Message/Author 


I am new to both Mplus and path analysis. I want to see if M mediates the relationship between X and Y. Both M and Y are coded dichotomously. I also have several control variables (CV) I want to include. I planned on using the ML estimator given that it is more efficient than WLSMV. Here is the code I have been using: Variable: names are Y X M CV1 CV2 CV3; usevariables Y X M CV1 CV2 CV3; categorical are M Y; Analysis: bootstrap=1000; Estimator=ml; Model: Y on X M CV1 CV2 CV3(p1); optimism1 on X CV1 CV2 CV3(p2); Model Constraint: NEW(XtoY); XtoY = p1*p2; OUTPUT: cinterval(bcbootstrap); However, I found that with the ML estimator, I cannot test for indirect effects specifically (hence the model constraint option). This did not provide a stated test of the indirect effect like when I used the default estimator. Instead, it provided the estimates (though not the OR) for the new parameter XtoY. How do you interpret these results? Since my main research question surrounds the indirect effect, would it be better to use the default estimator so I can include the stated indirect effects test (versus the model constraint)? Also: 2) Should I run the model without the potential mediator so I can compare the fit statistics across the two models? 3) Is my coding for the control variables correct? 


There are a couple of problems in this setup: (1) only one parameter label is allowed per line, so instead of saying y on x m (b); you should say y on x m (b); (2) With a binary M and Y, you should use "counterfactuallydefined" indirect and direct effects as described on our web page: http://www.statmodel.com/Mediation.shtml This is described in our 2016 Regression and Mediation book. Chapter 8 gives a thorough discussion of how to handle the control variables. 


Thank you for your prompt reply. I know this is a 'basic' question, but could you tell me what the second parameter is in the code "y on x m (b)"? Is it the (b)? I saw that on example code using the model constraint option and assumed I needed it, without knowing what the (b) referred to. 2) Thank you for that guidance. I saw that the book is not available for order right now do you know when it will be available? Do you know of any other sources that would be helpful in the interim? 


Have a look at mediation modeling in either our Topic 1 video and handout on the website, or get the Regression and Mediation book. 


Regarding the RMA book, US Postal Service has a problem right now with applying the correct shipping cost but this should be cleared up in a day. 


OK thank you. I was just able to order the book this morning. I will let you know if I have any further questions. 


Yes, the book ordering is now operating smoothly again. 


Dear Drs Muthen & Muthen, I have two questions regarding path analysis with manifest/observed variables. 1) I ran two simple mediation models with the same x and m but different ys, and the coefficient for the xm path had a slightly different value in each. Could this be an artefact of using FIML for missing data? If not, what else could explain the discrepancy? 2) Is there a recommended way to correct for multiple testing when running a path model? Thank you kindly, Ann 


1) Yes, it could be due to the missing data handling. 2) You may want to ask this on SEMNET. 


Thank You! 

Back to top 