四格表OR值点估计的条件与非条件最大似然估计比较  

The Comparison between Conditional and Unconditional Maximum Likelihood Estimate Methods for the Point Estimate of Odds Ratio in 2×2 Table

作　　者：潘建红[1] 陈靖[1] 何春拉[1] 陈平雁[1] 

机构地区：[1]南方医科大学公共卫生与热带医学学院生物统计学系,510515

出　　处：《中国卫生统计》2013年第1期12-14,共3页Chinese Journal of Health Statistics

基　　金：国家自然科学基金资助(30972554);广东省自然科学基金资助(915180200400001)

摘　　要：目的比较四格表OR值的点估计值的条件与非条件最大似然估计方法的优劣,并为其应用提供依据。方法采用MonteCarlo模拟方法,考虑样本量、OR值、对照组暴露概率、病例组与对照组比例等四种参数的不同组合,使用SAS9.2软件编程,比较两种估计方法的相对误差。结果在所有参数组合下,非条件最大似然估计的相对误差均明显小于条件最大似然估计方法。结论对于四格表资料,无论是实际资料的符合情况,还是OR值估计的准确性和计算的便捷性,非条件最大似然估计方法比基于超几何分布的条件最大似然估计方法更值得提倡。Objective To compare the conditional maximum likelihood estimate （CMLE） with the unconditional maximum likelihood estimate（UCMLE） for odds ratio estimation in 2×2 table. Methods Data with different setting of parameters N （ sample size）, OR, % （ exposure rate in control） and n1：n0（the ratio of case to control ） were generated using Monte Carlo in SAS9.2. The relative errors of the two methods were compared. Results Simulation results showed that the relative error of UCMLE method was much lower than CMLE whatever combination of parameters. Conclusion It prefers to apply UCMLE method to CMLE which is based on the hypergeometric distribution according to relative errors and calculation.

关 键 词：OR值 条件最大似然估计 非条件最大似然估计 二项分布 超几何分布 

分 类 号：R195[医药卫生—卫生统计学]