INTERMEDIATE VARIABLES IN CB SEM MODEL (PART 2)
This part continues with part 1. Part 1 details here: Click here.
In this part 2, we will not mention how to test regression estimates in SEM, but will discuss in more depth about intermediate variables in CB SEM. For the intermediate variable in PLS SEM, we will have a separate introductory analysis.
In part two, we will use a different model from part one. The goal of this part 2 is to guide how to determine the p-value of standardized relationships in a bootstrap SEM model.
The research model is as follows:
Many people ask the question, why do we look at the regression weight table after running the analysis and see that all relationships have p-value ≤ 5%, but why do we conclude “No mediation” (no mediation). This is because, we need to note, the model results after running SEM analysis are the results of the original model, but when analyzing the intermediate variables we need to rely on the results of the bootstrap analysis. For example, after running SEM, the regression weight analysis results (Figure 1) show that TV → SAT has a p-value ≤ 5%, SAT → LOY has a p-value ≤ 5%, and TV → LOY means having p-value ≤ 5%. So usually we will conclude that TV has a partial mediation role. Theoretically, the mediation analysis is correct, but we mistakenly believe that the regression weights need to be based on the bootstrap method for conclusions to be accurate.
Figure 1 – Results of regression weight analysis of the original model
Now, let’s run the bootstrap to test the relationships in SEM and evaluate mediation. The results of running bootstrap are as shown in Figure 2 (note that this is an unstandardized result, because this relationship is defined by the user, see more details in part 1).
Figure 2 – Results of user influence analysis
The results show that the indirect effect of TV is not statistically significant, so TV in this case does not exactly play a mediating role (No mediation). At this point, we can see the difference between the original model results and the bootstrap model results. Again, to evaluate the mediating relationship, we need to rely on the p-value of the bootstrap model.
Next, we will show you how to determine p-value for relationships with standardized regression weights.
Figure 3 – Steps to get the standardized regression weights of the bootstrap model
First, in the Output window of Amos, we select Matrices and choose one of the normalization effects (number 2). Next, under Estimates/Bootstrap, select Estimates. On the right will appear the corresponding regression weight. In Figure 3, the normalized direct effect results are shown.
Figure 4 – Steps to get p-value
The results are summarized in the following table:
It can be seen that the results of analyzing the role of the intermediate variable in the SEM model through the bootstrap method will give much different results than if you only rely on the original model to conclude.
How to evaluate the role of intermediaries in CB SEM is done using AMOS. We would like to stop here. If you need to analyze and process data to achieve expectations, please contact Le Minh Data Processing Services. We will strive to support you in achieving your research goals.
In the next topic, we will present in detail how to evaluate intermediate variables in PLS SEM using SmartPLS software.
