罗吉斯模式之适合度检定的比较.pdf

羅吉斯模式之適合度檢定的比較 羅吉斯模式之適合度檢定的比較 林國欽 台南女子技術學院會計系 副教授 摘要 本研究以無母數局部線性估計技巧為基礎，提出新的羅吉斯迴歸模式的適合度檢定 方法。此方法對傳統之Hosmer Lemeshow （2000 ）檢定法，以估計機率分組技巧的適 合度檢定法所可能產生的缺點做了一些改進。就估計本質而言，局部線性估計較優於一 般核估計法，新檢定法也改善Cessie Houwelingen （1991）所提出核估計（kernel ）技 巧為基礎的適合度檢定法。新檢定法可提供較高的檢定力。除推導新檢定統計量之實際 期望值與變異數形式並討論虛無假設之下的條件機率函數有未知參數時，其修正的期望 值與變異數。我們藉由模擬計算檢定統計量之近似分配函數，其特質有進一步的探討。 最後，我們應用新的檢定法說明及分析參個二元資料集。 關鍵詞：局部線性估計；適合度檢定；核估計；羅吉斯迴歸。 - 241 - 台南女院學報 第二十三期 Comparison of Goodness-of-Fit Tests for Logistic Models Kuo-Chin Lin Associate Professor, Department of Accounting, Tainan Woman’s College of Arts Technology Abstract A goodness-of-fit test based on nonparametric local linear smoothing residuals is proposed for logistic regression model with continuous covariates. The proposed test does not suffer from the disadvantages of the chi-square test suggested by Hosmer and Lemeshow (2000) that uses fixed groups of the estimated probabilities. The motivation of applying local linear smoother through cross-validation choosing a suitable bandwidth parameter is to take some advantages over the kernel methods, see Fan (1992). Therefore, the power of the proposed test is higher than that of the Cessie and Houwelingen (1991) test which is based on kernel smoothing with an ad hoc bandwidth. Moreover, the mean and variance of the test statistic are given in explicit expression as well as the adjusted mean and variance are provided when the probability function under test has unknown parameters. The a