Testing indirect effects PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Daniel posted on Wednesday, November 09, 2005 - 7:53 am
I ran a SEM with a combination of categorical and continuous indicator variables. I tested for several indirect effects with and without bootstrapping. Without bootstrapping, two out of three of my specific indirect effects were significant, p < .05, and the other wasn't close. With bootstrapping, however, one of the key paths was significant, p < .05, and the other was significant, p < .10. Which result is more valid, if such a thing exists, when modeling with continuous and categorical indicator variables, bootstrapping or not bootstrapping?
 Linda K. Muthen posted on Saturday, November 12, 2005 - 5:55 pm
What is your sample size?
 Daniel posted on Tuesday, December 06, 2005 - 8:41 am
Sorry it took sssso long to get back, but this analysis was put on hold for a while. The sample size is 868.
 Linda K. Muthen posted on Tuesday, December 06, 2005 - 10:38 am
Hmm. I think you should send your input, data, output, and license number at support@statmodel.com so we can see exactly what you are finding.
 Daniel posted on Tuesday, December 06, 2005 - 12:11 pm
Ok, it's on the way
 JOEL WONG posted on Friday, January 29, 2016 - 5:50 am
I am trying to understand when I can infer that two indirect effects in multigroup SEM (using bootstrapping) are significant different from each other. Assuming I have measurement invariance, if the Mplus output shows that the 95% confidence intervals of the indirect effect for group A (e.g., .01 - .05) doesn't overlap with the 95% confidence intervals for group B (e.g., (.06 - .10), is that sufficient evidence that the two indirect effects are significantly different?

Or do I need to use some other method, e.g., the Wald test to demonstrate that the two indirect effects are significantly different?
 Linda K. Muthen posted on Friday, January 29, 2016 - 9:05 am
You can create the indirect effects in MODEL CONSTRAINT and create a new parameter that is their difference. You will obtain a z-test and p-value for this.
 JOEL WONG posted on Saturday, January 30, 2016 - 4:31 am
Thank you, Linda. The MODEL CONSTRAINT is a good idea.

I guess what I want to know is, can we make an inference on significant differences between 2 indirect effects based solely on the Mplus output showing that the 95% confidence intervals for group A (e.g., .01 - .05) doesn't overlap with the 95% confidence intervals for group B (e.g., (.06 - .10)?
 Bengt O. Muthen posted on Saturday, January 30, 2016 - 6:35 am
No, only if those indirect effect estimates are independent as judged by TECH3. Instead, do what Linda suggested.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: