Apologies in advance if this is a repeat posting, my system hiccupped a while ago and I lost track of my posting.
I am trying to model a mediation model in which a X is a binary predictor (treatment vs. control), M is a mediator (continuous), and Y is a Binary outcome. I am interested in the indirect effect of X on Y as it passes through M. My syntax and ensuing error are shown below:
title: Mediation Dichotomous data: file= 'C:\test.txt'; variable: names=X M Y; USEVARIABLES ARE X M Y; CATEGORICAL ARE Y X;
MODEL: Y ON X; Y ON M X;
MODEL INDIRECT: Y VIA M X;
*** ERROR in Variable command CATEGORICAL option is used for dependent variables only. X is not a dependent variable.
The CATEGORICAL option is for dependent variables. The variable x is an independent variable. You should take x off of the CATEGORICAL list. The scale of independent variables does not affect model estimation.
Thanks, for the clarification on how to state the measurement quality of the variables. I'd like to continue this line with a more simple question about how to create my model.
How do I estimate a mediation model in which the effects of X on Y are mediated by M, and in which X and Y are binary? (I see from you earlier comment that I do not declare X as binary.) I want to obtain the estimates of the indirect effects of X on Y through M and the SE of the indirect estimate? In this model the regression of M on X is a linear regression and Y on X and Y on M & X are logistic regressions.
You can use MODEL INDIRECT to obtain the indirect effects. In the input you showed earlier, you will obtain probit regressions with the default weighted least squares estimator for categorical dependent variables and linear regressions for continuous dependent variables. The indirect effect is the product of the two regression coefficients. You will obtain standard errors using the Delta method and can also request bootstrap standard errors.
Linda I am still not hitting the mark. I ran the Syntax below, to estimate the indirect effects. But as can be seen from the output below, the estimate for the effect from X to Y via M is 0. The data input to this analysis was generated randomly, and was designed to have a net indirect effect.
I do not know if this is related, but the model Indirect statement is the only keyword with the colon that does not change to blue in the syntax file. All others like data:, variable:, Model:, etc change color as soon as the colon is applied.
Once again, Thanks in advance Bill
title: Mediation Dichotomous data: file= 'C:\test.txt'; variable: names=X M Y; USEVARIABLES ARE X M Y; CATEGORICAL ARE Y;
MODEL: Y ON X; Y ON M X;
MODEL INDIRECT: Y VIA M X;
Estimates S.E. Est./S.E.
Y ON X 0.391 0.084 4.675 M 0.382 0.041 9.393
Observed Residual Variable Variance R-Square
Y 1.000 0.198
TOTAL, TOTAL INDIRECT, SPECIFIC INDIRECT, AND DIRECT EFFECTS
You would need to send your input, data, output, and license number to firstname.lastname@example.org so I can see what is happening. I suspect you are not using Version 4.1 and that you are running into a bug from an earlier version.
I have a question concerning the standardized output in a mediator analysis, in which "X" is a binary (manifest) predictor (gender), "M" is a continuous latent (factor) mediator (based on categorical items), and "Y" is a continuous latent factor outcome variable (based on categorical items).
The problem is that I do not receive any error message, but the standardized outputs for SE, p-values, and indirect effects are not printed (in contrast to the unstandardized output). As the standardized outputs usually have to be reported in papers, I want to ask if you know a solution. Alternatively, if the standardized estimates are not estimated in MPlus by default, what would you suggest to report?
names=X M Y; USEVARIABLES ARE X M Y; Categorical ARE "indicators of M and Y"
A reviewer asked us to provide standardized estimates. Do you suggest to report unstandardized estimates to show significant influences of mediators and to explain that Mplus does not report standardized estimates of SE and p-values when using WLSMV and covariates? Do you have any other suggestions for alternative procedures or a reference we could cite?
I have: a) binary observed moderator B) latent predictor(4 continuous indicators) c) continuous observed outcome.
My model is a multigroup model testing these relations: a-->c B-->c B*a-->c (I created the interaction using XWITH).
When I ran this model, I got this message:
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS -0.884D-15. PROBLEM INVOLVING PARAMETER 23.
THE MODEL ESTIMATION TERMINATED NORMALLY
It seems this message appear when one of the variables used in the interaction term is binary since this did not happen when I used a continuous moderator. Is my understanding correct? If not, what could be contributing to this message?
I have one dichotomous/categorical variable (group; 0=group1, 1=group2) which I want to include as a predictor in my latent change score model. In my output, I see that there is a significant group effect (xd2 on group), but I'm unsure how to analyze whether the effect was for group 1 or 2. What extra syntax do I need to include for this information? I appreciate your help.