摘要：The Box-Cox (1964) transformation model (BC model) is widely used to examine various problems. The likelihood function under the normality assumption is misspecified, and the maximum likelihood estimator (BC MLE) cannot in general be consistent. However, under the “small sigma” assumption described in Bickel and Doksum (1981), the BC MLE can be consistent. It is, therefore, necessary to test whether this assumption is satisfied when the BC model is used. In this paper, we propose a new test of whether or not the BC MLE can be used based on the estimator proposed by Nawata (2013). We then analyze length of hospital stay for type 2 diabetes patients hospitalized for educational programs about managing diabetes at home. A dataset of 970 patients collected from 27 general hospitals in Japan is used.
关键词：Box-Cox transformation model; power transformation model; diabetes; length of stay (LOS)
