頁籤選單縮合
題 名 | 電腦模擬研究--遺傳迴歸D級模型之迴歸診斷分析=Regression Diagnostics for Class D Regressive Models-A Simulation Study |
---|---|
作 者 | 王曉玫; | 書刊名 | 中國統計學報 |
卷 期 | 37:2 1999.06[民88.06] |
頁 次 | 頁185-199 |
分類號 | 362.62 |
關鍵詞 | 遺傳迴歸D級模型; 迴歸診斷分析; 影響案例; Class D regressive model; Diagnostic analysis; Influential cases; |
語 文 | 中文(Chinese) |
中文摘要 | 本研究乃是針對Bonney (1986) 所提之遺傳迴歸D級模型尋找出與該最適模型不符之家庭。因傳統的統計個案消去法需要每次移去一個家庭,重覆作分析需時甚久。本研究以電腦模擬測試探討影響函數法、一步趨近法及近似一步趨近法等三種迴歸診斷分析方法的信賴度、有效性與精確性,以及取代個案消去法的可行性。結果顯示,影響函數法為最佳取代個案消去法的方法。另以表型之變異數做為診斷分析的參數,因為其可正確辨認出的異常家庭比例最高。以美國愛荷華州莫斯卡丁鎮上,122個至少有兩個子女就讀學校的家庭,所收集之402位有親戚關係個體的身體質量指數 (該資料經年齡和性別調整後) 作迴歸診斷分析。發現該資料可以隱性瘦因子而且父母與子女、兄弟與姐妹之相關相等的遺傳迴歸D級模型解釋之。影響函數法在多數為表型變異數和臨界值為3下,辨認出30.28%的異常家庭;而傳統χ2檢定則辨認40.92%的異常家庭。 |
英文摘要 | Regression diagnostic methods in identifying the families whose phenotypic distributions do not conform to the same genetic model as the majority of the families do are developed and investigated under the Class D regressive model proposed by Bonney (1986). A Monte-Carlo investigation on the reliability, validity, and accuracy of the three diagnostic methods (the empirical influence function, the one-step approximation, and the approximated one-step approach) was conducted to identify the best alternative relative to the exact case-deletion approach which based on examining the changes in each model parameter estimated by excluding one family at a time. Based on simulation studies, the EIF approach is the recommended alternative. The phenotypic variance which identified most frequently and correctly etiotic families is the parameter suggested to be associated with regression diagnostics. The body size families data in Muscatine, Iowa were analyzed. The data included 402 individuals in 122 families ascertained through at least two children in the schools. The phenotype was age-sex-adjusted body mass index which can be explained by a Class D regressive model with a recessive major locus (1eanerallele) and equal mother-offspring, father-offspring and sib-sib correlations. The proposed regression diagnostics identified up to 30.28% of the 122 families as outliers compared to the X2 tests (40.92%). |
本系統中英文摘要資訊取自各篇刊載內容。