Is Bootstrap Available for Indirect Effects in Structural Equation Models in JMP Student Edition?

[#1]

I am using JMP Student Edition to conduct a mediation analysis using the Structural Equation Models platform. My model contains three observed variables:

REG_PRESS as the predictor, GOV_STR5 as the mediator, and CLIENT_OUT as the outcome.

The model runs successfully, and I can obtain the Indirect Effects and Specific Indirect Effects results. For the specific mediated relationship through GOV_STR5, JMP reports an indirect effect of approximately 0.094, together with the standard error, Wald Z statistic and p-value.

For my analysis, I would like to obtain a 95% bootstrap confidence interval for the indirect effect using 5,000 bootstrap resamples. However, I cannot find a bootstrap option in the red-triangle menu of the Structural Equation Models platform in the Student Edition. I have attached a screenshot showing the available options.

Could someone please advise whether bootstrapping of indirect effects is available in JMP Student Edition? If it is available, where can I access the option and specify the number of bootstrap samples? If it is not directly available through the SEM platform, is there a recommended JSL approach or alternative procedure within JMP for obtaining a 5,000-resample bootstrap confidence interval for the mediation effect?

Many thanks for any guidance.

[#2]

Hi @Madhu_Acharyya, 

There are two ways to get bootstrap confidence intervals (CIs) for indirect effects. For your case, I recommend you follow option #1 below as it's optimized to be very fast.

OPTION #1

After fitting your model, to obtain bootstrap CIs for all indirect effects, the best approach is to go to the main red triangle menu (not the one specific to the fitted model, but the one above the Model Specification)--look under "Inference" and you'll see the option "Bootstrap Inference"

This brings up a dialog with options for a variety of statistics that can be bootstrapped. For the fastest performance, just check the Indirect Effects option. You can also change the number of samples to 5,000 and set a random seed for replication. The default is to provide a bias-corrected bootstrap,

After you click OK, you'll get the Bootstrap Inference information in the report:

This might be all you need. However, you'll also get a new data table with all the bootstrap results. Your example only has one indirect effect so the data table should only have two columns; one identifying the bootstrap sample and another with the indirect effect estimate for the given sample. There's a script saved to that data table called "Distribution" 

You can run this script to get a nice summary of the empirical distribution

OPTION #2

Every statistic in the SEM report can also be bootstrapped by using the right-click menu. However, this approach is slower than what's described in option #1. Thus, when possible, option #1 should be preferred.

In your particular example, the model only has ONE indirect effect, so the result of the Indirect Effects option and the Specific Indirect Effects option will be identical. With more complex mediation models, the Specific Indirect Effects option provides different estimates (because it decomposes indirect effects into specific components). 

For the case of SPECIFIC Indirect Effects, the right-click menu option is currently (version 19.2) the only way to get bootstrap CIs.

After fitting the model, hover over the estimate that you want to bootstrap. Right-click and select the Bootstrap option.

Click OK on the resulting dialog and you'll get the table of bootstrap samples with estimates. As before, run the Distribution script to see the empirical distribution.

HTH,

~Laura

[#3]

Thank you, Laura, for your comments and guidance. I tried Option 1 as you suggested; however, it does not allow me to select the Indirect Effect, as this item is not highlighted (please see the attached screenshot). Is there anything I might be missing?

[#4]

Could you please share the specifics of the model you're estimating? In your screenshot, I see the path diagram partially... and I think you might not have an indirect effect in this model (looks like 3 predictors and 2 outcomes but without mediators?). If you do have a mediator here, then did the model converge? Lastly, can you see the indirect effects in the report when you click on the red triangle menu of the fitted model?