Multiple Mediation And Multilevel Mediation Analysis Based On Structural Equation Modeling

Mediator is an important statistical concept,which causes the popularity of mediation analysis in psychology,medical science and education in recent years.The current mediation research has developed to the multiple mediation analysis and multi-level mediation analysis.However,some existing analysis of multiple mediation and multi-level mediation concentrates on traditional methods,such as causal-step method,product-of-coefficients method and Bootstrap method,rarely on MCMC method.This paper discusses the MCMC method in multiple mediation analysis based on structural equation modeling and compares it with the results of Bootstrap.The results show that in the multiple mediation analysis with two-specific-mediators,MCMC method with no information performs better than Bootstrap only when the contrast mediating effect and sample size are both large.Otherwise,Bootstrap performs better.In the multiple mediation analysis with two-intermediate-mediators,Bootstrap performs better than MCMC method with no information when the sample size is small,otherwise when the sample size is large.However,MCMC method with information performs better than MCMC method with no information and Bootstrap in both two multiple mediation models under any conditions.In the analysis of 2-1-1 multi-level mediation based on SEM,this paper also compares the estimation performance of MCMC and EM algorithm.The results show that in the 2-1-1 multi-level mediation analysis,EM algorithm performs better than MCMC with on information when J is small,otherwise when J becomes larger.In the same manner,MCMC with information performs better than MCMC with no information and EM algorithm in the 2-1-1 multi-level mediation models no matter what sample size is.In summary,this paper concludes that the MCMC method with information is the best method to do mediation analysis nowadays.