| Message/Author |
|
|
|
|
We are attempting to do a chi-square difference test for nested sub models using both weights and missing data. Looking at equations 119-120 in the technical manual it would seem that we need to fit the model using ML and also MLR to get the 4 bits of info we need for the calculation. But the ML model does
not estimate with weights. Any recommendations? It would also be helpful to know how to do the chi-square difference with weights not using missing data. |
|
|
|
|
| If Mplus prints a scaling factor, then you can multiply the MLR chi-square by the scaling factor to get the ML chi-square. |
|
|
|
|
Hi Linda,
I have a similar situation as the post above (missing data, complex survey data). However, my models mostly involve binary outcome variables, so there are no provided chi-square values on my output. My outputs include loglikelihoods (with scaling correction factors).
I have questions regarding the rest of the information I need for the difference tests: 1) number of parameters: Can this be found in the "number of free parameters" row in the "Information Criteria" section? 2) To test the significance of the computed chi-square difference, I need the df reflecting the difference in degrees of freedom between the H0 and H1 models. However, df are not printed for my models. Is there an option for the df to be printed on my outputs?
Thanks,
Jim |
|
|
|
|
1) Yes.
2) You use the difference in number of parameters. See the bottom section
Difference Testing Using the Loglikelihood
of the Chi-Square Diff testing section on our web site. |
|
|
|
|
| The difference in the number of free parameters is the same as the difference in the number of degrees of freedom. |
|
|
| Back to top |
