Abstract:
|
Test of non-constant variance in linear model has a long history and is still an important matter. Many types of tests exist, but they are limited in use to their own specific cases and sensitive to normality. Here we propose an unified approach to test non-constant variance in linear model. The proposed test overcomes the shortcomings of the existing methods, so that it is robust to normality and is unified in sense that it is applicable in linear model with multi-dimensional response. Simulation studies show that the proposed approach is favorable over the existing tests with moderate sample sizes.
|