有限样本场合下Meta分析中的假设检验问题

发布时间：2018-12-14 15:29

【摘要】：荟萃分析(meta-analysis)是一种合并针对同一科学问题的不同科研活动所获得之结果的统计推断方法。本文考察了荟萃分析中常用的随机效应模型的异质性方差的假设检验问题,依托谱表示,给出了原假设下了限制性似然比检验(restricted likelihood ratio test,即RLRT)统计量的精确分布的等价形式。使用参数型bootstrap法模拟了上述统计量的精确分布,并据此设计出了相应的有限样本下的假设检验方法。模拟研究表明,在有限样本场合,所提出的方法与基于大样本理论的方法检验和基于渐近分布的RLRT相比有更好的性质。在样本量较大时,所提出的方法与已有方法的表现非常类似。最后,作为演示,所提出的方法被用于分析一组卡介苗数据。该方法适用于小样本场合,有效地弥补了传统的基于大样本理论的方法之缺点,使得今后在遇到荟萃分析中异质性检验的问题时,能够获得一种新的解决方案。
[Abstract]:Meta-analysis (meta-analysis) is a statistical inference method that combines the results obtained from different scientific activities for the same scientific problem. In this paper, we investigate the hypothesis testing problem of heterogeneity variance of random effect models commonly used in meta-analysis. Based on spectral representation, we give the restrictive likelihood ratio test (restricted likelihood ratio test,) under the original hypothesis. That is, the equivalent form of exact distribution of RLRT) statistics. The accurate distribution of the above statistics is simulated by using the parametric bootstrap method, and the hypothesis testing method under finite samples is designed accordingly. Simulation results show that the proposed method has better properties than the method based on large sample theory and RLRT based on asymptotic distribution in the case of finite samples. When the sample size is large, the proposed method is very similar to the existing method. Finally, as a demonstration, the proposed method is used to analyze a set of BCG data. This method is suitable for the small sample situation, which effectively makes up for the shortcomings of the traditional method based on the large sample theory, and makes it possible to obtain a new solution in the future when the problem of heterogeneity test in the meta-analysis is encountered.
【学位授予单位】：云南财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O212.1

【参考文献】