Message/Author |
|
|
I have a question about mediation analyses. Normally I work with SAS, but because I want to do a mediation analysis with a dichotomous outcome and dichotomous mediator, I switch to Mplus and trying to learn and understand this program. For the mediation analyses, I am using WLSMV estimator, with BCbootstrap CI's. In the output I get indirect and direct results, but I have some difficulty interpreting these results. Are these direct and indirect results interpretable as an OR? Or do I have to standardize, and are the estimates then OR’s? Is it possible to say something about the effect size? I look forward to a response and thank you in advance. |
|
|
With WLSMV, you obtain probit regression coefficients which cannot be exponentiated to odds ratios. The indirect effects concern the latent response variables underlying the observed categorical variables. For logistic regression and odds ratios, you need to use maximum likelihood estimation. You can use MODEL CONSTRAINT to specify the indirect effects according to the formulas described in the following paper on the website: Muthén, B. (2011). Applications of causally defined direct and indirect effects in mediation analysis using SEM in Mplus |
|
|
Hi. I am trying to analyze a mediation model with binary mediator and binary outcome. I used to use STATA's "binary_mediation" commend but this time I need to apply sampling weight. I wonder if I can handle this issue with M-plus. I am pretty new to this program so your advice will be really helpful. Thank you in advance! |
|
|
The best way to do that is shown in: Muthén, B. (2011). Applications of causally defined direct and indirect effects in mediation analysis using SEM in Mplus. This paper is on our website under Mediation Modeling, including examples and Mplus scripts. |
|
db40 posted on Wednesday, April 09, 2014 - 4:08 am
|
|
|
Dr. Muthen, I have submitted a paper which used mediation with binary outcomes (using model constraint). One of the reviewers has flagged this an issue asking for more detail on this this technique has been implemented. However Im not sure how else I can write this. Is there any references I can use to point them to this technique? thank you, |
|
db40 posted on Wednesday, April 09, 2014 - 4:10 am
|
|
|
apologies for the double post - I meant to say can I use this paper as a reference? Muthén, B. (2011). Applications of causally defined direct and indirect effects in mediation analysis using SEM in Mplus. |
|
|
Yes. |
|
db40 posted on Thursday, April 10, 2014 - 4:47 am
|
|
|
Dear Dr Muthen, I have come across the problem where the product of my estimates is not significant although I do have significant AB paths. In an earlier post I was told I could use the Bayes estimator - which I have and it has produced what appear to be significant estimates for those paths. http://goo.gl/ZhJ6g1 However, in this study the use of the Bayes estimator after using MLR was to negate the issue of an nonsignificant indirect path - how would this be written into the method section? Would it be normal procedure to mention the use of MLR produced non-normal product estimates therefore Bayes was used to compensate? Apologies, I'm a primarily a psychologist not a statistician. |
|
|
See the following FAQ on the website: Indirect effect insignificant while both paths significant ML with bootstrapping would be comparable to Bayes. |
|
db40 posted on Friday, April 11, 2014 - 1:38 am
|
|
|
Apologies Linda, I should have mentioned I am using Model Constraint - Bootstrapping is out in this case. |
|
|
I just ran an example with MODEL CONSTRAINT and BOOTSTRAP and there was no problem. |
|
Back to top |