Can I Use a Correlation Matrix or Summary Data to Test Mediation in AMOS?

Up to this point, we have tested mediation using the raw data. AMOS will allow you to test mediation with a correlation/covariance matrix as your input, but it will require some addi- tional steps. For simplicity, let’s use our existing example, except we are going to test for mediation with a path model that has composite variables. If we are looking to test that Adap- tive Behavior has a significant indirect effect to Positive Word of Mouth, we need to make sure the summary data is in a format that AMOS can read. For a review of how to use a correlation or covariance matrix as input data, see Chapter 5, page 133.

Figure 6.12 Mediation Test With a Simple Path Model

Once you have formatted and saved your summary data, you need to make sure AMOS has the new data file as the input. In the Analysis Properties window, select the “Indirect, direct, and total effects” option again. With the Bootstrap tab, select “Perform Bootstrap” and change the number of bootstraps to 5,000. Select the “Bias-corrected confidence inter- vals” and change the confidence interval to 95. Since you do not have raw data, you also need to select “Monte Carlo (parametric) bootstrap” option. With a Monte Carlo boot- strap, the bootstrap sample’s means, variances, and covariances will match the summary data In essence, it runs a simulation that will match the summary data provided. If you do not select the Monte Carlo option, AMOS will give you an error message and say it cannot run the bootstrap without the raw data. After making these selections, you are ready to run the analysis.The format for the output with the Monte Carlo simulation will look exactly like the output for the mediation test that used the raw data.Thus, you will need to go to the same place in the output to find your results.

Figure 6.13 Monte Carlo Simulation Needed for Bootstrap With Summary Data Input

Source: Thakkar, J.J. (2020). “Procedural Steps in Structural Equation Modelling”. In: Structural Equation Modelling. Studies in Systems, Decision and Control, vol 285. Springer, Singapore.