首页 | 本学科首页   官方微博 | 高级检索  
检索        

REML法和MCMC法在数量性状核心家系遗传方差分量模型中参数估计的比较
引用本文:郜艳晖,姜庆五,孟炜,陈启明,赵耐青,沈福民.REML法和MCMC法在数量性状核心家系遗传方差分量模型中参数估计的比较[J].复旦学报(医学版),2003,30(4):299-303.
作者姓名:郜艳晖  姜庆五  孟炜  陈启明  赵耐青  沈福民
作者单位:1. 复旦大学公共卫生学院流行病学教研室,上海,200032
2. 复旦大学公共卫生学院卫生统计与社会医学教研室,上海,200032
基金项目:国家自然科学基金(39930160)资助课题
摘    要:目的 比较限制性最大似然估计(REML)和马尔可夫链蒙特卡罗(MCMC)方法在数量性状遗传方差分量模型中参数估计的偏差和精度。方法 计算机模拟50个数量性状核心家系数据集,运用SAS软件中的PROC MIXED程序和WinBUGS软件进行参数估计,比较不同样本含量时两法参数估计结果。结果 在本试验参数指定条件下,对固定效应参数,两法估计的相对偏差低于2%;对随机效应,估计偏差较高。小样本时,MCMC法估计结果比REML法更接近真值。结论 在模拟数量性状方差分量模型时,大样本家系资料可根据实际情况选用两法;小样本家系资料,MCMC法可得到比REML法更稳健的估计,推荐使用。

关 键 词:数量性状核心家系  遗传方差分量模型  参数估计  REML法  MCMC法  偏差  精度
修稿时间:2002年10月8日

Comparison of REML and MCMC for Genetic Variance Component Model of Quantitative Trait in Nuclear Family
GAO Yan-hui,JIANG Qing-wu,MENG Wei,CHEN Qi-ming,ZHAO Nai-qing,SHEN Fu-min.Comparison of REML and MCMC for Genetic Variance Component Model of Quantitative Trait in Nuclear Family[J].Fudan University Journal of Medical Sciences,2003,30(4):299-303.
Authors:GAO Yan-hui  JIANG Qing-wu  MENG Wei  CHEN Qi-ming  ZHAO Nai-qing  SHEN Fu-min
Abstract:Purpose To compare restricted estimation maximum likelihood (REML) to markov chain monto carlo (MCMC) for genetic variance component model of quantitative trait in nuclear family. Methods To simulate 50 data sets of nuclear family with quantitative traits, to estimate parameters by the PROC MIXED in SAS and WinBUGS, and to compare two methods on precision and bias of parameters estimation. Results Conditional on our given parameters values relative bias was lower than 2 per cent for fixed effect parameters,and bias for fixed effect parameters was larger than that for random effect parameters whichever method was used. Estimation of MCMC more approach to true vales than estimation of REML to small sample size. Conclusions Both methods can be chosen in simulating genetic variance component model to large sample family data. MCMC method is recommended to small sample family data due to its steadiness.
Keywords:genetic variance component model  quantitative trait  markov chain monto carlo  restricted estimation maximum likelihood  nuclear family
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号