國立政治大學統計學系
學 術 演 講
主講人：王俊毅教授(亞洲大學健康產業管理學系)學 術 演 講
題 目：Goodness-of-fit tests for the multinomial logistic regression model under
retrospective sampling
時 間：民國112年3月6日 (星期一) 下午1：30
地 點：國立政治大學逸仙樓050101教室
摘 要：
The multinomial logistic regression model is widely applied to explore the relationship of explanatory variables and a response variable with multiple categories. It is known that model fitting is always critical and should be assessed before looking into analysis results. However, limited goodness-of-fit tests are available for checking the multinomial logistic model under retrospective sampling. In this talk, I will show our proposed methods, including a Kolmogorov-Smirnov-type (KS-type) and a moment-type test, for assessing the fit of the multinomial logistic model. Moreover, we manifests that the generalized Hosmer-Lemeshow (GHL) goodness-of-fit test for the multinomial logistic model derived under retrospective sampling is equivalent to that derived under prospective sampling. Simulation studies were conducted to explore and compare the finite sample behavior of the studied tests. Results show that the KS-type and GHL tests have comparable powers and the moment-type test is obviously more powerful under the studied scenarios. Finally, an example will be used to illustrate the testing procedure.
Keywords: goodness-of-fit test; Hosmer-Lemeshow test; Kolmogorov-Smirnov test; moment test; multinomial logistic model; retrospective sampling.