頁籤選單縮合
題 名 | 在複雜抽樣下邏輯斯迴歸模式之適合度檢定=On the Chi-Squared Test for Goodness-of-fit of the Logistic Regression Model under Complex Survey Designs |
---|---|
作 者 | 沈葆聖; | 書刊名 | 中國統計學報 |
卷 期 | 36:1 1998.03[民87.03] |
頁 次 | 頁69-78 |
分類號 | 319.51 |
關鍵詞 | 卡方檢定; 邏輯斯迴歸; 設計效應; Chi-squared test; Logistic regression; Design effect; |
語 文 | 中文(Chinese) |
中文摘要 | 在複雜抽樣(complex survey)下,以獨立二項式分配為依據之卡方檢定X ,並 不適用於分析二元資料。Rao和Scott(1987)提出一階修正(first-order adjustment)卡方檢 定X 。之後,Rao和Scott(1992)從抽樣設計效應(design effect)的觀點提出‘擬’最 大概似估計值('pseudo' maximum likelihood estimates) ,用以估計集群抽樣(cluster sampling)下邏輯斯迴歸(logistic regression)模式之參數。本文證明以 為依據之卡方檢定 滿 足一階修正。此外,從證明中並得知當子族群(subpopulation)設計效應之變異數 大時, 在某些情況下, 比 能較正確的控制型一錯誤(type I error)。模擬結果支持此論點。 |
英文摘要 | The standard chi-squared test statisitc X for binomial proportions is inap-propriate for analysing data from complex sample designs. Rao and Scott (1987) derived a first-order adjusted statistic that takes account of the survey design. Furthermore, Rao and Scott (1992) proposed a 'pseudo' maximum likedlihood estimate for estimating the logistic regression parameters under cluster sampling. In this article, it is shown that the chi-squared test based on satisfies the first-order adjustment. From the proof, it becomes clear that under some conditions X can have a substantially smaller distortion of nominal significance level compared to when the variance of the design effect among subpopulations is large. Simulation results support the argument. |
本系統中英文摘要資訊取自各篇刊載內容。