I am new to Mplus and have recently ran a moderated mediation model. I had my predictors mean centered and the product terms in the equation. The main effect of one of the predictors on the dependent variable yielded a p value of .047. Initially, I took this as being < .05. However, I later learned about the indirect command and when using the predictor ind outcome command the pvalue for the indirect effet is .056. How do I report these findings? I know this question is basic, but I would appreciate your help.
There are several good references in the user's guide reference list under MacKinnon that can help you understand direct and indirect effects.
Yan Liu posted on Friday, October 26, 2012 - 10:42 pm
Dear Dr. Muthen
I am running a multilevel mediational model with random intercpet only. It has one dependent variable (Y, continuous), three mediators (M1, M2, M3)and one predictor (X). M3 is also the outcome for M1 and M2.
Whenever M1 and M2 entered into the model simultaneously, the relation between IV and M3 changed from positive to negative, but was positive if I only added either M1 or M2. I guess there is a suppresion effect. The bivariate correlations between mediators range from 0.42 to 0.66. The correlations between IV and mediators are 0.38, 0.52 and 0.77.
My questions are: (1) how can I deal with this suppresion effect? (Can I keep the suppressor in the model based on my theorectical model? If so, how can I address the negative relation between Y and X, which is supposed to be positive) (2) Why did the model fit become very good when I changed the estimator from MLR to WLSMV?
ANALYSIS: TYPE = TWOLEVEL; ESTIMATOR=WLSMV; MODEL: %WITHIN% M1 ON sex X ; M2 ON sex X; M3 ON sex X M1 M2; Y ON sex X M1 M2 M3; %BETWEEN% X M1 M2 M3 Y;
Regarding the suppression effect, I would ask that question on SEMNET or a general discussion forum.
Regarding the difference between MLR and WLSMV, please send the two outputs and your license number to email@example.com.
anonymous Z posted on Friday, December 04, 2015 - 1:49 pm
I am using cross-sectional data to fit a mediation model.
X - M - Y
I fit an alternative model by switching M and Y. The fit indices for both models are good. According to your experience, is this very common that fit indices for target model and alternative model are usually very similar? Have you done any writing on this or would you recommend any relevant readings?
Hello! Can I ask for clarification on how the model indirect command works? I can tell from the handbook that it uses a standard non-parametric bootstrapping method, but I canít figure out the default number of draws in Mplus. Would you happen to know this?