Note : the execution softwares are Matlab software and Stata software. According to the different test functions,the final execution parameters are all done in the mean value processing.
(3)The execution parameters are all processed by averaging, and the modeling steps and processes of structural equation modeling based on Bayesian estimation are referenced(Wuminglong, 2013; Rongtaisheng, 2009). Firstly, the structural equation model of the set exploration is adopted to identify the final suitable form of structural equation model, the index of goodness of fit index of structural equation model and the significant level corresponding to the path coefficient. Then, the confidence interval of path coefficients and mean of path coefficients are calculated by using the structural equation model based on Bayesian estimation, and this study processes the small sample date and correlation distribution. The Markov Chain Monte Carlo sample generation method is adopted for sample and population sampling, and verifies causal relationships among the simmelian ties, high performance work practice system, knowledge fermenting, and knowledge spiral in the theoretical framework, and reveals the transmission mechanism of the simmelian ties on knowledge spiral with the high performance work practice system and knowledge fermenting as conduction variables. Finally, Amos software is used to estimate the relationships among the four variables in the theoretical framework, and also estimate the path coefficients and mean values (whether both the confidence interval of 95% of the significant level and the mean value of significant level include 0 or not.) based on Bayesian estimates, and the path coefficients and standardized mean values (whether both the confidence interval of 95% of the significant level and the mean value of significant level include 0 or not.) based on Bayesian estimates which are analyzed in Table 5.
In order to verify the theoretical hypothesis better, this study uses SEM to highlight the significance of path coefficients and mean values, and to determine which levels they are significant at among 0.001, 0.01 and 0.05. The significances of the path coefficients and the mean values follow the significance level of the corresponding path coefficients of the modified structural equation model. Based on the empirical analysis of Amos software (as shown in Table 5) ,the strong simmelian ties have significantly positive effects on the high performance work practice system (mean value is 0.308, significance level is p<0.05), so the theoretical hypothesis is verified. The weak simmelian ties have positive effects on high performance work practice systems, but the effects are not significant (mean value is 0.017, significance level is p>0.1), and the theoretical hypothesis is verified. The high performance work practice system has significantly positive effects on the knowledge fermenting (mean value is 0.323, significance level is p<0.01), and the theoretical hypothesis is verified. The knowledge fermenting has significantly positive effects on the knowledge spiral (mean value is 0.4401, significance level is p<0.01), and the theoretical hypothesis is also finally verified. The fitting optimization index corresponds to the corrected SEM model.
Table5 The results of path coefficients estimation of SEM based on Bayesian estimation