Lucy Hobby posted on Tuesday, September 20, 2011 - 7:48 pm
I have some questions regarding a model I have been conducting for my research, and the output it is producing. In this model, I want to control for cultural background (Culture), and then using a contrast variable (contrast) containing two groups of interest, predict academic self-concept (ESCT) and academic achievement (EACH) at Time 1 and Time 2. The most relevant syntax is produced below: ESCT2 ON Culture ESCT1 EACHT1 contrast; ESCT1 ON Culture contrast; EACHT2 ON Culture EACHT1 ESCT1 contrast; EACHT1 ON Culture contrast; When I try to run this model with the culture and the contrast variable correlated (culture with contrast;), the following error message is produced: 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 0.133D-18. PROBLEM INVOLVING PARAMETER 101. continued
Lucy Hobby posted on Tuesday, September 20, 2011 - 7:50 pm
I’ve tried to get around this error by specifying alterations in the syntax (more specifically, how Culture and contrast are associated with each other). Following are a list of the models run with the exact same syntax except for what is listed below), all of which produce the same statistical output (chi-square, degrees of freedom, and predictive paths are all the same), but one model receives the above error. Focusing only on what is modified in the syntax: Model 1 – Culture with contrast; (error) Model 2 – !Culture with contrast; (no error) Model 3 – contrast on Culture; (no error) Model 4 – Culture on contrast; (no error) When I run the same model without the correlation between culture and contrast specified, I receive no error message and the same chi-square, df, and predictive path values. I am assuming this is because the correlation is a default in mplus. Now, if I run the same model with culture predicting contrast (Contrast on culture;) or contrast predicting culture (Culture on contrast;), I also receive no error message and the same chi-square, df, and predictive path values as the above two models. I’m wondering why I receive the error message when running the model where the variables are correlated, and why I’m getting the exact same statistical results across all four models?