论文部分内容阅读
目的 探讨方差比法在MCMC收敛性诊断中的应用以及收敛性诊断的重要性 ,以期为贝叶斯统计方法的研究和MCMC的应用提供有用的参考及帮助。方法 为演示方差比法诊断的结果 ,对一个正态混合分布的MCMC计算实例进行诊断 ,诊断结果由WinBUGS软件和CODA软件给出。结果 方差比法诊断结果反映出当没有对数据进行事先分组时 ,MCMC模拟结果经 10 0 0 0次迭代没有达到收敛状态 ,此时参数的MCMC估计值有较大偏倚。结论 方差比法能够对尚未收敛的马尔科夫链给出较为明确的诊断 ,其诊断结果简单明了易于应用
Objective To explore the application of variance ratio in the diagnosis of MCMC convergence and the importance of convergence diagnosis in order to provide useful reference and help for the study of Bayesian statistical methods and the application of MCMC. Methods In order to demonstrate the result of the Variance method, a case of normal mixed distribution MCMC was diagnosed. The diagnostic results were given by WinBUGS software and CODA software. Results Variance ratio diagnostic results showed that when the data were not grouped in advance, the MCMC simulation results did not reach the convergence state after 10 0 0 iterations, and the MCMC estimates of parameters were greatly biased. Conclusion The variance ratio method can give a more definite diagnosis of Markov chains that have not converged, and the diagnosis results are simple and easy to apply