Structural Zeros with combinations of... PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Frank Lawrence posted on Wednesday, August 01, 2001 - 9:36 am
I am analyzing responses to a substance abuse survey. The responses are dichotomous. The models posit several indicators of substance abuse including alcohol consumption. Some combinations of indicator variables have structural zeros in their respective 2 x 2 tables. For example, "Have you ever used alcohol?" and "Did you consume more than 6 alcoholic beverages at one sitting in the last month?". Using raw data and the wlsmv estimator, these combinations cause Mplus difficulty. How can I create a factor that represents substance abuse given the structural zeros?
 bmuthen posted on Wednesday, August 01, 2001 - 3:52 pm
I would avoid structural zeros by subsetting the data for analysis. For instance analyze alcohol items only for those who have ever used alcohol.
 Anonymous posted on Friday, February 11, 2005 - 2:54 pm
We have defined a semi-continuous model as indicated in the following syntax, where u is defined as 0 if the outcome is 0 and 1 if the outcome has a value; and v is defined as the value of the outcome if it is positive and missing otherwise. The definition follows Olsen and Schafer 2001 paper for a two-part model with Zeros and a semi-continuous part. My question is why Mplus refuses the statement “x1 WITH x2;” in the following program. When we omit this statement the programs does not complain, but we are not sure at Mplus is doing is this case. Is it permissible that the statement “x1 WITH x2;” should not be used when dealing with semi-continuous models?

VARIABLE: NAMES ARE ….;
USEVARIABLES x1 y2 x2 y1 v u;
CATEGORICAL = u;

ANALYSIS:
TYPE= GENERAL MISSING;

MODEL:
u ON y1 x1 x2 y2; ! Zero part
v ON x1 x2 y2; ! Semi continuous Part

y2 ON x1 x2;
y1 ON x1;

x1 WITH x2;

MODEL INDIRECT: u IND y2 x1;
u IND y2 x2;
u IND y1 x1;

v IND y2 x1;
v IND y2 x2;
 Linda K. Muthen posted on Friday, February 11, 2005 - 3:16 pm
With categorical outcomes, the model is estimated conditional on the observed exogenous variables. Therefore they are not part of the model and you should not mention their means, variances, or covariances in the MODEL command. If you want to know their means, variances, or covariances, you can obtain them by using TYPE=BASIC;
 Anonymous posted on Tuesday, February 15, 2005 - 10:54 am
So does this mean I don’t specify the correlation between x1 WITH x2 right? I understand what this response, but what about the semi-continuous part? Do we apply the same rule to the s-c part because technically it’s part of the larger model which includes categorical outcomes?
 Linda K. Muthen posted on Tuesday, February 15, 2005 - 11:09 am
You should remove the statement x1 WITH x2; You treat the entire model the same. If you have one categorical outcome, the entire model is estimated conditional on the x's.
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: