

SEM with dichotomous outcome 

Message/Author 


I used Mplus to test a SEM model of the relation between socioeconomic status (SES) and smoking relapse. SES and the mediators are latent variables and smoking relapse is binary. Relapse was coded as a categorical variable. It is my understanding that Mplus is appropriate for this type of model. However, a reviewer made this comment "One important remaining issue, is the model interpretation. Given that the outcome measure (i.e., smoking status) is binary, the output of the model estimation, especially in regards to the dependent variable, has been interpreted as if the dependent variable was continuous. This is inappropriate. It does not appear that Mplus can estimate such a model (i.e., latent variable models with a binary dependent variable) within the latent variable framework. Clarification is needed." I will cite the Mplus manual in my response to the reviewer, but are there any other citations that I can use to defend using Mplus to analyze this data? Thanks in advance. 


If smoking status was put on the CATEGORICAL list, then as a dependent variable a probit regression is estimated using weighted least squares estimation and a logistic regression is estimated using maximum likelihood estimation. If you want further information, send the full output from the analysis and your license number to support@statmodel.com. 


Thanks Dr. Muthen. To clarify, both probit and logistic regressions are estimated? 


No, probit is estimated for weighted least squares estimators and logistic is estimated for maximum likelihood estimators. 

Back to top 

