中华流行病学杂志
中華流行病學雜誌
중화류행병학잡지
CHINESE JOURNAL OF EPIDEMIOLOGY
2013年
9期
927-930
,共4页
梁融%周舒冬%李丽霞%张俊国%郜艳晖
樑融%週舒鼕%李麗霞%張俊國%郜豔暉
량융%주서동%리려하%장준국%고염휘
层次结构数据%组内相关系数%可信区间%Bootstrap自抽样
層次結構數據%組內相關繫數%可信區間%Bootstrap自抽樣
층차결구수거%조내상관계수%가신구간%Bootstrap자추양
Hierarchical data%Intraclass correlation coefficient%Confidence interval%Bootstrapping
探讨Bootstrap自抽样如何在层次结构数据中实现,为组内相关系数(ICC)可信区间的计算方法提供选择.文中利用混合效应模型估计重复测量数据和两阶段抽样数据的ICC及利用Bootstrap法估计ICC的可信区间,比较不同的自抽样模式下ICC可信区间结果.重复测量实例结果显示Bootstrap整群抽样估计的可信区间包含ICC真值,如忽视数据的层次结构特征,Bootstrap方法得到无效的可信区间估计;两阶段抽样实例结果显示整群Bootstrap自抽样方法估计的ICC均数与原样本ICC偏差最小,可信区间宽泛.表明对层次结构数据进行Bootstrap自抽样,需考虑数据的产生机制,即高水平Bootstrap自抽样的统计量估计更接近原样本统计量.
探討Bootstrap自抽樣如何在層次結構數據中實現,為組內相關繫數(ICC)可信區間的計算方法提供選擇.文中利用混閤效應模型估計重複測量數據和兩階段抽樣數據的ICC及利用Bootstrap法估計ICC的可信區間,比較不同的自抽樣模式下ICC可信區間結果.重複測量實例結果顯示Bootstrap整群抽樣估計的可信區間包含ICC真值,如忽視數據的層次結構特徵,Bootstrap方法得到無效的可信區間估計;兩階段抽樣實例結果顯示整群Bootstrap自抽樣方法估計的ICC均數與原樣本ICC偏差最小,可信區間寬汎.錶明對層次結構數據進行Bootstrap自抽樣,需攷慮數據的產生機製,即高水平Bootstrap自抽樣的統計量估計更接近原樣本統計量.
탐토Bootstrap자추양여하재층차결구수거중실현,위조내상관계수(ICC)가신구간적계산방법제공선택.문중이용혼합효응모형고계중복측량수거화량계단추양수거적ICC급이용Bootstrap법고계ICC적가신구간,비교불동적자추양모식하ICC가신구간결과.중복측량실례결과현시Bootstrap정군추양고계적가신구간포함ICC진치,여홀시수거적층차결구특정,Bootstrap방법득도무효적가신구간고계;량계단추양실례결과현시정군Bootstrap자추양방법고계적ICC균수여원양본ICC편차최소,가신구간관범.표명대층차결구수거진행Bootstrap자추양,수고필수거적산생궤제,즉고수평Bootstrap자추양적통계량고계경접근원양본통계량.
This paper aims to achieve Bootstraping in hierarchical data and to provide a method for the estimation on confidence interval (CI) of intraclass correlation coefficient (ICC).First,we utilize the mixed-effects model to estimate data from ICC of repeated measurement and from the two-stage sampling.Then,we use Bootstrap method to estimate CI from related ICCs.Finally,the influences of different Bootstraping strategies to ICC' s CIs are compared.The repeated measurement instance show that the CI of cluster Bootsraping containing the true ICC value.However,when ignoring the hierarchy characteristics of data,the random Bootsraping method shows that it has the invalid CI.Result from the two-stage instance shows that bias obsered between cluster Bootstraping's ICC means while the ICC of the original sample is the smallest,but with wide CI.It is necessary to consider the structure of data as important,when hierarchical data is being resampled.Bootstrapping seems to be better on the higher than that on lower levels.