Controlling for dichotomous independe... PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Luke Rusowicz-Orazem posted on Friday, September 20, 2019 - 12:22 pm
I am getting the following error: 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 0.688D-10. PROBLEM INVOLVING PARAMETER 18.

THIS IS MOST LIKELY DUE TO VARIABLE F BEING DICHOTOMOUS BUT
DECLARED AS CONTINUOUS.

I am trying to run a simple model regressing G on D while controlling for LZ, C, E and F. Please see code below:

Variable: Names are A-W;
UseV are A-G;
Missing = ALL (-999);
Analysis:

coverage = .000;
type=general;
estimator=ml;
convergence=.00001;

Model:
LZ by A B;
D with LZ;
D with C;
D with E;
D with F;
G on D;

OUTPUT: SAMPSTAT STANDARDIZED RESIDUAL TECH1 TECH4 cinterval;

F is a dichotimous variable. I am not sure how to correctly code this as it is not a dependent variable in the model but an independent variable I am trying to control for.
 Bengt O. Muthen posted on Saturday, September 21, 2019 - 11:41 am
"Controlling for" means that you include those variables in the regression, so

G ON D LZ C E F;

The message can be ignored because F is binary.
 Luke Rusowicz-Orazem posted on Saturday, September 21, 2019 - 12:01 pm
Thank you so much! Based on your response, it sounds like I do not need any with statements? Could you clarify how my code is currently being interpreted by Mplus? Thank you!
 Bengt O. Muthen posted on Saturday, September 21, 2019 - 1:45 pm
In addition to saying

G ON D LZ C E F;

you need to say that the factor LZ is correlated with the covariates (control variables) C, D, E, F and that the covariates are correlated (the latter is not needed when all predictors are observed). You do this by using WITH statement between them, e.g.


c-f with c-f;

c-f with lz;
 Luke Rusowicz-Orazem posted on Sunday, September 22, 2019 - 7:11 pm
all the other covariates besides LZ are observed. Then my code should be as follows?

Variable: Names are A-W;
UseV are A-G;
Missing = ALL (-999);
Analysis:

coverage = .000;
type=general;
estimator=ml;
convergence=.00001;

Model:
c-f with lz;
G ON D LZ C E F;

OUTPUT: SAMPSTAT STANDARDIZED RESIDUAL TECH1 TECH4 cinterval;

For my understanding, could you help me understand what my original code was commanding plus to do?
 Bengt O. Muthen posted on Monday, September 23, 2019 - 5:05 pm
Once you mention c-f in the WITH statement, these observed x variables turn into y variables and therefore you also have to specifically say that they are correlated. So add

c-f WITH c-f;
 Luke Rusowicz-Orazem posted on Wednesday, September 25, 2019 - 7:05 am
It is necessary to have any with statements? I am not sure that these variables are correlated. The only reason I included them in the code to begin with was because I thought that was how I could tell Mplus to control for these variables in my regression.

Also, I am noticing that when I run this code, MPLUS listwise deletes data from any participant that has any x variable missing. Is there anyway that I can stop MPLUS from doing this such that I can make use of all the data that I have?
 Bengt O. Muthen posted on Wednesday, September 25, 2019 - 3:10 pm
Q1: I think that is a good default approach since there may be correlations and your modeling isn't about that part.

Q2: When you turn x variables into y variables (using my terms earlier), then you won't have listwise deletion.

If this doesn't help, send your output to Support along with your license number.
 Luke Rusowicz-Orazem posted on Friday, September 27, 2019 - 9:32 am
I am really confused by what your response to Q2 means. If I turn my x variables into y variables then my entire research question is changed. My x variables are by definition independent. If I turn them into dependent variables then what would I be using to predict them?
 Bengt O. Muthen posted on Friday, September 27, 2019 - 11:23 am
No need to be alarmed by an x turned into a y. It doesn't mean that they have to be predicted by anything. Here is what we say in a FAQ on this ("bringing covariates into the model" is the same as turning x's into y's in our terminology):

Covariates - bringing them into the model

In regression, the model is estimated conditioned on the covariates. Their means, variances, and covariances are not model parameters and no distributional assumptions are made about them. You can find their means, variances, and covariances in the descriptive statistics for the data set. Any observation that has a missing value on one or more observed exogenous covariates is eliminated from the analysis. To avoid this, the covariates can be brought into the model by mentioning their means, variances, and/or covariances in the MODEL command. This can be done for maximum likelihood estimation and Bayes. It should not be done for weighted least squares estimation. When this is done, the means, variances, and covariances of the covariates become model parameters and distributional assumptions are made about them. This is not innocuous, however, especially when there are binary covariates. Pros and cons of this approach are discussed in Chapter 10 of our book Regression and Mediation Analysis using Mplus as well as at the end of our Short Course Topic 11 video and handout on our website.
 Luke Rusowicz-Orazem posted on Monday, September 30, 2019 - 7:11 am
Oh understood. I did this and when I run the model it tells me I must use montecarlo integration. But when I do this I see that MPlus is computing the code but then no output is produced and nothing happens. Do you know what is happening? My code has slightly changed so I am repeating it below:

Variable: Names are A-Z;
UseV are A-G;
Count is G (i);
Missing = ALL (-999);
Analysis:

estimator=ml;
integration=montecarlo;

Model:
LZ by A B;
D C E F with LZ;
D C E F with D C E F;
G on D C E F LZ;
G#1 on D C E F LZ;
D C E F LZ

OUTPUT: SAMPSTAT STANDARDIZED RESIDUAL TECH1 TECH4 cinterval;
 Bengt O. Muthen posted on Monday, September 30, 2019 - 12:34 pm
To be able to diagnose this, we need to run it using your input and data - send to Support along with your license number.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: