Abstract
In this paper, we propose double power-normal transormation that can effectively be used in the analysis of linear models and also in regression analysis, where achievement of the normality of transformed response variable and the linearity of its expected value as a predictor are treated separately. We examine the effectiveness of our transformation with some examples. To investigate the performances of our double powernormal transformation (DPNT), we compare it with the ordinary power-normal transformation (PNT) by Box and Cox (1964), and also with the marginal and joint powernormal transformations for explanatory and response variables proposed by Goto et al. (1983). From these results, we recommend the DPNT as a complementary transformation of the PNT, in the view of covering a certain kind of extended situation different from those seen for the PNT.
