Niba S posted on Friday, January 05, 2018 - 11:06 am
My model includes 2 xs and an interaction, two mediators, and one outcome. Script is below. Few questions:
1) Is it ok to add with statements between Xs? Model fit improves with these but coefficients don't change. Why is the model fit changing with or without these statements?
2) If the answer to 1 is yes, do I also include with statements for my interaction? I tried and got an 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 0.695D-17. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 67, INT
3) If you have controls (age sex born), do these have to be correlated with each other?
Define: CENTER x1 x2 (grandmean); INT = x1*x2; ANALYSIS: TYPE IS GENERAL; ESTIMATOR IS ML; ITERATIONS = 1000; CONVERGENCE = 0.00005; Bootstrap = 1000;
MODEL: s3Ceff on x1 x2 int born sex age; s3Anger on x1 x2 int born sex age; s4pact on s3ceff s3anger born sex age;
!correlating mediators s3Anger with s3CEFF; !do i correlate the xs? x1 with x2; x1 with int; x2 with int; !do i correlate controls? born with age;
If you add WITH statements for x's, you bring them into the model. That can be useful if x's have missing data as we discuss in our book. The model fit as judged by CFI/TLI gets a bit exaggerated. Note that x's get correlated as the default but their correlation parameters are not estimated in the model (they are obtained from a Type=Basic run). If you bring some x's into the model you have to bring in all of them to get all correlated. Yes, control variables should be correlated (and are correlated as the default like other x's if not brought into the model).
Niba S posted on Friday, January 05, 2018 - 1:05 pm
The N only improves by 4 if I explicitly include with statements between my xs. That doesn't seem like the reason why the model fit (including x square and rmsea values) would improve so much. Any other reasons as to why the model fit would improve if I explicitly include with statements between xs rather than not include them?
Can you clarify why the error term might have occurred? Is it incorrect to include with statement for interactions? (see question 2 above)?