Tsai, Chih-Ling (1988, ASQC and the American Statistical Association) University of California, Davis, CA
A two-stage procedure is proposed to achieve normality and homoscedasticity of the classical error assumptions and to remove nonlinearity of the regression function. This systematic approach involves power transformations and reparameterization. Test statistics are obtained to assess the necessity of power transformations and the validity of homoscedasticity of the errors. In addition, nonlinearity measures are provided to diagnose the accuracy of linearization-based approximate confidence regions for parameters. Numerical illustrations of the two-stage procedure are presented.