Vol. 3, Special Issue 1 (2018)
Analysis of saturated 2-level factorial experiments with gamma distributed response variable
Author(s): Sangita Birajdar (SC Patil)
Abstract: This paper assesses and compares the performance of Generalized linear models (GLM) with respect to the traditional transformation approach for saturated 2-level factorial experiments for Gamma distributed response variable. The evaluation is based on the coverage and expected length of Confidence Interval (CI) for the expected response for factorial effects under these approaches at the prescribed confidence coefficient. The traditional transformation approach outperforms GLM approach for its non-canonical link for two replications with respect to coverage probabilities. However the extent of out-performance does not attain the desired confidence level. An ad-hoc modification to the CI for transformation approach and non-canonical link of GLM approach is suggested, that shows decisive improvement over their performances for small shape parameters.