Point estimation and interval estimation of mediation effect: multiplicative integration method, Non-parametric Bootstrap and MCMC method

<strong><em><b><i>-</i></b></em></strong><em><i>Abstract In view of the problem that the sampling distribution of the mediating effect ab is often not a normal distribution, scholars have proposed three types of methods that do not require any restrictions on the sampling distribution of ab and are suitable for small and medium samples, including the multiplicative integration method, non-parametric Bootstrap and Markov Chain Monte Carlo (MCMC) method. The simulation technique is used to compare the performance of the three methods in the analysis of mediation effect. The results showed that: 1) The MCMC method with prior information has the most accurate ab point estimation; 2) The MCMC method with prior information has the highest statistical power, but it pays the price of underestimating the error rate of type I, and the non-parametric bias correction The centile Bootstrap method is second in terms of statistical power, but it pays the price of overestimating the type I error rate; 3) The MCMC method with prior information has the most accurate estimation of the intermediate effect interval. The results show that when there is prior information, it is recommended to use the MCMC method with prior information; when the prior information is not available, it is recommended to use the deviation-corrected non-parametric percentile Bootstrap method.</i></em> <strong><em><b><i> </i></b></em></strong> <strong><em><i>Type of Paper--- </i></em></strong><em><i>Review</i></em> &nbsp;