Message/Author |
|
Lynne W posted on Friday, March 09, 2018 - 4:33 am
|
|
|
Hello, I am fitting a mediation model using the Bayes estimator and include categorical covariates with missing data in the model as described in Muthen et al. 2016. However, with MEDIATOR=OBSERVED, no fit statistics are reported and in the MODEL FIT INFORMATION section only the number of free parameters is presented. Is there a way to obtain fit statistics? thank you kindly |
|
|
Not available. Are you sure the model is over-identified? |
|
Lynne W posted on Saturday, March 10, 2018 - 4:13 am
|
|
|
Yes, the model has a positive number of degrees of freedom. This is the model: VARIABLE: NAMES = y x1-x5; MISSING = ALL (-9999); CATEGORICAL = y x1; ANALYSIS: ESTIMATOR=BAYES; MEDIATOR=OBSERVED; MODEL: y ON x1-x5; x1 WITH x1; |
|
|
I assume you have missing on X1 and therefor add the WITH statement to bring X1 into the model. But you should bring all 5 x's into the model because otherwise X1 will be uncorrelated with the other x's. Your model is not over-identified when done this way. |
|
Lynne W posted on Sunday, March 11, 2018 - 10:47 am
|
|
|
Thank you, I assumed they would still be correlated. However, there is still no output of fit statistics. It works fine when neither y nor x1 is declared categorical. But as soon as y or x1 is declared categorical, the fit statistics disappear. I also tried freeing up parameters by fixing x2-x5 means to zero, but it doesn't help. |
|
|
Send the output with no fit statistics to Support along with your license number. |
|
Lynne W posted on Tuesday, March 13, 2018 - 10:49 am
|
|
|
I'll post Linda's reply here in case others wonder as well: For PREDICTOR = OBSERVED, PPP has not be developed statistically. It calls for new research. |
|
Back to top |