnormal Responses , 2001 .

[4] F. Garwood,et al. i) Fiducial Limits for the Poisson Distribution , 1936 .

[5] Eric R. Ziegel,et al. Generalized Linear Models , 2002, Technometrics.

[6] David C. Woods,et al. Designs for Generalized Linear Models With Several Variables and Model Uncertainty , 2006, Technometrics.

[7] P. McCullagh,et al. Generalized Linear Models, 2nd Edn. , 1990 .

[8] David M. Steinberg,et al. Fast Computation of Designs Robust to Parameter Uncertainty for Nonlinear Settings , 2009, Technometrics.

[9] R. H. Myers,et al. Confidence Interval Coverage for Designed Experiments Analyzed with GLMs , 2001 .